1College of Computer and Information Engineering Hohai University Nanjing Jiangsu 211100 China2School of Information Engineering Nanchang Institute of Technology Nanchang 330099 China
Correspondence should be addressed to Xijun Yan zllchl163com
Received 13 May 2015 Revised 24 August 2015 Accepted 15 September 2015
Academic Editor Jiri Jan
Copyright copy 2015 Lili Zhang et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
A new river flow measurement method based on graphic process has been proposed recently which gets the velocity in opticalimaging modality through measuring the continuous displacement of floating debris then reconstructs a two-dimensional riversurface velocity field by using the velocity of floating debris and computes the section flow at last However the surface opticalimages have not only lights of target information but also surface optical noise It is difficult for reliable and stable continuousdisplacement detection of complex small observation target which occupies only a small number of pixels comparing to a large fieldimaging area and has complex optical reflection properties To solve this problem this paper presents a background suppressionmethod based on soft morphology and Retinex theory Soft morphology is firstly used for the opening operation of the imageand then Retinex theory is used for optimal estimation of image incident component to suppress background of image Finally thesimulations show that ourmethod is superior to graymorphology and softmorphology on the performance of targets enhancementnoise filtering and background suppression and it has better background and targets discrimination quality subjective evaluationand higher signal-to-clutter ratio
1 Introduction
Image background suppression is the method of removingthe relativelymotionless background information from videoframes and retaining themoving target information of a scenefrom video frames Background suppression is widely used inimage processing applications and is usually the first stage ofthe detection applications
In the late 1990s Fujita used particle image velocimetry(LSPIV) to observe floods in theYodoRiver which has alwaysbeen used to put artificial tracer particles to the sink inlaboratory conditions The river flow measure method needsto capture images on the river bank of a tilt angle of the sur-face and measures large areas of natural rivers gets thevelocity in optical imaging modality through measuringthe continuous displacement of floating debris such as treebranches leaves and other water tracers then reconstructsa two-dimensional river surface velocity field by using thevelocity of floating debris and computes the section flow byvelocity-area method at last This measurement method was
called large-scale particle image velocimetry [1ndash8] and hasbeen applied widely in the field of hydrology
The existing LSPIV method only follows the particleimage enhancing technology of traditional PIV [9 10] so thatthe signal-to-noise ratio (SNR) improvement in image islimited However the river surface imaging environment isvery complexThere are many lighter components in the sur-face shooting image The image is a collection of directlight from sun atmospheric scattering light surface reflectedlight (flare) surface-emitting light (reflection) underwaterreflected light targets reflected light and so forth so thebackground of the image is undulate On the other hand theflowing tracers that meet the following requirement on thesurface have a smaller size The targets are weak and thereare small distinctions between background and targets so thetargets are difficult to be foundTherefore the research on themethods of target enhancement and background suppressionin the river surface image is important on the theoreticalsignificance and application value [11ndash15]
Hindawi Publishing CorporationJournal of Electrical and Computer EngineeringVolume 2015 Article ID 389487 11 pageshttpdxdoiorg1011552015389487
2 Journal of Electrical and Computer Engineering
This paper will propose one new background suppressionmethod of the river surface image based on soft morphologyand optimal estimation of image incidence component inRetinex theory The related works and main contributions ofthe current work are presented in Section 2 The backgroundsuppressionmethod of surface image based on softmorphol-ogy has been discussed in Section 3 Section 4 is dividedinto two subsections to explain processes of the method thatcombined soft morphology and Retinex theory and thenpropose ourmethod Section 5 gives the experiment results toverify the efficiency of ourmethod by comparing the differentmethods Finally the paper is concluded in Section 6
2 Related Works
Morphology was applied in background suppression severalyears ago and the background suppression method based onmathematical morphology has been proposed a long timeago The mathematical morphology method is developedfrom geometry and proposed by Serra and Soille [16] whichis a nonlinear filter essentiallyThemethod has unique advan-tages to remove the high-frequency components in the imageHowever the traditional background suppression method ofgraymorphology is limited by the size selection of structuringelement and it is difficult to achieve the optimal suppressioneffect against the different target size
Soft morphology [17] is developed from the gray mor-phology and proposed by Kuosmanen and Astola at firstThe operating methods of gray morphology such as erosiondilation opening and closing are based on the process witha single structuring element to achieve smooth backgroundHowever it is not good to use one single structuring elementin the complex background Soft morphology is to reformthe structuring element and let the single structuring elementbecome a structuring element system It uses the structuringelement system that considers the difference between targetsand target neighborhood to improve the effect of backgroundsuppression And its performance is significantly improvedcompared to the standard gray morphology filter
This paper will propose a new image background sup-pression method based on soft morphology The complexriver surface image was processed with soft morphologyoperation and then the optimal estimation of image incidentcomponent to the result image was made The results arealso compared with two kinds of existing morphologicalbackground suppression methods
3 Background SuppressionBased on Soft Morphology
Soft morphology is developed from basic gray morphologyGray morphology filtering method is usually used in theinfrared image process Now we analyze the effect of the graymorphology filtering method in the river image backgroundsuppression
31 Gray Morphology and Top-Hat Method Gray morphol-ogy is the basis of traditional mathematical morphology
The most typical background suppression method of gray-scale image is the top-hat transform [18ndash20] The white top-hat transform and black top-hat transform (WTH and BTH)of grayscale morphology are as follows
WTH = 119891 minus 119891 ∘ 119887
BTH = 119891 sdot 119887 minus 119891(1)
The first step of the process ofWTH and BTH transformsis to make opening operation and closing operation to theoriginal image and then make a difference between theoperated resultsThe computational process of119891∘119887 operationis always conducted below the original image 119891 which isequivalent to the result that uses the original image 119891 tosubtract the opening operation119891∘119887 resultThe computationalprocess of 119891 sdot 119887 operation is always conducted above theoriginal image 119891 which is equivalent to the result that uses119891 sdot 119887 operation result to subtract the original image 119891
Figure 1 shows the experimental results of WTH andBTH transforms They are operated with 3 times 3 pixelsrsquo squarestructuring element in surface grayscale images (256 times 256pixels) From Figures 1(a2) and 1(a3) we can find that WTHtransform is able to detect the peak of images and BTHtransform is able to detect the valley of images effectively Inthe background image of displacement detection the targetslook like small bright spots and will be lost in the image afterBTH transform Compared with the opening and closingoperations of grayscale morphology the result of imagebackground suppression after WTH transform is obvious Itnot only removes clutter effectively but also retains the targetintegrity The operation of 119891 ∘ 119887 can be regarded as a processof background estimation The physical meaning of 119891 minus 119891 ∘ 119887is a process that uses the original image to subtract the resultafter opening operation and retains the targets So the WTHtransform is a process of background suppression essentially
From the comparison of gray distribution images afterWTH and BTH transform we find that the gray intensityof image background tends to zero except for the targetsand noise Figure 1 shows the experimental results of WTHand BTH transform The original images are operated with3 times 3 pixelsrsquo square structuring element Figure 2 showsthe experimental results of WTH and BTH transform Theoriginal images are operated with 7 times 7 pixelsrsquo squarestructuring element The targets have greater visibility afterWTH transformwith 7times 7 pixelsrsquo square structuring elementThe targets that are smaller than the structuring element sizehave been filtered in Figure 1(a2) The comparison of graydistribution shows that the targets have higher gray intensityand the contrast of targets and background in the image ishigher Therefore the size and structure selection of struc-turing element are very important for background estimationprocess If the target size is too small it causes elimination ifthe target size is too large it causes excessive smooth Theseresults will disturb the displacement detection of targets andmotion vector estimation in the subsequent process
32 Soft Morphology In soft morphology the maximum andminimumoperation of standard graymorphology are instead
Journal of Electrical and Computer Engineering 3
(a1) Original image (b1) Gray distribution of original image
(a2) WTH transform result (b2) Gray distribution after WTH transform
(a3) BTH transform result (b3) Gray distribution after BTH transform
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200300
1000 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
Figure 1 WTH and BTH transform results and gray distribution images with 3 times 3 pixels
4 Journal of Electrical and Computer Engineering
(a1) Original image (b1) Gray distribution of original image
(a2) WTH transform result (b2) Gray distribution after WTH transform
(a3) BTH transform result (b3) Gray distribution after BTH transform
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200300
1000 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
Figure 2 WTH and BTH transform results and gray distribution images with 7 times 7 pixels
Journal of Electrical and Computer Engineering 5
b1
b2
Figure 3 Schematic diagram of soft morphology structuring ele-ment 119887
of sorting weighted statistical operation The determinationof weighted coefficients is associated with the structuringelement Unlike standard mathematical morphology thestructuring element 119887 in soft morphology consists of twoparts one is the center of the structure 119887
1(also called hard-
core) and theweight (number of iterations) of correspondingpixel is greater than 1 or equal to 1 the other is soft edge 119887
2 and
the weight of corresponding pixel is equal to 1The schematicdiagram is shown in Figure 3
Similar to the standard mathematical morphology softmorphology erosion and dilation operations are defined asfollows
119891Θ [1198871 1198872 119896] (119909) = 119896th min
119909+119910isin1198701
119909+119911isin1198702
119896
[119891 (119910) minus 1198872(119909 + 119910)] cup [119891 (119911) minus 119887
1(119909 + 119911)]
(2)
119891 oplus [1198871 1198872 119896] (119909) = 119896th max
Formula (2) is the erosion operation of soft morphologyand formula (3) is the dilation operation of soft morphology1198871is the hard-core of structuring element 119870
1sube 119885
2 is thedefinition domain of 119887
1 1198872is the soft edge of structuring
element 1198702sube 119885
2 is the definition domain of 1198872 the
structuring element 119887 = 1198871cup 1198872 and 119887
1cap 1198872= 119896 119891(119909) =
119891(119909) 119891(119909) 119891(119909) is a repeating set 119896 is the number ofiterations When 119896 = 1 119887
1= 1198872 and 119887 = the soft morphol-
ogy is degenerated to the standard graymorphologyThe cen-ter pixel of 119887
1is also the center pixel of 119887 the soft edge pixels of
1198872are also the soft edge pixels of 119887 the weights of 119887
1and 1198872are
determined by 119896 the value of 119896 reflects differences betweentarget area and the neighborhood We can change the softmorphology structuring element by adjusting the value of 119896and make the differences between targets and background inimage to achieve a better effect of background suppression
Based on erosion and dilation operation the definition ofsoft morphology opening and closing operation is as follows
Formula (4) is the soft morphology opening operationand formula (5) is the softmorphology closing operationTheprocess of opening operation can be interpreted as making
dilation operation to 119891 with soft edge 1198872first and then mak-
ing erosion operation to the result with 1198871and 1198872 The proc-
ess order of closing operation is contrasted with openingoperation Hard-core 119887
1and soft edge 119887
2are shown in
1198871= (
1 1 1
1 1nabla1
1 1 1
)
1198872=
(
(
(
1 1 1 1 1
1 0 0 0 1
1 0 0nabla0 1
1 0 0 0 1
1 1 1 1 1
)
)
)
(6)
Figure 4 shows the results after the soft morphologyoperation in which (b) is the opening operation result and(c) is the result of SWTH transform based on soft mor-phology opening operation By using soft morphology open-ing operation it can smooth the background image filtertargets and noise smaller than the structuring element andget an estimated image of the original image backgroundFrom the comparison between Figures 4(c) and 2(a2) thebackground suppression effect of soft morphology top-hat(SWTH) transform is better than gray morphology WTHrsquosAfter SWTH transform filtering in Figure 4(c) the gray gra-dation of target is decreased obviouslyThe small weak targetsalmost are suppressed In order to improve the backgroundsuppression effect of soft morphology we propose a newbackground suppression method based on soft morphologyfiltering and Retinex theory
4 Background Suppression Based on SoftMorphology and Retinex Theory
Based on the analysis in the previous section it can be foundthat the soft morphology operation has a good performancein suppressing background of the surface image In orderto enhance the image contrast and improve SCR in theimage wemake the background suppression of displacementmeasurement of targets and motion vector estimation in thesurface image with the Retinex theory
41 RetinexTheory Retinex is derived from the combinationof retina and cortex It is a color theory which can describethe color constancy of the human visual system In the studyof the principles of the human visual perception systemand psychophysical brightness Land [21] found that whenthe visual system processes the visual information someuncertain external factors such as light intensity and unevenlight will be excluded and some characteristics informationwhich can reflect the essence of objects will be retainedThen these characteristicsrsquo information will be delivered tothe cortex by neural network and form visual image Retinextheory was proposed in 1977 it can powerfully explainthe homeostatic mechanism the human visual system can
6 Journal of Electrical and Computer Engineering
(a) Original image (b) Result of soft morphology opening operation
(c) Result of soft morphology background suppres-sion
Figure 4 Result image of soft morphology operation
Incident light L
Reflecting object R
Observer
S(x y)
S(x y) = L(x y) middot R(x y)
Figure 5 Schematics of Retinex
achieve the same color of one object under different light byself-regulation
Figure 5 shows the schematic diagram of Retinex theoryAccording to Retinex theory the image 119878(119909 119910) is constitutedby two factors One factor is the illumination intensity 119871 ofthe object which corresponds to the low-frequency part ofthe image and presents the luminance image 119871(119909 119910) anotherfactor is the reflective brightness 119877 of the object whichcorresponds to the high-frequency part of the image and
presents the reflection image 119877(119909 119910) So the imaging processof the image can be expressed as
The illumination intensity 119871 determines the dynamicrange of pixel in an image and the reflection luminance 119877reflects the nature of the objectThe essence of Retinex theoryis casting aside the nature of the illumination intensity 119871 andobtains the inherent essential characteristics 119877 of the objectfrom the image 119878 Taking the logarithm of formula (7) intoaccount the complex operations can be translated into simpleaddition and subtraction and the formulas are as follows
where 119904 = ln[119878(119909 119910)] 119897 = ln[119871(119909 119910)] and 119903 = ln[119877(119909 119910)]Usually we cannot achieve the reflection luminance 119877 of
the object directly However we can estimate the illumination
Journal of Electrical and Computer Engineering 7
intensity 119871 firstly Then we use the image 119878 to subtract theillumination intensity 119871 In this way the reflection luminance119877 which can reflect the essential characteristics of the objectcan be achieved The formula can be expressed as
119903 = 119904 minus 119897 (9)
This is also equivalent to the concept of background sup-pression principle the high-frequency part (including targetand high-frequency noise) can be separated by comparing theoriginal image with low-frequency part of the image There-fore how to estimate the light intensity is the key of the issue
Ferwerda et al [22] showed that incident component inan image can be estimated andKimmel et al [23] showed thatthe incident component estimation problem (illuminationintensity 119871) can be formulated as a Quadratic Programmingoptimization problem and furthermore they showed theoptimization problemhaving a unique solutionWewill applythe above conclusion in our algorithm
The commonly used methods to estimate the incidentcomponent include look-up table and convolution methodsTo deal with the background suppression issue of the riverwater visual image it needs to face multiple different imagesApparently building a single gray look-up table cannot meetthe requirementsTherefore we use themethod of the convo-lution operation to estimate the optimal incident componentIn this method selecting the appropriate kernel functionto do the convolution operation is the key of the problemGaussian kernel function can highlight the center position ofweight value Meanwhile the influence of the surroundingpoints of the center position can be taken into accountAnd the estimated image has a good correlation with theoriginal image Based on the above reasons 3 times 3 Gaussiankernel function is chosen to do the optimal estimation ofthe incident component The values of 3 times 3 Gaussian kernelfunction are as follows
119870 =
[
[
[
[
[
[
[
[
[
[
1
16
1
8
1
16
1
8
1
4
1
8
1
16
1
8
1
16
]
]
]
]
]
]
]
]
]
]
(10)
The convolution operation to the image with Gaussiankernel function is equivalent to doing a low pass filterA new image will be achieved after each convolution andthe optimal estimated value of incident component can beachieved According to the literature [24 25] the average graylevel of the image tends to stability after three convolutionoperations So it is thought that the result after the thirdconvolution is the most suitable result to be the optimalestimation of the incident component of the image
42 Background Suppression Method In the process of targetmotion vector estimation with the river visible backgroundimage the obtained river image has a complex backgroundwhich includes many lights such as direct illumination fromthe sun atmospheric scattering light surface reflected light
Soft morphologyopening operation
Input originalimage
Outputresult
Result of openoperation
Originalimage
Optimal estimation ofincident component
++
minus
Figure 6 Background suppression flowchart based on soft mor-phology and Retinex theory
(flare) surface-emitting light (reflection) and target reflectedlight Therefore the image will present the uneven lightundulating background and unidentified target For thecomplex situation a method based on soft morphology andRetinex theory is proposed to realize image background sup-pression
According to the previous analysis we can find that theestimation of incident component can achieve an optimalestimation for low-frequency part of the image It has impor-tant practical significance of the surface visible backgroundimage with complex lighting conditions Through the softmorphology operations and optimal incident componentestimation we can achieve the optimal estimation of thebackground image Then by using the original image tosubtract the estimated image a background suppressionimagewith a higher signal-to-noise ratiowill be achievedTheflowchart of the proposed method is shown in Figure 6
Step 1 (opening operation) It has been shown that the opti-mized size of operator structure is generally equal to the halfof the maximal size of a small two-dimensional target [26]Therefore we chose the following central structuring element1198871and flexible edge structure element 119887
2[27ndash29]
1198871=
(
(
(
1 1 1 1 1
1 1 1 1 1
1 1 1nabla1 1
1 1 1 1 1
1 1 1 1 1
)
)
)
1198872=
(
(
(
(
(
(
(
(
(
1 1 1 1 1 1 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 0 0 0nabla0 0 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 1 1 1 1 1 1
)
)
)
)
)
)
)
)
)
(11)
8 Journal of Electrical and Computer Engineering
Table 1 Comparison of three background suppression methods
The structural element 119887 is composed of 1198871and 119887
2 It
takes soft morphological opening operation on the originalimage After the opening operation the noise and the targetwhich is smaller than the structuring element in the image areeliminated and the background image becomes smoother
Step 2 (estimating the optimal incident component) Accord-ing to the aforementioned method of optimal estimation ofimage incident component we make a cubic convolutionwith the result after Step 1 and Gaussian kernel function informula (10)ThroughGaussian convolution we filter out thehigh-frequency part of the image and achieve the optimalbackground estimation of the image
Step 3 (background suppression) Using the original imageto subtract the optimal estimated image obtained after Step 2we can achieve the background suppression result that filterslow-frequency part of the original image and enhancescontrast
119891119879+ 119891119873= 119891 minus (119891 ∘ [119887
1 1198872 119896])
1015840
(12)
where 119891119879is target component 119891
119873is noise component 119891 is
original image and (119891 ∘ [1198871 1198872 119896])
1015840 is the background imageafter the processing of soft morphology opening operationand the incident component optimal estimation Formula(12) can be as the soft morphology white top-hat transformbased on Retinex theory
5 Experiments and Simulations
In order to verify the effectiveness of the proposed methodwe make an experiment with the method which combineswith soft morphological opening operation and optimal esti-mation of incident component with background suppressionof the image based on Retinex theoryWemake a comparisonbetween WTH and SWTH The experiment is carried outwith the PCwhich is equipped with PentiumT4300 memory286GB and uses Matlab software platform To evaluate theperformance of this method we use pixel-level evaluationIn the study of image background suppression an indicatorknown as SCR (signal-to-clutter ratio) is used as follows
SCR =(119891119879minus 120583119887)
120590119887
(13)
where 119891119879is the target strength and it can be replaced by the
maximumgrayscale value of the image under the visible light
120583119887is the average grayscale value of the image and it reflects
the background DC component 120590119887is the grayscale standard
deviation of the image and it reflects the degree of clutterTheSCR Gain can be defined as
119866SCR = 20 log(SCRoutSCRin) (14)
where SCRin and SCRout denote the SCR of the original imageand the background suppression image SCR can be usedto describe the improvement of the original image with thebackground suppression method In experiment we chosethree screenshots from the actual shooting river scene videoas subjects (image size is 256 times 256 pixels) as shown inFigures 7(a1) 7(b1) and 7(c1) Figures 7(a2)ndash7(a4) 7(b2)ndash7(b4) and 7(c2)ndash7(c4) are the results of the experiments
Figures 7 and 8 and Table 1 respectively present the resultcharts three-dimensional grayscale distribution diagramsand SCRSCR Gain data statistics of the three surface imagebackground suppression methods Grayscale morphologicalWTH transform is computed with 5 times 5 square structuringelement while the soft morphology white top-hat (SWTH)transform and the method in this paper are using thestructuring element 119887 which is comprised of 119887
1and 119887
2in
formula (11) From Table 1 it can be found that the originalimage has a low SCR the targets have poor visibility inthe grayscale image and the grayscale values of backgroundare fluctuant WTH and SWTH transform can suppressbackground eliminate some of the background clutter andimprove the SCR However there is still some backgroundclutter residue in the image Through the proposed methodin this paper the background clutter in the image is filtered toa large extent The background is flattened and the grayscalevalue of background is low Moreover most of the targets areretained and have good visibility and the target enhancementis notable these can be reflected by SCR Gain in Table 1
Compared with the three methods mentioned in thispaper the proposed background suppression method has anotable improvement of SCR in the image and the SCR Gainis also the largest Grayscale morphology WTH transformhas a weak performance in background suppression and thesoftmorphology SWTH has amedium performance in back-ground suppression Therefore the experiment fully demon-strated that the proposed method has a better ability in back-ground suppression than the other two methods Howeverthis method also has some shortcomings After the imageprocessing of image 3 the effect of background suppressionhas no notable improvement compared with WTH and
Journal of Electrical and Computer Engineering 9
(a1) Img 1 original image (a2) Img 1 result of WTH (a3) Img 1 result of SWTH (a4) Img 1 result of our method
(b1) Img 2 original image (b2) Img 2 result of WTH (b3) Img 2 result of SWTH (b4) Img 2 result of our method
(c1) Img 3 original image (c2) Img 3 result of WTH (c3) Img 3 result of SWTH (c4) Img 3 result of our method
Figure 7 Comparison of three methods
SWTH and the three methods also have almost the sameperformance in SCR Gain The reason is that the pixelsize of the target in image 3 is large In this method thelarge size target with soft morphological opening operationcannot achieve an ideal result in background clutter residualTherefore the background image has a big fluctuation afterbackground suppression Although the improvement of SCRis not notable the target grayscale and image contrast have anotable improvement Moreover our method has the largestgrayscale value and optimal target visibility among the threemethods and the subjective evaluation of the quality indiscrimination between the background and target is also thebest
6 Conclusion
To overcome the shortcomings of surface noise and cluttersurface tracer optical reflection complexity difficulty in targetdisplacement detection and motion vector estimation we
present a background suppression method based on softmorphological filtering and Retinex theory in this paperIn order to improve the performance of surface imagebackground suppression method we use the Retinex theoryand make an optimal estimation of incident component ofthe background image through soft morphological openingoperation The experiments give the results of backgroundsuppression of surface image and make a comparison withgrayscale morphological WTH transform and soft morphol-ogy SWTH transform experiments The simulations showthat the proposed method has a notable improvement inbackground suppression of surface image Meanwhile ourmethod makes a good preparatory work for the next targetdisplacement detection and motion vector estimation
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
10 Journal of Electrical and Computer Engineering
(a1) Img 1 original image (b1) Img 2 original image (c1) Img 3 original image
Figure 8 Comparison of three methods on 3D grayscale distribution diagrams
Acknowledgments
This paper is partially supported by the National NaturalScience Foundation of China (no 61263029 no 61374019)a project funded by the Priority Academic Program Devel-opment (PAPD) of Jiangsu Higher Education Institutionsand Natural Science Foundation of Jiangsu Province (noBK20130851)
References
[1] X Hao Research of moving point targets detection method inimage sequences [PhD thesis] Shandong University JinanChina 2005
[2] L Xu Z Zhang X Yan H Wang and X Wang ldquoAdvances ofnon-contact instruments and techniques for open-channel flow
Journal of Electrical and Computer Engineering 11
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
This paper will propose one new background suppressionmethod of the river surface image based on soft morphologyand optimal estimation of image incidence component inRetinex theory The related works and main contributions ofthe current work are presented in Section 2 The backgroundsuppressionmethod of surface image based on softmorphol-ogy has been discussed in Section 3 Section 4 is dividedinto two subsections to explain processes of the method thatcombined soft morphology and Retinex theory and thenpropose ourmethod Section 5 gives the experiment results toverify the efficiency of ourmethod by comparing the differentmethods Finally the paper is concluded in Section 6
2 Related Works
Morphology was applied in background suppression severalyears ago and the background suppression method based onmathematical morphology has been proposed a long timeago The mathematical morphology method is developedfrom geometry and proposed by Serra and Soille [16] whichis a nonlinear filter essentiallyThemethod has unique advan-tages to remove the high-frequency components in the imageHowever the traditional background suppression method ofgraymorphology is limited by the size selection of structuringelement and it is difficult to achieve the optimal suppressioneffect against the different target size
Soft morphology [17] is developed from the gray mor-phology and proposed by Kuosmanen and Astola at firstThe operating methods of gray morphology such as erosiondilation opening and closing are based on the process witha single structuring element to achieve smooth backgroundHowever it is not good to use one single structuring elementin the complex background Soft morphology is to reformthe structuring element and let the single structuring elementbecome a structuring element system It uses the structuringelement system that considers the difference between targetsand target neighborhood to improve the effect of backgroundsuppression And its performance is significantly improvedcompared to the standard gray morphology filter
This paper will propose a new image background sup-pression method based on soft morphology The complexriver surface image was processed with soft morphologyoperation and then the optimal estimation of image incidentcomponent to the result image was made The results arealso compared with two kinds of existing morphologicalbackground suppression methods
3 Background SuppressionBased on Soft Morphology
Soft morphology is developed from basic gray morphologyGray morphology filtering method is usually used in theinfrared image process Now we analyze the effect of the graymorphology filtering method in the river image backgroundsuppression
31 Gray Morphology and Top-Hat Method Gray morphol-ogy is the basis of traditional mathematical morphology
The most typical background suppression method of gray-scale image is the top-hat transform [18ndash20] The white top-hat transform and black top-hat transform (WTH and BTH)of grayscale morphology are as follows
WTH = 119891 minus 119891 ∘ 119887
BTH = 119891 sdot 119887 minus 119891(1)
The first step of the process ofWTH and BTH transformsis to make opening operation and closing operation to theoriginal image and then make a difference between theoperated resultsThe computational process of119891∘119887 operationis always conducted below the original image 119891 which isequivalent to the result that uses the original image 119891 tosubtract the opening operation119891∘119887 resultThe computationalprocess of 119891 sdot 119887 operation is always conducted above theoriginal image 119891 which is equivalent to the result that uses119891 sdot 119887 operation result to subtract the original image 119891
Figure 1 shows the experimental results of WTH andBTH transforms They are operated with 3 times 3 pixelsrsquo squarestructuring element in surface grayscale images (256 times 256pixels) From Figures 1(a2) and 1(a3) we can find that WTHtransform is able to detect the peak of images and BTHtransform is able to detect the valley of images effectively Inthe background image of displacement detection the targetslook like small bright spots and will be lost in the image afterBTH transform Compared with the opening and closingoperations of grayscale morphology the result of imagebackground suppression after WTH transform is obvious Itnot only removes clutter effectively but also retains the targetintegrity The operation of 119891 ∘ 119887 can be regarded as a processof background estimation The physical meaning of 119891 minus 119891 ∘ 119887is a process that uses the original image to subtract the resultafter opening operation and retains the targets So the WTHtransform is a process of background suppression essentially
From the comparison of gray distribution images afterWTH and BTH transform we find that the gray intensityof image background tends to zero except for the targetsand noise Figure 1 shows the experimental results of WTHand BTH transform The original images are operated with3 times 3 pixelsrsquo square structuring element Figure 2 showsthe experimental results of WTH and BTH transform Theoriginal images are operated with 7 times 7 pixelsrsquo squarestructuring element The targets have greater visibility afterWTH transformwith 7times 7 pixelsrsquo square structuring elementThe targets that are smaller than the structuring element sizehave been filtered in Figure 1(a2) The comparison of graydistribution shows that the targets have higher gray intensityand the contrast of targets and background in the image ishigher Therefore the size and structure selection of struc-turing element are very important for background estimationprocess If the target size is too small it causes elimination ifthe target size is too large it causes excessive smooth Theseresults will disturb the displacement detection of targets andmotion vector estimation in the subsequent process
32 Soft Morphology In soft morphology the maximum andminimumoperation of standard graymorphology are instead
Journal of Electrical and Computer Engineering 3
(a1) Original image (b1) Gray distribution of original image
(a2) WTH transform result (b2) Gray distribution after WTH transform
(a3) BTH transform result (b3) Gray distribution after BTH transform
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200300
1000 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
Figure 1 WTH and BTH transform results and gray distribution images with 3 times 3 pixels
4 Journal of Electrical and Computer Engineering
(a1) Original image (b1) Gray distribution of original image
(a2) WTH transform result (b2) Gray distribution after WTH transform
(a3) BTH transform result (b3) Gray distribution after BTH transform
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200300
1000 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
Figure 2 WTH and BTH transform results and gray distribution images with 7 times 7 pixels
Journal of Electrical and Computer Engineering 5
b1
b2
Figure 3 Schematic diagram of soft morphology structuring ele-ment 119887
of sorting weighted statistical operation The determinationof weighted coefficients is associated with the structuringelement Unlike standard mathematical morphology thestructuring element 119887 in soft morphology consists of twoparts one is the center of the structure 119887
1(also called hard-
core) and theweight (number of iterations) of correspondingpixel is greater than 1 or equal to 1 the other is soft edge 119887
2 and
the weight of corresponding pixel is equal to 1The schematicdiagram is shown in Figure 3
Similar to the standard mathematical morphology softmorphology erosion and dilation operations are defined asfollows
119891Θ [1198871 1198872 119896] (119909) = 119896th min
119909+119910isin1198701
119909+119911isin1198702
119896
[119891 (119910) minus 1198872(119909 + 119910)] cup [119891 (119911) minus 119887
1(119909 + 119911)]
(2)
119891 oplus [1198871 1198872 119896] (119909) = 119896th max
Formula (2) is the erosion operation of soft morphologyand formula (3) is the dilation operation of soft morphology1198871is the hard-core of structuring element 119870
1sube 119885
2 is thedefinition domain of 119887
1 1198872is the soft edge of structuring
element 1198702sube 119885
2 is the definition domain of 1198872 the
structuring element 119887 = 1198871cup 1198872 and 119887
1cap 1198872= 119896 119891(119909) =
119891(119909) 119891(119909) 119891(119909) is a repeating set 119896 is the number ofiterations When 119896 = 1 119887
1= 1198872 and 119887 = the soft morphol-
ogy is degenerated to the standard graymorphologyThe cen-ter pixel of 119887
1is also the center pixel of 119887 the soft edge pixels of
1198872are also the soft edge pixels of 119887 the weights of 119887
1and 1198872are
determined by 119896 the value of 119896 reflects differences betweentarget area and the neighborhood We can change the softmorphology structuring element by adjusting the value of 119896and make the differences between targets and background inimage to achieve a better effect of background suppression
Based on erosion and dilation operation the definition ofsoft morphology opening and closing operation is as follows
Formula (4) is the soft morphology opening operationand formula (5) is the softmorphology closing operationTheprocess of opening operation can be interpreted as making
dilation operation to 119891 with soft edge 1198872first and then mak-
ing erosion operation to the result with 1198871and 1198872 The proc-
ess order of closing operation is contrasted with openingoperation Hard-core 119887
1and soft edge 119887
2are shown in
1198871= (
1 1 1
1 1nabla1
1 1 1
)
1198872=
(
(
(
1 1 1 1 1
1 0 0 0 1
1 0 0nabla0 1
1 0 0 0 1
1 1 1 1 1
)
)
)
(6)
Figure 4 shows the results after the soft morphologyoperation in which (b) is the opening operation result and(c) is the result of SWTH transform based on soft mor-phology opening operation By using soft morphology open-ing operation it can smooth the background image filtertargets and noise smaller than the structuring element andget an estimated image of the original image backgroundFrom the comparison between Figures 4(c) and 2(a2) thebackground suppression effect of soft morphology top-hat(SWTH) transform is better than gray morphology WTHrsquosAfter SWTH transform filtering in Figure 4(c) the gray gra-dation of target is decreased obviouslyThe small weak targetsalmost are suppressed In order to improve the backgroundsuppression effect of soft morphology we propose a newbackground suppression method based on soft morphologyfiltering and Retinex theory
4 Background Suppression Based on SoftMorphology and Retinex Theory
Based on the analysis in the previous section it can be foundthat the soft morphology operation has a good performancein suppressing background of the surface image In orderto enhance the image contrast and improve SCR in theimage wemake the background suppression of displacementmeasurement of targets and motion vector estimation in thesurface image with the Retinex theory
41 RetinexTheory Retinex is derived from the combinationof retina and cortex It is a color theory which can describethe color constancy of the human visual system In the studyof the principles of the human visual perception systemand psychophysical brightness Land [21] found that whenthe visual system processes the visual information someuncertain external factors such as light intensity and unevenlight will be excluded and some characteristics informationwhich can reflect the essence of objects will be retainedThen these characteristicsrsquo information will be delivered tothe cortex by neural network and form visual image Retinextheory was proposed in 1977 it can powerfully explainthe homeostatic mechanism the human visual system can
6 Journal of Electrical and Computer Engineering
(a) Original image (b) Result of soft morphology opening operation
(c) Result of soft morphology background suppres-sion
Figure 4 Result image of soft morphology operation
Incident light L
Reflecting object R
Observer
S(x y)
S(x y) = L(x y) middot R(x y)
Figure 5 Schematics of Retinex
achieve the same color of one object under different light byself-regulation
Figure 5 shows the schematic diagram of Retinex theoryAccording to Retinex theory the image 119878(119909 119910) is constitutedby two factors One factor is the illumination intensity 119871 ofthe object which corresponds to the low-frequency part ofthe image and presents the luminance image 119871(119909 119910) anotherfactor is the reflective brightness 119877 of the object whichcorresponds to the high-frequency part of the image and
presents the reflection image 119877(119909 119910) So the imaging processof the image can be expressed as
The illumination intensity 119871 determines the dynamicrange of pixel in an image and the reflection luminance 119877reflects the nature of the objectThe essence of Retinex theoryis casting aside the nature of the illumination intensity 119871 andobtains the inherent essential characteristics 119877 of the objectfrom the image 119878 Taking the logarithm of formula (7) intoaccount the complex operations can be translated into simpleaddition and subtraction and the formulas are as follows
where 119904 = ln[119878(119909 119910)] 119897 = ln[119871(119909 119910)] and 119903 = ln[119877(119909 119910)]Usually we cannot achieve the reflection luminance 119877 of
the object directly However we can estimate the illumination
Journal of Electrical and Computer Engineering 7
intensity 119871 firstly Then we use the image 119878 to subtract theillumination intensity 119871 In this way the reflection luminance119877 which can reflect the essential characteristics of the objectcan be achieved The formula can be expressed as
119903 = 119904 minus 119897 (9)
This is also equivalent to the concept of background sup-pression principle the high-frequency part (including targetand high-frequency noise) can be separated by comparing theoriginal image with low-frequency part of the image There-fore how to estimate the light intensity is the key of the issue
Ferwerda et al [22] showed that incident component inan image can be estimated andKimmel et al [23] showed thatthe incident component estimation problem (illuminationintensity 119871) can be formulated as a Quadratic Programmingoptimization problem and furthermore they showed theoptimization problemhaving a unique solutionWewill applythe above conclusion in our algorithm
The commonly used methods to estimate the incidentcomponent include look-up table and convolution methodsTo deal with the background suppression issue of the riverwater visual image it needs to face multiple different imagesApparently building a single gray look-up table cannot meetthe requirementsTherefore we use themethod of the convo-lution operation to estimate the optimal incident componentIn this method selecting the appropriate kernel functionto do the convolution operation is the key of the problemGaussian kernel function can highlight the center position ofweight value Meanwhile the influence of the surroundingpoints of the center position can be taken into accountAnd the estimated image has a good correlation with theoriginal image Based on the above reasons 3 times 3 Gaussiankernel function is chosen to do the optimal estimation ofthe incident component The values of 3 times 3 Gaussian kernelfunction are as follows
119870 =
[
[
[
[
[
[
[
[
[
[
1
16
1
8
1
16
1
8
1
4
1
8
1
16
1
8
1
16
]
]
]
]
]
]
]
]
]
]
(10)
The convolution operation to the image with Gaussiankernel function is equivalent to doing a low pass filterA new image will be achieved after each convolution andthe optimal estimated value of incident component can beachieved According to the literature [24 25] the average graylevel of the image tends to stability after three convolutionoperations So it is thought that the result after the thirdconvolution is the most suitable result to be the optimalestimation of the incident component of the image
42 Background Suppression Method In the process of targetmotion vector estimation with the river visible backgroundimage the obtained river image has a complex backgroundwhich includes many lights such as direct illumination fromthe sun atmospheric scattering light surface reflected light
Soft morphologyopening operation
Input originalimage
Outputresult
Result of openoperation
Originalimage
Optimal estimation ofincident component
++
minus
Figure 6 Background suppression flowchart based on soft mor-phology and Retinex theory
(flare) surface-emitting light (reflection) and target reflectedlight Therefore the image will present the uneven lightundulating background and unidentified target For thecomplex situation a method based on soft morphology andRetinex theory is proposed to realize image background sup-pression
According to the previous analysis we can find that theestimation of incident component can achieve an optimalestimation for low-frequency part of the image It has impor-tant practical significance of the surface visible backgroundimage with complex lighting conditions Through the softmorphology operations and optimal incident componentestimation we can achieve the optimal estimation of thebackground image Then by using the original image tosubtract the estimated image a background suppressionimagewith a higher signal-to-noise ratiowill be achievedTheflowchart of the proposed method is shown in Figure 6
Step 1 (opening operation) It has been shown that the opti-mized size of operator structure is generally equal to the halfof the maximal size of a small two-dimensional target [26]Therefore we chose the following central structuring element1198871and flexible edge structure element 119887
2[27ndash29]
1198871=
(
(
(
1 1 1 1 1
1 1 1 1 1
1 1 1nabla1 1
1 1 1 1 1
1 1 1 1 1
)
)
)
1198872=
(
(
(
(
(
(
(
(
(
1 1 1 1 1 1 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 0 0 0nabla0 0 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 1 1 1 1 1 1
)
)
)
)
)
)
)
)
)
(11)
8 Journal of Electrical and Computer Engineering
Table 1 Comparison of three background suppression methods
The structural element 119887 is composed of 1198871and 119887
2 It
takes soft morphological opening operation on the originalimage After the opening operation the noise and the targetwhich is smaller than the structuring element in the image areeliminated and the background image becomes smoother
Step 2 (estimating the optimal incident component) Accord-ing to the aforementioned method of optimal estimation ofimage incident component we make a cubic convolutionwith the result after Step 1 and Gaussian kernel function informula (10)ThroughGaussian convolution we filter out thehigh-frequency part of the image and achieve the optimalbackground estimation of the image
Step 3 (background suppression) Using the original imageto subtract the optimal estimated image obtained after Step 2we can achieve the background suppression result that filterslow-frequency part of the original image and enhancescontrast
119891119879+ 119891119873= 119891 minus (119891 ∘ [119887
1 1198872 119896])
1015840
(12)
where 119891119879is target component 119891
119873is noise component 119891 is
original image and (119891 ∘ [1198871 1198872 119896])
1015840 is the background imageafter the processing of soft morphology opening operationand the incident component optimal estimation Formula(12) can be as the soft morphology white top-hat transformbased on Retinex theory
5 Experiments and Simulations
In order to verify the effectiveness of the proposed methodwe make an experiment with the method which combineswith soft morphological opening operation and optimal esti-mation of incident component with background suppressionof the image based on Retinex theoryWemake a comparisonbetween WTH and SWTH The experiment is carried outwith the PCwhich is equipped with PentiumT4300 memory286GB and uses Matlab software platform To evaluate theperformance of this method we use pixel-level evaluationIn the study of image background suppression an indicatorknown as SCR (signal-to-clutter ratio) is used as follows
SCR =(119891119879minus 120583119887)
120590119887
(13)
where 119891119879is the target strength and it can be replaced by the
maximumgrayscale value of the image under the visible light
120583119887is the average grayscale value of the image and it reflects
the background DC component 120590119887is the grayscale standard
deviation of the image and it reflects the degree of clutterTheSCR Gain can be defined as
119866SCR = 20 log(SCRoutSCRin) (14)
where SCRin and SCRout denote the SCR of the original imageand the background suppression image SCR can be usedto describe the improvement of the original image with thebackground suppression method In experiment we chosethree screenshots from the actual shooting river scene videoas subjects (image size is 256 times 256 pixels) as shown inFigures 7(a1) 7(b1) and 7(c1) Figures 7(a2)ndash7(a4) 7(b2)ndash7(b4) and 7(c2)ndash7(c4) are the results of the experiments
Figures 7 and 8 and Table 1 respectively present the resultcharts three-dimensional grayscale distribution diagramsand SCRSCR Gain data statistics of the three surface imagebackground suppression methods Grayscale morphologicalWTH transform is computed with 5 times 5 square structuringelement while the soft morphology white top-hat (SWTH)transform and the method in this paper are using thestructuring element 119887 which is comprised of 119887
1and 119887
2in
formula (11) From Table 1 it can be found that the originalimage has a low SCR the targets have poor visibility inthe grayscale image and the grayscale values of backgroundare fluctuant WTH and SWTH transform can suppressbackground eliminate some of the background clutter andimprove the SCR However there is still some backgroundclutter residue in the image Through the proposed methodin this paper the background clutter in the image is filtered toa large extent The background is flattened and the grayscalevalue of background is low Moreover most of the targets areretained and have good visibility and the target enhancementis notable these can be reflected by SCR Gain in Table 1
Compared with the three methods mentioned in thispaper the proposed background suppression method has anotable improvement of SCR in the image and the SCR Gainis also the largest Grayscale morphology WTH transformhas a weak performance in background suppression and thesoftmorphology SWTH has amedium performance in back-ground suppression Therefore the experiment fully demon-strated that the proposed method has a better ability in back-ground suppression than the other two methods Howeverthis method also has some shortcomings After the imageprocessing of image 3 the effect of background suppressionhas no notable improvement compared with WTH and
Journal of Electrical and Computer Engineering 9
(a1) Img 1 original image (a2) Img 1 result of WTH (a3) Img 1 result of SWTH (a4) Img 1 result of our method
(b1) Img 2 original image (b2) Img 2 result of WTH (b3) Img 2 result of SWTH (b4) Img 2 result of our method
(c1) Img 3 original image (c2) Img 3 result of WTH (c3) Img 3 result of SWTH (c4) Img 3 result of our method
Figure 7 Comparison of three methods
SWTH and the three methods also have almost the sameperformance in SCR Gain The reason is that the pixelsize of the target in image 3 is large In this method thelarge size target with soft morphological opening operationcannot achieve an ideal result in background clutter residualTherefore the background image has a big fluctuation afterbackground suppression Although the improvement of SCRis not notable the target grayscale and image contrast have anotable improvement Moreover our method has the largestgrayscale value and optimal target visibility among the threemethods and the subjective evaluation of the quality indiscrimination between the background and target is also thebest
6 Conclusion
To overcome the shortcomings of surface noise and cluttersurface tracer optical reflection complexity difficulty in targetdisplacement detection and motion vector estimation we
present a background suppression method based on softmorphological filtering and Retinex theory in this paperIn order to improve the performance of surface imagebackground suppression method we use the Retinex theoryand make an optimal estimation of incident component ofthe background image through soft morphological openingoperation The experiments give the results of backgroundsuppression of surface image and make a comparison withgrayscale morphological WTH transform and soft morphol-ogy SWTH transform experiments The simulations showthat the proposed method has a notable improvement inbackground suppression of surface image Meanwhile ourmethod makes a good preparatory work for the next targetdisplacement detection and motion vector estimation
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
10 Journal of Electrical and Computer Engineering
(a1) Img 1 original image (b1) Img 2 original image (c1) Img 3 original image
Figure 8 Comparison of three methods on 3D grayscale distribution diagrams
Acknowledgments
This paper is partially supported by the National NaturalScience Foundation of China (no 61263029 no 61374019)a project funded by the Priority Academic Program Devel-opment (PAPD) of Jiangsu Higher Education Institutionsand Natural Science Foundation of Jiangsu Province (noBK20130851)
References
[1] X Hao Research of moving point targets detection method inimage sequences [PhD thesis] Shandong University JinanChina 2005
[2] L Xu Z Zhang X Yan H Wang and X Wang ldquoAdvances ofnon-contact instruments and techniques for open-channel flow
Journal of Electrical and Computer Engineering 11
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
(a1) Original image (b1) Gray distribution of original image
(a2) WTH transform result (b2) Gray distribution after WTH transform
(a3) BTH transform result (b3) Gray distribution after BTH transform
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200300
1000 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
Figure 1 WTH and BTH transform results and gray distribution images with 3 times 3 pixels
4 Journal of Electrical and Computer Engineering
(a1) Original image (b1) Gray distribution of original image
(a2) WTH transform result (b2) Gray distribution after WTH transform
(a3) BTH transform result (b3) Gray distribution after BTH transform
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200300
1000 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
Figure 2 WTH and BTH transform results and gray distribution images with 7 times 7 pixels
Journal of Electrical and Computer Engineering 5
b1
b2
Figure 3 Schematic diagram of soft morphology structuring ele-ment 119887
of sorting weighted statistical operation The determinationof weighted coefficients is associated with the structuringelement Unlike standard mathematical morphology thestructuring element 119887 in soft morphology consists of twoparts one is the center of the structure 119887
1(also called hard-
core) and theweight (number of iterations) of correspondingpixel is greater than 1 or equal to 1 the other is soft edge 119887
2 and
the weight of corresponding pixel is equal to 1The schematicdiagram is shown in Figure 3
Similar to the standard mathematical morphology softmorphology erosion and dilation operations are defined asfollows
119891Θ [1198871 1198872 119896] (119909) = 119896th min
119909+119910isin1198701
119909+119911isin1198702
119896
[119891 (119910) minus 1198872(119909 + 119910)] cup [119891 (119911) minus 119887
1(119909 + 119911)]
(2)
119891 oplus [1198871 1198872 119896] (119909) = 119896th max
Formula (2) is the erosion operation of soft morphologyand formula (3) is the dilation operation of soft morphology1198871is the hard-core of structuring element 119870
1sube 119885
2 is thedefinition domain of 119887
1 1198872is the soft edge of structuring
element 1198702sube 119885
2 is the definition domain of 1198872 the
structuring element 119887 = 1198871cup 1198872 and 119887
1cap 1198872= 119896 119891(119909) =
119891(119909) 119891(119909) 119891(119909) is a repeating set 119896 is the number ofiterations When 119896 = 1 119887
1= 1198872 and 119887 = the soft morphol-
ogy is degenerated to the standard graymorphologyThe cen-ter pixel of 119887
1is also the center pixel of 119887 the soft edge pixels of
1198872are also the soft edge pixels of 119887 the weights of 119887
1and 1198872are
determined by 119896 the value of 119896 reflects differences betweentarget area and the neighborhood We can change the softmorphology structuring element by adjusting the value of 119896and make the differences between targets and background inimage to achieve a better effect of background suppression
Based on erosion and dilation operation the definition ofsoft morphology opening and closing operation is as follows
Formula (4) is the soft morphology opening operationand formula (5) is the softmorphology closing operationTheprocess of opening operation can be interpreted as making
dilation operation to 119891 with soft edge 1198872first and then mak-
ing erosion operation to the result with 1198871and 1198872 The proc-
ess order of closing operation is contrasted with openingoperation Hard-core 119887
1and soft edge 119887
2are shown in
1198871= (
1 1 1
1 1nabla1
1 1 1
)
1198872=
(
(
(
1 1 1 1 1
1 0 0 0 1
1 0 0nabla0 1
1 0 0 0 1
1 1 1 1 1
)
)
)
(6)
Figure 4 shows the results after the soft morphologyoperation in which (b) is the opening operation result and(c) is the result of SWTH transform based on soft mor-phology opening operation By using soft morphology open-ing operation it can smooth the background image filtertargets and noise smaller than the structuring element andget an estimated image of the original image backgroundFrom the comparison between Figures 4(c) and 2(a2) thebackground suppression effect of soft morphology top-hat(SWTH) transform is better than gray morphology WTHrsquosAfter SWTH transform filtering in Figure 4(c) the gray gra-dation of target is decreased obviouslyThe small weak targetsalmost are suppressed In order to improve the backgroundsuppression effect of soft morphology we propose a newbackground suppression method based on soft morphologyfiltering and Retinex theory
4 Background Suppression Based on SoftMorphology and Retinex Theory
Based on the analysis in the previous section it can be foundthat the soft morphology operation has a good performancein suppressing background of the surface image In orderto enhance the image contrast and improve SCR in theimage wemake the background suppression of displacementmeasurement of targets and motion vector estimation in thesurface image with the Retinex theory
41 RetinexTheory Retinex is derived from the combinationof retina and cortex It is a color theory which can describethe color constancy of the human visual system In the studyof the principles of the human visual perception systemand psychophysical brightness Land [21] found that whenthe visual system processes the visual information someuncertain external factors such as light intensity and unevenlight will be excluded and some characteristics informationwhich can reflect the essence of objects will be retainedThen these characteristicsrsquo information will be delivered tothe cortex by neural network and form visual image Retinextheory was proposed in 1977 it can powerfully explainthe homeostatic mechanism the human visual system can
6 Journal of Electrical and Computer Engineering
(a) Original image (b) Result of soft morphology opening operation
(c) Result of soft morphology background suppres-sion
Figure 4 Result image of soft morphology operation
Incident light L
Reflecting object R
Observer
S(x y)
S(x y) = L(x y) middot R(x y)
Figure 5 Schematics of Retinex
achieve the same color of one object under different light byself-regulation
Figure 5 shows the schematic diagram of Retinex theoryAccording to Retinex theory the image 119878(119909 119910) is constitutedby two factors One factor is the illumination intensity 119871 ofthe object which corresponds to the low-frequency part ofthe image and presents the luminance image 119871(119909 119910) anotherfactor is the reflective brightness 119877 of the object whichcorresponds to the high-frequency part of the image and
presents the reflection image 119877(119909 119910) So the imaging processof the image can be expressed as
The illumination intensity 119871 determines the dynamicrange of pixel in an image and the reflection luminance 119877reflects the nature of the objectThe essence of Retinex theoryis casting aside the nature of the illumination intensity 119871 andobtains the inherent essential characteristics 119877 of the objectfrom the image 119878 Taking the logarithm of formula (7) intoaccount the complex operations can be translated into simpleaddition and subtraction and the formulas are as follows
where 119904 = ln[119878(119909 119910)] 119897 = ln[119871(119909 119910)] and 119903 = ln[119877(119909 119910)]Usually we cannot achieve the reflection luminance 119877 of
the object directly However we can estimate the illumination
Journal of Electrical and Computer Engineering 7
intensity 119871 firstly Then we use the image 119878 to subtract theillumination intensity 119871 In this way the reflection luminance119877 which can reflect the essential characteristics of the objectcan be achieved The formula can be expressed as
119903 = 119904 minus 119897 (9)
This is also equivalent to the concept of background sup-pression principle the high-frequency part (including targetand high-frequency noise) can be separated by comparing theoriginal image with low-frequency part of the image There-fore how to estimate the light intensity is the key of the issue
Ferwerda et al [22] showed that incident component inan image can be estimated andKimmel et al [23] showed thatthe incident component estimation problem (illuminationintensity 119871) can be formulated as a Quadratic Programmingoptimization problem and furthermore they showed theoptimization problemhaving a unique solutionWewill applythe above conclusion in our algorithm
The commonly used methods to estimate the incidentcomponent include look-up table and convolution methodsTo deal with the background suppression issue of the riverwater visual image it needs to face multiple different imagesApparently building a single gray look-up table cannot meetthe requirementsTherefore we use themethod of the convo-lution operation to estimate the optimal incident componentIn this method selecting the appropriate kernel functionto do the convolution operation is the key of the problemGaussian kernel function can highlight the center position ofweight value Meanwhile the influence of the surroundingpoints of the center position can be taken into accountAnd the estimated image has a good correlation with theoriginal image Based on the above reasons 3 times 3 Gaussiankernel function is chosen to do the optimal estimation ofthe incident component The values of 3 times 3 Gaussian kernelfunction are as follows
119870 =
[
[
[
[
[
[
[
[
[
[
1
16
1
8
1
16
1
8
1
4
1
8
1
16
1
8
1
16
]
]
]
]
]
]
]
]
]
]
(10)
The convolution operation to the image with Gaussiankernel function is equivalent to doing a low pass filterA new image will be achieved after each convolution andthe optimal estimated value of incident component can beachieved According to the literature [24 25] the average graylevel of the image tends to stability after three convolutionoperations So it is thought that the result after the thirdconvolution is the most suitable result to be the optimalestimation of the incident component of the image
42 Background Suppression Method In the process of targetmotion vector estimation with the river visible backgroundimage the obtained river image has a complex backgroundwhich includes many lights such as direct illumination fromthe sun atmospheric scattering light surface reflected light
Soft morphologyopening operation
Input originalimage
Outputresult
Result of openoperation
Originalimage
Optimal estimation ofincident component
++
minus
Figure 6 Background suppression flowchart based on soft mor-phology and Retinex theory
(flare) surface-emitting light (reflection) and target reflectedlight Therefore the image will present the uneven lightundulating background and unidentified target For thecomplex situation a method based on soft morphology andRetinex theory is proposed to realize image background sup-pression
According to the previous analysis we can find that theestimation of incident component can achieve an optimalestimation for low-frequency part of the image It has impor-tant practical significance of the surface visible backgroundimage with complex lighting conditions Through the softmorphology operations and optimal incident componentestimation we can achieve the optimal estimation of thebackground image Then by using the original image tosubtract the estimated image a background suppressionimagewith a higher signal-to-noise ratiowill be achievedTheflowchart of the proposed method is shown in Figure 6
Step 1 (opening operation) It has been shown that the opti-mized size of operator structure is generally equal to the halfof the maximal size of a small two-dimensional target [26]Therefore we chose the following central structuring element1198871and flexible edge structure element 119887
2[27ndash29]
1198871=
(
(
(
1 1 1 1 1
1 1 1 1 1
1 1 1nabla1 1
1 1 1 1 1
1 1 1 1 1
)
)
)
1198872=
(
(
(
(
(
(
(
(
(
1 1 1 1 1 1 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 0 0 0nabla0 0 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 1 1 1 1 1 1
)
)
)
)
)
)
)
)
)
(11)
8 Journal of Electrical and Computer Engineering
Table 1 Comparison of three background suppression methods
The structural element 119887 is composed of 1198871and 119887
2 It
takes soft morphological opening operation on the originalimage After the opening operation the noise and the targetwhich is smaller than the structuring element in the image areeliminated and the background image becomes smoother
Step 2 (estimating the optimal incident component) Accord-ing to the aforementioned method of optimal estimation ofimage incident component we make a cubic convolutionwith the result after Step 1 and Gaussian kernel function informula (10)ThroughGaussian convolution we filter out thehigh-frequency part of the image and achieve the optimalbackground estimation of the image
Step 3 (background suppression) Using the original imageto subtract the optimal estimated image obtained after Step 2we can achieve the background suppression result that filterslow-frequency part of the original image and enhancescontrast
119891119879+ 119891119873= 119891 minus (119891 ∘ [119887
1 1198872 119896])
1015840
(12)
where 119891119879is target component 119891
119873is noise component 119891 is
original image and (119891 ∘ [1198871 1198872 119896])
1015840 is the background imageafter the processing of soft morphology opening operationand the incident component optimal estimation Formula(12) can be as the soft morphology white top-hat transformbased on Retinex theory
5 Experiments and Simulations
In order to verify the effectiveness of the proposed methodwe make an experiment with the method which combineswith soft morphological opening operation and optimal esti-mation of incident component with background suppressionof the image based on Retinex theoryWemake a comparisonbetween WTH and SWTH The experiment is carried outwith the PCwhich is equipped with PentiumT4300 memory286GB and uses Matlab software platform To evaluate theperformance of this method we use pixel-level evaluationIn the study of image background suppression an indicatorknown as SCR (signal-to-clutter ratio) is used as follows
SCR =(119891119879minus 120583119887)
120590119887
(13)
where 119891119879is the target strength and it can be replaced by the
maximumgrayscale value of the image under the visible light
120583119887is the average grayscale value of the image and it reflects
the background DC component 120590119887is the grayscale standard
deviation of the image and it reflects the degree of clutterTheSCR Gain can be defined as
119866SCR = 20 log(SCRoutSCRin) (14)
where SCRin and SCRout denote the SCR of the original imageand the background suppression image SCR can be usedto describe the improvement of the original image with thebackground suppression method In experiment we chosethree screenshots from the actual shooting river scene videoas subjects (image size is 256 times 256 pixels) as shown inFigures 7(a1) 7(b1) and 7(c1) Figures 7(a2)ndash7(a4) 7(b2)ndash7(b4) and 7(c2)ndash7(c4) are the results of the experiments
Figures 7 and 8 and Table 1 respectively present the resultcharts three-dimensional grayscale distribution diagramsand SCRSCR Gain data statistics of the three surface imagebackground suppression methods Grayscale morphologicalWTH transform is computed with 5 times 5 square structuringelement while the soft morphology white top-hat (SWTH)transform and the method in this paper are using thestructuring element 119887 which is comprised of 119887
1and 119887
2in
formula (11) From Table 1 it can be found that the originalimage has a low SCR the targets have poor visibility inthe grayscale image and the grayscale values of backgroundare fluctuant WTH and SWTH transform can suppressbackground eliminate some of the background clutter andimprove the SCR However there is still some backgroundclutter residue in the image Through the proposed methodin this paper the background clutter in the image is filtered toa large extent The background is flattened and the grayscalevalue of background is low Moreover most of the targets areretained and have good visibility and the target enhancementis notable these can be reflected by SCR Gain in Table 1
Compared with the three methods mentioned in thispaper the proposed background suppression method has anotable improvement of SCR in the image and the SCR Gainis also the largest Grayscale morphology WTH transformhas a weak performance in background suppression and thesoftmorphology SWTH has amedium performance in back-ground suppression Therefore the experiment fully demon-strated that the proposed method has a better ability in back-ground suppression than the other two methods Howeverthis method also has some shortcomings After the imageprocessing of image 3 the effect of background suppressionhas no notable improvement compared with WTH and
Journal of Electrical and Computer Engineering 9
(a1) Img 1 original image (a2) Img 1 result of WTH (a3) Img 1 result of SWTH (a4) Img 1 result of our method
(b1) Img 2 original image (b2) Img 2 result of WTH (b3) Img 2 result of SWTH (b4) Img 2 result of our method
(c1) Img 3 original image (c2) Img 3 result of WTH (c3) Img 3 result of SWTH (c4) Img 3 result of our method
Figure 7 Comparison of three methods
SWTH and the three methods also have almost the sameperformance in SCR Gain The reason is that the pixelsize of the target in image 3 is large In this method thelarge size target with soft morphological opening operationcannot achieve an ideal result in background clutter residualTherefore the background image has a big fluctuation afterbackground suppression Although the improvement of SCRis not notable the target grayscale and image contrast have anotable improvement Moreover our method has the largestgrayscale value and optimal target visibility among the threemethods and the subjective evaluation of the quality indiscrimination between the background and target is also thebest
6 Conclusion
To overcome the shortcomings of surface noise and cluttersurface tracer optical reflection complexity difficulty in targetdisplacement detection and motion vector estimation we
present a background suppression method based on softmorphological filtering and Retinex theory in this paperIn order to improve the performance of surface imagebackground suppression method we use the Retinex theoryand make an optimal estimation of incident component ofthe background image through soft morphological openingoperation The experiments give the results of backgroundsuppression of surface image and make a comparison withgrayscale morphological WTH transform and soft morphol-ogy SWTH transform experiments The simulations showthat the proposed method has a notable improvement inbackground suppression of surface image Meanwhile ourmethod makes a good preparatory work for the next targetdisplacement detection and motion vector estimation
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
10 Journal of Electrical and Computer Engineering
(a1) Img 1 original image (b1) Img 2 original image (c1) Img 3 original image
Figure 8 Comparison of three methods on 3D grayscale distribution diagrams
Acknowledgments
This paper is partially supported by the National NaturalScience Foundation of China (no 61263029 no 61374019)a project funded by the Priority Academic Program Devel-opment (PAPD) of Jiangsu Higher Education Institutionsand Natural Science Foundation of Jiangsu Province (noBK20130851)
References
[1] X Hao Research of moving point targets detection method inimage sequences [PhD thesis] Shandong University JinanChina 2005
[2] L Xu Z Zhang X Yan H Wang and X Wang ldquoAdvances ofnon-contact instruments and techniques for open-channel flow
Journal of Electrical and Computer Engineering 11
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
(a1) Original image (b1) Gray distribution of original image
(a2) WTH transform result (b2) Gray distribution after WTH transform
(a3) BTH transform result (b3) Gray distribution after BTH transform
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200300
1000 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
300
50
100
150
200
250
300
Gra
ysca
le in
tens
ity
200100
0 0
0
100200
300
y (pixel)x (pixel)
Figure 2 WTH and BTH transform results and gray distribution images with 7 times 7 pixels
Journal of Electrical and Computer Engineering 5
b1
b2
Figure 3 Schematic diagram of soft morphology structuring ele-ment 119887
of sorting weighted statistical operation The determinationof weighted coefficients is associated with the structuringelement Unlike standard mathematical morphology thestructuring element 119887 in soft morphology consists of twoparts one is the center of the structure 119887
1(also called hard-
core) and theweight (number of iterations) of correspondingpixel is greater than 1 or equal to 1 the other is soft edge 119887
2 and
the weight of corresponding pixel is equal to 1The schematicdiagram is shown in Figure 3
Similar to the standard mathematical morphology softmorphology erosion and dilation operations are defined asfollows
119891Θ [1198871 1198872 119896] (119909) = 119896th min
119909+119910isin1198701
119909+119911isin1198702
119896
[119891 (119910) minus 1198872(119909 + 119910)] cup [119891 (119911) minus 119887
1(119909 + 119911)]
(2)
119891 oplus [1198871 1198872 119896] (119909) = 119896th max
Formula (2) is the erosion operation of soft morphologyand formula (3) is the dilation operation of soft morphology1198871is the hard-core of structuring element 119870
1sube 119885
2 is thedefinition domain of 119887
1 1198872is the soft edge of structuring
element 1198702sube 119885
2 is the definition domain of 1198872 the
structuring element 119887 = 1198871cup 1198872 and 119887
1cap 1198872= 119896 119891(119909) =
119891(119909) 119891(119909) 119891(119909) is a repeating set 119896 is the number ofiterations When 119896 = 1 119887
1= 1198872 and 119887 = the soft morphol-
ogy is degenerated to the standard graymorphologyThe cen-ter pixel of 119887
1is also the center pixel of 119887 the soft edge pixels of
1198872are also the soft edge pixels of 119887 the weights of 119887
1and 1198872are
determined by 119896 the value of 119896 reflects differences betweentarget area and the neighborhood We can change the softmorphology structuring element by adjusting the value of 119896and make the differences between targets and background inimage to achieve a better effect of background suppression
Based on erosion and dilation operation the definition ofsoft morphology opening and closing operation is as follows
Formula (4) is the soft morphology opening operationand formula (5) is the softmorphology closing operationTheprocess of opening operation can be interpreted as making
dilation operation to 119891 with soft edge 1198872first and then mak-
ing erosion operation to the result with 1198871and 1198872 The proc-
ess order of closing operation is contrasted with openingoperation Hard-core 119887
1and soft edge 119887
2are shown in
1198871= (
1 1 1
1 1nabla1
1 1 1
)
1198872=
(
(
(
1 1 1 1 1
1 0 0 0 1
1 0 0nabla0 1
1 0 0 0 1
1 1 1 1 1
)
)
)
(6)
Figure 4 shows the results after the soft morphologyoperation in which (b) is the opening operation result and(c) is the result of SWTH transform based on soft mor-phology opening operation By using soft morphology open-ing operation it can smooth the background image filtertargets and noise smaller than the structuring element andget an estimated image of the original image backgroundFrom the comparison between Figures 4(c) and 2(a2) thebackground suppression effect of soft morphology top-hat(SWTH) transform is better than gray morphology WTHrsquosAfter SWTH transform filtering in Figure 4(c) the gray gra-dation of target is decreased obviouslyThe small weak targetsalmost are suppressed In order to improve the backgroundsuppression effect of soft morphology we propose a newbackground suppression method based on soft morphologyfiltering and Retinex theory
4 Background Suppression Based on SoftMorphology and Retinex Theory
Based on the analysis in the previous section it can be foundthat the soft morphology operation has a good performancein suppressing background of the surface image In orderto enhance the image contrast and improve SCR in theimage wemake the background suppression of displacementmeasurement of targets and motion vector estimation in thesurface image with the Retinex theory
41 RetinexTheory Retinex is derived from the combinationof retina and cortex It is a color theory which can describethe color constancy of the human visual system In the studyof the principles of the human visual perception systemand psychophysical brightness Land [21] found that whenthe visual system processes the visual information someuncertain external factors such as light intensity and unevenlight will be excluded and some characteristics informationwhich can reflect the essence of objects will be retainedThen these characteristicsrsquo information will be delivered tothe cortex by neural network and form visual image Retinextheory was proposed in 1977 it can powerfully explainthe homeostatic mechanism the human visual system can
6 Journal of Electrical and Computer Engineering
(a) Original image (b) Result of soft morphology opening operation
(c) Result of soft morphology background suppres-sion
Figure 4 Result image of soft morphology operation
Incident light L
Reflecting object R
Observer
S(x y)
S(x y) = L(x y) middot R(x y)
Figure 5 Schematics of Retinex
achieve the same color of one object under different light byself-regulation
Figure 5 shows the schematic diagram of Retinex theoryAccording to Retinex theory the image 119878(119909 119910) is constitutedby two factors One factor is the illumination intensity 119871 ofthe object which corresponds to the low-frequency part ofthe image and presents the luminance image 119871(119909 119910) anotherfactor is the reflective brightness 119877 of the object whichcorresponds to the high-frequency part of the image and
presents the reflection image 119877(119909 119910) So the imaging processof the image can be expressed as
The illumination intensity 119871 determines the dynamicrange of pixel in an image and the reflection luminance 119877reflects the nature of the objectThe essence of Retinex theoryis casting aside the nature of the illumination intensity 119871 andobtains the inherent essential characteristics 119877 of the objectfrom the image 119878 Taking the logarithm of formula (7) intoaccount the complex operations can be translated into simpleaddition and subtraction and the formulas are as follows
where 119904 = ln[119878(119909 119910)] 119897 = ln[119871(119909 119910)] and 119903 = ln[119877(119909 119910)]Usually we cannot achieve the reflection luminance 119877 of
the object directly However we can estimate the illumination
Journal of Electrical and Computer Engineering 7
intensity 119871 firstly Then we use the image 119878 to subtract theillumination intensity 119871 In this way the reflection luminance119877 which can reflect the essential characteristics of the objectcan be achieved The formula can be expressed as
119903 = 119904 minus 119897 (9)
This is also equivalent to the concept of background sup-pression principle the high-frequency part (including targetand high-frequency noise) can be separated by comparing theoriginal image with low-frequency part of the image There-fore how to estimate the light intensity is the key of the issue
Ferwerda et al [22] showed that incident component inan image can be estimated andKimmel et al [23] showed thatthe incident component estimation problem (illuminationintensity 119871) can be formulated as a Quadratic Programmingoptimization problem and furthermore they showed theoptimization problemhaving a unique solutionWewill applythe above conclusion in our algorithm
The commonly used methods to estimate the incidentcomponent include look-up table and convolution methodsTo deal with the background suppression issue of the riverwater visual image it needs to face multiple different imagesApparently building a single gray look-up table cannot meetthe requirementsTherefore we use themethod of the convo-lution operation to estimate the optimal incident componentIn this method selecting the appropriate kernel functionto do the convolution operation is the key of the problemGaussian kernel function can highlight the center position ofweight value Meanwhile the influence of the surroundingpoints of the center position can be taken into accountAnd the estimated image has a good correlation with theoriginal image Based on the above reasons 3 times 3 Gaussiankernel function is chosen to do the optimal estimation ofthe incident component The values of 3 times 3 Gaussian kernelfunction are as follows
119870 =
[
[
[
[
[
[
[
[
[
[
1
16
1
8
1
16
1
8
1
4
1
8
1
16
1
8
1
16
]
]
]
]
]
]
]
]
]
]
(10)
The convolution operation to the image with Gaussiankernel function is equivalent to doing a low pass filterA new image will be achieved after each convolution andthe optimal estimated value of incident component can beachieved According to the literature [24 25] the average graylevel of the image tends to stability after three convolutionoperations So it is thought that the result after the thirdconvolution is the most suitable result to be the optimalestimation of the incident component of the image
42 Background Suppression Method In the process of targetmotion vector estimation with the river visible backgroundimage the obtained river image has a complex backgroundwhich includes many lights such as direct illumination fromthe sun atmospheric scattering light surface reflected light
Soft morphologyopening operation
Input originalimage
Outputresult
Result of openoperation
Originalimage
Optimal estimation ofincident component
++
minus
Figure 6 Background suppression flowchart based on soft mor-phology and Retinex theory
(flare) surface-emitting light (reflection) and target reflectedlight Therefore the image will present the uneven lightundulating background and unidentified target For thecomplex situation a method based on soft morphology andRetinex theory is proposed to realize image background sup-pression
According to the previous analysis we can find that theestimation of incident component can achieve an optimalestimation for low-frequency part of the image It has impor-tant practical significance of the surface visible backgroundimage with complex lighting conditions Through the softmorphology operations and optimal incident componentestimation we can achieve the optimal estimation of thebackground image Then by using the original image tosubtract the estimated image a background suppressionimagewith a higher signal-to-noise ratiowill be achievedTheflowchart of the proposed method is shown in Figure 6
Step 1 (opening operation) It has been shown that the opti-mized size of operator structure is generally equal to the halfof the maximal size of a small two-dimensional target [26]Therefore we chose the following central structuring element1198871and flexible edge structure element 119887
2[27ndash29]
1198871=
(
(
(
1 1 1 1 1
1 1 1 1 1
1 1 1nabla1 1
1 1 1 1 1
1 1 1 1 1
)
)
)
1198872=
(
(
(
(
(
(
(
(
(
1 1 1 1 1 1 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 0 0 0nabla0 0 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 1 1 1 1 1 1
)
)
)
)
)
)
)
)
)
(11)
8 Journal of Electrical and Computer Engineering
Table 1 Comparison of three background suppression methods
The structural element 119887 is composed of 1198871and 119887
2 It
takes soft morphological opening operation on the originalimage After the opening operation the noise and the targetwhich is smaller than the structuring element in the image areeliminated and the background image becomes smoother
Step 2 (estimating the optimal incident component) Accord-ing to the aforementioned method of optimal estimation ofimage incident component we make a cubic convolutionwith the result after Step 1 and Gaussian kernel function informula (10)ThroughGaussian convolution we filter out thehigh-frequency part of the image and achieve the optimalbackground estimation of the image
Step 3 (background suppression) Using the original imageto subtract the optimal estimated image obtained after Step 2we can achieve the background suppression result that filterslow-frequency part of the original image and enhancescontrast
119891119879+ 119891119873= 119891 minus (119891 ∘ [119887
1 1198872 119896])
1015840
(12)
where 119891119879is target component 119891
119873is noise component 119891 is
original image and (119891 ∘ [1198871 1198872 119896])
1015840 is the background imageafter the processing of soft morphology opening operationand the incident component optimal estimation Formula(12) can be as the soft morphology white top-hat transformbased on Retinex theory
5 Experiments and Simulations
In order to verify the effectiveness of the proposed methodwe make an experiment with the method which combineswith soft morphological opening operation and optimal esti-mation of incident component with background suppressionof the image based on Retinex theoryWemake a comparisonbetween WTH and SWTH The experiment is carried outwith the PCwhich is equipped with PentiumT4300 memory286GB and uses Matlab software platform To evaluate theperformance of this method we use pixel-level evaluationIn the study of image background suppression an indicatorknown as SCR (signal-to-clutter ratio) is used as follows
SCR =(119891119879minus 120583119887)
120590119887
(13)
where 119891119879is the target strength and it can be replaced by the
maximumgrayscale value of the image under the visible light
120583119887is the average grayscale value of the image and it reflects
the background DC component 120590119887is the grayscale standard
deviation of the image and it reflects the degree of clutterTheSCR Gain can be defined as
119866SCR = 20 log(SCRoutSCRin) (14)
where SCRin and SCRout denote the SCR of the original imageand the background suppression image SCR can be usedto describe the improvement of the original image with thebackground suppression method In experiment we chosethree screenshots from the actual shooting river scene videoas subjects (image size is 256 times 256 pixels) as shown inFigures 7(a1) 7(b1) and 7(c1) Figures 7(a2)ndash7(a4) 7(b2)ndash7(b4) and 7(c2)ndash7(c4) are the results of the experiments
Figures 7 and 8 and Table 1 respectively present the resultcharts three-dimensional grayscale distribution diagramsand SCRSCR Gain data statistics of the three surface imagebackground suppression methods Grayscale morphologicalWTH transform is computed with 5 times 5 square structuringelement while the soft morphology white top-hat (SWTH)transform and the method in this paper are using thestructuring element 119887 which is comprised of 119887
1and 119887
2in
formula (11) From Table 1 it can be found that the originalimage has a low SCR the targets have poor visibility inthe grayscale image and the grayscale values of backgroundare fluctuant WTH and SWTH transform can suppressbackground eliminate some of the background clutter andimprove the SCR However there is still some backgroundclutter residue in the image Through the proposed methodin this paper the background clutter in the image is filtered toa large extent The background is flattened and the grayscalevalue of background is low Moreover most of the targets areretained and have good visibility and the target enhancementis notable these can be reflected by SCR Gain in Table 1
Compared with the three methods mentioned in thispaper the proposed background suppression method has anotable improvement of SCR in the image and the SCR Gainis also the largest Grayscale morphology WTH transformhas a weak performance in background suppression and thesoftmorphology SWTH has amedium performance in back-ground suppression Therefore the experiment fully demon-strated that the proposed method has a better ability in back-ground suppression than the other two methods Howeverthis method also has some shortcomings After the imageprocessing of image 3 the effect of background suppressionhas no notable improvement compared with WTH and
Journal of Electrical and Computer Engineering 9
(a1) Img 1 original image (a2) Img 1 result of WTH (a3) Img 1 result of SWTH (a4) Img 1 result of our method
(b1) Img 2 original image (b2) Img 2 result of WTH (b3) Img 2 result of SWTH (b4) Img 2 result of our method
(c1) Img 3 original image (c2) Img 3 result of WTH (c3) Img 3 result of SWTH (c4) Img 3 result of our method
Figure 7 Comparison of three methods
SWTH and the three methods also have almost the sameperformance in SCR Gain The reason is that the pixelsize of the target in image 3 is large In this method thelarge size target with soft morphological opening operationcannot achieve an ideal result in background clutter residualTherefore the background image has a big fluctuation afterbackground suppression Although the improvement of SCRis not notable the target grayscale and image contrast have anotable improvement Moreover our method has the largestgrayscale value and optimal target visibility among the threemethods and the subjective evaluation of the quality indiscrimination between the background and target is also thebest
6 Conclusion
To overcome the shortcomings of surface noise and cluttersurface tracer optical reflection complexity difficulty in targetdisplacement detection and motion vector estimation we
present a background suppression method based on softmorphological filtering and Retinex theory in this paperIn order to improve the performance of surface imagebackground suppression method we use the Retinex theoryand make an optimal estimation of incident component ofthe background image through soft morphological openingoperation The experiments give the results of backgroundsuppression of surface image and make a comparison withgrayscale morphological WTH transform and soft morphol-ogy SWTH transform experiments The simulations showthat the proposed method has a notable improvement inbackground suppression of surface image Meanwhile ourmethod makes a good preparatory work for the next targetdisplacement detection and motion vector estimation
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
10 Journal of Electrical and Computer Engineering
(a1) Img 1 original image (b1) Img 2 original image (c1) Img 3 original image
Figure 8 Comparison of three methods on 3D grayscale distribution diagrams
Acknowledgments
This paper is partially supported by the National NaturalScience Foundation of China (no 61263029 no 61374019)a project funded by the Priority Academic Program Devel-opment (PAPD) of Jiangsu Higher Education Institutionsand Natural Science Foundation of Jiangsu Province (noBK20130851)
References
[1] X Hao Research of moving point targets detection method inimage sequences [PhD thesis] Shandong University JinanChina 2005
[2] L Xu Z Zhang X Yan H Wang and X Wang ldquoAdvances ofnon-contact instruments and techniques for open-channel flow
Journal of Electrical and Computer Engineering 11
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
Figure 3 Schematic diagram of soft morphology structuring ele-ment 119887
of sorting weighted statistical operation The determinationof weighted coefficients is associated with the structuringelement Unlike standard mathematical morphology thestructuring element 119887 in soft morphology consists of twoparts one is the center of the structure 119887
1(also called hard-
core) and theweight (number of iterations) of correspondingpixel is greater than 1 or equal to 1 the other is soft edge 119887
2 and
the weight of corresponding pixel is equal to 1The schematicdiagram is shown in Figure 3
Similar to the standard mathematical morphology softmorphology erosion and dilation operations are defined asfollows
119891Θ [1198871 1198872 119896] (119909) = 119896th min
119909+119910isin1198701
119909+119911isin1198702
119896
[119891 (119910) minus 1198872(119909 + 119910)] cup [119891 (119911) minus 119887
1(119909 + 119911)]
(2)
119891 oplus [1198871 1198872 119896] (119909) = 119896th max
Formula (2) is the erosion operation of soft morphologyand formula (3) is the dilation operation of soft morphology1198871is the hard-core of structuring element 119870
1sube 119885
2 is thedefinition domain of 119887
1 1198872is the soft edge of structuring
element 1198702sube 119885
2 is the definition domain of 1198872 the
structuring element 119887 = 1198871cup 1198872 and 119887
1cap 1198872= 119896 119891(119909) =
119891(119909) 119891(119909) 119891(119909) is a repeating set 119896 is the number ofiterations When 119896 = 1 119887
1= 1198872 and 119887 = the soft morphol-
ogy is degenerated to the standard graymorphologyThe cen-ter pixel of 119887
1is also the center pixel of 119887 the soft edge pixels of
1198872are also the soft edge pixels of 119887 the weights of 119887
1and 1198872are
determined by 119896 the value of 119896 reflects differences betweentarget area and the neighborhood We can change the softmorphology structuring element by adjusting the value of 119896and make the differences between targets and background inimage to achieve a better effect of background suppression
Based on erosion and dilation operation the definition ofsoft morphology opening and closing operation is as follows
Formula (4) is the soft morphology opening operationand formula (5) is the softmorphology closing operationTheprocess of opening operation can be interpreted as making
dilation operation to 119891 with soft edge 1198872first and then mak-
ing erosion operation to the result with 1198871and 1198872 The proc-
ess order of closing operation is contrasted with openingoperation Hard-core 119887
1and soft edge 119887
2are shown in
1198871= (
1 1 1
1 1nabla1
1 1 1
)
1198872=
(
(
(
1 1 1 1 1
1 0 0 0 1
1 0 0nabla0 1
1 0 0 0 1
1 1 1 1 1
)
)
)
(6)
Figure 4 shows the results after the soft morphologyoperation in which (b) is the opening operation result and(c) is the result of SWTH transform based on soft mor-phology opening operation By using soft morphology open-ing operation it can smooth the background image filtertargets and noise smaller than the structuring element andget an estimated image of the original image backgroundFrom the comparison between Figures 4(c) and 2(a2) thebackground suppression effect of soft morphology top-hat(SWTH) transform is better than gray morphology WTHrsquosAfter SWTH transform filtering in Figure 4(c) the gray gra-dation of target is decreased obviouslyThe small weak targetsalmost are suppressed In order to improve the backgroundsuppression effect of soft morphology we propose a newbackground suppression method based on soft morphologyfiltering and Retinex theory
4 Background Suppression Based on SoftMorphology and Retinex Theory
Based on the analysis in the previous section it can be foundthat the soft morphology operation has a good performancein suppressing background of the surface image In orderto enhance the image contrast and improve SCR in theimage wemake the background suppression of displacementmeasurement of targets and motion vector estimation in thesurface image with the Retinex theory
41 RetinexTheory Retinex is derived from the combinationof retina and cortex It is a color theory which can describethe color constancy of the human visual system In the studyof the principles of the human visual perception systemand psychophysical brightness Land [21] found that whenthe visual system processes the visual information someuncertain external factors such as light intensity and unevenlight will be excluded and some characteristics informationwhich can reflect the essence of objects will be retainedThen these characteristicsrsquo information will be delivered tothe cortex by neural network and form visual image Retinextheory was proposed in 1977 it can powerfully explainthe homeostatic mechanism the human visual system can
6 Journal of Electrical and Computer Engineering
(a) Original image (b) Result of soft morphology opening operation
(c) Result of soft morphology background suppres-sion
Figure 4 Result image of soft morphology operation
Incident light L
Reflecting object R
Observer
S(x y)
S(x y) = L(x y) middot R(x y)
Figure 5 Schematics of Retinex
achieve the same color of one object under different light byself-regulation
Figure 5 shows the schematic diagram of Retinex theoryAccording to Retinex theory the image 119878(119909 119910) is constitutedby two factors One factor is the illumination intensity 119871 ofthe object which corresponds to the low-frequency part ofthe image and presents the luminance image 119871(119909 119910) anotherfactor is the reflective brightness 119877 of the object whichcorresponds to the high-frequency part of the image and
presents the reflection image 119877(119909 119910) So the imaging processof the image can be expressed as
The illumination intensity 119871 determines the dynamicrange of pixel in an image and the reflection luminance 119877reflects the nature of the objectThe essence of Retinex theoryis casting aside the nature of the illumination intensity 119871 andobtains the inherent essential characteristics 119877 of the objectfrom the image 119878 Taking the logarithm of formula (7) intoaccount the complex operations can be translated into simpleaddition and subtraction and the formulas are as follows
where 119904 = ln[119878(119909 119910)] 119897 = ln[119871(119909 119910)] and 119903 = ln[119877(119909 119910)]Usually we cannot achieve the reflection luminance 119877 of
the object directly However we can estimate the illumination
Journal of Electrical and Computer Engineering 7
intensity 119871 firstly Then we use the image 119878 to subtract theillumination intensity 119871 In this way the reflection luminance119877 which can reflect the essential characteristics of the objectcan be achieved The formula can be expressed as
119903 = 119904 minus 119897 (9)
This is also equivalent to the concept of background sup-pression principle the high-frequency part (including targetand high-frequency noise) can be separated by comparing theoriginal image with low-frequency part of the image There-fore how to estimate the light intensity is the key of the issue
Ferwerda et al [22] showed that incident component inan image can be estimated andKimmel et al [23] showed thatthe incident component estimation problem (illuminationintensity 119871) can be formulated as a Quadratic Programmingoptimization problem and furthermore they showed theoptimization problemhaving a unique solutionWewill applythe above conclusion in our algorithm
The commonly used methods to estimate the incidentcomponent include look-up table and convolution methodsTo deal with the background suppression issue of the riverwater visual image it needs to face multiple different imagesApparently building a single gray look-up table cannot meetthe requirementsTherefore we use themethod of the convo-lution operation to estimate the optimal incident componentIn this method selecting the appropriate kernel functionto do the convolution operation is the key of the problemGaussian kernel function can highlight the center position ofweight value Meanwhile the influence of the surroundingpoints of the center position can be taken into accountAnd the estimated image has a good correlation with theoriginal image Based on the above reasons 3 times 3 Gaussiankernel function is chosen to do the optimal estimation ofthe incident component The values of 3 times 3 Gaussian kernelfunction are as follows
119870 =
[
[
[
[
[
[
[
[
[
[
1
16
1
8
1
16
1
8
1
4
1
8
1
16
1
8
1
16
]
]
]
]
]
]
]
]
]
]
(10)
The convolution operation to the image with Gaussiankernel function is equivalent to doing a low pass filterA new image will be achieved after each convolution andthe optimal estimated value of incident component can beachieved According to the literature [24 25] the average graylevel of the image tends to stability after three convolutionoperations So it is thought that the result after the thirdconvolution is the most suitable result to be the optimalestimation of the incident component of the image
42 Background Suppression Method In the process of targetmotion vector estimation with the river visible backgroundimage the obtained river image has a complex backgroundwhich includes many lights such as direct illumination fromthe sun atmospheric scattering light surface reflected light
Soft morphologyopening operation
Input originalimage
Outputresult
Result of openoperation
Originalimage
Optimal estimation ofincident component
++
minus
Figure 6 Background suppression flowchart based on soft mor-phology and Retinex theory
(flare) surface-emitting light (reflection) and target reflectedlight Therefore the image will present the uneven lightundulating background and unidentified target For thecomplex situation a method based on soft morphology andRetinex theory is proposed to realize image background sup-pression
According to the previous analysis we can find that theestimation of incident component can achieve an optimalestimation for low-frequency part of the image It has impor-tant practical significance of the surface visible backgroundimage with complex lighting conditions Through the softmorphology operations and optimal incident componentestimation we can achieve the optimal estimation of thebackground image Then by using the original image tosubtract the estimated image a background suppressionimagewith a higher signal-to-noise ratiowill be achievedTheflowchart of the proposed method is shown in Figure 6
Step 1 (opening operation) It has been shown that the opti-mized size of operator structure is generally equal to the halfof the maximal size of a small two-dimensional target [26]Therefore we chose the following central structuring element1198871and flexible edge structure element 119887
2[27ndash29]
1198871=
(
(
(
1 1 1 1 1
1 1 1 1 1
1 1 1nabla1 1
1 1 1 1 1
1 1 1 1 1
)
)
)
1198872=
(
(
(
(
(
(
(
(
(
1 1 1 1 1 1 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 0 0 0nabla0 0 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 1 1 1 1 1 1
)
)
)
)
)
)
)
)
)
(11)
8 Journal of Electrical and Computer Engineering
Table 1 Comparison of three background suppression methods
The structural element 119887 is composed of 1198871and 119887
2 It
takes soft morphological opening operation on the originalimage After the opening operation the noise and the targetwhich is smaller than the structuring element in the image areeliminated and the background image becomes smoother
Step 2 (estimating the optimal incident component) Accord-ing to the aforementioned method of optimal estimation ofimage incident component we make a cubic convolutionwith the result after Step 1 and Gaussian kernel function informula (10)ThroughGaussian convolution we filter out thehigh-frequency part of the image and achieve the optimalbackground estimation of the image
Step 3 (background suppression) Using the original imageto subtract the optimal estimated image obtained after Step 2we can achieve the background suppression result that filterslow-frequency part of the original image and enhancescontrast
119891119879+ 119891119873= 119891 minus (119891 ∘ [119887
1 1198872 119896])
1015840
(12)
where 119891119879is target component 119891
119873is noise component 119891 is
original image and (119891 ∘ [1198871 1198872 119896])
1015840 is the background imageafter the processing of soft morphology opening operationand the incident component optimal estimation Formula(12) can be as the soft morphology white top-hat transformbased on Retinex theory
5 Experiments and Simulations
In order to verify the effectiveness of the proposed methodwe make an experiment with the method which combineswith soft morphological opening operation and optimal esti-mation of incident component with background suppressionof the image based on Retinex theoryWemake a comparisonbetween WTH and SWTH The experiment is carried outwith the PCwhich is equipped with PentiumT4300 memory286GB and uses Matlab software platform To evaluate theperformance of this method we use pixel-level evaluationIn the study of image background suppression an indicatorknown as SCR (signal-to-clutter ratio) is used as follows
SCR =(119891119879minus 120583119887)
120590119887
(13)
where 119891119879is the target strength and it can be replaced by the
maximumgrayscale value of the image under the visible light
120583119887is the average grayscale value of the image and it reflects
the background DC component 120590119887is the grayscale standard
deviation of the image and it reflects the degree of clutterTheSCR Gain can be defined as
119866SCR = 20 log(SCRoutSCRin) (14)
where SCRin and SCRout denote the SCR of the original imageand the background suppression image SCR can be usedto describe the improvement of the original image with thebackground suppression method In experiment we chosethree screenshots from the actual shooting river scene videoas subjects (image size is 256 times 256 pixels) as shown inFigures 7(a1) 7(b1) and 7(c1) Figures 7(a2)ndash7(a4) 7(b2)ndash7(b4) and 7(c2)ndash7(c4) are the results of the experiments
Figures 7 and 8 and Table 1 respectively present the resultcharts three-dimensional grayscale distribution diagramsand SCRSCR Gain data statistics of the three surface imagebackground suppression methods Grayscale morphologicalWTH transform is computed with 5 times 5 square structuringelement while the soft morphology white top-hat (SWTH)transform and the method in this paper are using thestructuring element 119887 which is comprised of 119887
1and 119887
2in
formula (11) From Table 1 it can be found that the originalimage has a low SCR the targets have poor visibility inthe grayscale image and the grayscale values of backgroundare fluctuant WTH and SWTH transform can suppressbackground eliminate some of the background clutter andimprove the SCR However there is still some backgroundclutter residue in the image Through the proposed methodin this paper the background clutter in the image is filtered toa large extent The background is flattened and the grayscalevalue of background is low Moreover most of the targets areretained and have good visibility and the target enhancementis notable these can be reflected by SCR Gain in Table 1
Compared with the three methods mentioned in thispaper the proposed background suppression method has anotable improvement of SCR in the image and the SCR Gainis also the largest Grayscale morphology WTH transformhas a weak performance in background suppression and thesoftmorphology SWTH has amedium performance in back-ground suppression Therefore the experiment fully demon-strated that the proposed method has a better ability in back-ground suppression than the other two methods Howeverthis method also has some shortcomings After the imageprocessing of image 3 the effect of background suppressionhas no notable improvement compared with WTH and
Journal of Electrical and Computer Engineering 9
(a1) Img 1 original image (a2) Img 1 result of WTH (a3) Img 1 result of SWTH (a4) Img 1 result of our method
(b1) Img 2 original image (b2) Img 2 result of WTH (b3) Img 2 result of SWTH (b4) Img 2 result of our method
(c1) Img 3 original image (c2) Img 3 result of WTH (c3) Img 3 result of SWTH (c4) Img 3 result of our method
Figure 7 Comparison of three methods
SWTH and the three methods also have almost the sameperformance in SCR Gain The reason is that the pixelsize of the target in image 3 is large In this method thelarge size target with soft morphological opening operationcannot achieve an ideal result in background clutter residualTherefore the background image has a big fluctuation afterbackground suppression Although the improvement of SCRis not notable the target grayscale and image contrast have anotable improvement Moreover our method has the largestgrayscale value and optimal target visibility among the threemethods and the subjective evaluation of the quality indiscrimination between the background and target is also thebest
6 Conclusion
To overcome the shortcomings of surface noise and cluttersurface tracer optical reflection complexity difficulty in targetdisplacement detection and motion vector estimation we
present a background suppression method based on softmorphological filtering and Retinex theory in this paperIn order to improve the performance of surface imagebackground suppression method we use the Retinex theoryand make an optimal estimation of incident component ofthe background image through soft morphological openingoperation The experiments give the results of backgroundsuppression of surface image and make a comparison withgrayscale morphological WTH transform and soft morphol-ogy SWTH transform experiments The simulations showthat the proposed method has a notable improvement inbackground suppression of surface image Meanwhile ourmethod makes a good preparatory work for the next targetdisplacement detection and motion vector estimation
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
10 Journal of Electrical and Computer Engineering
(a1) Img 1 original image (b1) Img 2 original image (c1) Img 3 original image
Figure 8 Comparison of three methods on 3D grayscale distribution diagrams
Acknowledgments
This paper is partially supported by the National NaturalScience Foundation of China (no 61263029 no 61374019)a project funded by the Priority Academic Program Devel-opment (PAPD) of Jiangsu Higher Education Institutionsand Natural Science Foundation of Jiangsu Province (noBK20130851)
References
[1] X Hao Research of moving point targets detection method inimage sequences [PhD thesis] Shandong University JinanChina 2005
[2] L Xu Z Zhang X Yan H Wang and X Wang ldquoAdvances ofnon-contact instruments and techniques for open-channel flow
Journal of Electrical and Computer Engineering 11
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
(a) Original image (b) Result of soft morphology opening operation
(c) Result of soft morphology background suppres-sion
Figure 4 Result image of soft morphology operation
Incident light L
Reflecting object R
Observer
S(x y)
S(x y) = L(x y) middot R(x y)
Figure 5 Schematics of Retinex
achieve the same color of one object under different light byself-regulation
Figure 5 shows the schematic diagram of Retinex theoryAccording to Retinex theory the image 119878(119909 119910) is constitutedby two factors One factor is the illumination intensity 119871 ofthe object which corresponds to the low-frequency part ofthe image and presents the luminance image 119871(119909 119910) anotherfactor is the reflective brightness 119877 of the object whichcorresponds to the high-frequency part of the image and
presents the reflection image 119877(119909 119910) So the imaging processof the image can be expressed as
The illumination intensity 119871 determines the dynamicrange of pixel in an image and the reflection luminance 119877reflects the nature of the objectThe essence of Retinex theoryis casting aside the nature of the illumination intensity 119871 andobtains the inherent essential characteristics 119877 of the objectfrom the image 119878 Taking the logarithm of formula (7) intoaccount the complex operations can be translated into simpleaddition and subtraction and the formulas are as follows
where 119904 = ln[119878(119909 119910)] 119897 = ln[119871(119909 119910)] and 119903 = ln[119877(119909 119910)]Usually we cannot achieve the reflection luminance 119877 of
the object directly However we can estimate the illumination
Journal of Electrical and Computer Engineering 7
intensity 119871 firstly Then we use the image 119878 to subtract theillumination intensity 119871 In this way the reflection luminance119877 which can reflect the essential characteristics of the objectcan be achieved The formula can be expressed as
119903 = 119904 minus 119897 (9)
This is also equivalent to the concept of background sup-pression principle the high-frequency part (including targetand high-frequency noise) can be separated by comparing theoriginal image with low-frequency part of the image There-fore how to estimate the light intensity is the key of the issue
Ferwerda et al [22] showed that incident component inan image can be estimated andKimmel et al [23] showed thatthe incident component estimation problem (illuminationintensity 119871) can be formulated as a Quadratic Programmingoptimization problem and furthermore they showed theoptimization problemhaving a unique solutionWewill applythe above conclusion in our algorithm
The commonly used methods to estimate the incidentcomponent include look-up table and convolution methodsTo deal with the background suppression issue of the riverwater visual image it needs to face multiple different imagesApparently building a single gray look-up table cannot meetthe requirementsTherefore we use themethod of the convo-lution operation to estimate the optimal incident componentIn this method selecting the appropriate kernel functionto do the convolution operation is the key of the problemGaussian kernel function can highlight the center position ofweight value Meanwhile the influence of the surroundingpoints of the center position can be taken into accountAnd the estimated image has a good correlation with theoriginal image Based on the above reasons 3 times 3 Gaussiankernel function is chosen to do the optimal estimation ofthe incident component The values of 3 times 3 Gaussian kernelfunction are as follows
119870 =
[
[
[
[
[
[
[
[
[
[
1
16
1
8
1
16
1
8
1
4
1
8
1
16
1
8
1
16
]
]
]
]
]
]
]
]
]
]
(10)
The convolution operation to the image with Gaussiankernel function is equivalent to doing a low pass filterA new image will be achieved after each convolution andthe optimal estimated value of incident component can beachieved According to the literature [24 25] the average graylevel of the image tends to stability after three convolutionoperations So it is thought that the result after the thirdconvolution is the most suitable result to be the optimalestimation of the incident component of the image
42 Background Suppression Method In the process of targetmotion vector estimation with the river visible backgroundimage the obtained river image has a complex backgroundwhich includes many lights such as direct illumination fromthe sun atmospheric scattering light surface reflected light
Soft morphologyopening operation
Input originalimage
Outputresult
Result of openoperation
Originalimage
Optimal estimation ofincident component
++
minus
Figure 6 Background suppression flowchart based on soft mor-phology and Retinex theory
(flare) surface-emitting light (reflection) and target reflectedlight Therefore the image will present the uneven lightundulating background and unidentified target For thecomplex situation a method based on soft morphology andRetinex theory is proposed to realize image background sup-pression
According to the previous analysis we can find that theestimation of incident component can achieve an optimalestimation for low-frequency part of the image It has impor-tant practical significance of the surface visible backgroundimage with complex lighting conditions Through the softmorphology operations and optimal incident componentestimation we can achieve the optimal estimation of thebackground image Then by using the original image tosubtract the estimated image a background suppressionimagewith a higher signal-to-noise ratiowill be achievedTheflowchart of the proposed method is shown in Figure 6
Step 1 (opening operation) It has been shown that the opti-mized size of operator structure is generally equal to the halfof the maximal size of a small two-dimensional target [26]Therefore we chose the following central structuring element1198871and flexible edge structure element 119887
2[27ndash29]
1198871=
(
(
(
1 1 1 1 1
1 1 1 1 1
1 1 1nabla1 1
1 1 1 1 1
1 1 1 1 1
)
)
)
1198872=
(
(
(
(
(
(
(
(
(
1 1 1 1 1 1 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 0 0 0nabla0 0 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 1 1 1 1 1 1
)
)
)
)
)
)
)
)
)
(11)
8 Journal of Electrical and Computer Engineering
Table 1 Comparison of three background suppression methods
The structural element 119887 is composed of 1198871and 119887
2 It
takes soft morphological opening operation on the originalimage After the opening operation the noise and the targetwhich is smaller than the structuring element in the image areeliminated and the background image becomes smoother
Step 2 (estimating the optimal incident component) Accord-ing to the aforementioned method of optimal estimation ofimage incident component we make a cubic convolutionwith the result after Step 1 and Gaussian kernel function informula (10)ThroughGaussian convolution we filter out thehigh-frequency part of the image and achieve the optimalbackground estimation of the image
Step 3 (background suppression) Using the original imageto subtract the optimal estimated image obtained after Step 2we can achieve the background suppression result that filterslow-frequency part of the original image and enhancescontrast
119891119879+ 119891119873= 119891 minus (119891 ∘ [119887
1 1198872 119896])
1015840
(12)
where 119891119879is target component 119891
119873is noise component 119891 is
original image and (119891 ∘ [1198871 1198872 119896])
1015840 is the background imageafter the processing of soft morphology opening operationand the incident component optimal estimation Formula(12) can be as the soft morphology white top-hat transformbased on Retinex theory
5 Experiments and Simulations
In order to verify the effectiveness of the proposed methodwe make an experiment with the method which combineswith soft morphological opening operation and optimal esti-mation of incident component with background suppressionof the image based on Retinex theoryWemake a comparisonbetween WTH and SWTH The experiment is carried outwith the PCwhich is equipped with PentiumT4300 memory286GB and uses Matlab software platform To evaluate theperformance of this method we use pixel-level evaluationIn the study of image background suppression an indicatorknown as SCR (signal-to-clutter ratio) is used as follows
SCR =(119891119879minus 120583119887)
120590119887
(13)
where 119891119879is the target strength and it can be replaced by the
maximumgrayscale value of the image under the visible light
120583119887is the average grayscale value of the image and it reflects
the background DC component 120590119887is the grayscale standard
deviation of the image and it reflects the degree of clutterTheSCR Gain can be defined as
119866SCR = 20 log(SCRoutSCRin) (14)
where SCRin and SCRout denote the SCR of the original imageand the background suppression image SCR can be usedto describe the improvement of the original image with thebackground suppression method In experiment we chosethree screenshots from the actual shooting river scene videoas subjects (image size is 256 times 256 pixels) as shown inFigures 7(a1) 7(b1) and 7(c1) Figures 7(a2)ndash7(a4) 7(b2)ndash7(b4) and 7(c2)ndash7(c4) are the results of the experiments
Figures 7 and 8 and Table 1 respectively present the resultcharts three-dimensional grayscale distribution diagramsand SCRSCR Gain data statistics of the three surface imagebackground suppression methods Grayscale morphologicalWTH transform is computed with 5 times 5 square structuringelement while the soft morphology white top-hat (SWTH)transform and the method in this paper are using thestructuring element 119887 which is comprised of 119887
1and 119887
2in
formula (11) From Table 1 it can be found that the originalimage has a low SCR the targets have poor visibility inthe grayscale image and the grayscale values of backgroundare fluctuant WTH and SWTH transform can suppressbackground eliminate some of the background clutter andimprove the SCR However there is still some backgroundclutter residue in the image Through the proposed methodin this paper the background clutter in the image is filtered toa large extent The background is flattened and the grayscalevalue of background is low Moreover most of the targets areretained and have good visibility and the target enhancementis notable these can be reflected by SCR Gain in Table 1
Compared with the three methods mentioned in thispaper the proposed background suppression method has anotable improvement of SCR in the image and the SCR Gainis also the largest Grayscale morphology WTH transformhas a weak performance in background suppression and thesoftmorphology SWTH has amedium performance in back-ground suppression Therefore the experiment fully demon-strated that the proposed method has a better ability in back-ground suppression than the other two methods Howeverthis method also has some shortcomings After the imageprocessing of image 3 the effect of background suppressionhas no notable improvement compared with WTH and
Journal of Electrical and Computer Engineering 9
(a1) Img 1 original image (a2) Img 1 result of WTH (a3) Img 1 result of SWTH (a4) Img 1 result of our method
(b1) Img 2 original image (b2) Img 2 result of WTH (b3) Img 2 result of SWTH (b4) Img 2 result of our method
(c1) Img 3 original image (c2) Img 3 result of WTH (c3) Img 3 result of SWTH (c4) Img 3 result of our method
Figure 7 Comparison of three methods
SWTH and the three methods also have almost the sameperformance in SCR Gain The reason is that the pixelsize of the target in image 3 is large In this method thelarge size target with soft morphological opening operationcannot achieve an ideal result in background clutter residualTherefore the background image has a big fluctuation afterbackground suppression Although the improvement of SCRis not notable the target grayscale and image contrast have anotable improvement Moreover our method has the largestgrayscale value and optimal target visibility among the threemethods and the subjective evaluation of the quality indiscrimination between the background and target is also thebest
6 Conclusion
To overcome the shortcomings of surface noise and cluttersurface tracer optical reflection complexity difficulty in targetdisplacement detection and motion vector estimation we
present a background suppression method based on softmorphological filtering and Retinex theory in this paperIn order to improve the performance of surface imagebackground suppression method we use the Retinex theoryand make an optimal estimation of incident component ofthe background image through soft morphological openingoperation The experiments give the results of backgroundsuppression of surface image and make a comparison withgrayscale morphological WTH transform and soft morphol-ogy SWTH transform experiments The simulations showthat the proposed method has a notable improvement inbackground suppression of surface image Meanwhile ourmethod makes a good preparatory work for the next targetdisplacement detection and motion vector estimation
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
10 Journal of Electrical and Computer Engineering
(a1) Img 1 original image (b1) Img 2 original image (c1) Img 3 original image
Figure 8 Comparison of three methods on 3D grayscale distribution diagrams
Acknowledgments
This paper is partially supported by the National NaturalScience Foundation of China (no 61263029 no 61374019)a project funded by the Priority Academic Program Devel-opment (PAPD) of Jiangsu Higher Education Institutionsand Natural Science Foundation of Jiangsu Province (noBK20130851)
References
[1] X Hao Research of moving point targets detection method inimage sequences [PhD thesis] Shandong University JinanChina 2005
[2] L Xu Z Zhang X Yan H Wang and X Wang ldquoAdvances ofnon-contact instruments and techniques for open-channel flow
Journal of Electrical and Computer Engineering 11
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
intensity 119871 firstly Then we use the image 119878 to subtract theillumination intensity 119871 In this way the reflection luminance119877 which can reflect the essential characteristics of the objectcan be achieved The formula can be expressed as
119903 = 119904 minus 119897 (9)
This is also equivalent to the concept of background sup-pression principle the high-frequency part (including targetand high-frequency noise) can be separated by comparing theoriginal image with low-frequency part of the image There-fore how to estimate the light intensity is the key of the issue
Ferwerda et al [22] showed that incident component inan image can be estimated andKimmel et al [23] showed thatthe incident component estimation problem (illuminationintensity 119871) can be formulated as a Quadratic Programmingoptimization problem and furthermore they showed theoptimization problemhaving a unique solutionWewill applythe above conclusion in our algorithm
The commonly used methods to estimate the incidentcomponent include look-up table and convolution methodsTo deal with the background suppression issue of the riverwater visual image it needs to face multiple different imagesApparently building a single gray look-up table cannot meetthe requirementsTherefore we use themethod of the convo-lution operation to estimate the optimal incident componentIn this method selecting the appropriate kernel functionto do the convolution operation is the key of the problemGaussian kernel function can highlight the center position ofweight value Meanwhile the influence of the surroundingpoints of the center position can be taken into accountAnd the estimated image has a good correlation with theoriginal image Based on the above reasons 3 times 3 Gaussiankernel function is chosen to do the optimal estimation ofthe incident component The values of 3 times 3 Gaussian kernelfunction are as follows
119870 =
[
[
[
[
[
[
[
[
[
[
1
16
1
8
1
16
1
8
1
4
1
8
1
16
1
8
1
16
]
]
]
]
]
]
]
]
]
]
(10)
The convolution operation to the image with Gaussiankernel function is equivalent to doing a low pass filterA new image will be achieved after each convolution andthe optimal estimated value of incident component can beachieved According to the literature [24 25] the average graylevel of the image tends to stability after three convolutionoperations So it is thought that the result after the thirdconvolution is the most suitable result to be the optimalestimation of the incident component of the image
42 Background Suppression Method In the process of targetmotion vector estimation with the river visible backgroundimage the obtained river image has a complex backgroundwhich includes many lights such as direct illumination fromthe sun atmospheric scattering light surface reflected light
Soft morphologyopening operation
Input originalimage
Outputresult
Result of openoperation
Originalimage
Optimal estimation ofincident component
++
minus
Figure 6 Background suppression flowchart based on soft mor-phology and Retinex theory
(flare) surface-emitting light (reflection) and target reflectedlight Therefore the image will present the uneven lightundulating background and unidentified target For thecomplex situation a method based on soft morphology andRetinex theory is proposed to realize image background sup-pression
According to the previous analysis we can find that theestimation of incident component can achieve an optimalestimation for low-frequency part of the image It has impor-tant practical significance of the surface visible backgroundimage with complex lighting conditions Through the softmorphology operations and optimal incident componentestimation we can achieve the optimal estimation of thebackground image Then by using the original image tosubtract the estimated image a background suppressionimagewith a higher signal-to-noise ratiowill be achievedTheflowchart of the proposed method is shown in Figure 6
Step 1 (opening operation) It has been shown that the opti-mized size of operator structure is generally equal to the halfof the maximal size of a small two-dimensional target [26]Therefore we chose the following central structuring element1198871and flexible edge structure element 119887
2[27ndash29]
1198871=
(
(
(
1 1 1 1 1
1 1 1 1 1
1 1 1nabla1 1
1 1 1 1 1
1 1 1 1 1
)
)
)
1198872=
(
(
(
(
(
(
(
(
(
1 1 1 1 1 1 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 0 0 0nabla0 0 1
1 0 0 0 0 0 1
1 0 0 0 0 0 1
1 1 1 1 1 1 1
)
)
)
)
)
)
)
)
)
(11)
8 Journal of Electrical and Computer Engineering
Table 1 Comparison of three background suppression methods
The structural element 119887 is composed of 1198871and 119887
2 It
takes soft morphological opening operation on the originalimage After the opening operation the noise and the targetwhich is smaller than the structuring element in the image areeliminated and the background image becomes smoother
Step 2 (estimating the optimal incident component) Accord-ing to the aforementioned method of optimal estimation ofimage incident component we make a cubic convolutionwith the result after Step 1 and Gaussian kernel function informula (10)ThroughGaussian convolution we filter out thehigh-frequency part of the image and achieve the optimalbackground estimation of the image
Step 3 (background suppression) Using the original imageto subtract the optimal estimated image obtained after Step 2we can achieve the background suppression result that filterslow-frequency part of the original image and enhancescontrast
119891119879+ 119891119873= 119891 minus (119891 ∘ [119887
1 1198872 119896])
1015840
(12)
where 119891119879is target component 119891
119873is noise component 119891 is
original image and (119891 ∘ [1198871 1198872 119896])
1015840 is the background imageafter the processing of soft morphology opening operationand the incident component optimal estimation Formula(12) can be as the soft morphology white top-hat transformbased on Retinex theory
5 Experiments and Simulations
In order to verify the effectiveness of the proposed methodwe make an experiment with the method which combineswith soft morphological opening operation and optimal esti-mation of incident component with background suppressionof the image based on Retinex theoryWemake a comparisonbetween WTH and SWTH The experiment is carried outwith the PCwhich is equipped with PentiumT4300 memory286GB and uses Matlab software platform To evaluate theperformance of this method we use pixel-level evaluationIn the study of image background suppression an indicatorknown as SCR (signal-to-clutter ratio) is used as follows
SCR =(119891119879minus 120583119887)
120590119887
(13)
where 119891119879is the target strength and it can be replaced by the
maximumgrayscale value of the image under the visible light
120583119887is the average grayscale value of the image and it reflects
the background DC component 120590119887is the grayscale standard
deviation of the image and it reflects the degree of clutterTheSCR Gain can be defined as
119866SCR = 20 log(SCRoutSCRin) (14)
where SCRin and SCRout denote the SCR of the original imageand the background suppression image SCR can be usedto describe the improvement of the original image with thebackground suppression method In experiment we chosethree screenshots from the actual shooting river scene videoas subjects (image size is 256 times 256 pixels) as shown inFigures 7(a1) 7(b1) and 7(c1) Figures 7(a2)ndash7(a4) 7(b2)ndash7(b4) and 7(c2)ndash7(c4) are the results of the experiments
Figures 7 and 8 and Table 1 respectively present the resultcharts three-dimensional grayscale distribution diagramsand SCRSCR Gain data statistics of the three surface imagebackground suppression methods Grayscale morphologicalWTH transform is computed with 5 times 5 square structuringelement while the soft morphology white top-hat (SWTH)transform and the method in this paper are using thestructuring element 119887 which is comprised of 119887
1and 119887
2in
formula (11) From Table 1 it can be found that the originalimage has a low SCR the targets have poor visibility inthe grayscale image and the grayscale values of backgroundare fluctuant WTH and SWTH transform can suppressbackground eliminate some of the background clutter andimprove the SCR However there is still some backgroundclutter residue in the image Through the proposed methodin this paper the background clutter in the image is filtered toa large extent The background is flattened and the grayscalevalue of background is low Moreover most of the targets areretained and have good visibility and the target enhancementis notable these can be reflected by SCR Gain in Table 1
Compared with the three methods mentioned in thispaper the proposed background suppression method has anotable improvement of SCR in the image and the SCR Gainis also the largest Grayscale morphology WTH transformhas a weak performance in background suppression and thesoftmorphology SWTH has amedium performance in back-ground suppression Therefore the experiment fully demon-strated that the proposed method has a better ability in back-ground suppression than the other two methods Howeverthis method also has some shortcomings After the imageprocessing of image 3 the effect of background suppressionhas no notable improvement compared with WTH and
Journal of Electrical and Computer Engineering 9
(a1) Img 1 original image (a2) Img 1 result of WTH (a3) Img 1 result of SWTH (a4) Img 1 result of our method
(b1) Img 2 original image (b2) Img 2 result of WTH (b3) Img 2 result of SWTH (b4) Img 2 result of our method
(c1) Img 3 original image (c2) Img 3 result of WTH (c3) Img 3 result of SWTH (c4) Img 3 result of our method
Figure 7 Comparison of three methods
SWTH and the three methods also have almost the sameperformance in SCR Gain The reason is that the pixelsize of the target in image 3 is large In this method thelarge size target with soft morphological opening operationcannot achieve an ideal result in background clutter residualTherefore the background image has a big fluctuation afterbackground suppression Although the improvement of SCRis not notable the target grayscale and image contrast have anotable improvement Moreover our method has the largestgrayscale value and optimal target visibility among the threemethods and the subjective evaluation of the quality indiscrimination between the background and target is also thebest
6 Conclusion
To overcome the shortcomings of surface noise and cluttersurface tracer optical reflection complexity difficulty in targetdisplacement detection and motion vector estimation we
present a background suppression method based on softmorphological filtering and Retinex theory in this paperIn order to improve the performance of surface imagebackground suppression method we use the Retinex theoryand make an optimal estimation of incident component ofthe background image through soft morphological openingoperation The experiments give the results of backgroundsuppression of surface image and make a comparison withgrayscale morphological WTH transform and soft morphol-ogy SWTH transform experiments The simulations showthat the proposed method has a notable improvement inbackground suppression of surface image Meanwhile ourmethod makes a good preparatory work for the next targetdisplacement detection and motion vector estimation
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
10 Journal of Electrical and Computer Engineering
(a1) Img 1 original image (b1) Img 2 original image (c1) Img 3 original image
Figure 8 Comparison of three methods on 3D grayscale distribution diagrams
Acknowledgments
This paper is partially supported by the National NaturalScience Foundation of China (no 61263029 no 61374019)a project funded by the Priority Academic Program Devel-opment (PAPD) of Jiangsu Higher Education Institutionsand Natural Science Foundation of Jiangsu Province (noBK20130851)
References
[1] X Hao Research of moving point targets detection method inimage sequences [PhD thesis] Shandong University JinanChina 2005
[2] L Xu Z Zhang X Yan H Wang and X Wang ldquoAdvances ofnon-contact instruments and techniques for open-channel flow
Journal of Electrical and Computer Engineering 11
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
The structural element 119887 is composed of 1198871and 119887
2 It
takes soft morphological opening operation on the originalimage After the opening operation the noise and the targetwhich is smaller than the structuring element in the image areeliminated and the background image becomes smoother
Step 2 (estimating the optimal incident component) Accord-ing to the aforementioned method of optimal estimation ofimage incident component we make a cubic convolutionwith the result after Step 1 and Gaussian kernel function informula (10)ThroughGaussian convolution we filter out thehigh-frequency part of the image and achieve the optimalbackground estimation of the image
Step 3 (background suppression) Using the original imageto subtract the optimal estimated image obtained after Step 2we can achieve the background suppression result that filterslow-frequency part of the original image and enhancescontrast
119891119879+ 119891119873= 119891 minus (119891 ∘ [119887
1 1198872 119896])
1015840
(12)
where 119891119879is target component 119891
119873is noise component 119891 is
original image and (119891 ∘ [1198871 1198872 119896])
1015840 is the background imageafter the processing of soft morphology opening operationand the incident component optimal estimation Formula(12) can be as the soft morphology white top-hat transformbased on Retinex theory
5 Experiments and Simulations
In order to verify the effectiveness of the proposed methodwe make an experiment with the method which combineswith soft morphological opening operation and optimal esti-mation of incident component with background suppressionof the image based on Retinex theoryWemake a comparisonbetween WTH and SWTH The experiment is carried outwith the PCwhich is equipped with PentiumT4300 memory286GB and uses Matlab software platform To evaluate theperformance of this method we use pixel-level evaluationIn the study of image background suppression an indicatorknown as SCR (signal-to-clutter ratio) is used as follows
SCR =(119891119879minus 120583119887)
120590119887
(13)
where 119891119879is the target strength and it can be replaced by the
maximumgrayscale value of the image under the visible light
120583119887is the average grayscale value of the image and it reflects
the background DC component 120590119887is the grayscale standard
deviation of the image and it reflects the degree of clutterTheSCR Gain can be defined as
119866SCR = 20 log(SCRoutSCRin) (14)
where SCRin and SCRout denote the SCR of the original imageand the background suppression image SCR can be usedto describe the improvement of the original image with thebackground suppression method In experiment we chosethree screenshots from the actual shooting river scene videoas subjects (image size is 256 times 256 pixels) as shown inFigures 7(a1) 7(b1) and 7(c1) Figures 7(a2)ndash7(a4) 7(b2)ndash7(b4) and 7(c2)ndash7(c4) are the results of the experiments
Figures 7 and 8 and Table 1 respectively present the resultcharts three-dimensional grayscale distribution diagramsand SCRSCR Gain data statistics of the three surface imagebackground suppression methods Grayscale morphologicalWTH transform is computed with 5 times 5 square structuringelement while the soft morphology white top-hat (SWTH)transform and the method in this paper are using thestructuring element 119887 which is comprised of 119887
1and 119887
2in
formula (11) From Table 1 it can be found that the originalimage has a low SCR the targets have poor visibility inthe grayscale image and the grayscale values of backgroundare fluctuant WTH and SWTH transform can suppressbackground eliminate some of the background clutter andimprove the SCR However there is still some backgroundclutter residue in the image Through the proposed methodin this paper the background clutter in the image is filtered toa large extent The background is flattened and the grayscalevalue of background is low Moreover most of the targets areretained and have good visibility and the target enhancementis notable these can be reflected by SCR Gain in Table 1
Compared with the three methods mentioned in thispaper the proposed background suppression method has anotable improvement of SCR in the image and the SCR Gainis also the largest Grayscale morphology WTH transformhas a weak performance in background suppression and thesoftmorphology SWTH has amedium performance in back-ground suppression Therefore the experiment fully demon-strated that the proposed method has a better ability in back-ground suppression than the other two methods Howeverthis method also has some shortcomings After the imageprocessing of image 3 the effect of background suppressionhas no notable improvement compared with WTH and
Journal of Electrical and Computer Engineering 9
(a1) Img 1 original image (a2) Img 1 result of WTH (a3) Img 1 result of SWTH (a4) Img 1 result of our method
(b1) Img 2 original image (b2) Img 2 result of WTH (b3) Img 2 result of SWTH (b4) Img 2 result of our method
(c1) Img 3 original image (c2) Img 3 result of WTH (c3) Img 3 result of SWTH (c4) Img 3 result of our method
Figure 7 Comparison of three methods
SWTH and the three methods also have almost the sameperformance in SCR Gain The reason is that the pixelsize of the target in image 3 is large In this method thelarge size target with soft morphological opening operationcannot achieve an ideal result in background clutter residualTherefore the background image has a big fluctuation afterbackground suppression Although the improvement of SCRis not notable the target grayscale and image contrast have anotable improvement Moreover our method has the largestgrayscale value and optimal target visibility among the threemethods and the subjective evaluation of the quality indiscrimination between the background and target is also thebest
6 Conclusion
To overcome the shortcomings of surface noise and cluttersurface tracer optical reflection complexity difficulty in targetdisplacement detection and motion vector estimation we
present a background suppression method based on softmorphological filtering and Retinex theory in this paperIn order to improve the performance of surface imagebackground suppression method we use the Retinex theoryand make an optimal estimation of incident component ofthe background image through soft morphological openingoperation The experiments give the results of backgroundsuppression of surface image and make a comparison withgrayscale morphological WTH transform and soft morphol-ogy SWTH transform experiments The simulations showthat the proposed method has a notable improvement inbackground suppression of surface image Meanwhile ourmethod makes a good preparatory work for the next targetdisplacement detection and motion vector estimation
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
10 Journal of Electrical and Computer Engineering
(a1) Img 1 original image (b1) Img 2 original image (c1) Img 3 original image
Figure 8 Comparison of three methods on 3D grayscale distribution diagrams
Acknowledgments
This paper is partially supported by the National NaturalScience Foundation of China (no 61263029 no 61374019)a project funded by the Priority Academic Program Devel-opment (PAPD) of Jiangsu Higher Education Institutionsand Natural Science Foundation of Jiangsu Province (noBK20130851)
References
[1] X Hao Research of moving point targets detection method inimage sequences [PhD thesis] Shandong University JinanChina 2005
[2] L Xu Z Zhang X Yan H Wang and X Wang ldquoAdvances ofnon-contact instruments and techniques for open-channel flow
Journal of Electrical and Computer Engineering 11
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
(a1) Img 1 original image (a2) Img 1 result of WTH (a3) Img 1 result of SWTH (a4) Img 1 result of our method
(b1) Img 2 original image (b2) Img 2 result of WTH (b3) Img 2 result of SWTH (b4) Img 2 result of our method
(c1) Img 3 original image (c2) Img 3 result of WTH (c3) Img 3 result of SWTH (c4) Img 3 result of our method
Figure 7 Comparison of three methods
SWTH and the three methods also have almost the sameperformance in SCR Gain The reason is that the pixelsize of the target in image 3 is large In this method thelarge size target with soft morphological opening operationcannot achieve an ideal result in background clutter residualTherefore the background image has a big fluctuation afterbackground suppression Although the improvement of SCRis not notable the target grayscale and image contrast have anotable improvement Moreover our method has the largestgrayscale value and optimal target visibility among the threemethods and the subjective evaluation of the quality indiscrimination between the background and target is also thebest
6 Conclusion
To overcome the shortcomings of surface noise and cluttersurface tracer optical reflection complexity difficulty in targetdisplacement detection and motion vector estimation we
present a background suppression method based on softmorphological filtering and Retinex theory in this paperIn order to improve the performance of surface imagebackground suppression method we use the Retinex theoryand make an optimal estimation of incident component ofthe background image through soft morphological openingoperation The experiments give the results of backgroundsuppression of surface image and make a comparison withgrayscale morphological WTH transform and soft morphol-ogy SWTH transform experiments The simulations showthat the proposed method has a notable improvement inbackground suppression of surface image Meanwhile ourmethod makes a good preparatory work for the next targetdisplacement detection and motion vector estimation
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
10 Journal of Electrical and Computer Engineering
(a1) Img 1 original image (b1) Img 2 original image (c1) Img 3 original image
Figure 8 Comparison of three methods on 3D grayscale distribution diagrams
Acknowledgments
This paper is partially supported by the National NaturalScience Foundation of China (no 61263029 no 61374019)a project funded by the Priority Academic Program Devel-opment (PAPD) of Jiangsu Higher Education Institutionsand Natural Science Foundation of Jiangsu Province (noBK20130851)
References
[1] X Hao Research of moving point targets detection method inimage sequences [PhD thesis] Shandong University JinanChina 2005
[2] L Xu Z Zhang X Yan H Wang and X Wang ldquoAdvances ofnon-contact instruments and techniques for open-channel flow
Journal of Electrical and Computer Engineering 11
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
Figure 8 Comparison of three methods on 3D grayscale distribution diagrams
Acknowledgments
This paper is partially supported by the National NaturalScience Foundation of China (no 61263029 no 61374019)a project funded by the Priority Academic Program Devel-opment (PAPD) of Jiangsu Higher Education Institutionsand Natural Science Foundation of Jiangsu Province (noBK20130851)
References
[1] X Hao Research of moving point targets detection method inimage sequences [PhD thesis] Shandong University JinanChina 2005
[2] L Xu Z Zhang X Yan H Wang and X Wang ldquoAdvances ofnon-contact instruments and techniques for open-channel flow
Journal of Electrical and Computer Engineering 11
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
measurementsrdquoWater Resources Informatization no 3 pp 37ndash44 2013
[3] MMuste D Kim andVMerwadeModernDigital Instrumentsand Techniques for Hydrodynamic and Morphologic Character-ization of River Channels in Gravel-Bed Rivers John Wiley ampSons New York NY USA 2012
[4] G Dramais J Le Coz B Camenen and A Hauet ldquoAdvantagesof a mobile LSPIV method for measuring flood dischargesand improving stage-discharge curvesrdquo Journal of Hydro-Envi-ronment Research vol 5 no 4 pp 301ndash312 2011
[5] L Xu X Li and S X Yang ldquoIntelligent information processingand system optimizationrdquo Intelligent Automation and Soft Com-puting vol 17 no 7 pp 829ndash831 2011
[6] M Xu and C Wei ldquoRemotely sensed image classification bycomplex network eigenvalue and connected degreerdquo Computa-tional and Mathematical Methods in Medicine vol 2012 ArticleID 632703 9 pages 2012
[7] M Xu F Xu C Huang and M Li ldquoImage restoration usingmajorization-minimizaiton algorithm based on generalizedtotal variationrdquo Journal of Image and Graphics vol 16 no 7 pp1317ndash1325 2011
[8] M Muste I Fujita and A Hauet ldquoLarge-scale particle imagevelocimetry formeasurements in riverine environmentsrdquoWaterResources Research vol 44 no 4 Article IDW00D19 14 pages2008
[9] M Jodeau A Hauet A Paquier J Le Coz and G DramaisldquoApplication and evaluation of LS-PIV technique for the mon-itoring of river surface velocities in high flow conditionsrdquo FlowMeasurement and Instrumentation vol 19 no 2 pp 117ndash1272008
[10] A A Harpold S Mostaghimi P P Vlachos K Brannan andT Dillaha ldquoStream discharge measurement using a large-scaleparticle image velocimetry (LSPIV) prototyperdquo Transactions ofthe ASABE vol 49 no 6 pp 1791ndash1805 2006
[11] Z Zhang Z Chen L Lv X Wang and L Xu ldquoAdaptivebackground suppression method based on visual receptivefieldrdquo Chinese Journal of Scientific Instrument vol 35 no 1 pp191ndash199 2014
[12] L-Z Xu M Li A-Y Shi M Tang and F-C Huang ldquoFea-ture detector model for multi-spectral remote sensing imageinspired by insect visual systemrdquoActa Electronica Sinica vol 39no 11 pp 2497ndash2501 2011
[13] Z Zhang F Xu J Shen L Han and L Xu ldquoPlane measure-ment method with monocular vision based on variable-heighthomographyrdquo Chinese Journal of Scientific Instrument vol 35no 8 pp 1860ndash1867 2014
[14] Z Zhang L Xu and H Wang ldquoReview of natural flow tracersfor river surface imaging velocimetryrdquo Advances in Science andTechnology of Water Resources vol 34 no 3 pp 81ndash88 2014
[15] F Xu Z Sun R Wang X Ding F Huang and L Xu ldquoSuper-resolution reconstruction using kernel regression and feature-driven prior in a charge-coupled device sensor systemrdquo SensorLetters vol 12 no 2 pp 374ndash379 2014
[16] J Serra and P SoilleMathematical Morphology and Its Applica-tions to Image and Signal Processing Kluwer Academic Publish-ers Boston Mass USA 1986
[17] P Kuosmanen and J Astola ldquoSoft morphological filteringrdquoJournal of Mathematical Imaging and Vision vol 5 no 3 pp231ndash262 1995
[18] T Wen J Gu Z Zhang and L Wang ldquoScale selection for mor-phological top-hat transformation based on mutual informa-tionrdquo in Proceedings of the IEEE 3rd International Congress on
Image and Signal Processing (CISP rsquo10) pp 1092ndash1096 YantaiChina October 2010
[19] C Zhao J Wang and P Ji ldquoDetection of small target in IR greyimage based on mathematical morphology by GA optimizedrdquoJournal of Shenyang Ligong University vol 1 p 4 2011
[20] V T Tom T Peli M Leung and J E Bondaryk ldquoMorphology-based algorithm for point target detection in infrared back-groundsrdquo in Signal and Data Processing of Small Targets vol1954 of Proceedings of SPIE pp 2ndash11 Orlando Fla USA April1993
[21] E H Land ldquoAn alternative technique for the computation of thedesignator in the Retinex theory of color visionrdquo Proceedings ofthe National Academy of Sciences of the United States of Americavol 83 no 10 pp 3078ndash3080 1986
[22] J A Ferwerda S N Pattanaik P Shirley and D P GreenbergldquoA model of visual adaptation for realistic image synthesisrdquo inProceedings of the Computer Graphics Conference (SIGGRAPHrsquo96) pp 249ndash258 New Orleans La USA August 1996
[23] R Kimmel M Elad D Shaked R Keshet and I Sobel ldquoA vari-ational framework for retinexrdquo International Journal of Com-puter Vision vol 52 no 1 pp 7ndash23 2003
[24] X Bai and F Zhou ldquoInfrared small target enhancement anddetection based onmodified top-hat transformationsrdquoComput-ers and Electrical Engineering vol 36 no 6 pp 1193ndash1201 2010
[25] U Braga-Neto M Choudhary and J Goutsias ldquoAutomatictarget detection and tracking in forward-looking infrared imagesequences usingmorphological connected operatorsrdquo Journal ofElectronic Imaging vol 13 no 4 pp 802ndash813 2004
[26] B Ye and J Peng ldquoSmall target detection method based onmorphology top-hat operatorrdquo Journal of Image and Graphicsvol 7 no 7 pp 638ndash642 2002
[27] X Bai F Zhou Y Xie and T Jin ldquoNew top-hat transformationand application on infrared small target detectionrdquo Journal ofData Acquisition and Processing vol 24 no 5 pp 643ndash6492009
[28] E R Dougherty An Introduction to Morphological Image Proc-essing SPIE Optical Engineering Press 1992
[29] M Zeng and J Li ldquoThe small target detection in infraredimage based on adaptive morphological top-hat filterrdquo Journalof Shanghai Jiaotong University vol 40 no 1 pp 90ndash93 2006
