+ All Categories
Home > Documents > Impulse Noise Detection & Reduction From Color Images Using Fuzzy Logic

Impulse Noise Detection & Reduction From Color Images Using Fuzzy Logic

Date post: 03-Jan-2016
Category:
Upload: quinlan-gregory
View: 27 times
Download: 2 times
Share this document with a friend
Description:
Impulse Noise Detection & Reduction From Color Images Using Fuzzy Logic. 1.Introduction. Fuzzy Set: If X is a universe of discourse & x is a particular element of X , then a fuzzy set A de fined on X is A={(x,µ A (x)), x€X} & µ A (x):X->[0,1] Noise of Digital Images: - PowerPoint PPT Presentation
Popular Tags:
26
IMPULSE NOISE DETECTION & REDUCTION FROM COLOR IMAGES USING FUZZY LOGIC 2 0 - 0 2 - 2 0 0 8 1 O r i s s a I n s t i t u t e o f T e c h n o l o g y ( O I T ) I N D I A
Transcript
Page 1: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

IMPULSE NOISE DETECTION & REDUCTION FROM COLOR IMAGES USING FUZZY LOGIC

20

-02

-20

08

1

Orissa

Institu

te o

f Tech

nolo

gy(O

IT) IN

DIA

Page 2: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

1.INTRODUCTION

Fuzzy Set:

If X is a universe of discourse & x is a particular element of X , then a fuzzy set A de fined on X is

A={(x,µA(x)), x€X} & µA(x):X->[0,1]

Noise of Digital Images: Digital images are corrupted by impulse noise

during acquision or transmission through communication cahannel.

3-types of noises(impulse, additive & multiplicative)

Removal of impulse noise & enhance image quality done filters like median filter, FIRE, DS-FIRE, MFF(contain 3 subfilters),IFCF, Modified IFCF etc.

20

-02

-20

08

2

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 3: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

IMPULSE NOISE FOR (RGB) COLOR IMAGES:

Let O:Original color digital imageO(i,j,1) , O(i,j,2) , O(i,j,3) be red,green,blue

component of a pixel at O(i,j)Then, Impulse noise for color images as[A(i,j,1)A(i,j,2)A(i,j,3)]= [O(i,j,1)O(i,j,2)O(i,j,3)], with prob.

1-pr Noisy with prob.

pr

Where, pr-Probability that a pixel corrupted & A-corrupted image

20

-02

-20

08

3

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 4: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

IMPULSE NOISE FOR COLOR IMAGES CONTD . . .An impulse noise pixel for color images is R G B

Noise pixel=[pk O(i,j,2) O(i,j,3)] or

[O(i,j,1) p’k O(i,j,3)] or

[O(i,j,1) O(i,j,2) p”k] or

[O(i,j,1) p’k p”k] or

[pk p’k O(i,j,3)] or

[pk O(i,j,2) p”k] or

[pk p’k p”k]

with k€{1,......,n}; n≤2m-1, where pk,p’k,p”k€(0,2m-1)B.FTSCF Method: 2 steps i.e. Detection(local & global), filtering

After detection method ,the filtering is concentrated on impulse noise pixels taking into account the color information of a certain neighborhood around a given central pixel.

20

-02

-20

08

4

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 5: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

II.FUZZY IMPULSE NOISE DETECTION Use fuzzy gradient values to test noisy of

pixelA. Fuzzy gradient Values:

If AR red component of i/p image then

▽(k,l)AR(i,j)=

AR(i+k,j+l)-AR(i,j) with k,l€{-1,0,1}

[Neighbourhood of a central pixel ]

20

-02

-20

08

5

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 6: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

FUZZY IMPULSE NOISE DETECTION CONTD . . .

The pair(i,j): center of gradient 8 gradient values : basic gradient values If value of one basic gradient is large

then some impulse noise at the central red component pixel(i,j) (... Partially true)

1) One of neighbour is noisy, while central pixel is uncorrupted or

2) An edge or a texture element is presentCase-I solved by 8 different values rather

than one gradient value

20

-02

-20

08

6

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 7: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

FUZZY IMPULSE NOISE DETECTION CONTD . . .

Case-II (edge image pixel) solved by one basic and two related gradient values for each direction.

The two related gradient values in the same direction, are determined by the centres making a right-angle with the direction of the first (basic) gradient.

e.g. in the NW-direction (i.e. for (k, l) = (−1,−1)) we calculate the basic gradientvalue ▽(−1,−1)A(i, j) plus the two related gradientvalues ▽(−1,−1)A(i−1, j+1) and ▽(−1,−1)A(i+1, j−1).

20

-02

-20

08

7

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 8: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

INVOLVED GRADIENT VALUE TO CALCULATEFUZZY GRADIENT 2

0-0

2-2

00

8

8

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 9: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

HOW TO CALCULATE ▽FRA(I, J)

▽FRA(i, j) for direction R (R ∈ {NW, N,NE, E, SE, S, SW,W}), is calculated by

IF |▽R A(i, j)| is large AND |▽′R A(i, j)| is small

OR

|▽R A(i, j)| is large AND |▽′′R A(i, j)| is small

OR

▽R A(i, j) is big positive AND (▽′R A(i, j) AND ▽′′R A(i, j) are big negative)

OR

▽R A(i, j) is big negative AND (▽′R A(i, j) AND ▽′′R A(i, j) are big positive)

Then ▽FR A(i, j) is large

where ▽RA(i, j) - basic gradient value & ▽′RA(i, j) and

▽′′RA(i, j)- two related gradient values for the direction R.

“large”, “small”, “big negative” and “bigpositive” are fuzzy sets.

20

-02

-20

08

9

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 10: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

IMPLEMENTATION WITH MEMBERSHIP FUNCTION Fuzzy sets can be

represented by a membership function.

In the fig. Horizontal axes all possible gradient values

[-255,255] & Vertical axes-membership degree(∈ [0,1])

Fuzzy Rule-I In fuzzy logic triangular

norms and co-norms are used to represent conjunctions (roughly the equivalent of AND

operators) and disjunctions (roughly the equivalent of OR

operators).e.g. “|▽R A(i, j)| is large AND |▽′R A(i,

j)|is small” as minLARGE(| ▽R A(i, j)|),SMALL(| ▽′RA(i, j)|)

20

-02

-20

08

10

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 11: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

HOW TO KNOW IMPULSE QUANTITY IN AN IMAGE 1)▽values in R€[0,40] –

most likely nonedge & non-noisy pixels e.g.fig(b)

2)▽values in R€[40,125]-most likely edge pixels or noise pixels eg.fig(c)

3) ▽values in R€[125,255]- most likely noise pixels e.g. Fig(d)

Fuzzy sets big positive and big negative is that if the basic gradient and the two related gradients are both large but have different signs then it is a good indication that noise is present.

20

-02

-20

08

11

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 12: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

DETECTION METHOD

Fuzzy Rule2: If a centralpixel (non-border pixel)is an impulse noise pixel

IF most of the eight ▽FRA(i, j)

are large THEN the central pixel A(i, j) is an impulse noise pixel

If a pixel (i, j) is detected as an impulse noise pixel, then we store the corresponding greyscale value in a histogram

20

-02

-20

08

12

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 13: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

PSEUDO-CODE OF THE DECISION PROCEDURE.

HIST: the noise histogram matrixHIST(i): indicates the amount of selected noise pixels with

greyscale value iIf((i < 0) or (i > 255)) HIST(i) = 0TOT: the total quantity of detected noise pixelsHIST(m):the maximum histogram value assumed to be located at

greyscale value m (m ∈ [0, 255])(1) IF (HIST(m)/TOT) × 100 < THR1(2) NO IMPULSE NOISE DETECTED(3) ELSE(4) IF HIST(i)/(HIST(i−1)+HIST(i)+HIST(i+1)) ≥ 0.95(5) ImpulseNoise(TOT, HIST)(6) ELSE(7) ImpulseNoise&Other(TOT, HIST)(8) END IF(9) END IF

20

-02

-20

08

13

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 14: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

PSEUDO-CODE OF THE IMPULSENOISE FUNCTION.

If impulse noise is detected we perform the ImpulseNoise function

Function ImpulseNoise(real TOT, matrix HIST)(1) temp = 0(2) FOR k = 1 to 5(3) kmax = select the kth greatest y−value of the noise

histogram(4) kpos = the corresponding greyscale value where the

maximumis reached(5) IF ((kmax/TOT) + cum) ≥ THR2(6) pk = kpos(7) set the 4 parameters (ak, bk, ck, dk) to (pk, pk, pk, pk)(8) temp = temp + (kmax/TOT)(9) ELSE(10) break(11) END IF(12) END FOR

20

-02

-20

08

14

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 15: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

PSEUDO-CODE OF THE IMPULSENOISE&OTHER FUNCTION.

Function ImpulseNoise&Other(real TOT, matrix HIST)(1) temp = 0(2) FOR k = 1 to 2(3) kmax = select the kth greatest y−value of the

noise histogram(4) kpos = the corresponding greyscale value where

the maximumis reached(5) IF ((kmax/TOT) + cum) ≥ THR2(6) pk = kpos(7) selecting 4 parameters (ak, bk, ck, dk)(8) temp = temp+ (kmax/TOT)(9) ELSE(10) break(11) END IF(12) END FOREND Function

20

-02

-20

08

15

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 16: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

FUZZY SET IMPULSE NOISE It is a set of all color component

of each pixels in term of membership function.

Each M.F take zero or more parameters i.e. ak,bk,ck,dk & are used to define slope of each selected intensity value pk

µimpulse(AR(i,j))=max(µnoise-

pk(AR(i,j))) k€{1,...,n}

Where,AR(i,j)red component intensity value at posn (i,j)

The parameters are calculated as ak=pk-THRa bk=k-THRb THRb=(2/3)THRa ck=k+THRc THRc=(2/3)THRd dk=k+THRd

20

-02

-20

08

16

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 17: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

FILTERING PHASE µR

impulse, µGimpulse & µB

impulse are used in filtering phase to reduce impulse noise & preserving color diffreence and edges.

The 2-D Array matrices for different color pixel RG(i,j)=AR(i,j)-AG(i,j) RB(i,j)=AR(i,j)-AB(i,j) GR(i,j)=-RG(i,j) GB(i,j)=AG(i,j)-AB(i,j) BR(i,j)=-RB(i,j) BG(i,j)=-GB(i,j)A.First Iteration:Fuzzy Rule 3:(Combination red & green)IF AR(i,j) is impulse noise OR AG(i,j) is impulse noiseTHEN ( RG(i,j) is impulse noise ) AND (GR(i,j) is

impulse noise)Fuzzy Rule 4:(Red+Blue)IF AR(i,j) is impulse noise OR AB(i,j) is impulse noiseTHEN (RB(i,j) is impulse noise ) AND (BR(i,j) is

impulse noise)

20

-02

-20

08

17

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 18: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

FILTERING PHASE CONTD . . .

Fuzzy Rule 5: (Green+Blue)

IF AG(i,j) is impulse noise OR AB(i,j) is impulse noise)

THEN (GB(i,j) is impulse noise) AND (BG(i,j) is impulse noise)

How to improve the impulse component: If red component of a pixel is corrupted (indicated

by large membership degree µRimpulse(AR(i,j)) &

green,blue component are noise free are indicated by small m.d.s then color difference µRG(i,j), µRB(i,j), µGR(i,j), µBR(i,j) have higher degree.

CASE A :If green & blue component of a pixel A(i,j) is noise free(i.e. µG(AG(i,j)= µB(AB(i,j)=0) then filtered red component at posn(i,j) becomes

FR(i,j)=(1/2)(AG(i,j)+ΔRG(i,j))+(1/2)(AB(i,j)+ΔRB(i,j))

20

-02

-20

08

18

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 19: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

FILTERING PHASE CONTD . . .

CASE B: If the green component of pixel AG(i,j) is noisy & blue component AB(i,j) is noise free (i.e. µG(AG(i,j))>0 & µB(AB(i,j))=0 then the filtered red component at posn(i,j) bocomes

FR(i,j)=AB(i,j)+ ΔRB(i,j)

CASE C: If the green component AG(i,j) is noise free & blue component AB(i,j) is noisy(i.e. µG(AG(i,j))=0 & µB(AB(i,j))>0) then the filtered red component at posn

( i,j) bocomesFR(i,j)=AG(i,j)+ ΔRG(i,j)

20

-02

-20

08

19

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 20: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

FILTERING PHASE CONTD . . .Otherwise:

The filtered red component at position(i,j) becomes

B.Next Iterations:

Successive iteration are necessary to reduce the side effect of previous iterations.

C.Stop Criteria: If #e = Amount of pixel values which belong to the support of the fuzzy set more or less impulse noise in the eth iteration, then stop criteria:

1) If there are no pixel values in the support of the fuzzy set (#e = 0) in any iteration e (e ≥ 2).

2) #e is equal to #e−1.

3) Since #1 ≥ #2 ≥ ... ≥ #e−1 ≥ #e hold we can define e = #e−1 − #e. When e is (very) small then we can decide to stop too.

20

-02

-20

08

20

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 21: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

FUTURE WORK This method filters out the impulse noise

rather than Gaussian noise, So Gaussian filter can be applied afterwards.

Our future research will be concentrated on the construction of other fuzzy filtering methods for color images to supress other type noises like speckle noise,stripping noise etc.

All algorithms described here have liner time complexity(i.e.,O(n)).This means if amount of pixels doubled then time complexity will also be doubled (i.e.,2O(n)).

Our future aim is to minimise the time comlexity.

20

-02

-20

08

21

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 22: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

H/W & S/W SPECIFICATION

H/W Specification: Intel Pentium 4 CPU @3.0GHz 512MB DDRAM Mother Board-Intel 945 GZ Chipset Integrated graphics @32bit@1280X1124 Integrated Azallia HD Sound card Hard Disk WD SATA/80GB@7200 rpmS/W Specification: Windows XP Media Center with SP-2 ANSI-C Compiler supporting 32 bit variables MATLAB version 7.0

20

-02

-20

08

22

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 23: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

CONCLUSION

Our fuzzy two step filter reduce impulse noise very effectively without degrading the image structure.

Fuzzy rules are fired every direction around the processed pixel.

The shape of membership function are adapted according to remaining amount of noise after each iterations.

20

-02

-20

08

23

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 24: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

REFERENCES 1.www.research-thesis.org/fuzzyfilter.pdf 2.www.elsevier.com/locate/fss 3.Stefan Schulte, Mike Nachtegael, Val´erie

De Witte, “A Fuzzy Impulse Noise Detection and Reduction Method,” vol-I & II, Fuzziness and Uncertainty Modelling Research Unit B-9000 Gent, Belgium(pdf virsion)

20

-02

-20

08

24

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 25: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

THANKS!!!

20

-02

-20

08

25

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA

Page 26: Impulse Noise Detection &  Reduction From Color Images Using Fuzzy Logic

QUERIES???2

0-0

2-2

00

8

26

Orissa

Institu

te o

f Tech

nolo

gy(O

IT)

IND

IA


Recommended