Computers in Industry 64 (2013) 1383–1389

Multidimensional imaging for skin tissue surface characterization

Jiuai Sun *, Melvyn Smith

Machine Vision Laboratory, University of the West of England, Bristol BS16 1QY, UK

A R T I C L E I N F O

Article history:

Received 3 September 2012

Received in revised form 22 April 2013

Accepted 4 June 2013

Available online 10 July 2013

Keywords:

Skin

Spectral

Shape

Photometric stereo

Chromophore

Wide-field

A B S T R A C T

Human skin, the outer and largest organ covering our body, can be described in terms of both its 3D

spatial topography and its 2D spectral reflectance. Such a characterization normally requires the

application of separate procedures using different kinds of equipment, where spectral reflectance can

only be obtained from a small patch of the skin surface. This paper investigates the integration of

multiple imaging modalities to simultaneously capture both spectral and spatial information from the

skin surface over a wide area. By extending the imaging spectrum from the visible to the near-infrared

(NIR), we improve general recovery, obtain a more detailed skin profile, and are able to identify the

distribution of various principal chromophores within the deeper dermal layers. Experiments show that

new dimensions of skin characterization can be generated through the recovered geometrical and

spectral information, so that an enhanced visibility of important skin physiological phenomena can be

achieved

� 2013 Elsevier B.V. All rights reserved.

Contents lists available at SciVerse ScienceDirect

Computers in Industry

jo ur n al ho m epag e: ww w.els evier . c om / lo cat e/co mp in d

1. Introduction

Covering our entire body, our skin allows us to interact with theexternal environment, while at the same time protecting us fromenvironmental damage – for example through mechanical friction,noxious chemicals, and radiation exposure or temperatureextremes. Our skin appearance also serves to reflect internalchanges, such as mood, ageing and various pathologies relating toinfection and metastasis. Although these mechanisms are still notfully understood, knowledge concerning dermal molecular andskin anatomical structures have been understood and found usefulin characterizing certain functional and cosmetic skin attributes invarious applications. For example, some cosmetic products aim tomoisturize and so smooth the skin microstructure to achieve ahealthy and attractive impression. Computer graphics and visionexperts also use knowledge concerning the skin structure to bettermodel its reflective properties in order to achieve more realisticrendering of human faces or in identifying individuals [1].

The skin surface at a micro-scale is not simply flat but able to becharacterized by a particular relief format, representing theevolving keratinization progress of the epidermis and general3D organization of dermis and subcutaneous tissue [2]. Theperiodical cellular updating and frequent mechanical interactionwith the external environment give skin a net-like topographicprofile composed of regular triangles, polygons and line patterns,which may later evolve progressively into large scale wrinkles. The

* Corresponding author. Tel.: +44 1173282069.

E-mail address: jiuai2.sun@uwe.ac.uk (J. Sun).

0166-3615/$ – see front matter � 2013 Elsevier B.V. All rights reserved.

http://dx.doi.org/10.1016/j.compind.2013.06.004

roughness and palpation parameters obtained from this pattern ofmicro-relief have been used for evaluating the effectiveness ofcosmetic products and in assessing the status of tumours beneaththe skin surface [3]. The skin subsurface is itself rich in variouspigments whose interaction with incident photons gives skin itscolourful impression. Among these are melanin and haemoglobin –the two main chromophores embedded within the epidermal anddermal layers respectively. These have their own selectiveabsorption and scatter characteristics at different regions of thelight spectrum. The optical characteristics, spatial density anddistribution of these chromophores jointly determine the exactskin reflectance/colour. The reflectance/colour is most accessibleand is frequently used for assessing relevant skin physiologicalphenomena, such as an erythema caused by blood perfusion ofburnt/grafted tissue, or a variegated pigment lesion produced by adeeper invasion of malignant melanin [4]. To objectively charac-terize the skin profile or its reflectance information, severalapproaches have already been developed.

Skin surface profiles are mainly obtained through mechanical oroptical techniques. Mechanical profilometry works by moving andrecording a stylus across a silicone polymer replica of the skinsurface and then reconstructing the surface profile from therecorded data of the vertical movements of the stylus. Althoughthis has been used in the field for several years, it should be notedthat the contact pressure from the weight of the stylus mayintroduce some deformation as the stylus moves across the softsilicon replica. It is also a time-consuming procedure to move thestylus in very small steps to achieve high resolution measurements[5] and of course there is the need to produce the silicon replica inthe first place. In contrast to mechanical techniques, the ‘stylus’

J. Sun, M. Smith / Computers in Industry 64 (2013) 1383–13891384

used in various optical methods take the form of spots, lines orpatterns of structured lights. For example, a time multiplexingoptical system frequently used in dermatology projects a series ofparallel white stripes with a constant phase shift onto the tissuesurface from which the 3D topography is calculated based on theobserved deflection of the stripes [6]. There is no direct contactpressure on the skin in the case of optical measurement techniquesand this is an important consideration that serves to make opticalmeasurement techniques more appealing. Moreover, in addition tothe skin profile, an optical approach offers the potentiality tocapture skin reflectance, another important dimension in charac-terizing the skin surface.

The reflectance/colour of skin is normally measured throughspectrometers or tristimulus colorimeters. Spectrometers capturethe skin spectral response within a number of spectral bands over arelatively small skin area, and the reflectance of the skin iscalculated from the collected abundant spectral response data. Thespectral bands are normally selected to be able to reflectconcentrations of melanin and haemoglobin, according to theirabsorption and scattering characteristics [7]. The approach canachieve some reproducibility in quantifying skin reflectance byconstraining the light spectrum and the working environment,though it is not a format offering a direct visual impression.Tristimulus colorimeters quantify skin colour through specifyingthe illumination and viewing standards which are able to representall possible perceivable colours in three values, i.e. the tristimulussystem based on the International Commission on Illumination(CIE) standard. This enables measurements obtained using atristimulus colorimeter to be more traceable and are usefullycomparable to that perceived by the human eye. It has also beenverified that readings from colorimeters are able to correlate withthose from spectrometers. However both spectrometers andtristimulus colorimeters have difficulties in quantifying skinreflectance over a relative large area (�5 cm2), since they needto assume the area inspected is flat and homogeneous [8].

Measuring skin profile and reflectance separately in the abovemethods may be problematic, especially for a relatively large area,where skin topography is neither flat nor of uniform reflection. Thisis because the appearance of skin is jointly determined by both theintrinsic variables of the skin surface profile and also itsreflectance. A reflection bias can easily be produced whenmeasuring curved body regions if the shape factor has not beentaken into account. Such bias could then be transferred into thecalculation of the real distribution of optical features forquantifying the physiological status of the skin [9]. In order toremove artefacts caused by surface curvature, an additional 3Dimaging modality is often used to recover an object’s shape forreliable spectral mapping measurements. For example, themultiplexing optical system mentioned above, along with otherstructured lighting approaches, has been employed to capture theskin profile for removing intensity artefacts [6,10,11].

To reconstruct the skin micro-relief using structured lightingapproaches requires high spatial resolution of any light stripepattern. Difficulties may be experienced in scanning those heavilypigmented lesion such as melasma and melanoma as the projectedstrips will tend to appear blended with the pigmentation of theskin. In order to solve these problems we have proposed aphotometric stereo (PMS) technique to recover the topographicshape of the skin tissue by using multiple images, illuminated fromdifferent directions, but viewed from one single location [12].Because most photons entering the skin and interacting with thechromophores will undergo hundreds of absorptions and scatter-ings before emitting from the skin, the spatial distribution of theradiation becomes isotropic. Also, the micro-facet structure of thestratum corneum makes the skin somewhat diffusive [13].Therefore a photometric stereo approach based on a diffusive

reflection model is appropriate for extraction of both the surfacereflectance and shape, and suffers less from any coloured texture atthe tissue surface.

This paper will extend our early work and aims to exploit thesynergy of a multimodal approach by combining a multispectralcapability with photometric stereo imaging technology for theanalysis of skin tissue reflection and geometrical dimensioningwithin the visible and NIR spectrum. As far as we are aware, only afew systems have tried to fuse these two separate forms ofinformation [14,15]. Superior to the existing systems, thecombined multispectral imaging and the PMS in our approach,share a single lighting path, therefore the geometrical and spectralreflection measurements are aligned exactly, without involvingany extra registration procedure. This means that the shape factoraffecting the calculation of spectral mapping can be easilyeliminated. Furthermore, we also later recombine the recoveredspectral and geometrical information together in order to enhancethe visibility of more subtle physiological phenomena.

2. Methodology

2.1. Simplified skin optical reflection model

Skin is commonly recognized as a thin structure withdistinctive multiple sub-layers, often simplified into epidermis,dermis and the basal collagen/fat layer [1,13]. When a beam ofwhite light is projected onto this thin structure, most of the lightwill enter the skin and interact with unique biological chromo-phores embedded within the layered anatomical structures. Morespecifically, the light firstly passing through the epidermis isabsorbed by melanin, which demonstrates a higher absorptivecoefficient at shorter wavelengths. The total amount of thisabsorption is determined by the range of the illuminationspectrum, together with the concentration and distribution ofmelanin. Then the light further penetrating into the dermisinteracts with the haemoglobin pigments and collagen fibres in theform of absorption and scatter. Finally, the basal collagen/fat layerunder the dermis acts as a highly diffusive surface, and isapproximately independent of wavelength.

The above simplification of the skin optical transportationmechanism allows the development of a three layer optical model,i.e. two uniform media layers over an ideal diffusive surface [16].The first and second layers are the epidermis and dermis, and thelowest surface is equivalent to a perfect diffusive fat layer,containing no chromophores. The direct reflection on theepidermis surface and the interface between the epidermis anddermis are negligible. Although light scatter does exist, especiallywithin the dermal layer, this optical model assumes absorption asthe single most significant manner of light transportation withinthe two layers. This simplified theory has been successfully appliedto explain and detect important physiological phenomena.

The simplified skin reflection Rskin can be represented as:

Rlskin ¼ Tl

e pid2Tl

derm2Rl

di f f (1)

where Tle pid and Tl

derm are the transmittance coefficients of thesimulated epidermis and dermis layers; the square operator meansthe lights pass through the layers two times; and Tl

e pid is thediffusive reflectance coefficient of the deepest fat layer.

The transmittance through transparent layer satisfies theLambert-Beer law, i.e.

Tle pid ¼ e�ml

m�le pid�cm (2)

Tlderm ¼ e�ml

h�lderm�ch (3)

J. Sun, M. Smith / Computers in Industry 64 (2013) 1383–1389 1385

where Tle pid and ml

h are the wavelength-dependent absorptivecoefficients of melanin and haemoglobin; lepid and lderm are thelength of light paths within epidermis and dermis; and cm and ch

are the concentrations of melanin and haemoglobin chromophoresrespectively.

Therefore, the general reflectance of the skin surface at thewavelength l is:

Rlskin ¼ ðTl

e pidÞ2ðTl

dermÞ2Rl

di f f ¼ e�2ðmlmle pidcmþml

hldermchÞRl

di f f (4)

Incorporating knowledge concerning the penetration depth ofthe incident light and the unique absorption patterns of melaninand haemoglobin, the above expression can be further simplified ata specific range of wavelengths. For example, the approximatedepths of light penetration on a fair Caucasian skin type at typicalwavelengths of 450 nm (blue), 500 nm (green) and 600 nm (red)are 150 mm, 230 mm and 550 mm respectively. However, thethickness of the epidermis and dermis are around 100 mm and200 mm. Therefore the blue light is just able to penetrate throughthe epidermal layer, while the green light nearly reaches thediffusive reflectance layer. Note that the absorbance of haemoglo-bin at a longer wavelength becomes weak, so that only the melanineffect need be considered in that case.

2.2. Multiple spectral photometric stereo

A Lambertian model can be used to approximate the reflectionproperties of human skin as the skin exhibits a predominatelydiffusive characteristic after the specular reflection is removed.Therefore the intensity of light reflected from the skin surface canbe expressed as:

Il ¼ jlRl

skinðl � nÞ ¼ jlRl

skin cosðuÞ (5)

where jlis a composite factor determined by the intensity and

spectral of the illumination together with the sensitivity of theoptical sensor, which can be calibrated for by using a standardwhite flat reflectance object [17]; Rl

skin is the skin reflectance,sometimes called as absolute spectral reflectance, which can betreated as a combination of transmittance and diffusive reflection[18]; l and n are the incident light direction and skin surfacenormal; and u is the angle between surface normal and lightdirection.

In order to recover skin surface orientation and reflectance, onepotential way is to take more than one image using the sameimaging system from the same observation point, but illuminatedby a single light source from a different direction lj (j = 1, 2, . . ., N,N � 3). Therefore a grouping of Eq. (5) can be found and rewrittenas:

Il ¼ jlRl

skinðL � nÞ (6)

where L is the illumination matrix, formed by arranging theindividual light directions as L = (l1, l2, . . ., lN)T. Similarly a vector ofimage intensity can be compiled as Ij = (I1, I2, . . ., IN)T. The unknownreflectance and surface normal can now be uniquely calculatedthrough a least-squares method, providing the calibrated illumi-nation matrix L is available. This is known as a traditionalphotometric stereo approach. Due to accessibility of colourphotography, an improved solution has been achieved throughaveraging the traditional PMS results across three channels [12].

Theoretically the surface normal as calculated using the threeseparate colour channels should be similar and its average shouldbe closer to a true value. Similarly the calculated reflectance shouldbe consistent and independent from the illumination spectrum[18]. However, the major light transportation within the skin is allwavelength dependent due to the variations of optical transmis-

sion characteristics and anatomical structure of the skin subsur-face. Therefore the characterization of the skin surface should betreated differently at each channel if a colour camera is used.

2.3. Chromophore separation with additional spectral response

As light reflected from the skin is dependent on the illuminationspectrum and optical characteristics of the chromophores withrespect to the spectrum, each image taken under each specificlighting condition will be different from each other, though thebiological and anatomical features underneath skin remain thesame. Given the availability of the spectral response at redundantchannels, some biological features, such as the concentration ofchromophores, may be extracted.

As the light attenuation at longer a wavelength is dominated bythe absorption of melanin, while the absorption caused by thehaemoglobin is negligible, expressions (4) and (5) can be furthersimplified into:

Iinf ¼ jinfe�2minf

m le pidcm Rinfdi f f cosðuÞ (7)

Ired ¼ jrede�2mred

m le pidcm Rreddi f f cosðuÞ (8)

Igreen ¼ jgreene�2ðmgreen

m le pidcmþmgreenh

ldermchÞRgreendi f f cosðuÞ (9)

Iblue ¼ jbluee�2ðmblue

m le pidcmþmblueh

ldermchÞRbluedi f f cosðuÞ (10)

The concentration of melanin cm is readily deduced by dividingEq. (7) by (8), providing the absorption coefficient and depth ofpenetration of melanin are prior known. It is also worth noting thatthe factor of surface shape is cancelled out automatically due tothis division operation. Therefore the melanin component is able tobe calculated from the response of two long wavelengths (i.e. redand NIR). This conclusion agrees with early work [21].

The concentration of haemoglobin ch can be recovered bysubstituting the calculated melanin concentration into Eq. (9). Theresponse at a blue wavelength has not been used because thepenetration capability of blue light in the dermal layer is uncertain.

2.4. Viewing blood vessels through the NIR spectrum

A new combination of spectral reflections at differentwavelengths may be used to provide a unique impression ofhuman skin, able to usefully reveal the subtle intensity anddistribution of interesting skin chromophore components. Inprevious related work, NIR images have been fused with visibleimages by professional photographers and remote surveyors inorder to identify particular plants and vegetable regions [19,20].The depth dependent absorptions of chromophores has also beenobtained from skin reflectance captured by a multispectraltransillumination imaging device [22]. Similarly, here we useboth recovered spectral and shape information, integrated toenhance interesting physiological conditions of the skin tissuesurface.

The NIR spectrum can penetrate deeper into skin than thevisible spectrum because the absorption and scatter effect withinthe optical window is weaker. Also the relative high density of theblood component within the blood vessels absorbs a greater degreeof the incident light. Therefore a high contrast image of the bloodvessels can be obtained by employing NIR illuminations. The850 nm NIR light used in our new imaging approach approximatesto those optimal wavelengths necessary for viewing the vascularstructure [23]. In contrast to others who have used a mercury arclamp for general spectral illumination and structured lighting forobtaining the 3D topography of the object surface [11], here four

J. Sun, M. Smith / Computers in Industry 64 (2013) 1383–13891386

pairs of visible and NIR LEDs are sequentially triggered to projectlight onto the skin surface in our new imaging approach (Fig. 1(a)).An enhanced colour blood vessel image is obtained by fusing thebase layer of the NIR image with that of a visible image, calculatedfrom a bilateral filter [24]. After suitable enhancement, even thosefine blood vessels with small size below the pigmented skin areacan be clearly visually identified.

3. Experiments

3.1. Instrumental setup

A multi-spectral camera (JAI’s AD-080GE) shown in Fig. 1(a) hasbeen used as the optical sensor for building the new imagingsystem. Fig. 1(b) demonstrates how the new system works: a prismbuilt in to the camera splits the incident light into both the visibleand NIR spectrum; two separated light sources are projected ontotwo CCD sensors with same resolution but different spectralresponse; reflection and geometrical properties are recovered fromthe PMS approach; finally, meaningful physiological phenomenaare revealed through different visualization and enhancementpipelines.

For implementing multiple spectral PMS, four pairs of whiteand NIR LEDs are selected as the lighting sources. A high-resolutionSchneider lens is used to collect images using both visible and NIRspectrum lights. The lighting directions relative to optical centreare estimated by using a specular sphere with known size [25]. Inorder to eliminate the composite spectral effect of the illuminationand optical sensor, an Edmund white balance target having 99%uniform reflectance from ultraviolet to NIR spectrum is used tosample the light field. The combination effect of lighting and

Fig. 1. Setup (a) and schematics (b)

optical sensor can be compensated for when the objects to berecovered are approximately located around the calibrated space[17].

3.2. Reflectance recovery and 3D reconstruction

Previous work has taken the average value of the surfacenormal recovered across three colour channels as the final solutionand then used it to calculate the reflectance at every channel. Thisexperiment investigated the individual channels separately andfinds their contribution to the final result. Fig. 2(a) is one image of avolunteer’s wrist taken using the new system and Fig. 2(b) is amagnified view of the small skin region outlined by the black solidlines in Fig. 2(a). Four pairs of colour and NIR images are taken forthe recovery of the reflectance and surface normal data from theskin surface using the PMS technique.

The recovered reflectance in Fig. 2(c, e, g, i) demonstrates anoticeable variation across the visible and NIR spectrum. Thereflectance maps recovered in the blue and NIR channels look darkerthan that in green and red channels. This could be caused by thestrong absorption of melanin at the shorter wavelength within theepidermis and light scattering effect at a longer wavelength withinthe dermis. In addition, the NIR reflectance exhibits a morehomogenous distribution, which may reflect the uniform diffusionat the collagen/fat layer. The 3D reconstruction results in Fig. 2(d, f, h,j) show that the green and red channels offer good recovery of finerskin line net structures, while the NIR channel can only provide thegeneral smooth shape of the overall skin surface. Therefore, we cansee that selection of the green and red spectrum becomes significantfor imaging conditions such as wrinkle and psoriasis assessment,where the fine skin structure is of particular interest.

of the new imaging modalities.

Fig. 2. A volunteer’s wrist (a) with small skin region (b) outlined, its reflectance (c, e, g, i) and 3D surfaces (d, f, h, j) recovered at blue, green, red and NIR channels. (For

interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)

J. Sun, M. Smith / Computers in Industry 64 (2013) 1383–1389 1387

3.3. Chromophore mappings

Melanin and haemoglobin embedded within skin surface arethe two main chromophores reflecting important skin physiologi-cal phenomena, such as skin tanning, expression, alcohol abuse,physical abuse and skin cancers. An objective extraction of the twopigmentation components from photographs across a large area isalways desired, as compared to conventional spectrometerapproaches working only over a limited and small skin area.

Fig. 3. A photo (courtesy of Prof. Sabine Susstrunk, EPFL) in the visible (a) and its cou

Fig. 3 shows the implementation of the approach described inSection 2.3 on photographs captured with the same optical sensorby a different research group [24]. The freckles distributed aroundthe man’s forehead and the region under his eyes can be seen moreclearly by looking at the separation result in Fig. 3(c). Similarly, theconcentration of haemoglobin can be found to be denser aroundlips and cheeks from Fig. 3(d). This shows the fusion of the NIR andvisible images captured from using a multiple spectral imagingmodality can be used to quantitatively isolate aspects of the skin

nterpart NIR spectrum (b), separated melanin (c) and haemoglobin mappings (d).

Fig. 4. A fist under visible (a) light and its counterpart NIR spectrum (b), enhanced blood vessel detail (c) and its overlay on the 3D reconstruction result from the PMS (d).

J. Sun, M. Smith / Computers in Industry 64 (2013) 1383–13891388

condition (melanin and haemoglobin mappings) that are under-neath the skin tissue surface and so normally subtle and withoutany complicated human intervention.

3.4. Enhancement and visualization of blood vessels

Blood vessels within the subcutaneous layer are only poorlyvisible under white light as most of the incident light is normallyattenuated by the absorption of chromophores and dermalscattering before reaching the subcutaneous tissues, includingblood vessels. However, the employment of NIR light gives our newimaging modality an additional sensitivity, able to revealsubcutaneous anatomic structures deeper within the skin.

Fig. 4(a) and (b) are images of a volunteer’s fist acquired usingthe new imaging system within the visible and NIR spectrum. It canbe seen that the smaller fine blood vessels are hidden and almostinvisible under the white lighting condition, while the tree-likestructure of the vascular structure is clearly revealed in the colourenhancement format of Fig. 4(c) Finally, the enhanced vesselstructure is overlaid onto the 3D surface reconstructed from thePMS approach in Fig. 4(d) to give an enhanced augmented view.

4. Conclusions

Due to the complexity of the various light interactionmechanisms, partial translucency and complex sub-surfaceanatomical microstructures, human skin tissue is significantlymore difficult to characterize than most artificial object materials,whose surfaces normally consists of an established composition

with well-known optical properties. This paper describes anapproach that exploits the synergy of a multimodal approach,combining multispectral and photometric stereo imaging techni-ques to simultaneously capture both a spectral and geometricaldescription of the skin. The extracted geometric and spectralinformation is later synthetically recombined to usefully enhancethe visibility of particular physiological conditions associated withhuman skin. Initial experimental results demonstrate that thesenew imaging modalities are able to reconstruct skin tissuetopography registered with various enhanced spectral informa-tion. This demonstrates the potentiality of the new approach forapplications in the biomedical, forensic and computer graphicsfields, such as documenting and assessing the tissue surface,assisting in tissue feature catheterization and guiding endoscopicinterventions requiring enhanced or augmented visibility of finesurface and subsurface features.

Although the approach has been verified as effective inrecovering the geometry and spectral reflection of the skin tissuesurface over a relative large area, and enhancing the visibility ofchromophore maps by fusing the recovered information together,further work is still required to quantitatively evaluate itssensitivity to melanin, haemoglobin and other important chro-mophores.

References

[1] T. Igarashi, K. Nishino, S.K. Nayar, The appearance of human skin, TechnicalReport, Dept. of Computer Science, Columbia University, 2005 CUCS-024-05.

[2] J.L. Leveque, EEMCO guidance for the assessment of skin topography, Journal ofthe European Academy of Dermatology and Venereology 12 (1999) 103–114.

J. Sun, M. Smith / Computers in Industry 64 (2013) 1383–1389 1389

[3] M. Egawa, M.H. Oguri, M. Takahashi, M. Miyakawa, The evaluation of skin frictionusing a frictional feel analyzer, Skin Research and Technology 8 (2002) 41–51.

[4] Z. Liu, J. Sun, L. Smith, M. Smith, R. Warr, Distribution quantification on dermo-scopy images for computer-assisted diagnosis of cutaneous melanomas, Medical& Biological Engineering & Computing 50 (5) (2012) 503–513.

[5] K. Articus, G. Khazaka, K.P. Wilhelm, The Skin Visiometer – a photometric devicefor the measurement of skin roughness, in: K.P. Wilhelm, P. Elsner, E. Berardesca,H.I. Maibach (Eds.), Bioengineering of the Skin: Skin Surface Imaging and Analysis,CRC Press, Boca Raton, 1997, pp. 59–71.

[6] S. Jaspers, Rapid in vivo measurement of the topography of human skin by activeimage triangulation using a digital micro-mirror device, Skin Research andTechnology 5 (3) (1999) 195–207.

[7] G.N. Stamatas, B.Z. Zmudzka, N. Kolias, J.Z. Beer, Non-invasive measurements ofskin pigmentation in situ, Pigment Cell Research 17 (6) (2004) 618–626.

[8] H. Takiwaki, L. Overgaard, J. Serup, Comparison of narrowband reflectance spec-tro-photometric and tristimulus colorimetric measurement of skin color – 23anatomical sites evaluated by the DermaSpectrometer and the Chroma Meter CR-200, Skin Pharmacology 7 (1994) 217–225.

[9] J.M. Kainerstorfer, et al., Direct curvature correction for noncontact imagingmodalities applied to multispectral imaging, Journal of Biomedical Optics 15(4) (2010) 046013.

[10] M. Westhauser, et al., Optimizing color reproduction of a topometric measure-ment system for medical applications, Medical Engineering & Physics 30 (8)(2008) 1065–1070.

[11] V.C. Paquit, K.W. Tobin, J.R. Price, F. Meriaudeau, 3D and multispectral imaging forsubcutaneous vein detection, Optics Express 17 (2009) 11360–11365.

[12] J. Sun, M. Smith, L. Smith, L. Coutts, R. Dabis, C. Harland, J. Bamber, Reflectance ofhuman skin using color photometric stereo: with particular application topigmented lesion analysis, Skin Research and Technology 14 (2) (2008)173–179.

[13] A. Krishnaswamy, G.V.G. Baranoski, A study on skin optics. Technical Report CS-2004-01, School of Computer Science, University of Waterloo, Canada, 2004.

[14] G.N. Stamatas, et al., In vivo monitoring of cutaneous edema using spectralimaging in the visible and near infrared, Journal of Investigative Dermatology126 (8) (2006) 1753–1760.

[15] J.M. Kainerstorfer, P. Smith, A.H. Gandjbakhche, Non-contact wide-field multi-spectral imaging for tissue characterization, IEEE Journal of Selected Topics inQuantum Electronics 18 (4) (2012) 1343–1354.

[16] B. Dawson, J. Barker, J. Ellis, E. Grassam, A. Cotterill, W. Fisher, W. Feather, Atheoretical and experimental study of light absorption and scattering by in vivoskin, Physics in Medicine & Biology 25 (1980) 695–709.

[17] J. Sun, M.L. Smith, L.N. Smith, A. Farooq, Compensation of illumination radiance inphotometric stereo, in: IEEE International Conference on Machine Vision, Decem-ber, Hong Kong, China, 2010.

[18] N. Tsumura, Y. Miyake, F.H. Imai, Medical vision: measurement of skin abso-lute spectral-reflectance-image and the application to component analysis,in: Third International Conference on Multispectral Colour Science, 2001,25–28.

[19] M. Vilaseca, J. Pujol, M. Arjona, F.M. Martinez-Verdu, Color visualization systemfor near-infrared multispectral images, Journal of Imaging Science and Technol-ogy (49 (3),2005) 246–255.

[20] D. Lu, E. Moran, M. Batistella, Linear mixture model applied to Amazonianvegetation classification, Remote Sensing of Environment (87 (4),2003) 456–469.

[21] H. Takiwaki, et al., Quantification of erythema and pigmentation using a video-microscope and computer, British Journal of Dermatology 131 (1984) 85–92.

[22] B.D. Alessandro, A.P. Dhawan, 3D volume reconstruction of skin lesions formelanin and blood volume estimation and lesion severity analysis, IEEE Transac-tions on Medical Imaging 31 (11) (2012) 2083–2092.

[23] V.P. Zharov, S. Ferguson, J.F. Eidt, P.C. Howard, L.M. Fink, M. Waner, Infraredimaging of subcutaneous veins, Lasers in Surgery and Medicine 34 (2004) 56–61.

[24] C. Fredembach, N. Barbuscia, S. Susstrunk, Combining visible and near-infraredimages for realistic skin smoothing, in: IS&T/SID 17th Color Imaging, 2009.

[25] N. Tsumura, M.N. Dang, Y. Miyake, Estimating the directions to light sources usingimages of eye for reconstructing 3D human face, in: IS&T/SID’s 11th ColorImaging, 2003.

Dr Jiuai Sun is a lecturer working at the Centre forMachine Vision with the University of the West ofEngland, Bristol. Currently his research focuses onobject surface reconstruction applicable to healthcareand general industry.

Melvyn L. Smith is professor of machine vision andDirector of the Centre for Machine Vision (CMV) atUWE. He received his B.Eng. (Hons) degree in mechani-cal engineering from the University of Bath in 1987,M.Sc. in robotics and advanced manufacturing systemsfrom the Cranfield Institute of Technology in 1988 andPh.D. from the UWE in 1997. He acts as associate editorfor four leading international journals, includingcomputers in industry, for which he is currently guestediting a special issue on 3D imaging. He has publisheda book on computer vision for surface inspectiontogether with numerous book chapters, patents andjournal/conference papers in connection with his work.He has been a member of the EPSRC Peer Review

College since 2003 and is currently a programme and evaluator/review/monitoringexpert for the EU Framework 7 Programme. Professor Smith is a chartered engineerand an active member of the IET.