OPEN ACCESS
Dermatologic lesions are monitored among several months or years by the expert without robust acquisition system, reproductible during time. Taking into account the dermatologist constraints, we search to produce a simple acquisition system with few using constraints. Unfortunately some colours variations appeared and must be corrected. In this work, we compare and analyze several basic colour corrections used in the literature, and we show that in our using case the colour normalization is a non linear transformation.
Extended Abstract
Dermatologic lesions are monitored among several months or years by the expert without robust acquisition system, reproductible during time. Taking into account the dermatologist constraints, we search to produce a simple acquisition system with few using constraints. Colour contrast measures and texture features are processed from the images, and consequently results are highly sensible to acquisition conditions changes. As shown in the proposed examples, unfortunately the colour variations can be important and must be corrected.
Classically, the colour correction is defined as a linear transformation searching to modify linearly the image colour set into another ones in relation to a reference image. To simplify this process, a colour chart is used for each image acquisition. The normalisation parameters are processed using the same colour from the colour chart in the reference and the image to correct. Then all the colour of the image to correct are modify using the linear transform and the processed parameters. Three linear normalisation are compared. The first one is based on the hypothesis that the colour average is not changed, only the colour dynamic. Three colour coordinates in correspondance in the 2 images . The second one free this hypothesis, allowing the colour average and the colour dynamic corrections. The third one allows the same corrections, but uses an optimisation scheme with more than 4 colour rather than a direct processing.
To compare results, a perceptual error criterion is defined based on the sum of the perceptualdistance (standard CIE ∆E) between the colour reference and the normalized corresponding colours. Normalized images are corrected and useable for comparison by the dermatologist. Nevertheless the perceptual criterion shows that the hypothesis of linearity for the colour changes is not respected, even with the third approach when a great number of colour are used for the normalization. This result prove that the linear normalization is not adapted in our case and that a more constraint acquisition system must be developed. In addition another normalization scheme using physical constraint must be defined to obtain a sufficient level of accuracy.
RÉSUMÉ
Les lésions dermatologiques sont suivies sur plusieurs mois ou années par le dermatologue, sans système d’acquisition d’images robuste et reproductible dans le temps. Compte tenu des contraintes d’usage du dermatologue, nous voulons construire un système d’acquisition faiblement contraint. Malheureusement, des variations de couleurs apparaissent et doivent être corrigées. Dans ce travail, nous comparons et analysons quelques modèles basiques de correction utilisés dans la littérature et nous montrons que la normalisation couleur est une transformation non linéaire dans des cadres d’acquisition insuffisamment contraints.
colour, normalization, colour distance, perception.
MOTS-CLÉS
couleur, normalisation, distance couleur, perception.
Barnard K., Gabbur P. (2003). Color and color constancy in a translation model for object recognition. In 11th color imaging conference, p. 364–369.
Bianco S., Schettini R. (2012, June). Color constancy using faces. In Ieee conference on computer vision and pattern recognition (cvpr), p. 65-72.
Buchsbaum G. (1980). A spatial processor model for object colour perception. Journal of the Franklin Institute, vol. 310, no 1, p. 1–26.
CIE. (2008). Methodsforderivingcolourdifferencesinimages.Rapporttechniqueno18x:2008. Auteur.
Finlayson G., Schiele B., Crowley J. (1998). Comprehensive colour image normalization. In 5th european conference on computer vision, p. 475–490.
FuntB.,ShiL. (2010). Theeffectofexposureonmaxrgbcolorconstancy. InProc.spiehuman vision and electronic imaging xv, vol. 7527.
Gijsenij A., Gevers T., Weijer J. van de. (2011). Computational color constancy: Survey and experiments. IEEE Transactions on Image Processing, vol. 20, no 9, p. 2475-2489.
Harville M., Baker H., Bhatti N., Süsstrunk S. (2005). Consistent image-based measurement andclassificationofskincolor. InIeeeinternationalconferenceonimageprocessing(icip), vol. 2, p. 374-377.
Horaud R., Monga O. (1995). Vision par ordinateur : outils fondamentaux. Hermès.
Iyatomi H., Celebu M., Schaefer G., Tanaka M. (2010). Automated color normalization for dermoscopy images. In Ieee international conference on image processing(icip), p. 4357– 4360.
Land E. (1977). The retinex theory of color vision. Scientific American, vol. 237, no 6, p. 108– 128.
Ledoux A., Richard N., Capelle-Laizé A. (2011). Limitations et comparaisons d’ordonnancement utilisant des distances couleur. Traitement et Analyse de l’Information, Méthodes et Applications (TAIMA), vol. 2, p. 325–330.
Ledoux A., Richard N., Capelle-Laizé A. (2012, May). The fractal estimator : A validation criterion for the colour mathematical morphology. In Proceeedings of the 6th european conference on colour in graphics, p. 206-210. Amsterdam, IS&T.
Lin Z., Wang J., Ma K.-K. (2002). Using eigencolor normalization for illumination-invariant color object recognition. Pattern Recognition, vol. 35, no 11, p. 2629 - 2642.
Marguier J., Bhatti N., Baker H., Harville M., Süsstrunk S. (2007a). Assessing human skin colorfromuncalibratedimages. InternationalJournalofImaging,SystemsandTechnology, special issue on Applied Color Image Processing, vol. 17, no 3, p. 143 – 151. Consulté sur http://www3.interscience.wiley.com/journal/116327590/abstract
Marguier J., Bhatti N., Baker H., Harville M., Süsstrunk S. (2007b). Color correction of uncalibrated images for the classification of human skin color. In Proceedings of the 15th IS&T/SID Color Imaging Conference, p. 331–335. Albuquerque, NM. Consulté sur http:// www.imaging.org/conferences/cic15/
Mezzana P., Anniboletti T., Curinga G., Onesti M. (2007). A new method of skin erythrosis evaluation in digital images. In Acta chirurgiae plasticae, vol. 49, p. 21–26.
MooreE.H. (1920). Onthereciprocalofthegeneralalgebraicmatrix. BulletinoftheAmerican Mathematical Society, vol. 26, p. 394–395.
Penrose R. (1955). A generalized inverse for matrices. Mathematical Proceedings of the CambridgePhilosophicalSociety,vol.51,no 03,p.406–413. Consultésurhttp://dx.doi.org/ 10.1017/S0305004100030401
WeijerJ.V.D.,GeversT.,GijsenijA. (2007). Edge-BasedColorConstancy. IEEETransactions on Image Processing, vol. 16, no 9, p. 2207–2214.