Visual evaluation of cutaneous lesions is the analysis the most commonly performedby dermatologists. This diagnostic is mainly done by naked eye and is based on criterionsuch as the size, shape, symmetry but principally on colour of the lesions. However, thisanalysis is subjective because it depends on the practician experience and the acquisitionconditions. We propose in this dissertation (1) the development of a multispectralcamera specifically dedicated for dermatological use. This device is based on a filterwheel composed of interferential filters and a neural network-based algorithm, generatinga hyperspectral cube of cutaneous data. This setting combines advantage of both spectrophotometer(spectral information) and digital camera (spatial information). Its maininterest is also to provide reproducible information which is independent of the acquisitionconditions. The setting-up of an acquisition protocol of healthy skin data from five of thesix exisiting skin phototypes allows the validation of our system by comparing spectragenerated by our system and theoretical spectra acquired by professional spectrophotometer.(2) Skin spectral reflectance provides precious information because it is directly linkedto the skin chromophore composition. Quantitative measure of cutaneous tissue opticalproperties can be based on the modelisation of light propagation in skin. For this purpose,we based our method on Kubelka-Munk model with which we associated an optimizationmethod based on evolutionary algorithm. This method helps for the model inversion.Using this approach, quantification of diverse parameters of skin can be obtained such asmelanin and haemoglobin. (3) The validation of this model is performed on disease skindata (vitiligo and melasma) and allows to quantify difference between healthy and affectedskin area within a single image.
