The main objective of this PhD thesis is to develop a score that measures the skin pigmentation using spectral images. The ultimate goal is to build a more objective and at least as powerful as clinical methods for evaluation of treatment effects acting on the skin hyper-pigmentation. This tool is intended to be used in clinical trials. The work focuses on melasma that is a disease mainly due to hormonal disorders and sunlight exposure. To assess the severity of this disease and its evolution under treatment, we proposed two types of classification. The first one is a binary classification between healthy tissue and pathological tissue. The second one consists in defining different levels of severity for pathological tissues. The first classification concerns high dimensional spaces. An algorithm for dimensionality reduction associated with a classification by support vector machines has been developed. This method comes with a comparison of projection pursuit and source separation, as well as automated methods to estimate the dimension of the arrival space, and the estimation of different groups of spectral bands in the case of projection pursuit. The second classification criterion aims at qualifying a clinical severity criterion of hyperpigmentation. This clinical criterion includes three components: area, contrast and homogeneity. The surface component arises from the classification between healthy and pathological tissues. A methodology for estimating combination of spectral bands taking into account the spectral information and the kinetics of the treatment effect on a clinical study is proposed to obtain a contrast criterion. To get a spatial homogeneity criterion, an approach based on multiscale analysis of Gaussian fields adapted from the methodology of statistical parametric mapping is used between two acquisition dates.
