Texture and multifractals : new tools for image analysis
Résumé
We present some new ideas for image analysis and texture segmentation. The two important points are the introduction of the multifractal theory for image understanding in a rigorous manner, which allows to do all the computations directly on the discrete signal, and the definition of a precise framework for texture analysis, in which we do not choose any a priori model for the study but instead let the peculiar data in each case decide which model is the most appropriate. We define an approach general enough as to handle a great variety of textures and we propose a new scheme, based on learning and optimal cooperation, that gives good results on a lot of segmentation experiments presented here.