The general issue of our work is the elaboration of region-based active contours models for image and video segmentation. We propose to segment regions or objects by minimizing a general functional including domains and boundaries integrals. In this framework, the functions characterizing regions or boundaries are named "descriptors''. The minimum is searched using the propagation of a region-based active contour. The associated evolution equation is computed using shape derivation tools. Besides, we take into account the case of region-dependent descriptors that evolve during the curve propagation. We show that this variation induces supplementary terms in the evolution equation.Region-based active contours models are then applied to various applications of segmentation. First, statistical descriptors based on the matrix covariance determinant are proposed for face segmentation. Statistical parameters estimation is performed jointly to the segmentation. Second, we propose statistical descriptors using a distance to a reference histogram. Endly, detection of moving objects in sequences acquired either by a static or a mobile camera is performed using a hierarchical association of motion-based descriptors and spatial ones.
