We describe in this thesis three key steps of image processing algorithms. We first study the reflexion models which describe the image formation process. These models are used to obtain a segmentation of the image into materials and to reconstruct the surface of some of the regions previously segmented. The materials studded for the reconstruction stage are metallic ones. We also study quantization and inverse colormap operations. These operations are used to display an image onto low cost terminals. Such processes May also be applied into the image compression or image segmentation framework. We finally describe a new hierarchical model based on a topological representation of an image partition. The model named Combinatorial Pyramid is the only hierarchical model currently developed which allows to encode all the topological information.