Probabilistic Methods for Image Segmentation

Image segmentation is an important problem, with many applications in image processing and in computational vision. In abstract terms, it consists in partitioning the pixel space of an image into a set of regions that are homogeneous, in the sense that the variation of a given attribute may be represented by a relatively simple model. In this talk, this problem will be formulated in probabilistic terms, and the solution will be specified in terms of Bayesian Estimation Theory. The derivation of efficient estimation algorithms will be discussed, and some applications to several problems, such as medical image processing, 3-D digitalization and interactive segmentation will be presented.