A general approach to the problem of analysis of images which are piecewise-constant with respect to some feature is described. Algorithms for solving this problem are presented. The algorithm proposed is illustrated by the following examples: the problem of noise suppression on images with piecewise-polynomial brightness; the problem of localization of an unknown object in an image with a given background image, which may be obtained under different lighting conditions; and the problem of random field segmentation. The results obtained by these algorithms on real images are not worse than those obtained by existing algorithms.
展开▼