Стационарные процессы на графах и анализ изображений

Abstract

The method of a classification of images concerning to a substance propagation process is proposed. The image is considered as a lattice formed by pixels of given intensity. Then an oriented graph corresponding to the image is constructed in the following way: every vertex (pixel) is connected with N neighbours. For a given vertex all outcoming edges have a value (pixel intensity/N), for boundary vertex – pixel intensity/(N – 1). The constructed flow is normed. For obtained markov chain by the Sheleihovsky-Bregman method a stationary distribution is constructed, which maximizes weighted entropy. It is weighted entropy that is used as a classifying sign when images with different doses of a substance are analyzed. The results of numerical experiments are given.