A recognition problem of binary images by the methods of percolation theory

Recently the problem of recognition of bitmaps gained the special importance for automation of sorting railway cars and sea containers. The main direction of researches is the development of methods of extraction of a noise from the initial image. Note, the problems of improvement of the quality of recognition of images with distortions, caused by movement and change of an angle of eyesight, were less investigated. Especially enormous problems arise for images smudged along some direction. The modern specialized programs of recognition solve hardly such problems. Particular attention is oriented to the following approaches improving the binary image: neural network method, and cellular automata. In this paper we suggest considering a noisy bitmap as a set of percolation clusters. The mathematical theory of percolation provides a number of the results concerning the behavior of the function n (s) that is the density of the clusters number of a mass s. Existence of an image in the noise can change the behavior of this function. At that, if a priori we know what image may appear, the Bayesian approach allows to give the answer on the presence or lack of this object in the noisy image. A new approach to preprocess monochrome images, based on the analysis of distribution function of the weight of a cluster of the noisy image is proposed. Novelty of this approach consists of the decision on belonging of this cluster to the initial image is taken on the basis of calculation of a divergence between the graphs of distribution functions of weights of a cluster, and the decision has probabilistic character. Thus, the algorithm is not strictly determined and related to the family of the Monte-Carlo methods. Using computer experiments, we calculate the value n(s) for the noise of the different intensity. The main result of this work is the realization of the proposed method of improvement of images, based on the extremely specific way of application of the percolation theory. At that, the suggested method is rather unexpensive for its apparatus realization. The results of the simulations register the special efficiency of this algorithm at the processing smudged images because of the object movement. The principal possibility of its application for color images is shown.

Keywords

Authors

References