A.I. Sherstobitov, V.I. Marchuk, D.V. Timofeev, R.A. Sizyakin
In this paper we study methods for classifying objects in textured images. The nonlinear model of the mathematical component interaction processed signal. Introduced a nonlinear objective function of processing solutions for the problem of estimating the alpha channel, which separates the set of points from a variety of background object points by way its minimization for each pixel of the image. We study the method of classification algorithm set of points of the object and the background of the image by determining the alpha channel in a canonical form. Consider methods of classification based on Bayesian estimation method, gradient method, robust and spectral method. Are the advantages and disadvantages of the methods considered classification. The examples of the solution of the problem classification of the object on a color image and its separation on a textured background by different method, are presented. A comparative analysis of the processing, obtained by the methods of classification of objects on a set of textured images. As a criterion for evaluation of the values the probability of correct detection and false alarm probability, standard deviation of the true contour of the object from the estimate. The numerical values are presented assess the effectiveness of edge detection of objects in the processing of textured digital images in the table.