EVALUATION OF THE QUALITY OF THE EVOLUTION METHOD OF SEGMENTATION OF THE OPTICAL-ELECTRONIC OBSERVATION SYSTEMS IMAGE

Abstract

It is established that the use of known methods of image segmentation obtained from on-board optical-electronic surveillance systems does not ensure the resistance of segmentation methods to variations in various image parameters. Theoretically, the possibility of using the roving method of optical-electronic image segmentation was substantiated, and the objective function was determined when solving the problem of clustering a multispectral image. Segmentation by the evolutionary method of the optical-electronic surveillance systems image. To evaluate the effectiveness of the evolutionary segmentation method, the method of isolating Canny contours is chosen as a comparative method. Analyzed are the known indicators for assessing the quality of segmentation of optical-electronic images. As an indicator of the segmentation estimation by the evolutionary method and the Canny method, the information indicator Kulbak-Leibner distance is chosen. The dependence of Kulbak-Leibner's information divergence (gain) on the change in image scale for two methods of segmentation is given. It is established that segmentation by the evolutionary method provides a gain in the value of the information indicator by 7-16%.