АНАЛІЗ ШЛЯХІВ ВИРІШЕННЯ ПРОБЛЕМИ СЕГМЕНТАЦІЇ ВИСОКОТЕКСТУРОВАНИХ ОБ’ЄКТІВ

Abstract

Increment of speed and reliability of aerospace images processing is directly related to solution of the task of automation of images interpretation process, which is achieved by minimizing search areas, detecting masked objects and defining the dynamics of changes in surveillance areas. The primary stage that in general determines the quality of results received by automated processing and interpretation is thematic segmentation of the image. In the process of thematic segmentation it is necessary to take into account presence of a large number of textured objects. The paper analyzes the ways of solving the segmentation problem for highly textured objects with large range of variation of possible color values. The research included separation of woodlands and single plants from meadows, steppes, etc., which are characterized by similar color characteristics, but differ in texture. It also included separation of residential areas from forests, which are characterized by the same grain size of texture and different color characteristics. The method of texture description, which is based on calculation of the number of differences in brightness per unit area of the image, the method of description and measurement of texture, characterized by the length of the series, the methods of texture description based on calculation of their fractal dimension have been investigated. In order to describe the texture by different methods, first of all, an aperture of the analysis window was defined. That aperture ensures separation of different classes of objects. The analyzed methods of texture description showed that areas of false identification are always present in the result images. It was determined that the best results were obtained with two of the discussed methods. The first one was the method of texture description and measurement based on calculation of the number of brightness differences per unit area of the image. The second one was the method of texture description based on calculation of fractal dimension by searching the area of the pyramid which covers image fragments. To obtain a more accurate segmented map of an image containing highly textured fragments, a combination of the two methods is suggested.