Segmentation of statistical texture images using the metric space theory

Abstract

This paper proposes a new segmentation algorithm for statistical texture using the metric space concept. The coordinate space, region space and texture image space are modeled in terms of the metric space, a special form of the topology space. A texture segmentation technique based on the metric space modeling is proposed and its multiresolution extension is also presented. To show the effectiveness of the proposed algorithm, its segmentation results for synthesized statistical texture images and a real aerial image are compared with those of the conventional methods such as the Gauss-Markov random field (GMRF) method and the spatial gray level difference method (SGLDM). Computer simulation with several statistical texture images shows that the proposed algorithm gives better performance than the conventional ones, based on the subjective evaluation and the quantitative measure. Also its multiresolution extension is observed to be effective.