Color Image Segmentation Using Statistical Features and Dempster-shafer Evidence Theory

ABSTRACT In this paper, we propose a new color image segmentation method based on a possibilistic clustering algorithm and data fusion techniques. The general idea of mass functions estimation in the Dempster-Shafer evidence theory is to link, at the image pixel level, the notion of membership in fuzzy logic. For segmentation, we proceed in two steps. In the first step, we begin by identifying the most significant attribute images of the tristimuli (R, G and B) and automatically determining the mass functions. In the second step, the evidence theory is employed to merge several attribute images which characterized the three component images, in order to get a final reliable and accurate segmentation result. The mass functions assigned to each pixel covered the images to be combined, is obtained from the membership degree of the current pixel and those of its neighboring pixels. The membership degree of each pixel is determined by applying the possibilistic clustering approach to the representative attribute images, and the final segmentation is achieved, on an input image, characterized by different attribute images, by using the DS combination rule and decision. Experimental segmentation results on medical and textured color images demonstrate the value of introducing the fuzzy clustering combined with the statistical features in the evidence theory for color image segmentation. The obtained results show the robustness of the proposed method.