Texture-based Color Image Segmentation Using Local Contrast Information

A novel texture-based segmentation algorithm for color images is proposed in this paper. Many powerful color image segmentation algorithms such as J-Segmentation (JSEG)suffer from over segmentation. Analysis of undesirable results from JSEG segmentation has led to the development of an improved JSEG method called improved contrast JSEG, or IC-JSEG. The proposed method first constructs a contrast map to obtain the basic contours of the homogeneous regions in the image. Filtering is applied to remove noise and enhance the strength of edges found by a noise-protected edge detector. A seed growing-merging method uses both the improved contrast map and the original J-map constructed by JSEG to segment the image. Experiments on natural color-texture images and color medical images show improved results.