A Texture and Color Based Method for Color Image Segmentation

A new method of fully automatic color image segmentation is proposed in this paper. First, colors in the image are quantized to several representing classes in the HSI color space. And then the uniform surface uncertainty of every pixel is computed which will be used to generate the fuzzy image. Then seeds are detected and region growing is implemented to form the initial regions. A region merging algorithm base on the color information is used to merge the over-segmented regions. Utilized in natural image and texture image processing, the new method has got a good performance.