An improved approach for image segmentation based on color and local homogeneity features

In this paper, we propose an improved approach for image segmentation based on color and local homogeneity features. A given image is transformed into a quantized image by a self-constructing fuzzy clustering. Then, a color-based region image and an initial seeded region image are obtained from the quantized image by color-based and homogeneity-based region growing methods, respectively. After that, we combine these two images to generate a refined seeded region image and obtain an initial segmented image by a region-based region growing. Finally, merging based on color similarities and sizes of regions is performed for avoiding the problem of over-segmentation. Compared with the other method, experimental results show that the segmented regions obtained by our approach are more reasonable and precise.