An improved graph-based image segmentation algorithm

This paper proposes an efficient representation framework for image segmentation based on perceptual organization. Specifically, a new hierarchy of image representation with three levels is presented. In a bottom-up way, the framework transforms representation between the three levels, and takes the advantages of different organization techniques. A region-based organization approach is developed in the framework for image segmentation. The approach improves previous organization techniques, and makes image segmentation results globally optimum and greedily efficient. The experimental results have proved the effectiveness and efficiency of the image representation framework.