A Segmentation Algorithm Combined with Non-local Information and Graph Cut

The interactive image segmentation algorithm based on graph cut segments a foreground object from its background, which has drawn great interest in image processing and computer vision. In order to further improve the accuracy of the segmentation, this paper presents an interactive segmentation algorithm by introducing the non-local information into graph cut framework. To model the non-local information of the image, we set a fixed-size search window for each pixel, and hence each pixel only needs to consider the relationships with the pixels in the search window. Instead of using intensity of each pixel, the image patches are utilized to compute the similarity between pixels when considering the similarities between non-local pixels. By introducing the non-local information into the graph cut framework, a new energy term combining the local and non-local information is constructed by merging the local and non-local information in the boundary term of the conventional energy function. A new set of non-local edges is added in the graph to represent the non-local information of the image. The segmentation results can be obtained by solving the mincut/maxflow algorithm. Finally, the experimental results demonstrate the effectiveness and feasibility of the proposed algorithm.