Iterated Graph Cut Integrating Texture Characterization for Interactive Image Segmentation

Graph cuts based interactive segmentation has drawn a lot of attention in recent years. In original graph cuts, the extraction of foreground object from its background often leads to many mistakes and the histogram distribution for energy function is not enough. In this paper, an iterated graph cut algorithm integrating texture characterization is proposed. We utilize user intervention to cycle the object approximately in the beginning, and the image is divided into superpixels by "SLIC" method. After initialization Gaussian mixture model (GMM) by RGB colors, we use a vector which combines color model and texture description for the estimation of GMM parameters. Then min-cut algorithm is applied in the graph for energy minimization, so GMM adjust their clusters and recompute the parameters. The process iterates until min-cut algorithm converges. Finally, we give a comparison between our method and "GrabCut". The experiments show that our have good results.