Research On Image Segmentation Based On Graph Cuts

Image segmentation is the foundation of object recognition and image interpretation. Generally, image segmentation is proceeded by partitioning it into different parts according to some criteria, such as color, intensity and so on. However, image information is so complicated that we may get very different segmentation result for the same image just because we have used different features as segmentation criteria. So we can get a conclusion that image segmentation is a complicated problem.As an optimization algorithm to minimize the energy function, Graph Cuts has been widely used in the area of computer vision. For its good characteristics, the image segmentation algorithm based on Graph Cuts has been a hot research topic. When the user specifies the foreground seeds and background seeds, Graph Cuts will automatically segment the image. Graph Cuts algorithm is a kind of global optimization segmentation method. In the process of image segmentation, super pixel will be applied to the following process. Super pixel can divide the image into different parts by using color, texture and other similarities of image information,which can decrease the complexity of the algorithm.The research contents are summarized as follows:(1) This paper has researched the basic theory and knowledge of Graph Cuts algorithm to segment the image. The correspondence between graph and image, the edge weights setting, the structure of energy function and the basic framework of segmentation based on Graph Cuts are all deep researched.(2) The segmentation algorithm of Graph Cuts has the shortages of relying on the selection of seed points and the model establishment. Interactive correction is also essential to get a good result. A kind of image automatic segmentation algorithm based on Graph Cuts and Kmeans clustering is proposed. First, the method initializes image to build an image map. It also segments the image to several regions using Kmeans clustering method. Then replace pixels by the regions. Second, we set the weights for each side to constructs the energy function. Finally, we solve the optimal energy function to get the optimal image segmentation. The advantage of the improved method in this paper is user can obtain the final segmentation result without the guidance of user’s interactive operation.Experiment results show that the proposed algorithm can automatically get a prior knowledge of the foreground and background without human interaction and increase the accuracy of segmentation. The segmentation results of grayscale images and color images show that the improved method is more effective and accurate than the former.