Research On Image Segmentation Methods Based On Graph Theory

Human's information requirement is huge in today's information society. Vision is the main way to get information, and it's also the most advanced sense of human. As the main way to use information, image plays an important role in human sense due to its advantage. Image segmentation is a classical and important research in image processing, the result of which would have decisive influence on the subsequent image analysis and image understanding. Although much progress has been made after years, domestic and international scholars are still exploring new theory and methods of image segmentation to meet the requirements of actual engineering application.This paper focuses on the research of image segmentation based on graph theory, which is a branch of mathematics. Graph is a basic research object of graph theory, and it is made up of vertices and edges. The keystone of the research on graph theory is the relations between vertices, edges, and their own properties. The main work is as follows: Chapter one introduces the significance of the research on image segmentation in the field of computer vision, outlines the common image segmentation method. In chapter two, a typical algorithm of Graph Cut, the Swendsen-Wang cuts(SWC) algorithm is studied. Based on SWC algorithm, chapter three realizes an image segmentation method under the framework of random forest random field(RFRF). This method uses SWC to perform inference of conditional random field(CRF), and meanwhile trains random forest(RF) to compute the two ratios of probability distributions needed by the inference of CRF. Compared with other traditional CRF based segmentation, this method reduces the computational complexity. To improve the shortcoming of RF, chapter four use hough forest(HF) instead of RF, and realizes an image segmentation method based on hough forest random field. Experiment results show the effectiveness of the image segmentation methods based on graph theory discussed in this paper.