The Application Of Graph Cuts In Image Matching

In recent years, graph cut as a new method to solve energy function minimization is increasingly popular. It has an advantage in solving the image matching problem. This paper mainly studies image matching of the three-dimensional space, and the referenced images always have been rectified. The main contents are summarized as follows:1,Image matching methods outlined. Firstly, a brief introduction to the research background is showed. Secondly, some classical image matching algorithms are introduced, namely ABS algorithm, NC algorithm, image matching based image torque, feature points based ones; It introduces the appearance of graph cut, and describes the many applications of graph cut in image matching.2,The research status and applications of graph cut. It mainly introduces the development and application of graph cut, and points out public problems of graph cut. In recent years, graph cut as an optimized method which minimizes the energy function, because it can make the solutions of energy function converge to the globally optimal solutions and it can get stronger robustness of solutions. These make graph cut in computer vision have been widely applied.3,A research method of image matching based on graph cut. The image matching problem convert into the energy function minimization problem is the key of graph cut. However, energy function minimization problem is NP-hard problem. The paper finds a new energy function which uses parallax for the label. The energy function in the selection of the smooth items is different from the others. In this paper, graph cut method is used to minimize the energy function, and uses the new maximum flow algorithm to solve the minimum cut of graph structure network. Comparing the effect of image matching with the energy function based on Potts model, experimental results prove the energy function of this paper well balanced data items and smooth items. The effect of image matching make the signal-to-noise ratio and the correlation with real parallax image are slightly better than Potts model. The last results of the experiment are rendering pixels less and parallax contour clear.