Research Of Stereo Matching Disparity Estimation Using Globle Algorithm

As the development of image processing,communication and video encoding,the research of stereo video comes into a new rapid period.Stereo matching is to find the relationship in different perspective for one scene.According to the relationship we can get the disparity map,to rebuild a three-dimensional model of the scene.Stereo matching technology completely simulate the visual mechanism of human being,which relies the difference from different angle,to get the disparity information of different objects.On the other hand,it is also beneficial for humans to understand visual mechanisms of human being.Recalling the recent decades of technology development of stereo matching,the early researchers attempted to find the matching one discrete point by point,tried to find the best match for each pixel,but the result is not satisfactory.In recent years,with the development of Markov random field theory,which provide better research tools for computer vision and image processing.This article mainly carry from a global perspective based on Markov random field theory of matching algorithms to expand research and discussion,includes dynamic programming algorithm and the belief propagation algorithm.Both two algorithms are optimized in order to obtain better effect and faster speed of the algorithm.In this paper,the dynamic programming algorithm in stereo matching field is analyzed.We pointedly optimized the algorithm in pre-and post-processing steps according to the characteristics of dynamic programming.The algorithm uses the left and right images as the front view respectively,and then iterating the matching cost to calculate the cost for the corresponding disparity map.After that,the interpolation optimal is used to correct the mismatching point,and then,the accurate disparity map is obtained.In calculating the matching cost,both pixel similarity and texture similarity are considered.Dynamic programming is used in the iteration step,both real-time and accuracy.After that,because of the accuracy of the dynamic programming algorithm and the stripes feature the same scanning line right,the interpolation optimization algorithms is used to extract the precision region from the two disparity map area to synthesis the final disparity map.Experimental results show that,compared with conventional dynamic programming algorithm based on interpolation optimized dynamic programming stereo matching algorithm has an outstanding performance in the blocked area,and still retain real-time.In this paper,we analyze the theoretical basis of belief propagation algorithm,and describe the related theories and models.A large number of redundant iterations caused high time complexity and space complexity are pyramid optimized,effectively reducing the redundancy algorithm occupy space and accelerate the computing time and did not affect the algorithm results.According to the characteristic of belief propagation algorithm,a fast image sequence disparity map estimation algorithm is raised.Because the algorithm is usually gradual approach from zero disparity map and sequence the images themselves have almost the same background,it is used directly in stereo matching sequence of images in a frame on the disparity map as an iterative current frame will start greatly accelerate the speed of iteration.Experimental results show that compared to single image belief propagation,sequence belief propagation has a much higher speed,while the effect has not declined at the same time algorithm.