Research Of Stereo Matching Based On Coloured Image Segmentation

The stereo matching is the most important and difficult problem in the computervision, which is calculated the corresponding pixel in another image for each pixel inthe reference image, in other word is obtain the best disparity for each pixel. Stereomatching has been used widely in the field of three-dimensional reconstruction,virtual reality, visual navigation and industrial measurement, which has become a hotresearch focus in recent years. For the research of binocular testing equipment, wecommence the study to solve the widespread problem in the matching algorithm,which is about effectiveness and speed. The main contents are as follows:1. This paper summarizes the computer vision theory and the overseas and domesticresearch status, in addition, sum up the existing problems.2. For binocular stereo vision, this paper has a detailed description of relatedconcepts on matching algorithm, such as algorithm framework, methodsclassification, constraint condition and evaluation criterion. And focus onanalyzing the advantages and disadvantages of the existing matching algorithm,and described the current mainstream algorithm.3. Traditional stereo matching based on global optimization was of computationalcomplex which was poor to get accuracy matching result for the pixels inocclusion and depth discontinuity region. An efficient method of stereo matchingwas proposed, which was based on Tao stereo matching framework, namely: anefficient matching algorithm based on color image segmentation framework. Thealgorithm is summarized as follows:Firstly, used Mean Shift method to divide the reference image into sections.Then，got the initial disparity by the enhanced local method.Then got occlusion and mismatched pixels from reliable pixels using therobust method and named unreliable pixels, then used reliable pixels andpresupposition of disparity plane to refine the unreliable disparity. Finally, in order to improve the disparity accuracy in low texture region, anenhanced belief propagation method was used to optimize the refined initialdisparity, which had adaptive convergence threshold.Experimental results demonstrate that our method can reduce the error matchingrate effectively, improve the matching accuracy in occlusion and depth discontinuityregion. Reduce the computational complexity as well as improve matching speed.