Research Of Binocular Image Matching Technologies Based On Image Segmentation

Binocular vision technology has been widely used in many fields such as national defense military, machinery manufacturing, Large curved surface shape detection, protection of historical relics, pattern recognition, three-dimensional reconstruction, unmanned vehicle navigation, industrial inspection and so on. Binocular image matching is one of the key technologies in binocular vision, many research results haves been achieved by many researchers, but the matching strategy and other issues still need to be improved. Binocular matching as the core of binocular vision technology, the accuracy and efficiency of the algorithm directly affect the performance of binocular vision in practical application. Therefore, it is very important for the research of binocular image matching technology.In order to effectively extract the region of interest in the image, an algorithm of image segmentation based on region is proposed in the second chapter. Besides, the image segmentation based on facet model fitting, and accurate algorithm for color image segmentation based on region growing are proposed, respectively. On the basis of a lot of literatures and many experiment results, we found that it has a direct relationship between the accuracy of local matching algorithm and the size of matching window. The binocular image matching based on color and segmentation is proposed in the third chapter. The RGB information is firstly applied to automatically adjust the matching window in the process of initial matching. Secondly, on the basis of image segmentation, the chrominance in the HSV space is applied to correct the initial disparity map. In order to solve the influence caused by the weak texture region and the uneven illumination, an algorithm of binocular matching based on region which depends on the segmented image is proposed in fourth chapter. The proposed regional matching cost function is firstly applied to implement the initial matching in the algorithm. Secondly, the relative position constraint is combined to correct the matching. Finally, the disparity map is calculated by using the similarity measure function. The accuracy of local matching algorithm is very low, the global algorithm applying the BP (Believe Propagation) to four neighborhood domain is proposed in fifth chapter. The algorithm based on the assumption that the disparity value of each pixel in the image is only related to its four neighboring pixels, but not related to the other pixels. On the basis of it, a new energy function is constructed by this method. The transport mechanism of message and confidence of BP are applied to optimize the disparity map which obtained in the third chapter and the fourth chapter. The proposed method not only improves the accuracy of the matching, but also solves the problem of discontinuous regions. At the same time, the accuracy of the proposed algorithm has been validated by our experimental results using the standard image called Tsukuba and Venus that coming from the American College of Middlebury.