| Stereo video is a very important development direction in video technologies in future. Compared with single-view video, stereo video usually has two views and its data amount is much greater than single-view's, so efficient compression is very important.Disparity prediction is a commonly used method to realize the data redundancy and also a key technology for stereo video coding. In order to get the excellent matching results, big searching load is usually needed (Sometimes the disparity vector is as large as 100 pixels). This made the coding less efficient. Therefore, how to decrease matching expenses becomes very important. According to the difference of the matching element, 3-D matching can be divided into area matching,feature matching,and phase matching three categories. The measurement of similarity includes relevant,distance and probability. "Change size block method", "Search window method" and "Disparity estimation based on object" are proposed successively.Geometric constraint is a basic nature constrains for 3D picture, and it can be used to decrease the searching range. In this thesis, human vision characteristics and camera imaging principle are studied, and parallel,convergent system are also analyzed. Geometric constraint and the resolve method for fundamental matrix are focused on. The experiment uses eight algorithms, normalized algorithms and weighted shift normalized method. The precision and anti-noise abilities are improved in weighted shift normalized method.As primary condition for visual measurement, camera calibration is focused on how to build the relation between image coordinates and spatial coordinates. Basically speaking, camera calibration is a process to determine the camera internal and external parameter. The disparity prediction for calibrated camera is an algorithm which project spatial coordinates to image coordinates. For un-calibrated camera, if the points are matching, geometric constraint is the only information. A new method is proposed in the thesis which uses the image information to optimize disparity prediction. It includes the steps: Harris corner detect, rough matching, accurate matching, fundamental matrix solving, and block position prediction. Experiments show that this method can achieve high precision, reduce the residual value.Based on MPEG2 standard frame, IBP modes are extended. There are main 3D video coding reference method: movement compensation prediction (MCP), disparity compensation prediction (MCP), Motion estimation and disparity prediction combining method (MDCP).In this thesis, MDCP is used and the fast search method is added in un-calibrated camera disparity prediction method. Improving Adaptive Cross Pattern Search(ARPS)is be focused on. This method gets the best quality and time consuming balance. |
