A Region Based Stereo Matching Algorithm Using Cooperative Optimization And Application Of A Three-dimensional Display System

The traditional binocular stereo video which is based on binocular parallax principle Call from a single three-dimensional sense of perspective, and multi-view video Can increase the viewer’s sense of depth and reality. It will be widely used in the fields of 3DTV,free·viewpoint TV, immersive videoconferencing,virtual reality, etc. Multi-view video’s feature focuses on interactive function which is provided for the viewers to select view point interactively. The synthesis of virtual views is the key technique in multi-view video.This thesis topics is based on a researching project of the author engaging corporate, exploring the technology of multi-view reconstruction through binocular stereo image. The main research contents of this thesis is stereo matching algorithm, and using it in the multiple viewpoints synthesis technique. Our method mainly includes stereo disparity estimation and view synthesis. Some relevant algorithms have been presented and been implemented, and valuable experimental results have been obtained.Stereo disparity estimation is one of the fundamental and significant problems in stereo vision. This technique makes it possible to reproduce a three-dimensional stereo display by getting the distance information through correspondence between the image pairs. After making a summary of representative methods for disparity estimation, this thesis mainly focuses on the area-based matching method. Then a improved algorithm is proposed which is a region based stereo matching algorithm using cooperative optimization. And the variable window strategy is used in updating disparity values. To further enhance the accuracy of disparity map,the occlusion detection is applied in the process of disparity map. Experimental results show that accurate disparity map Can be obtained,and the effect is better than normal block-based stereo match.View synthesis technique is one of the key techniques in multi-view video processing and application. This thesis is aimed to extend the single view binocular stereo video into multi-view stereo video through double disparity images. In the process of view synthesis, it is necessary to smooth the disparity map in order to avoid big holes. Then a improved synthesis algorithm based on backward non-linear image interpolation is proposed under small-baseline condition. This method chooses weighting factor that varies with the local quality of corresponding pixels. In addition, a improved synthesis algorithm based on visibility analysis is presented under large-baseline condition.This method solves the occlusion problem well.And then our view synthesis algorithms are implemented in MATLAB software platform. The experimental results show that Our methods can obtain virtual views with higher quality and PNSR.