EPI And Color Segmentation Based Stereo Matching Algorithm And Image Synthesis

Stereo Matching algorithm is a frequently discussed topic in the area of computer vision. It is used to synthesize images from novel views and reconstruct 3-D information. Moreover, as the third generation of new technology, it has been applied to the area of video processing. This paper focuses on its application in image synthesis. At present, many algorithms can be used for image synthesis. This paper introduces two main categories of stereo matching algorithms, the local algorithms and global algorithms, such as the fixed window algorithm and adapted window algorithm, which belong to the local algorithms. And some global algorithms as dynamic programming and layer-based color segmentation algorithm are also introduced. As the main content of this paper, a synthesis system is constructed by using the technology of mean-shift based color segmentation and EPI, which instinctively avoids problems in intensity-incontinuous regions met by correlation-based algorithms. First the feature points are extracted through EPI. And according to those points, Delaunay triangulation is applied in every color segment to construct several planes, and then every plane function is calcualted using the disparity value of the pinnacles. Image warping and morphing technology is abandoned in this paper. Instead, Z-buffer technology is used to synthesis new images, through dual-direction compensation with two calculated disparity maps, ensuring the visibility of foreground object when occlusion happened. An evaluation system that specially constructed for evaluating stereo matching algorithms is introduced to evaluate the results in this paper.In this paper, the form of EPI, edge detection, extraction of feature points, Delaunay triangulation and disparity calculation are all completed in the environment of Matlab, and the color segmentation is finished by C++ code.