Research On View Synthesis Based On Binocular Vision

Image-Based Rendering (IBR) technique is a new approach to create virtual reality scene, and becomes an important research field in computer vision in recent years. IBR technique embraces many theories and results of different specialty including computer vision, computer graphics, and image processing and photogrammetric. It substitutes traditional geometric model with images of real world, and makes rendered image natural. The consumed time for rendering model is independent of scene complexity. In this paper, the methods for image synthesis to a new viewpoint are researched based on binocular vision. The related keys include the fundamental matrix computation, image rectification, stereo matching with relaxing symmetric multiple windows and view synthesis based on relative affine structure.The fundamental matrix is a cornerstone of image synthesis, and encodes the information of intrinsic and extrinsic parameters of camera. In this paper, the application of the data normalization algorithm for estimating fundamental matrix is discussed. This dissertation studies the application of data normalization algorithm to estimating Fundamental Matrix. Experiments show that anisotropic data normalization algorithm makes the fundamental matrix linear estimator algorithm more accurate and stable.Image rectification is the process of warping a pair of stereo images in order to align the epipolar lines with scan-lines of the images, and reduces stereo matching from a 2-D to a 1 -D search. A method is presented to rectify stereo pairs, which best preserves the sampling of the original images, and brings little image distortion.The stereo matching obtains a correspondence which points on the left and right images are projections of the same scene point. This dissertation uses Relaxing Symmetric Multiple Windows stereo based on Markov Random (R-SMW). The experiments show that R-SMW is robust. View synthesis consists in rendering images of a scene as if they were taken from a virtual view point different from all the viewpoints of the real views, and uses photographs of real scenes as a basis to create very realistic images. A method is used to synthesize new viewpoint image based on relative affine structure. The method expands the range of synthesis images not only on the baseline.