Research On Virtual View Synthesis

Virtual view synthesis is a new technology which synthesizes a clear image ofthe virtual viewpoint of scene by two or more images of other viewpoints obtained bycameras. Virtual view synthesis is one of the current hot topic in the field of computergraphics and computer vision, and has broad application prospects in virtual meetings,virtual roaming,3DTV, remote surgery and so on.This paper presents a virtual view synthesis method, and it consists of four steps:(1) Feature extraction: obtaining feature points by extracting feature in the images ofthe left and right view.(2)Feature matching: matching feature points and acquiringthe corresponding relation of the features in the left and right view. When the anglebetween the left and right view is large, say, greater than15degree, the featurematching may be hard to do. To solve this problem, we propose a hierarchicalmatching algorithm, which uses traditional Euclidean distance-based matchingalgorithm as the coarse matching and makes the points obtained above matched finelyby factorized graph matching algorithm.(3) Calculation of fundamental matrix: Thefundamental matrix which represents the relation of epipolar geometry is calculatedby8-Point algorithm and corresponding relation of the features that extracted from theleft and right view.(4) View morphing: images rectification, stereo matching, imagesynthesis and post-processing will be taken, then we get the image of the virtualviewpoint. In addition, neighborhood filling treatment is taken to improve the visualeffects.In the end, this paper takes experiments using C/C++as programming languagein the basis of public data sets (CMU house data sets) and the image and videocollected. The existing view morphing algorithm is sensitive to mismatching, even ifonly5%of mismatching will greatly affect the quality of virtual view image. Theexperimental results show that the hierarchical matching algorithm proposed in thispaper can improve the matching accuracy and get better visual effects in theapplications. The system developed by the proposed algorithm is capable to meet theneeds of practical real-time application and get ideal virtual image.