Multi-view Stereo Virtual Viewpoint Image Generation Algorithm

In recent years, with the rapid development of technology about multimedia andnetwork communication, what based on the application of three-dimensional videocommunication is increasingly widespread, such as three-dimensional videoconferencing, virtual stereoscopic cinema system which is gradually matured inbusiness and market. The technology what enable people to have a "spot" experiencetechnology began to enter the office and home. These applications promote relevantresearch about stereoscopic video, including the acquisition, production, views,reproduce, display, compression, communications and applications. It becomeresearch focus of computer vision, computer graphics, video processing andcommunication engineering.The virtual viewpoint is one of the key technologies togenerate stereoscopic display.The principle of stereo vision, stereoscopic display and the technology of virtualviewpoint generation are introduced in this paper, and it detailed multi-view stereomatching and image rendering technology. Virtual viewpoint rendering is to capturetwo-view image or video to generate multi-view image or video by model-based andimage-based rendering method. Based on this theory, we propose virtual viewsynthesis algorithm based on sparse disparity. In this paper, the sparse parallaxbetween reference image is firstly calculated which relies on image salient features.Then virtual viewpoint image based on the left and right view is generated by usingthe nonlinear mapping function to calculate the coordinates of the virtual viewpoint.In order to get the final virtual viewpoint image, the images obtained above are usedto blend. Experimental results show that synthetic view image has good quality, whichcan maintain the salient features of image, and it can rightly reflect the content of theoriginal views. It also has a good visual effect.