Research On Generating Virtual View Schemes In Natural Three-Dimesion TV

Natural Three-dimension Television (3DTV) can show people the stereo effect from several angles, and then more videos flow must be provided to TV receriver for realizing several angles stereo feeling. However, the present TV bandwidth can’t meet transmitting large number of videos flow simultaneously, and only rendering new virtual view after video decoding can realize the more matching for left and right eyes to meet stereo feeling from several angles. Depth image based rendering (DIBR) is considered as the main method for rendering virtual view in3DTV because of its simple and reliable.DIBR can generate any virtual view color image by using one color image and its associated depth image, but the virtual view image has low qulity and large exposed areas, which is called as "holes", and the traditional method for reducing holes is smoothing depth image, finally image inpainting algorithm is used on remaining holes. Smoothing depth image can introduce image geometric distortion; image inpainting algorithms are commonly complicated and fail to generate natural and ture texture information.This paper begins from non-preprocessing depth image and obtaining more real virtual view image, and deeply analyzes the DIBR problems, finally proposes a non-proprecessing depth image algorithm for double-sided DIBR and a searching real image texture to inpaint holes algorithm based on motion vector in reference view image sequence for unidirectional DIBR.(1) Double-sided DIBR: rendering and merging virtual view image by using left and right reference view images and their associated depth images. The innovations of proposed algorithm are:proposing a one pixel to four dot algorithm (OPFD) to solve the overlapping and resampling problems in the course of3D warping, which can greatly decrease the exposed areas and mentime introduce no image geometric distortion; and finally proposing a sub-regional merging to removing virtual view image edge ghosting algorithm in the course of image merging and its simplified algorithm. Owing to introduce no image geometric distortion and its effective merging algorithm, proposed algorithm can generate real and ideal virtual view image.(2) Unidirecional DIBR: rendering any view image by using one reference view color image and its associated depth image. Firstly, proposed algorithm solving overlapping and resampling proplems by using OPFD algorithm; secondly, precisely matching the holes edge coordinates of virtual view image from reference view image; and finally levelly inpainting the holes using the best texture pixles which come from searching in reference view image sequence based on motion vector. Owing to its real and natural texture pixel, proposed algorithm can obtain high quality virtual view image.Finally, problems to be solved in this field and future research topics are summarized.