Research On Virtual View Synthesis For3D Video

3D video technology has been developing rapidly in recent years.3D movie and3DTV give people immersive viewing effect in their entertainment life. However,3D video can’t completely simulate the natural viewing scenario due to the lack of depth perception sources compared with human eyes, leading to visual fatigue. Some flaws in the3D video, such as vertical disparity between left and right views, can make it worse. Image rendering is a possible way to improve the viewing comfort by adjusting the relative position between the viewer and the scene. View synthesis based rendering is one of the key technique to achieve that.This paper studies view synthesis technology based on Depth Image Based Rendering (DIBR) and proposes a new view synthesis method. In order to eliminate the infiltration phenomenon from foreground to background, a depth preprocessing method has been executed. A new method of pixel mapping is carried out to render the synthetic view. The new mapping method is divided into two steps:1) Depth map forward mapping and median filtering of the mapping results.2) Color image reverse mapping. Due to the wrong depth value and occlusion, there will be holes in the mapping result. Holes in the synthetic depth map are filled first. Then holes in the synthetic color image are filled based on the depth information in the filled synthetic depth map. The hole-filling method has also taken the relative position between real camera and virtual camera into consideration, making it more suitable for view synthesis.Stereo image rectification is also a crucial technology in3D video processing. Traditional rectification methods are based on epipolar geometry and need to apply projection transformations to the left and right images in stereo video. The projection transformations will bring distortion to images, resulting in a decline in viewing experience. This paper presents a new idea of stereo image rectification, which does not depend on projection transformation, but generate rectified images by view synthesis method. First, the original stereo images is rectified by traditional stereo rectification method. Then stereo matching is performed to get the horizontal disparity map of the rectified stereo image. Next, the original views’horizontal and vertical disparity maps are figured out through the reverse transformation of the traditional rectification. Finally, a synthesized right view is obtained through the proposed view synthesis method under the disparity constraints. The original left view and the synthesized right view are the rectification results of our method. Since the epipolar geometry between the camera only affect to the intermediate results, arbitrary positional relationship between cameras can be rectified by out method.