Research On 3D Video Post-processing

Ongoing applications in digital technology have led to an increasing development of 3D display technologies. People are no longer satisfied with 2D films and videos, and they want to feel as if they had been participating in tridimensionality when watching. Among the aided viewing technologies (such as handed observer,3D glasses and head mounted display) and free viewing technologies (such as autostereoscopic display, holographic and volumetric), the simplest and fastest growing stereoscopic display mode is based on binocular parallax, and 3D glasses and autostereoscopic display are the most common display modes. In this paper,3D video post-processing technologies, such as disparity map computation, view synthesis, and so on, are researched based on the discussion about the impact of display screen size changing on objects'depth perceived by human eyes. To disparity computation, a stereo matching method based on image color segmentation is proposed. Affine transformation model is applied to guarantee disparity consistency in time domain. Depth adjusting scheme combined disparity adjustment and view synthesis is given to reduce the impact of screen resizing.Stereo matching method based on color segmentation can make full use of smoothness assumption within segments, and can well combine seed disparity expansion, plane fitting and disparity histogram analysis within segments to keep the smoothness of disparity map. What's more, color Euclidean distance segmentation is easy to control segment size, and thus it is used to pass disparity messages between neighboring segments to rectify unreliable disparity distributions in segments. And Belief Propagation, a global energy minimization method, can keep the overall smoothness of disparity map. Experimental comparison on Middlebury datasets shows the effectiveness of our algorithm.To guarantee the disparity consistency of 3D video, affine transformation model between consecutive frames is used based on stereo matching using color segmentation, which can constraint the disparity distribution in current frame. In addition, the relationship between segment inconsistency and disparity inconsistency is discussed, and disparities in segments of current frame will be refined with segment correspondences of consecutive frames.The process of 3D video capture to 3D video display can be seemed as a coordinate transformation. According to the recurrence relation of coordinates, disparity adjustment method is given when 3D display screen is resized. And in order to change the objects'depth perceived by human, a new stereo video is produced with our view synthesis algorithm based on adjusted disparity map.