Research On The Optimization Technology Of View Synthesis From Single Viewpoint

With the development of 3D video technology,a number of 3D video applications such as three-dimensional television(3D-TV)and free-viewpoint television(FTV)has gradually stepped into our lives,changing the way we observe the world and increase our virtual experience.This current DIBR method employs the disparity correlation between views to create a so-called synthesized view which is displayed at receiver side to help the audience enjoy stereoscopic videos.The quality of the virtual views is affected by multiple factors.A common problem introduced in traditional synthesis methods is that there are some regions in the synthesized view are invisible in the captured views and it manifest as holes or so-called dis-occlusions in the synthesized image.Result of the DIBR method can be affected by the depth map quality.In addition to coding distortion,the depth map usually does not exactly match the texture map and may contain noise.This may lead to different types of artifacts in the synthesized image.Common problems are the texture-depth alignment,which may cause pixels belonging to foreground objects to be distorted as if they were part of the background.Another problem is related to the accuracy of the depth map,which may result in a small ”crack” in the synthetic image due to the incorrect displacement of the pixel.There are a number of view synthesis algorithms to reduce the distortion caused by hole or dis-occlusion problems in the virtual perspective synthesis process,thereby improving image quality.For example,in 3D-HEVC standard,an effective blending algorithm based on frame interpolation is employed to obtain the final synthesized view,by combining two rendered views of the same point.This method can achieve significant result,but make extra cost of bandwidth and complexity as it requires two captured views information.In this paper,we take a new strategy to deal with the problems.The depth map is compressed in coding stage,so when transmitted to the receiver side the edge portion of the depth map becomes blurred and the quality of synthesized view will be reduced.We employ edge sharpening algorithm to depth map,help restore it clear and sharp boundary.In addition,we preprocess the depth map to ensure that all foreground pixels in the texture map correspond to the foreground region in the depth map,thus avoiding the difficulty of hole-filling.At last,we use the background-based hole-fill algorithm for image post-processing.The experimental results show that the proposed method can express good performance in visual quality.