Study On Depth Image Coding Algorithm Based On Virtual Viewpoint Rendering

As a kind of advanced way of visual media, the FVV brings people a chance offeeling the immersive3D perception through the video. The technology is a brand newvideo media mode which is interactive and with stereo feeling. It can show us a real3Dscene through a more real and natural method. This new method can be applied to livingentertainment, exhibition display, medical institutions, education training,3Dsurveillance,3D reconstruction and the virtual reality etc. The virtual view renderingtechnology as the key part of the FVV is attracting the experts and scholars to do thein-depth research.With the growing up of the number of view, the amount of data increases rapidly.The limited bandwidth of practical application limits the storage and transmission of themulti-view video data. The solution of this matter is using DIBR in the terminal of thefree3D video system to realize choosing any viewpoint. As the DIBR uses the depthimage and the color image to generate the new viewpoint image, the quality of the depthimage takes a significant place in the virtual viewpoint drawing. Therefore how tocompress the depth image with high quality is the key point in the FTV system.Based on the above idea, a depth image coding algorithm based on virtual viewrendering is proposed in this paper. The main work content is as follows:(1) An adaptive quadtree segmentation algorithm is proposed which segments thedepth image with appropriate threshold according to a large number of experiments. Theflat areas are segmented as big blocks to encode while the texture areas are segmentedwith small blocks to encode.(2) The appropriate model functions are selected to fit the edge status of the image.Three types of model functions are used to fit the image. The3functions are singlevalue function, linear function and dual-area function. This method can increase thecompression ratio while ensure the accuracy of the decoded image. So as to reduce thechannel transmission cost with high virtual viewpoint rendering quality.(3) The decoded depth image is used to render the virtual viewpoint and theoptimal decoding performance with a large number of experiments is obtained. And theperformance of the proposed method is given by analyzing comparison results withH.264JM10.2..