The Study Of Virtual View Synthesis Optimization And Prediction Technology In 3D-HEVC

3DV can provide the depth information of the realistic scene, to make users experience stereoscopic vision. Recently, MPEG regards multi-view video plus depth(MVD) as the data format of 3DV, and develops the new generation 3DV coding standard 3D-HEVC based on HEVC. Compared to traditional 2D video, MVD brings extra video and depth information, but increases the encoding rate greatly. Because different views are taken from the same scene, they have strong inter-view correlation. Through the correlation, coding efficiency can improve. At the same time, depth information is mainly used to synthesize virtual view at the decoder and not to watch directly. So the Rate-Distortion Optimization of depth information is different from traditional video coding. This paper mainly studies view synthesis prediction of texture coding and view synthesis optimization of depth coding in 3D-HEVC.View synthesis prediction is a special inter-view prediction coding method, using inter-view matching block and depth map to synthesize the prediction block of current block. Divide the virtual depth block into 4×8 or 8×4 sub-block. Then compare the depth values of leftup rightup leftdown rightdown in each sub-block to choose the maximum depth value. Because there is no deblocking filter in depth coding of 3D-HEVC, blocking boundary has a bad effect on choosing the maximum value. Based on the reason, two methods are proposed to decrease the effect. One method is that the maximum value is elected from middle position in sub-block, another method is that the maximum value is elected from four positions inward one pixel to avoid the effect of boundary. Experimental results show that the two suggested methods improve the average coding bitrate of 0.07% and 0.18%, respectively. What’s more, the chosen DV in the second method can predict current block exactly.Different from texture pictures, the purpose of depth map coding is mainly to synthesize virtual views at the decoder. Hence, depth coding algorithm based on virtual view quality optimization is proposed in 3D-HEVC. When coding middle depth, six middle synthesized view distortion are computed. Currently, six virtual views use equal-weighted mechanism in rate-distortion criterion. Because the positions of 0.25 1.25 and 0.75 1.75 are close to left view and right view, the distortions are smaller. Butthe middle distortions of 0.5 and 1.5 are bigger, so the two distortions are given a bigger weight in order to choose the best coding model exactly. In this paper, eleven different weights groups are designed to make experiments. Preliminary results show that increasing the weights of 0.5 and 1.5 positions can improve the whole synthesis quality in the larger QP. The chosen weights improves the quality of synthesized view 0.05%.Through a lot of experiments, the correlation of different QP and better weight is fitted to enhance the quality of synthesized view.