Reasearch On Depth Coding For 3D-HEVC

In 2013,JCT-3V released the next generation 3D high efficiency video Coding.Compared with traditional 3D encoding technology,3D-HEVC provides higher compression efficiency and achieves higher quality for pictures,but also brings the problem of increase in the complexity of the encoder.Therefore,the method which can efficiently reduce the coding complexity of 3D-HEVC under the condition of guaranteeing the quality of 3D video compression has become a focus technology.In this thesis,for the purpose of reducing the complexity for depth map coding,the 3D-HEVC depth video coding algorithm is studied,and both the partition of the Coding Unit and the Depth Modeling Modes are improved.In order to reduce the 3D-HEVC coding complexity,an improved intra prediction mode decision algorithm for depth maps is proposed in this thesis,including early DMMs termination and fast wedgelet patterns algorithm.The early DMMs termination algorithm performs Laplacian edge detection,and determines whether it traverses depth modeling modes or not.To reduce the number of traversing wedgelet patterns,the fast wedgelet patterns algorithm utilizes the correlation between the wedgelet and 33 angle modes,and only traverses the wedgelet pattern related to angle modes.Compared with depth modeling modes,the average coding time of the algorithm is reduced by 53.6%,the bit-rate is increased by 1% and the PSNR is reduced by 0.15 dB.For the high complexity of Coding Unit in depth map,an improved fast coding unit decision algorithm is proposed for depth map.To reduce the CU depth range and the coding time,the depth of the current CU is predicted by the correlation between the the adjacent CU Then,the algorithm determines the depth range of the CU splitting,according to the predicted value instead of the original depth range(0~3).The performance tests show that,even if the bitrate is increased by 0.61%,the PSNR is reduced by 0.17 dB,and the average coding time is reduced by 31.5%,the subjective quality of the virtual viewpoints still maintains the same.