Study On Error Resilient Coding Of HEVC And 3D Fast Video Coding

To improve the error resilient ability of High Efficiency Video Coding(HEVC),a layered multiple description video coding algorithm based on 3D dual-tree discrete wavelet transform(3D-DDWT)for HEVC is proposed.Each description is composed of one base layer and an enhancement layer: The base layer is the bitstream of HEVC to provide basic information of video;the enhancement layer is generated by two subbands of 3D dual-tree discrete wavelet transform(3D DT-DWT).Based on the directional selection and the correlation of four 3D DT-DWT subbands,the proposed multiple description coding for HEVC can effectively provide a robust transmission of video,even when only a description is received in the decoder.For 3D video coding,3D-HEVC supports the multi-view video plus depth(MVD)format,which consists texture and depth sequences for each view,and the coding process of depth map follows the principle of texture sequence,which increase the computational complexity of 3D-HEVC.Therefore,two fast algorithms for 3D-HEVC depth map coding are proposed.The first one is to prejudge the CU segmentation depth,the second one is to skip the depth modeling modes(DMM)checking.The first algorithm named GLSM-FICUSD(Gray Level Co-occurrence Matrix Based Fast Intra Depth CU Size Decision).By calculating the gray level co-occurrence matrix of each CTU and counting numbers of nonzero values,the depth of CU segmentation can be prejudged.Experimental results show that compared with 3D-HEVC platform HTM-16.0,the proposed algorithm achieves a reduction of computational complexity by 19.1% on average,while only 0.0591% BD-rate loss on synthesis views.The second algorithm is called GLSM-FIMD(Gray Level Co-occurrence Matrix Based Fast Intra Depth Mode Decision).By calculating the gray level co-occurrence matrix of each CU with different size,such as 64×64,32×32,16×16 and 8×8,four directions,0°,45°,90°,and 135° are specified.Then,values of angular second moment(ASM)and correlation(COR)along the four specific directions are figured out for selectively skipping the DMM mode and classifing 35 prediction modes according to different directions.Experimental results show that compared with 3D-HEVC platform HTM-16.0,GLSMFIMD can achieve 20.2% computational complexity reduction on average while only 0.0951% BD-rate loss on synthesis views.These two fast depth map coding for 3DHEVC are with relatively low computational load while maintaining the synthesized quality of 3D video.