Research On Error Concealment For Three Dimensional Video Transmission

As a new generation of multimedia video system, Multi-view Video Coding (MVC) is one ofthe core development directions of Three Dimensional (3D) video coding. MVC needs more thantwo times video data compared with traditional mono-view video. Human eyes are the finalreceivers of a stereoscopic scene, human perception plays a key role in designing high efficientcoding algorithms for3D video. This paper conducted a study of MVC based on HVS to reducethecoding complexity andimproveefficiency ofvideo compression.(1) The Just Noticeable Distortion (JND) model can accurately describe the human visualcharacteristics. This paper uses stereoscopic characteristics of the HVS to establish a JND model inMVC. The proposed model includes spatiotemporal JND and depth JND. With the new modeladaptively adjusting Quantization Parameter (QP), an asymmetric stereoscopic video codingalgorithm is proposed. Experimental results show that, this algorithm averagely savings rate24.18%~36.24%with a variety of different attributes test sequences compared with the standardmethod,maintainingthesamevisualsubjectivequality.(2) This paperproposes a newfast macro block mode selection algorithm of MVC to decreasecomplexity in MVC through utilizing JND. It gains the correlation of MVC macro block modeselection with JND base on HVS properties to get the threshold. Then, the threshold can be used tooptimize variable size mode selection in MVC. On the basis of the relation between the thresholdand the JND of the current macro block, the new algorithm can adaptively reduce macro blocksegmentation selecting to enhance the efficiency. Experimental results indicate that the newalgorithm can decrease70.81%~79.84%of the encoding time compared with JMVC and almostkeep the same rate distortion performance, while showing the adaptability to various properties ofscenes.(3) This paper proposes a new algorithm to reduce the extremely high encoding complexity inmultiple reference frame prediction based on Binocular Just Noticeable Difference (BJND). Thenecessity for full searching all reference frames is decided on the statistical features betweenmultiple reference frame selection and BJND. Experimental results show that the proposedalgorithm can reduce averagely76.00%of the encoding time compared with JMVC and almostmaintainthesamecoding efficiency.