Three Dimensional Video Error Concealment For Network Transmission

With the rapid progress and development of science and technology, multimedia videotechnology experiences a transition from analog video technology to digital video technology.Now high-definition TV has entered ordinary families. Compared with the traditional two-dimensional video,3D video adds the depth information, providing immersive visual experience tothe viewers. Therefore, the related technology of3D video is bound to become a new topic andattracts great attention from domestic and foreign experts.However, linearly increasing number of viewpoints inevitably leads to great amount of data in3D video.The coding predict structure of3D video makes use of the correlation among and withinviewpoints and the spatial correlation to remove the redundant information and compress thetremendous data to the greatest extent. But once an error occurs in the transmission network, it notonly affects the quality of the current frame, but also propagates the error to the frame which takesthe current frame as a reference and affects its subjective quality. Thus, it is particularly importantto study the three-dimensional video error control techniques. The error concealment technology ofthree-dimensional video mainly uses the neighboring correctly received information to recover thelost image information at the decoder side. Moreover, the error concealment technology does notdegrade the coding performance and increase delay. This paper mainly researches errorconcealment techniques of three-dimensional video for network transmission. And the main resultsof this paper are as follows:(1)This paper analyzed the codec standard principles and key technologies of H.264/AVC andsome theoretical knowledge, such as3D video capture, perception mechanism, showing andpredicting structure of encoding. Also this paper summarized the current error control techniques indetail.(2)This paper proposed a motion and stillness characteristics judgment model based onstereoscopic vision. The model uses the inherent correlation between the video pictures ofneighboring viewpoints or neighboring frames and the current video pictures with loss information.The motion or static characteristics of loss information in current video pictures are predicted bythe motion characteristics of corresponding position in temporally adjacent frame or in the frame ofneighboring viewpoint with same time. The part of4.2.2and5.2.2gives a detailed introduction to the model.The model performs well in the error concealment algorithms of both full frame loss andpart macroblock loss, verifying the feasibility and effectiveness of model.(3)This paper proposed an error concealment algorithm with full frame loss for stereoscopicvideo based on region classification. Firstly, the frames with loss information are classified intomoving regions and static regions based on the proposed motion and stillness characteristicsjudgment model. The static regions are copied from corresponding information in temporallyadjacent frames. The moving regions are further segmented into obscured boundary regions andnon-obscured boundary regions by using the global disparity. The obscured boundary regions arerecovered by motion compensated prediction method, while the non-obscured boundary regionsare restored by disparity compensated prediction method. Experimental results show that theproposed algorithm can improve the PSNR with1.56dB to9.35dB, compared with the traditionalerror concealment algorithms.(4)This paper proposed an error concealment algorithm with part macroblock loss instereoscopic video based on visual sensitivity. Firstly, the characteristics of the lost macroblock areanalyzed in detail. Then based on the shielding characteristics, the lost macroblock are divided intoobscured boundary macroblock and non-obscured boundary macroblock. As for the non-obscuredboundary macroblock, after analyzing the motion characteristics based on the proposed motion andstillness characteristics judgment model, they are be marked as either moving macroblock or staticmacroblock. The moving macroblock are segmented into sensitive macroblock and insensitivemacroblock based on the analysis of sensitivity. And the sensitive macroblock are furthersegmented into texture macroblock and smooth macroblock based on the texture and smoothcharacteristics. Finally, obscured boundary macroblock, static macroblock, insensitive macroblock,texture macroblock and smooth macroblock are recovered by temporal motion compensatedprediction method, temporal frame copy method, disparity compensated prediction method, multi-direction interpolation method and linear interpolation method respectively. Compared with thetraditional error concealment algorithms, the proposed algorithm significantly improves thesubjective and objective quality of the lost video images and improves the PSNR ranging from0.31dB to6.52dB at the packet loss rate of10%.