Study On Video Error Cocealment Technology Based On Facial Features

H.264/AVC is the most recent international standard of video compression, designed by ITU-T VCEG and ISO/IEC MPEG. In the latest years, H.264 has been introduced into practical applications .But the video stream compressed by H.264 is more sensitive to transmission errors. Once an error occurred in transmission, the effects of fault information would be propagated in spatial and temporal rapidly which would result in the serious decline of video quality.This article contains major research aimed at people face region error concealment for video transmission technology, in the full study and understanding of domestic and international video error concealment technology, complete the human face detection algorithm based on H.264 joint model (JM) decoder, then proposed the spatial and temporal error concealment algorithm based on the face region. The detail points are as follows:First we introduce the face detection algorithm and add the face detection in the JM decoder .We combine the skin and adaboost detection algorithm to detect the picture of the decoder rebuild fast and accurate.An adaptive spatial error concealment algorithm based on face features(BFASEC) is proposed in chapter four .The missing macroblocks are classified into face macroblocks and background macroblocks based on the face detection .For the face macroblocks, they can be concealed by the level bilinear interpolation algorithm(BI). The edge macroblocks can be concealed by the weighted average of multi-direction interpolation method. Experimental results show that the BFASEC algorithm has much better error concealment ability than other spatial error concealment algorithms, and it achieves 0. 5～2. 53dB average PSNR increase for recovered images compared with the BI algorithm employed in the JM decoder especially.A two-step multi-weighted boundary-matching base on face features algorithm(BFTMBMA) for temporal error concealment is proposed in chapter five. The missing macroblocks are classified into face macroblocks and background macroblocks based on the face detection. The face macroblocks has most priority to be concealed. The missing face macroblocks can be concealed by means of 4 x 4 mode for refine the face features to be concealed, add the deblocking filter after concealment .Experimental results show that the BFTMBMA algorithm further improves the concealment performance for the corrupted video sequences than other temporal error concealment algorithms. Especially, 1. 94～2. 76dB average PSNR increase for recovered images is gained by the BFTMBMA algorithm compared with the SMA algorithm employed in the JM decoder and acquire more subjective quality.