Research On Error Resilience Video Coding For Face Area

Over the past few decades, with the high-speed development of Internet and wireless communication networks, digital video compression technique was extensively used in many fields. As a key technology ,Face detection has an important application values in image searching based on content, video monitoring, video meeting, videophone, human-machine interaction and so on. Facing with the emergence of video data, how to locate the face region exactly,has become an urgent problem.On the other hand, the Internet and wireless network nowadays can not provide reliable data transmission,and the packet loss and the bit error are liable to occur at the moment of delivering video streaming over these networks.The low bit-rate video stream compressed by motion compensated hybrid codecs is extremely vulnerable against transmission errors. How to improve error resilience of the required information in video, is become one of the urgent technologies of the current video field.The paper first introduces the main content of H.264 encoding technology which is closely related to the thesis, then analyzes the rate distortion algorithm model and RDO method in JM test model.Face detection technology in video is researched, Combining with motion vector of video encoding, interframe correlation, face-feature and frame difference, we propose a face detection algorithm based on video sequence which can detect face fast and exactly. The proposed algorithm can detect the position of face region effectively and remedy the face region which is false judged and missed judged in AdaBoost algorithm. Experimental results show that the algorithm improves both speed and accuracy of face detection, compared with AdaBoost algorithm.Finally, the paper researches rate-distortion optimized model. synthetically analyzes propagation distortion and concealment distortion in the process of video transmission. In particular, the video transmission distortion of end-to-end on pixel level precision is studied, realizing a general transmission distortion estimated model. Combining the transmission distortion of face region, a face region refreshed algorithm based on distortion is proposed and implemented. The algorithm can choose face region refreshed in dynamic ways in accordance with the missed package rate and the error propagation of transmission distortion. Simulation experiments for video transmission over packet loss channel indicate that the proposed algorithm can guarantee a better subjective and objective quality of the reconstructde facial area at the decoder and stable bit rate at the encoder, compared with fixed cycle of face region refreshed algorithm in whole frame.