The Research Of HVS Based Multi-description Video Compression Compression Codec System

In real-time video communication systems, it is a key role for robust video codectechnology to play in fighting against bad network environment. Multi-descriptioncoding is an unreliable channel for error recovery techniques. It encodes a single sourceto generate two or more of the description can be decoded independently, therefore thistechnology can solve the serious decline problems of video image quality due to thedelays, errors and losses in the transmission of data, thus improving data transmissionover unreliable channels robustness. However, multi-description coding may resultmismatching problems caused by which loss of information leads to the reference frameused in coding and decoding inconsistently, thus affecting the visual quality of decodedsignals.In the paper, chose x264as the encoder, ffmpeg as the decoder. A modifiedmultiple description scalar quantization (MDSQ)，the key of which lies in the design ofthe label distribution function，was transplanted into the video codec system. This paperpresents a distribution program based on four scans (north, west and south east) of thecodec for video compression system. Then, for the mismatching problem ofmulti-description video compression codec system, this paper proposes a combinationof human visual characteristics of the intra-refresh algorithm. The principle is thatintra-refresh the high visual sensitivity and strongly associated with subsequent frameareas, involving visual sensitivity and impact of the macro block weight. Visualsensitivity weighted normalized to strike by activity prediction and texture complex ofmacro blocks, while the impact weight of macro blocks deduced by backwarddependency in the area of high visual sensitivity. At last, the weighted control modulewhich determined by visual sensitivity and backward dependency improve the framerefresh rate in the region of the high visual sensitivity and strong correlation withfollow-up frame, in order to strike a balance between the introduction of dataredundancy and visual quality in specific regions.In this paper, using C language achieved the HVS-based multiple description videocompression codec system in the Windows XP platform. The results show that, whenthe overall redundancy of this program is about20%within an acceptable range, thePSNR of reconstructed image of the single-channel description is about24dB, andimprove significantly subjective visual quality in particular area of this reconstructedimage. While reconstructed video image quality of two-way description is36dB or so, closing to Single Description. At the same time, the time of average decoding andcombining a frame is less than40ms,meetting the needs of real-time applications.