Research On Error Resilient Video Coding For Low Bit-Rate Transmission

Information loss and bandwidth limitation are the main reasons for restricting video transmission quality. The low bit-rate video stream compressed by motion compensated hybrid codecs is extremely vulnerable against transmission errors. Once an error occurred in transmission, not only the effects of fault information would be propagated in spatial and temporal rapidly which may result in very poor objectionable visual quality severely, but also the video communication system could be in failure. To achieve the best video quality with the limited available bit rate, some resilient coding strategies against error propagation and a concealment method are proposed in this dissertation.This thesis first reviews the principle of rate-distortion(R-D) optimal for video coding, discusses the aspects relevant for microblock(MB) mode selection techniques briefly. In particular, the end-to-end error propagation distortion on pixel level precision is analyzed based on temporal concealment and channel error rate. Subsequently, a new general transmission distortion estimation model for various temporal concealments is proposed. The accuracy of this estimation is demonstrated via simulation results for video transmission over packet loss channel, which illustrates that actual distortion can be estimated by proposed expectation model approximately. The proposed model provides an efficient method for MB coding parameters optimal selection over error prone environment. The simulation results also show that the MB mode INTRA-update algorithm based on proposed model has better substantial and consistent gains when comparing with traditional R-D optimal INTRA-update algorithm. It is suitable for half-pixel accuracy and various temporal concealments.To overcome the shortcoming of computational complexity for traditional R-D optimal MB INTRA-update method, the solution of INTRA-update R-D problem is simplified in this proposed process. By analyzing the overall distortion of end-to-end and the rate control algorithm in TMN8, a fast INTRA-update algorithm which estimates the rough propagation and quantization distortion is presented. The fast INTRA-update selection algorithm is splited into two steps. By comparing the increased quantization distortion and reduced propagation distortion of one frame approximately, the encoder can quickly determine whether it is necessary to INTRA-update a MB. Then R-D based rata control strategy is fulfilled by TMN8 algorithm. Simulation results based on TMN8 codec, demonstrate that the performances of proposed algorithm achieves the same or even better outcomes inPSNR than traditional R-D based INTRA-update method. The computational complexity is also decreaseed significantly without preforming actual coding operations.The reference picture selection (RPS) methods for error resilient coding are researched in thesis. Based on extending concept of non-feedback RPS, a novel non-feedback key reference picture selection (KRPS) algorithm is proposed. Combined with proposed general transmission distortion estimation model and R-D optimal strategy, KRPS algorithm can determine which reference picture should be selected through comparing the distortion expectation of previous key picture and previous picture. This 'best selection' can limit error propagation between two key reference picture and minimizes error diffused distortion. Simulation results show that KRPS algorithm improves video quality efficiently and enhances the robustness of video coding observably.The temporal concealment and boundary matching methods are also studied in the thesis. A temporal concealment method that estimates lost motion vector by extrapolating motion vectors of MB and matching MB boundary is proposed. By estimating overlapped areas damaged blocks and the motion extrapolation MBs, the best estimation motion vector of damage block is obtained. After further choosing based on boundary matching, the corrupted block is concealed with the best suitable block in previous frame. This method overcomes the disadvantages of incorrect MB displac...