Research On Packet Loss Resilience In Video Communications

In nowdays, with the development of IP based network techniques including radio IP, the IP video application has become one of the main services in communications. However, there are many kinds of errors in the transmission channel. For instance, the multi-path fading may cause the bit error in wireless channel; the congestion in IP channel may cause packet loss. Because of the bandwidth limitation, generally, only the compressed video data can be transmitted. The compressed data is very sensitive to the transmission errors and packet loss which may severely impair the reconstructed image. Therefore, the effective error resilience methods should be used to guarantee the quality of service of the video communications.In this dissertation, the video encoding and transmission techniques for packet loss resilience in IP networks are studied in detail. The contents of this paper are:1. In the study of channel coding, the overview of erasure codes is given which includes the research and applications of the codes in IP video communications. The main items and relating problems of erasure codes are analyzed also, i.e. the complexity and delay of encoding and decoding, the packetization overhead and erasure coding redundancy.Due to the fact that erasure correction efficiency of Tornado codes is critically affected by the short cycles in its bipartite graph, two algorithms are proposed: one is short cycle eliminating algorithm (SCEA) , which is used to reduce the number of the short cycles while generating the bipartite graph; The other is the short cycle effects removing algorithm (SCERA), which can take advantage of the characteristic of XOR computation and remove the decoding efficiency degradation in decoding process. The experiment results show that the proposed methods improve the erasure correcting performance of Tornado codes remarkably.To fulfill the requirements of real-time in video communications, and to make trade-off between the packetization overhead and the packet loss recover ability, the construction method of a novel [2m-1,m,3] code is proposed. The [9,5,3] code, one special example of the [2m-1,m,3] code, improves loss recovering capability in contrast with the [7,4,3] Hamming codes at a tiny redundancy cost. Combined with NAL of H.264, the packetization and transmission scheme is presented, which has the same capability of packet loss resilience against the burst packet loss as the RS code.2. In the joint source-channel coding, we present a novel partitioning method of video object that divides the video image into two objects: the foreground objects and the background objects. The macroblock encoding mode, decided by theRate-Distortion Optimization coding of H.264, is the object partitioning rule which obviously reduces the complexity of the general partitioning methods. By using the flexible macroblock ordering techniques of H.264, the video objects are encoded separately and enables the video objects to be decoded and transmitted independently.After the influences of the factors, such as macroblock mode selection, transmission environment and error concealment, on the reconstructed video are comprehensively analyzed, a video object-based macroblock adaptive Intra refreshing (OBMAIR) algorithm is presented. The proposed OBMAIR improves the quality of reconstructed video image by applying more Intra updating rates on foreground video object without any extra bit rate cost.Combined with the [9,5,3] code and OBMAIR mentioned above, we proposed a unequal joint source-channel coding algorithm which applies adaptive intra updating rates on different video objects and changes the number of check packet to be send with the packet loss rates. This algorithm has a good packet-loss resilience capability and is easily implemented.3. In the video post processing, a novel spatial error concealment is proposed. In the packet loss channel, the Intra frame of video sequence is divided by pixel lines into two fields, which are independently encoded and transmitted, by using the H.264 field coding techniques. When the image data in one field is lost during transmission, the pixels in its neighbor fields are used to interpolate the lost one. Experiment results show that the proposed method has much better concealment results than the error conceal method in the H.264。An image header. protection method is proposed, which uses the syntax of video picture enhanced information field. By using the duplications of current image header, we correct the erroneous image header according to the Majority Decision theory. The emphasized protection on the image header intensified the error resilience capability of the decoder. This method has been adopted in a video conference product.An analysis of the two calculation methods for the average PSNR is made both in theory and through experiments. The analysis shows that the PSNR calculated by the arithmetic average of MSE is better for measuring the video quality in error prone environments. We point out the weakness of the JVT's computation method of the average PSNR and improve the accuracy of the statistics value of a video sequence.