Study On Multiple Description Of The Image And Video Coding

Multiple Description Coding (MDC) offers an image and video coding framework with very strong error resilient and error concealment ability. It can well conquer the data error or packet loss problem due to the Internet or wireless network channel failure. This thesis first reviews the development of MDC, concentrates its study on Multiple Description (MD) subband image coding and Multiple Description video coding, and then propose 3 novel MD subband coding schemes and one 3D MD video coding scheme.The first MD image coding scheme proposed is based on the individuality of the subband coefficient, named Structure Unanimity MDC image packeted coding. In this method, each significant wavelet coefficient is split into odd position and even position coefficients by bitwise down sample, and form odd phase and even phase. The coded odd position coefficients and the protecting data for the even position coefficients form one description, and the coded even position coefficients and the protecting data for the odd position coefficients form the other description. Two descriptions are dispatched from different channels. Since the main information and redundancy have the same structure, the code of redundancy structure can be eliminated to improve the coding efficiency. Two descriptions are then separated into packets by row and column respectively to reduce the distortion introduced by cross packet loss, and improve the error resilient and error concealment ability.The second MD image coding scheme presented is Unequal Error Protected MD image subband packeted coding, in which, the information of the wavelet transformed image can be divided into two parts, the structure information and residues of significant coefficients. The structure information, including positions, signs and bit lengths of significant coefficients, is crucial and should be duplicated into both descriptions. The residues are mapped into couples of indices by a multiple description scalar quantizers, and sent to separate descriptions. Two descriptions are dispatched from different channels. Two kinds of packet methods, simple packetedmethod and cross packet method are used to resist the cross packet loss. The algorism is not only of high coding efficiency and strong anti-packet-loss ability, but also of convenience in the redundancy rate adjustment.The third novel MD image coding scheme proposed in this paper is Layer Spanning Tree Based MD Image Coding. It provides flexible protections according to the importance of information. Based on the character of the embedded coding, a novel data structure called Layer Spanning Tree (LST) is presented and two kinds of MDC methods with high coding efficiency and strong packet loss resistant ability are devised based on the LST. This scheme can offer flexible error protection according to the significance of the code.Finally, a 3D lifting wavelet based MD video coding scheme with motion compensation is presented by the integration of the structure unanimity MD image coding and the motion compensate 3D lifting wavelet coding algorithm. This method has high compression efficiency and strong anti-packet-lost ability.Experiments have testified the correctness and validity of these proposed algorithms. They offer reference schemes for a high efficient and robust image and video coding.