Research On Wavelet Transform Based Multiple Description Image Coding

With the rapid development of wireless network techniques, packet losing has emerged as a serious problem which threats the security of transmission, real-time performance and effectiveness. Much attention has been paid on seeking for effective schemes to ensure the quality of transmission. As an efficient way to solve this fading problem, Multiple Description Coding (MDC) has been developed rapidly in these years. In this dissertation, we investigate the applications of wavelet transform in the MDC techniques. As mathematical microscope, wavelet has been used in most kinds of image and video coding standards and plays more and more important role in all kinds of information processing application areas. An extensive study of wavelet coding in MDC is described in this dissertation. The MDC techniques based on embedded and non-embedded wavelet coding are proposed. Besides, a MDC algorithm based on wavelet and vector quantization (VQ) is also designed.Firstly, in the study of embedded wavelet coding algorithms based MDC techniques, the multiresolution property of the wavelet images has been deeply exploited. New MDC framework based on odd-even division and the redundancy reconstruction methods is proposed at first. Combined with the Set Partitioning In Hierarchical Trees (SPIHT) embedded coding algorithm, novel embedded MDC algorithms are introduced. The first proposed algorithm is to construct a three-description coding frame by using the redistributed DCT image with a three-layer wavelet image structure. In the second algorithm, we propose a new embedded coding algorithm in light of the new distributed image which has the similar energy distribution as the three-layer wavelet image but has different structure. Excellent performance oberved in simulation results demonstrates the efficienty of these two algorithms. As both our principle and basic coding theory come from wavelet, these two proposed algorithms can be viewed as embedded"pseudo wavelet"coding, and the multiple description coding based on which should be catergoried into the family of embedded wavelet based schemes.Secondly, non-embedded wavelet coding based MDC schemes are peoposed. We develop the Space-Frequency Quantization (SFQ) based MDC method creatively. According to this joint optimization model of zero tree and scalar quantization, two algorithms are proposed. One is an odd-even sub optimization based two-description algorithm. The other one is an orientation information fine and coarse quantization based three-desciption coding algorithm. Good simulation results also show their superiority.Finally, different from the former lattice vector quantization based MDC schemes, the wavelet image and VQ are combined, based on which two new MDC schemes are proposed. The first is orientation wavelet vector based MDC algorithm. The other is the wavelet tree structure vector and"coarse and fine"code book introducing redundancy method based MDC algorithm. Simulation results show their efficiency. The sucessful applications of these two algorithms solve the index distribution problem of VQ in MDC from a new prospect, they have given some other ideas to avoid this problem and shed new lights in this kind of research.