| The Internet of today is a packet-switched, best-effort and heterogeneous communication infrastructure, which does not promise any guarantee on bandwidth, real time service and quality of service (QoS). Hence, for realtime service such as video conference and voice over Internet Protocl (VoIP), it is difficult to meet the requirement with the existed network structure alone. Multiple description coding (MDC) is a robust coding scheme for solving packet loss and error. This scheme provides satisfying QoS and real time service without changing the existed network structure, which obtains many researchers'interests in recent years.This paper focuses on studying and designing effective MDC scheme on image and video, which is the key to multimedia communication. For MDC on image coding, four different schemes are proposed to improve the state-of-art MDC scheme based on quantization. The four schemes are, MDC scheme based on stagger quantization, MDC scheme with optimized deadzone region of quantizer, MDC based on special lattice vector quantization and two-stage MDC scheme with TCQ. Then, the other two multiple description schemes on image coding are proposed. The first is based on classification, which can be seen as a different way to generate MDC system. The second is two-stage multiple description image coding with diversity, it has the best performance in the existed literature. For video coding, a scheme with redundancy tuning on marco-block level is proposed, which is compatible with H.264/AVC standard. The details are as follows.â‘ Four different schemes are proposed to improve the MDC coder based on quantization. Firstly, an MDC scheme based on stagger quantization is proposed. In this scheme, the side descriptions are formed by scalar quantizer with the same quantization step and different deadzone widths. When more than one description is received, the staggered quantizer is employed to reconstruct the signal with better performance than that of side description. By tuning the step and deadzone of the quantizer, different redundancy can be obtained, which is very simple in implementing and can be extended to more than two-description case. Secondly, to improve the MDC scheme with polyphase subsampling and different quantizers, we propose a fused quantization scheme that fuses any two different quantizers to form a new finer quanizer. With this fused quantizer, a better central performance can be obtained. According to the statistics of the source, an optimized scheme is proposed to get the best deadzone width of the quantizer to improve the central performance. Then the whole scheme is generalized to more than two description case. Thirdly, a three-channel MDC scheme based on lattice vector quantization (LVQ) is proposed to extend the existed two-channel MDC scheme based on LVQ. In this scheme, Z3 lattice is employed for its simplicity.. The scheme can be extended to other LVQs as well. Then an MDC image coding scheme is realized using this scheme. Finally, a two-stage MDC scheme with trellis coded quantization (TCQ) is proposed, in which the redundancy tuning is flexible and the performance is better due to the good performance of TCQ. In addition, an MDC image coding scheme is proposed by employing the quadtree classification scheme with TCQ.â‘¡A classified multiple description image coding scheme is proposed, in which the wavelet coefficients in each class can be processed separately according to its own source distribution. All the classes will be divided into two larger classes again. One of them is quantized by MDSQ and the other is subsampled by quadtree and quantized subsequently. It is the number of class quantized by the above two ways that determines the tradeoff between side and central performance. Most important of all, by solving the deadzone problem, the central performance can get improved even though the side performance is close to its corresponding single description one.â‘¢A two-stage diversity based MDC image coding scheme is proposed. Different from other MDC schemes, the proposed one employs the inherent redundancy in the encoder and the signal, thus generates the MDC streams without introducing extra redundancy. Therefore, the proposed central performance can be better than its side performance even when the side one is close to that of its single description. For other MDC schemes, however, the central performance will be the same as its side performance without any improvement at this case. Due to the employing of two stages, the redundancy tuning is flexible and it is easy to be extended to more than two description case. Most important of all, the reconstructed residual of the second stage that contributes to the central reconstruction has been testified to be useful to the side performance as well.â‘£A novel approach to insert and control the redundancy for MDC video coding scheme is proposed, which is implemented at marco block level. The basic idea behind this approach is to determine the relative importance of each macro block of the compressed video stream, which is achieved by practically analyzing the error propagation paths that depend on the video sequence itself and the adopted video encoder. This allows unequally protecting the video data according to the importance of the macro block and to the network status, consequently permitting better exploitation of the redundancy. Then the proposed scheme is implemented in H.264/AVC by employing its coding options, thus generating descriptions that are fully compatible with 11.264/AVC syntax. Moreover, a simple yet effective approach for the variable rate coding case has been devised based on the optimal selection of the quantization parameter. Finally, the corresponding redundancy tuning scheme for the known and unknown packet loss rate are proposed. The results are tested at Bernoulli and burst packet loss modle with better performance.
|
PDF Full Text Request