Study On Fine Granularity Scalable Coding Technology And Transmission Technology

In this paper, we study the statistical characteristic of DCT residues, and achieved some progress on the relationship between Rate and Distortion of enhancement layer of FGS coding, the scan way of residuals and the transmission method. The research of this paper is focused on the following 3 aspects.First, we studied the relationship between Rate and Distortion of FGS enhancement layer. In FGS encoding process, the DCT residues, difference between original image and the reconstructed image in the base layer, are regarded as the encoding source of the encoder in the enhancement layer. When we study statistical distribution of DCT residues, we found the decreasing speed is too great in the tail of approximating distribution than the real DCT residues distribution if Laplacian distribution is used to model the DCT residues, and this will cause some mismatches in the tail of approximating model. The first proposed is one optimized RD based on single Laplacian distribution. The second proposal is a model based on mixed Laplacian and Uniform distributions. In order to slow down the dropping speed in the tail of the approximation distribution, we use the mixed distribution to model the real DCT residues distribution. Uniform distribution approximates the average of dropping speed in the tailing part of real DCT residues.And then we studied the streaming framework for compressed FGS bit-stream over the TCP/IP network. In order control the output rate of FGS encoder precisely; we used mixed linear approximating and exponential approximating method to model the relationship between Rate and Distortion for FGS bit-streams and proposed new rate controlling algorithm to adapt the variation of available bandwidth between streaming sever and the playback client. The first proposal is based quality tags and the jointed linear and exponential model and the second transmission method is based on the mixed Laplacian and Uniform distribution RD model and the joint Linear and Exponential approximation RD model. Experimental results show the two proposed method could smooth the variation of quality along the decoded picture sequences.Moreover, we studied the application scenarios when FGS coding systems are deployed in home network in which the bandwidth between any two end-points is variable and different. Based on the OSGI service framework and UPNP AV specification, we design one novel transmission system structure for FGS coding. In the proposed structure, all the devices in home could be seamlessly connected together through OSGI service gateway and their built-in services could be shared with other devices in home network through OSGI service framework. Integated with UPNP AV, device networking and media transport and control became feasible and efficient. New tranmssition system also has a better robusticity.The innovations in this thesis are listed as the below:1. We proposed one improved RD model based on the single Laplacian distribution, and the improved model has a better performance2. We used mixed Laplacian and uniform distribution method to model the real DCT residues distribution, and proposed one novel RD model for FGS encoder. The proposed model can characterize the relationship between R and D very well.3. We used mixed linear approximating and exponential approximating method to model the relationship between Rate and Distortion for FGS bit-streams and proposed new rate controlling algorithm to adapt the variation of available bandwidth between streaming sever and the playback client4. Based on the OSGI service framework and UPNP AV specification, we design one new novel transmission system structure for FGS coding. Proposed transmission structure has a better robusticity and performance...