Abstract
Fine granularity scalability (FGS) is a useful technique to support video streaming through heterogeneous networks. According to available network bandwidth between the sender and the receiver, the FGS enhancement layer bitstream can be truncated at any point to provide relative video quality. By utilizing the characteristic of enhancement layer bitstream, this thesis proposes three schemes to smooth the quality of FGS video. Our target is to reduce the inter-frame and intra-frame quality variation. We propose to label each bit-plane according to its quality level so that the server can decide how many bit-planes should be transmitted to keep every frame has the same quality. Also, we change the block transmission order according to its sum of DCT residuals. By this approach, the lower quality parts in a frame can be enhanced earlier. We also present a predicting method by referring to the enhancement layer data. This method not only smoothes the video quality at low bitrate but also improve the coding efficiency of FGS. Experimental results show our methods take greatly effects on reducing the inter-frame and intra-frame quality variation. Further, the computational complexity of these three methods is very small so that our methods can be applied to the real time constrained systems.