Abstract
In this thesis, we present several novel GOP (Group of Pictures) structures by reorganizing the coding dependencies among frames in a GOP. In the conventional GOP structure, each frame except I-frame is predicted by the previous one. Once any of these frames is requested, we need to transmit not only the requested frame, but also all frames before the target frame. Therefore, when a client performs VCR operations such as Fast Forward Scan (FFS) or Jump, there may be only few frames for display, but all frames in a GOP have to be sent and decoded. Besides, the transmission rate of these frames needs corresponding to the playback rate so that every required frame can be shown on the screen in time. Consider the video streaming systems, if numerous users request for VCR operations at the same time, it may cause heavy server load. However, most of these transmitted data are redundant but inevitable. In our GOP structures, no redundant frame will be transmitted in most of cases when a VCR operation is performed. Thus even for the VCR operations with a high-speed factor, the server still sends data at normal transmission rate. The video server load and network bandwidth can significantly reduce as well. In addition, the proposed structures also provide error resilient mechanism. Similar motion or pixel information can be reconstructed without feedback information if packet loss or packet damage occurs during transmission. And error propagation phenomenon caused by coding dependencies among frames will reduce as well.