Research On Video Transcoding Techniques For Mobile Video Terminals

With the development of network and multimedia communication technology, the necessarity of delivering varied video information over different network is increasing, because of the diversity of network and terminal equipments. The encoded video information may not be the most optimal. In order to transmit video streams to those terminals with different capabilites of computation and display, video steam must be dynamically adjusted according to the user environment. Video transcoding is this kind of technique to change fhe format of video among different video coding standard, bit rate or resolution, so as to meet the special requirement of the transmission channel and decoder. There are two main problems in video transcoding. Firstly, the computational complexity should be low to meet the requirements of real-time processing. Secondly, because of the reduction of video resolution, the loss of important detail information will greatly affect the user's visual experience. Thus, region-of-interest based video adaptation should be developed.After a brief introduction of video coding standard, the techniques of video transcoding are overviewed. By taking the characteristics of mobile video terminals into consideration, this thesis puts emphasis on the spatial-downscaling transcoding and ROI-based transcoding. The main work of this thesis is summarized as follows:Firstly, for the spatial resolution reduction of H.264/AVC stream, a motion re-estimation algorithm based on linear regression model is proposed. It exploits the correlation between the motion vectors of original video stream and those of down-sampled video, which is modelled by linear regression model to obtain the estimated motion vectors. Experimental results demonstrate that the proposed approach can significantly reduce the computational complexity of the transcoder with only slight sacrifice of visual quality.Secondly, to display football sports video on mobile terminals, a video adaptation scheme is proposed for football sports video on mobile terminals. For football sports video on mobile terminals, the detailed information is easy to be lost due to its small screen size and low resolution, and the visual quality is degraded. According to the motion features, the football object is rapidly located and it is tracked in successive frames by utilizing the properties of football object and the relationship between successive frames. Moreover, the position and size of ROI which contains football is adaptively adjusted. Simulation results demonstrate that the proposed approach can accurately locate the football object and compromise well between processing speed and visual quality.