Research On Content-aware Network Transmitted Video Quality Estimation And Resource Allocation Scheme

Real-time video transmission is emerging as one of the most potential growing applications in the Internet. However, today's best-effort networks can only provide limited quality of service. Excessive delay, high delay jitter and packet loss will impair the video transmission performance due to insufficient network resource. Based on the nature of different importance existed in the video bitstream, the video quality can be effectively improved by unequally allocating available resource between video data. Resource allocation for video transmission depends on three key components: modeling the importance of different video data which is treated as resource allocation unit; estimating the end-to-end video transmission distortion; according to the importance of the video data, finding the optimal resources allocation strategy to minimize the end-to-end video transmission distortion. In order to achieve better utilization ratio of the resource, in this dissertation, we research on content-aware based importance calculation for different video data and end-to-end video transmission distortion estimation, also, we study the bandwidth allocation scheme at transport layer and packets scheduling algorithm in Diffserv enabled network.Firstly, video data in one frame usually be treated as the basic resource allocation unit, however, estimating the importance of video frame use the coding type or the position in the GOP (Group of Picture) can not get accurate result because of the distortion caused by loss of different video frame with variant content may be different. We propose an importance calculation scheme for H.264 coded video frame by using coding parameters and weight estimation, called RFDI (Relative Frame-lose-induced Distortion Index). The coding parameters of the macroblocks, including coding pattern, motion vector and reference picture selection are used to determine weights for the macroblocks. The weights can reflect the loss sensitivity of the macroblocks and are further used for calculating the importance of the video frame in the GOP. The proposed scheme can achieve better accuracy when estimating the relative importance of video frames. Because the parameters used in the scheme can be extracted directly from the video bitstream, the computation complex is quite low.On the other hand, before transmission, video frame is usually divided into several packets to achieve better error resilience. Modeling video importance at frame level ignores the fact that the packets belong to the same frame may have different importance. To overcome this limitation, a video packet importance estimation scheme is proposed based on macroblocks classification. This scheme first classifies and encodes the macroblocks based on their importance, and then estimates the importance of the video packet. The benefit of this approach is that the resource can be allocated more effectively according to the importance of the video packets.Secondly, one of the aims of the resource allocation is to minimize the end-to-end video transmission distortion. However, an accurate estimation of end-to-end video transmission distortion requires very complex computation. In order to reduce the computation overhead, a scheme which estimate the video transmission distortion by modeling the error propagation is proposed. This scheme calculates the distortion caused by error propagation by introducing "Error Propagation Index", avoids the high complexity brought by recursive calculation. Our scheme establishes the mapping relations between packets loss and actual video data loss by take both packetization strategy and error handling manner in the decoder into consideration. Simulation results show that the proposed scheme can estimate the end-to-end transmission distortion accurately.Finally, based on the previous research work, resource allocation algorithm for video transmission is further studied, including two aspects: First, a RFDI model based FEC redundant allocation scheme is proposed. This scheme finds the optimal FEC redundant packets allocation strategy between the BOP (Block of Packets) in the GOP to minimize the video quality degradation. Second, when streaming over the Diffserv network, the traditional two rate three color marking (trTCM) algorithm will mark the important video packets with low priority when the network overload is high. In order to avoid this, we improve the traditional trTCM packet marking algorithm. By using the proposed video packet priority estimation scheme based on macroblock classification, the importance of the content in the video packet is jointly considered when marking the packets. The improved trTCM algorithm can reduce the drop probability of the important video packets when the network is congested.