No-Reference Objective Quality Assessment For Networked Video

With the development of network and video coding technologies, the networked multimedia has been widely used in people's life. Since the video can provide intuitive and vivid information by stimulating human visual system, the networked video services have attracted extensive attention and achieved rapid development. As the high resolution video become more popular and the video standard H.265/HEVC is lunched, it raises a higher requirement for the transmission network. The IP network provides best-effort service and can't guarantee the quality of network service. Therefore, it becomes increasingly important to adjust the strategies of the video coding and transmission by evaluating the quality of the networked video, which is of great significance for improving the human experience. In this thesis, the quality assessment for networked video is studied, a parameter planning model and a content-adaptive packet-layer assessment model are proposed, respectively.The parametric planning model is designed for service planners to formulate a crude estimate of the perceptual video quality, which can avoid over-engineering the application and network. In this paper, a parametric planning model for video quality assessment combining the video coding and transmission characteristics is proposed. Firstly, the coding distortion is estimated according to the coded bitrate and the frame rate. Then, considering that the video transmission distortion is highly relies on the characteristics of both the video coding and the transmission network, the related impairment parameters, such as the frequency of the lost frame, the average number of the lost frame in one GOP, the average impairment rate of the GOP and the average impairment rate of one frame, are deduced by the statistical analysis method. Finally, the video quality combining the coding distortion and the transmission distortion is obtained.The packet-layer model allows the use of the packet header information to predict the video quality without resorting to any media-related payload information, which can enables real-time and detailed quality monitoring. To obtain more accurate quality of the H.265/HEVC encoded videos, a content-adaptive packet-layer model is proposed for the coding distortion assessment of the networked H.265/HEVC videos. Firstly, the coding parameters such as bitrate, frame rate and the bytes of each video frame are extracted in terms of the information provided by the packet header. Then, the frame type of each video frame is determined by the frame typed detection and the content complexity is estimated according to the characteristics of I frame, P frame and B frame. Finally, the content-adaptive packet-layer model for the networked video quality assessment is established by combining the video bitrate, frame rate and the content complexity. The proposed model can obtain a more accurate video quality with low computational complexity and is adaptive to different video content.