Prediction of video perceptual quality in the presence of packet loss

There is an increasing interest in transmitting video services over IP network. As a result, packet loss becomes one important parameter in determining the service quality. In this paper, we focus on the packet loss problem of MPEG video streaming service and its influence on the observed quality (Quality of Experience). We introduce a novel NR (No Reference) objective method, which is impairment event based, to estimate the short-term perceptual video quality in the presence of packet loss. By this method, the packet losses measured in a network will be classified into GOPs (Group Of Pictures) to evaluate the impact of the deduced video impairment. The second step of this proposal estimates the overall video quality by taking the impairment pattern within the entire measurement period under consideration. We conducted a series of subjective experiments covering various factors of packet loss, e.g., loss quantity, loss type, loss distribution, and investigated their perceptual impacts on H.264/AVC videos. We explored the relation between the packet losses of a GOP and their deduced single impairment event. We found that the behaviour of video degradation according to different packet loss ratio can be described by a logistic function. We also examined the video quality with different times of impairment events occurring within a short period. We found that more than two times impairment events can lead the video quality to be unacceptable (MOS value below 2.5). Furthermore, a GLM (Generalized Linear Model) model has been proposed to estimate the video quality in the case of dual impairments.