Quality Assessment For Adaptive Streaming Based On HTTP

With the development of the network and the multi-media technology, there is a popular tendency to watching video via the Internet. Dynamic adaptive streaming over HTTP(DASH) has become the most-promising technology in the video-demand domain owing to its reliability, adaptability and realizability. However, there is still no well-developed method to evaluate the influences of the rebuffering and quality fluctuation on the quality of DASH. The thesis analyzed the impact of these influential factors to the users’ subjective experience during their watching period, based on this analysis, two objective quality assessment models were established for evaluating the influence of the rebuffering and the quality fluctuation events of the DASH video, respectively.The thesis analyzed the impact caused by the rebuffering event to the DASH video quality firstly. Considering the status characteristics of the rebuffering event, the rebuffering event were divided into two stages: rebuffering stage and recovery stage, and then determined the influential factors of each stage. For the rebuffering stage, the influential factors were video perceptive quality and the length of rebuffering event. As for the recovery stage, the influential factors were determined as the initial perceptive quality of this stage, the length of the recovery stage and the video code quality. To check the relationship between the quality of user experience and each factor, a subjective test was carried out. Utilizing the results of subjective experiments, the objective video quality models of both the rebuffering stage and recovery stage were established.Moreover, the influence of video quality fluctuation on user experience was studied in this paper as well. The single video quality fluctuation event was also divided into two stages: quality fluctuation stage and quality stationary stage. In the quality fluctuation stage, the influential factors included the video perceptive quality and the level of the fluctuation. While in the stationary stage, the influential factors were determined as the intial perceptive quality of this stage, the length of this stage and the video code quality. To determine the relationship between the quality of user experience and each factor, another subjective test was carried out for investigating each stage. The final videoquality models were established according to the results of subjective experiments.User experience in each event would be affected by the result of last event, what was mean that the user experience should be acquired by analyzing the accumulation result of every rebuffering and quality fluctuation event from the very beginning. Therefore, the multiple rebuffering and quality fluctuation events needed to be considered in reality. This thesis further expanded the proposed two single event models to multiple events models. The verification test results showed that the models proposed in this thesis had excellent coherence with the subjective test results. The models could estimate the DASH video quality accurately.