QoE Perception And Optimization Of Video Service Over Multi-layer Overlay Networks

Multilayer overlay network allows mobile devices accessing multi-ple network at the same time. This indicates that it's possible to ensure the Quality of Experience (QoE) of video service through the cooperation between networks. Nevertheless, because of the uncertainty of network status and the complexity of access method, the difficulty of network re-source management will also increase. It's a topic worth studying that,according to the descending QoE value, how to timely adaptively adjust the distribution of network resources through a feedback mechanism to optimize it in this scenario.In order to study the video service QoE perception method that in line with the human's subject perception, this thesis first analyzes the visual perceptual selection mechanism, models the sight trajectory with maximum entropy random walk theory, proposes three steps i.e. su-per-pixel segmentation, two-tier neighbor searching and belief diffusion to improve accuracy and reduce computational complexity, and finally establishes a saliency detection method which is consistent with human perception.Based on this, this thesis detects multidimensional Region of Interest(ROI) weighted characteristics of frames, and joint characteristics of each frame with different weight. Then ensemble learning methods are adopted to training this video QoE perception model, which makes the predicted results basically consistent with the subjective scores.After the accurate perception of QoE, a bionic attractor component based optimization method is proposed. This method adaptively adjusts the video traffic proportion in each access network according to the changing of QoE timely, eventually find an appropriate allocation propor-tion to optimize video QoE.