VCFNet: Video Clarity-Fluency Network for Quality of Experience Evaluation Model of HTTP Adaptive Video Streaming Services.

In HTTP adaptive video streaming service, video clarity and fluency are the two most important Influencing Factors (IFs) affecting user’s Quality of Experience (QoE). In this paper, a Video Clarity-Fluency Network (VCFNet) is proposed to establish a QoE evaluation model, which focuses on characterizing the clarity-fluency characteristics of video. Firstly, the Harmonic-ResNeXt101 network is constructed by introducing the Harmonic Network into ResNeXt101 to capture the clarity information of video frames. The output of the Fully Connected (FC) layer of the Harmonic-ResNeXt101 network is fed into the Gated Recurrent Unit (GRU), which is used to perform short-term temporal modeling to capture the fluency information of video chunk. Then, the final output of GRU is extracted as the clarity-fluency features, which are concatenated with the statistical features of other IFs (including video quality level, re-buffering duration, re-buffering frequency, etc.) to form the feature parameter vector of IFs. Finally, a neural network composed of One-Dimensional Convolutional Neural Network (1D CNN) layer and two FC layers is designed to establish the mapping relationship model between the feature parameter vector of IFs and Mean Opinion Score (MOS) to predict user’s QoE. Experimental results on SQoE-III and SQoE-IV datasets demonstrate that the proposed VCFNet can effectively capture the clarity-fluency information of video, and the resulted QoE model can achieve the state-of-the-art performance compared with the existing QoE evaluation models.