| In recent years,with the rapid development of mobile communication technology,mobile video services have shown an explosive growth trend,and various mobile video services are constantly emerging.How to ensure the users' Quality of Experience(QoE)of mobile video services has become a key issue that requires to be addressed in the mobile video industry and the entire mobile Internet industry.To this end,QoE of mobile video services needs to be accurately evaluated.There are many factors affecting user's QoE,including subjective factors and objective factors.The essence of QoE assessment lies in establishing the mapping relationship between various influencing factors and the user's final subjective feelings,and characterizing the internal relationship between QoE and its influencing factors.Due to the interaction between subjective and objective factors of QoE,it raises a great challenge to establish a QoE assessment model.In this paper,the research work of QoE evaluation method of mobile video has been carried out in depth.Using the machine learning method,three QoE assessment models of mobile video services have been established.The specific research contents include:Firstly,an H.264 video quality assessment model based on video content propoerty has been proposed.The bitstream structure of H.264 is analyzed deeply,and Quality Parameter(QP),Motion Vectors(MVs),coding type of the macroblock in the video frame,ICT coefficients and other bitstream parameters are extracted from H.264 bitstream.Next,the proposed bitstream parameters are processed,and two indicators describing the property of the video content are proposed,which are defined as motion complexity and texture richness.Compared with the currently widely used TI and SI indicators describing the temporal and spatial property of video,the proposed indicators can better characterize the temporal and spatial property of video and are consistent with the subjective perception of the human visual.Finally,these two video content property indicators are combined with different H.264 bitstream parameters to form a feature parameter vector.Taking this as input and subjective MOS value as output,a video quality assessment model based on H.264 is established by using Deep Neural Network(DNN)to train a data set containing 180 videos.The experimental results show that Pearson Linear Correlation Coefficient(PLCC)and the Spearman Rankorder Correlation Coefficient(SROCC)can reach 0.9941 and 0.9893 respectively,which proves that the model can accurately predict the QoE of H.264 video bitstream.Secondly,a QoE assessment model of mobile video based on DASH transmission protocol is proposed.The model uses DASH transmission protocol to extract the initial buffer time of the video,the number of times and time of the pause,the number of times that the bitrate is switched up or down,and the video content characteristic parameters-motion complexity and texture richness.The feature parameter vector is formed as an input,and MOS value is used as the output.The DNN is adopted to establish a mapping model between the input feature parameter vector and the output,and the QoE assessment model is obtained by training on the public Waterloo SQoE-III data set.The experimental results show that PLCC and the SROCC of this model can reach 0.9632 and 0.9574 respectively,which proves that the model can accurately predict the QoE of mobile video.Finally,a client video quality assessment model based on deep spatiotemporal characteristics is proposed.The model adopts two DNNs of 3DCNN and LSTM(Long Short-Term Memory)together to extracts the deep spatiotemporal characteristics of the viewing video at the client side as the input feature parameter vector,and uses MOS value as the output to establish the assessment model of video quality.The experimental results show that the predictive accuracy of the model can reach 99.16% and the RMSE is only 0.1104,which proves that the model can accurately predict the viewing video quality at the client. |
PDF Full Text Request