Research On Evaluation Model Of IPTV Video Quality

IPTV is a kind of broadband television.Digital TV services are transmitted to subscribers through the broadband Internet protocol.With the growing television business and escalating customer demands,high-definition television programs gradually become the mainstream of IPTV development.Increasing high-quality high-definition programs in IPTV can significantly enhance user experience about video services.Meanwhile,operators can attract more users.Therefore,it is necessary to establish an evaluation system that can accurately measure video quality.It can help the video content providers,content delivery platforms,streaming media service platforms and major operators monitor and take actions in time to improve real-time video quality.At the present stage,there are two main methods to evaluate video quality: objective quality assessment and subjective quality assessment.Objective evaluation objectively scores the video quality according to a certain mathematical model.Subjective evaluation evaluates the image quality according to direct observation of human eyes.After comparing the applicability of these two methods,full reference quality model is chosen.Full reference model is the most accurate model in objective evaluation system.Traditional full reference evaluation models such as peak signal to noise ratio(PSNR)are gradually unable to meet the demand of increasingly rich video content and constantly updated video encoding format,thus proposed a new evaluation model: multi-element video evaluation model.Firstly,an experiment video set is established which can represent video quality under different network conditions.The observers then score the video subjectively,thereby create subjective impression scores for the video set.After establishing subjective impression score set,basic factors like Visual Information Fidelity(VIF),Detail Loss Metrics(DLM),Motion,and Video Quality Model with Variable Frame Delay(VQM-VFD),are selected to extract the objective features of video set.Through data preprocessing and feature selection,basic objective score set is established.Deeply customized open source project VMAF,the tool integrates Libsvm,Scikit-learn Python library,which is used for model training and comparison.Then train the training set with support vector machine(SVM)whose kernel function is Radial Basis Function,assign a certain weight to basic factors so the final model can retain the advantages of each basic model to get a more accurate final Score.Finally,use the test set to validate the model.By comparing the scores of Root mean squared error(RMSE),Pearson product-moment correlation coefficient(PCC)and Spearman‘s rank correlation coefficient(SRCC)of multi-element video evaluation model with traditional evaluation models and model‘s basic factors.Prove that multi-element video evaluation model has obvious advantages.Moreover,the experiment also compared multi-element video evaluation model with traditional evaluation models on three popular public data sets.In addition to the real-time video data set,which is inferior to VQM-VFD,the performance of multi-element video evaluation model is superior to traditional assessment models on all other data sets.Demonstrate that multi-element video evaluation model can provide a reference score closer to human perception.