No-reference cross-layer video quality estimation model over wireless networks

This paper proposes a no-reference cross-layer video quality estimation model for low-resolution video over wireless networks. The estimation model contains two parts as feature processing and quality prediction. The first part covers the content-aware features, network layer features and application layer features. As for the content-awareness, by using the weighted Euclidean distance, the temporal and spatial features are extracted and coalesced into one parameter representing the complexity of video. For network layer features, Principal Component Analysis (PCA) is used to reduce the dependence among parameters to make a trade-off of accuracy and costs. Frame Rate and Sent Bit Rate are selected as the application features. The second part is provided by Support Vector Machine (SVM) using all cross-layer parameters to make an excellent balance between accuracy and learning ability. Results show that the prediction values are well correlated with the mean opinion scores (MOS) associated to the subjective assessment.