A No-Reference Video Quality Estimation Model over Wireless Networks

This paper proposes a no-reference video quality estimation model over burst loss wireless networks. The estimation model is an end-to-end, cross-layer framework which considers two parts as feature extraction and quality prediction. The first part considers content-aware parameters obtaining from rebuilt video sequences, transmission-aware parameters getting from network layer and encoding parameters which are set at application layer. Firstly, for the content-dependent parameters, the temporal and spatial features are extracted representing the videos' nature. Then, for transmission-aware features, Principal Component Analysis (PCA) is used to reduce the number of parameters to give high prediction accuracy in test with low training costs. Frame Rate (FR) and Sent Bit Rate (SBR) are selected as the encoding features which bring the effects from quantization. In the second part, Support Vector Machine (SVM) is provided by using all cross-layer parameters to make a tradeoff between accuracy and learning ability. Results show that the prediction values are well correlated with subject scores with Pearson coefficient of 0.86 at least.