Packet-based PSNR time series prediction for video teleconferencing

It has been shown that various statistics of the peak signal-to-noise ratio (PSNR) time series of a video sequence can be used to construct fairly accurate Quality of Experience (QoE) models. To predict QoE, it is sufficient to predict the PSNR time series. The possibility of predicting QoE further enables QoE-based network resource allocation. We propose two approaches to packet-based prediction of PSNR time series to overcome the limitations of frame-based approaches. The proposed first approach adopts a parametric model for the impact on the video quality due to losing a packet, while the second proposed approach is parameter-free. Simulation results show that both approaches significantly outperform the simple mean or median algorithms and are close to their respective performance bounds.