Blind Video Quality Assessment Based on Human Visual Speed Perception and Nature Scene Statistic

In this paper, we incorporate human visual speed perception model into a NSS based VQA method for video sequences transmitted via wireless network. The human visual speed perception contains two parts: one is motion information, which is calculated by using the prior probability distribution of the relative motion in each frame of the video; the other one is the perception noise, derived from the background motion. The weighting factors are defined as perceptual information that minus perception noise from motion information. We extract both spatial and temporal statistical features in videos (NVS-S and NVS-T), and measure their deviations from pristine statistical features. Consequently, the deviations can be synthesized with perceptual information based weighting coefficients to get the video quality score. The proposed blind VQA model is trained and tested in the LIVE database and EPFL-PoliMI database. The experimental results indicate that our model outperforms other blind VQAs.