A perceptually motivated no-reference video quality assessment algorithm for packet loss artifacts

Packet loss artifacts are perhaps the most commonly occurring distortion type in video streaming applications. We present a perceptually motivated no-reference video quality assessment (NR-VQA) algorithm for assessing the quality of videos subject to IP and wireless distortions. The proposed algorithm is composed of a spatial quality assessment stage, a temporal quality assessment stage, followed by a spatio-temporal pooling stage. Each of these stages are perceptually motivated - the spatial stage is inspired by the sparse representation of natural scenes in the human visual system, the temporal stage is motivated by optical flow statistics, and the pooling stage by the sensitivity of the human visual system of spatio-temporal stimulus. We show that the spatio-temporal pooling results in significantly higher performance relative to the stand-alone performance of the spatial and temporal assessment stages. The performance of the algorithm is shown to be promising on a subset of the LIVE dataset.