Video quality assessment via supervised topic model

Video quality assessment (VQA) plays a very important role in many video processing and communication systems. Since video signals are ultimately delivered to human observers, an accurate objective video quality metric should agree well with judgment of human visual system (HVS). In this paper, a novel full-reference VQA scheme is developed to measure the perceived video quality in both local and global aspects. First, to account for the crucial impact of motion on perception, effective quality features are extracted from the local spatio-temporal volumes which are generated around the motion trajectories in the video. Second, a statistical model is utilized to discover the latent relation between local quality and global perceived quality. Experimental results on LIVE database demonstrate promising performance of the proposed metric in comparison with state-of-the-art VQA metrics.