Video quality assessment based on motion dissimilarity

Algorithms for video quality assessment (VQA) attempt to predict the quality of a video in a manner that agrees with subjective ratings of quality from human observers. In this paper, we present an approach to VQA that combines the frame-based distortion measurement with a spatiotem-poral analysis of motion dissimilarity. Our VQA algorithm, FDMD, applies our previous IQA algorithm, MAD [Larson and Chandler, 2010], to compute the frame-based distortion and develop a spatiotemporal model to capture motion dis-similarity through the STS images. The overall estimate of video quality is then determined based on a combination of these results collapsed across space-time. Testing on vari-ous video-quality databases demonstrates that FDMD per-forms well in predicting video quality. 1