Quality of Experience Assessment for Free-viewpoint Video

The evaluation of free-viewpoint video (FVV) quality is essential for improving the quality of experience (QoE). Prior deep video quality assessment (VQA) approaches for FVV typically focused on either spatial or temporal distortions and lacked a comprehensive assessment considering the two aspects. In this paper, we provide an end-to-end no-reference video quality assessment (NRVQA) model for FVV that predicts video quality scores based on both spatial and temporal features. It consists of a spatial feature perception module, a temporal motion feature perception module and a quality score fusion module. In order to provide a quality score that is highly relevant to the mean opinion score (MOS) from the subjective quality assessment experiment, the quality-related features in the spatial and temporal domains of FVV are effectively utilized and merged. Experimental results show that the PLCC and SRCC improved by 25.0% and 18.5%, respectively, compared to state-of-the-art method. Moreover, the ablation experiments demonstrate the importance of both spatial features and temporal motion features.