Abstract
Both mobile and egocentric videos contain much larger motion than broadcast videos. To estimate the quality of videos with huge motion, the temporal pooling strategy should be adaptive to the specific content. Existing methods focus more on the values and variations of frame quality scores and ignore the masking effect of motion. In this paper, a temporal pooling strategy using a visibility measure is proposed to estimate the quality of videos containing large motion, where the imperceivable details during motion are not considered. We then introduce a strategy to measure the influence of measured visibility on pooling and design a subjective test to gather data for the strategy by synthetically creating shaky videos. Our pooling method is demonstrated to be more effective than existing strategies at pooling frame scores estimated by different image quality metrics.