Spatio-Temporal Interactive Laws Feature Correlation Method to Video Quality Assessment
Kuan-Hsien Liu, Tsung-Jung Liu, Hsin‐Hua Liu, Soo‐Chang Pei · 2018
In this work, we proposed a full-reference method to estimate video quality. First, we decompose the video into one spatial image and two spatiotemporal slice images. Then for each one of them, sixteen Laws texture filters are applied to generate nine different Laws feature maps. In order to compare the similarity degree of these feature maps obtained from both original and distorted videos, we compute the two-dimensional correlation coefficients. Since the correlation coefficients are computed for each frame and spatiotemporal slice, we only choose four statistical values to represent them to reduce the complexity. Lastly, the regression approach is chosen to learn the mapping relationship between the selected features and subjective quality scores. The extensive experiments in the LIVE Video Quality Database suggest our proposed video quality assessment model has superior correlation performance with human visual perception than other state-of-the-art methods.