Efficient Video Quality Assessment via 3D-Gradient Similarity Deviation

• Published in: Intelligence Science and Big Data Engineering. Image and Video Data Engineering Vol. 9242; pp. 493 - 502
• Main Authors: Jia, Changcheng, He, Lihuo, Lu, Wen, Hao, Lei, Gao, Xinbo
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2015; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: 3D-gradient; Graphics programming; Image processing; Pattern recognition; Video quality assessment; Worst-case pooling
• ISBN: 9783319239873; 3319239872
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-319-23989-7_50

The description of the spatio-temporal distortion is highly important for video quality assessment (VQA). However, traditional VQA metrics do not have enough ability to capture the spatio-temporal distortion in the video and suffer high computational complexity. Hence, this paper proposed a novel efficient VQA metric via 3D-gradient similarity deviation. Firstly, 3D-gradient is introduced to extract features of the spatio-temporal distortion in the video. In that, 3D-gradient kernels in three directions are constructed to convolute a group of frames in a video sequence to obtain 3D-gradient blocks. And then the 3D-gradient similarity indices between the reference and the distorted video are calculated to describe the local degradation of video quality. After that, the standard deviation of the local gradient similarity map is calculated to predict perceptual video quality of a group of frames. Finally, the worst-case pooling strategy is applied to pool all the quality indices of the groups of frames into a final quality score. Experimental results show that the proposed metric has a good consistency with the subjective perception and perform better than traditional metrics.