Video quality assessment by compact representation of energy in 3D-DCT domain

摘要

Video quality assessment (VQA) aims to pfedict the perceptual quality for improving the performance of practical application systems. However, the traditional methods consider the video as a sequence of two-dimensional images, which conflicts with the fact that a video signal is a three-dimensional volume data. This operation ignores the temporal information and results in a poor consistency with human perception. Hence, the paper presents a novel VQA model by exploring and exploiting the compact representation of energy in the three-dimensional discrete cosine transform (3D-DCT) domain. First, the video is transformed by 3D-DCT for every group of frame (GOF). Then three types of statistical features are derived from the 3D-DCT coefficients to represent energy compaction properties for simulating the process of human visual system (HVS). The parameters of the generalized Gaussian distribution (GGD) are estimated to imitate the marginal distribution of the 3D-DCT coefficients. Three energy ratios are calculated to depict how the video energy distributes over different frequency components. And the mean and variance value of absolute 3D-DCT coefficients are employed to measure the frequency variation of the video. Finally, the differences between the feature of reference video and the feature of distorted video are calculated to predict the quality score of the distorted video. Experimental results show that the proposed VQA method has a good consistency with human perception and is competitive with the state-of-the-art methods. (C) 2017 Elsevier B.V. All rights reserved.