Tucker Decomposition and Log-Gabor Feature-Based Quality Assessment for the Screen Content Videos

With the rapid development of mobile communication and digital multimedia technology, the evaluation metric dedicated to visual information processing toward the screen content is in urgent need. In this article, a full-reference video quality assessment (VQA) model, called the tucker decomposition and Log-Gabor feature-based model (TDLGM), is presented for the perceptual quality evaluation of the screen content video (SCV). Generally speaking, the human visual system (HVS) is highly sensitive to the abrupt edge information and irregular motion information inherent in the SCVs, and it contains huge visual redundancy between the successive video frames during the HVS perception process. With these motivations, the dimensionality reduction operations together with feature extraction based on the HVS perception are adopted in the proposed TDLGM. Specifically, tucker decomposition is utilized to fuse spatiotemporal information for generating the luminance and chrominance spatiotemporal information slices of the SCVs. Followed by the spatiotemporal confluent Log-Gabor features of these slices are extracted to calculate the similarity between the reference and distorted SCVs. Finally, the luminance and chrominance quality scores are combined to arrive at the distorted SCV quality scores. The experimental results on the SCVD and CSCVQ databases prove that the proposed TDLGM is higher consistent with the HVS perception on the SCVs, compared with multiple classic and latest IQA/VQA models. In specific, it achieves the best results on the weight average performance of these two commonly-used databases, outperforming the state-of-the-art model-HSFM with 2.3289% in Spearman rank order correlation coefficient (SROCC). © 2024 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.