Date of Publication :4th March 2024

Abstract:— A key challenge in no reference video quality assessment (NR-VQA) is how to effectively mimic human visual system (HVS) in a data-driven way. A spatiotemporal Slice no-reference video quality assessment model using 2DLOG Filter and Support Vector Machine based on frame-level unsupervised feature learning and temporal pooling is presented in this paper. Given that the spatial and temporal channels of the human visual system both include the second derivative of the Gaussian function, first a two-dimensional LOG (2D LOG) filter was constructed to simulate a human visual filter and to extract perceptual-aware features for the design of VQA algorithms in order to filter the STS images and use support vector machine (SVM) to perform a successful NR video quality evaluation in this dissertation. Secondly, the successful features of STS images of video such as the statistical feature maps, orientation feature map from the gradient magnitude and filtering response of Laplacian of Gaussian were extracted to characterize the motion statistics of videos. Finally, the extracted perceptual features were fed in SVM to perform training and testing. The performance of proposed algorithms shows that the methods are better than that of most mainstream VQA methods.

Reference :