Blind Image Quality Assessment Based on Human Visual Perception

The human visual system (HVS) is important for guiding the blind image quality assessment method (BIQA). Inspired by the free-energy principle, an NR-IQA method that simulates human visual perception is proposed. The whole model consists of an image restoration network and a multi-stream quality prediction network. Firstly, the distorted image is fed into the image restoration network to generate the restored images and discrepancy map, in which the quality perception constraint and the structural similarity discrepancy map-based constraint are both considered during the optimization to improve the recovery performance. Then, the distorted images, the restored images, and the perceptual discrepancy maps are utilized as inputs for the multi-stream quality prediction network to obtain their fused features. Finally, the fused features are input into the patch-based attention module to obtain the final image patch scores. Extensive experiments demonstrate that our proposed model is effective and achieves competitive performance when compared with other related state-of-the-art methods.