Convolutional neural network according to HVS for NR-IQA
Jun Wang, Yue Li, Yunlong Jiang, Xiaoli Jiang · 2025
Since human is the receiver of digital image information, the result of image quality assessment should be consistent with the human visual system (HVS). Inspired by the HVS, we propose a no-reference image quality assessment (NR-IQA) model based on convolutional neural network (HVS-CNN). In addition, we add the inverse residual block into the CNN, which improves the calculation speed of the model. The model obtains the image quality score and the gradient response map (GRM) conforming to the HVS. Experimental results show that the accuracy of the model is higher than other methods in all kinds of distortion types of LIVE dataset. The model is cross validated on the TID 2008, TID 2013, and CIQA datasets to obtain high accuracy.