No-reference image quality assessment based on residual CNN

The captured image is easy to introduce distortion in the transmission process, resulting in image damage. Therefore, in order to ensure that the image looks clear enough, the design of image quality assessment methods is necessary. We propose a no-reference IQA model, which is a convolutional neural network model. The input of this model are image patches, and there is no need to manually extract features. There are six convolutional layers and an optimized residual module in the feature extraction network. The image prediction score is taken by linear regression through three fully connected layers. The metric data show that our method shows better quality score prediction performance on the LIVE database, and the test results are good on most of the single images.