Reduced-Reference Image Quality Assessment for Single-Image Super-Resolution by Convolutional Neural Network
Yuxia Sheng, Yaru Wu, Liangkang Yang, Dan Xiong · 2022 41st Chinese Control Conference (CCC) · 2022
Single-Image Super-Resolution (SISR) aims to improve the image resolution with good visual quality, which is a classical problem in the field of image processing. How to assess the SISR image quality is still a challenging problem, although many SISR algorithms have been proposed. In this paper, we design a convolutional neural network (CNN) to predict the image quality of SISR by taking the low resolution (LR) image as the reference image. The proposed network consists of six convolution layers, four fully connected layers and one regression layer. This reduced-reference method uses CNN to extract the features of LR and SR image patches, and predict the quality of super-resolution reconstruction image patches by random forest regression. Experimental results show that the proposed method can provide evaluation that is more consistent with the subjective assessment score, and outperforms other image quality assessment methods.