Quality Assessment for Contrast-Distorted Images Based on Convolutional Neural Network

In this paper, we devise a dedicated quality evaluation scheme to automatically predict the quality of contrast-changed images, which is based on the newly presented convolutional neural network (CNN) as CNN has been proved quite effective in dealing with computer vision tasks. The designed CNN includes three convolution layers and two fully connected layers. By training the CNN on existing image quality databases, we can evaluate the quality of the contrast-distorted images. Experimental results demonstrate the proposed method delivers high prediction performance and outperform mainstream image quality methods.