Convolutional neural network for blind quality evaluator of image super-resolution

Image super-resolution tends to increase the resolution of images with good visual experience. In the previous decade, there have been many image super-resolution algorithms proposed for various multimedia processing applications. However, how to predict the perceptual scores of high-resolution images generated by image super-resolution methods is still challenging. In this work, we proposed a Deep Convolutional Neural Network (DCNN) to evaluate the visual quality of image super-resolution. The proposed network is constructed by two convolutional layers, two pooling layers including average, min and max pooling, three fully connected layers and one regression layer. The proposed method by DCNN can extract high-level intrinsic features more effectively than handcrafted features for super-resolution images and can predict the visual quality more effectively. Furthermore, the proposed method divides the super-resolution image into small patches, which not only increases the number of training data but also considers the local information for the super-resolution image quality. Extensive comparison experiments demonstrate that the proposed method can predict visual quality scores for image super-resolution more accurately than other state-of-the-art algorithms.