SINGLE IMAGE SUPER-RESOLUTION USING A NON-LOCAL 3D CONVOLUTIONAL NEURAL NETWORK

Single image super-resolution (SR), which intends to recover a high-resolution (HR) image from a single low-resolution (LR) image, has attracted increasing attentions with a wide range of applications. In this paper, we propose a novel non-local scheme based on a 3D convolutional neural network (3DCNN) for image super-resolution. Different from most previous methods, our scheme takes the inherent non-local self-similarity property of natural images into account. Specifically, the non-local similar patches are searched and extracted from low-resolution images. Then a 3DCNN is constructed to jointly sharpen these non-local patches, which can make full use of the non-local similarity in natural images. Finally, the super-resolved image is reconstructed from the sharpened patches. Experiments show that the proposed non-local method achieves the superior reconstruction accuracy compared with several state-of-the-art methods.