Single-frame image super-resolution through gradient learning

In this paper, we propose a novel method for solving single-frame image super-resolution. Based on the traditional magnification methods that learn through training sets, our method increases the priority of low-frequency images´ gradient region and learns from gradient region. We extract features from every gradient image fragment, and then find the closest matching fragments to determine the high-frequency of target image. At first, we get horizontal and vertical gradient images of the input image, then combine these two images together to get the final high-resolution image. By this way, the target image has sharper edges and higher quality. Experiments show that our method is very flexible and gives good results.