Super-resolution Reconstruction Based on Capsule Generative Adversarial Network

Abstract Using each part of the image's spatial information to generate better local details image is a key problem that super-resolution reconstruction has been facing. At present, mainstream networks are all built based on convolutional neural (CNN). Some these methods Generative Adversarial Networks (GAN) have good performance in high-frequency and visual effects. However, because CNN lacks necessary attention information, method prone problems such as excessive brightness unnatural pixel regions image. Therefore, using capsule network's excellent perception hierarchical feature relationships, author proposes network CSRGAN. The experiment's final result shows compared with pure convolution RDN, PSNR value CSRGAN increased by 0.14, which closer original