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