Research on Image Reconstruction Method Based on Generative Adversarial Network

The traditional SRGAN image super-resolution reconstruction algorithm has a poor reconstruction quality and does not consider the detailed information at different scales. To address these issues, improvements were made to the generation network and discrimination network of the traditional SRGAN model. Channel attention mechanism and spatial attention mechanism were introduced into the residual block and upsampling module of the generation network to enhance the network's perceptual ability and reconstruction quality. In addition, a pyramid structure was introduced into the discriminator, allowing it to simultaneously consider the detailed information at different scales and better understand the spatial structure of the entire image. Test results on the Set5, Set14, and Urban100 datasets show that the improved SRGAN algorithm has better PSNR and SSIM than the traditional SRGAN algorithm. Furthermore, the improved SRGAN algorithm can better reconstruct the detailed parts of the image, making the reconstructed image clearer.