Two-stage processing of image restoration problem based on generative adversarial network
Qiong Wu, Zhanjun Jiang · 2023
Aiming at the problem that texture details and edge information of images are easily lost in the process of image recovery, a two-stage processing algorithm is proposed that combines denoising processing and super-resolution reconstruction processing. The algorithm is divided into two stages, the first stage uses the improved DEGAN network to denoise the image, and the training set and the test set are denoising the training set and the test set respectively through the optimal path obtained by training, and the result is directly transferred to the dataset for the second stage of super-resolution processing. In the second stage, the denoising training set is reconstructed with the improved SRGAN network, and a clean image dataset with the noise processing needs to be added as the target image for super-resolution processing. Compared with single denoising and super-resolution reconstruction, the images recovered by two-stage training processing have significantly improved subjective perception and objective evaluation indicators.