GAN-based image deblurring method for multilayer recurrent networks

Image deblurring is one of the research hotspots in the field of computer vision in recent years. Aiming at the problems that the images recovered by the traditional DeblurGan are not rich in detail information and the images are not clear enough, this paper proposes a multi-layer cyclic residual network image deblurring method based on GAN. The main contributions of the method are two: on the one hand, it effectively improves the ability of extracting detail information by introducing a multilayer recurrent residual network (MAB) module, which consists of multiple attention residual blocks nested together and fuses local and global information. On the other hand, the channel attention mechanism is introduced into the MAB module, which enhances the extraction of effective regions by modelling the correlation between the channels, and further improves the de-blurring effect. The experimental results show that compared with the original model, the resultant image obtained by this paper's method achieves better deblurring effect, and the PSNR and SSIM indexes are better than the comparison method.