Research on Image Motion Deblurring Based on Generative Adversarial Networks

As the carrier of information, images can be used to obtain all kinds of important information. With the gradual progress of science and technology, people have higher and higher demand for image quality. However, affected by various reasons, the image will show varying degrees of information loss, resulting in image degradation. Therefore, this paper has done the following research on the motion blur of the image. This research takes the Generative Adversarial Networks as the main framework. First, a shallow feature extraction is carried out on the blurred image, and the Adaptive Residual Block is used as our encoder. The Adaptive Residual Block combines the Deformable Convolution Module, which can automatically adjust the size of the convolution kernel according to the deformable information of the motion blurred image, greatly improving the reconstruction ability of the image. Finally, the image is decoded by Transposed Convolution, and finally the blurred image is generated. In this study, PatchGAN network is used as our discriminator, which has good discriminating ability compared with other discriminator networks. The results of comparative tests show that the PSNR and SSIM of this algorithm are improved by 4.4% and 3.4% compared with other algorithms.