Image deblurring method based on GAN with a channel attention mechanism

Motion blur in images arises from a combination of object motion and camera shake, resulting in uneven blurring and varying blur scales. Although larger convolutional kernels and deeper networks can enhance the accuracy of motion deblurring, they also increase network complexity, prolong training time, and diminish efficiency. To address these challenges, we propose a fast and effective image-deblurring method based on generative adversarial networks (GAN). To augment the feature-extraction capabilities of the generator we introduce a multi-scale feature-extraction structure. In addition, we propose an improved residual structure to simplify the training process and accelerate feature information propagation. Furthermore, we incorporate a channel attention mechanism into the residual structure, creating a channel attention residual structure that enhances the feature representation ability of the network. By employing adversarial and perceptual losses, the generator and discriminator collaboratively produce deblurred images with rich and detailed textures. The experimental results on the GoPro, REDs, HIDE, and RealBlur-J datasets demonstrate that the proposed method outperforms existing deblurring techniques in terms of both efficiency and quality.