Image Dehazing Algorithm Based on Generative Adversarial Networks

Haze in images has long been a major challenge in the field of computer vision, significantly affecting visual clarity and weakening the performance of applications such as autonomous driving and monitoring systems, Defogging of images is particularly important, Traditional defogging methods, such as dark channel prior (DCP) and atmospheric scattering model (ASM), often encounter difficulties in complex scenes and may be limited by low computational efficiency or loss of details, In order to solve these problems and improve defogging effectiveness, this paper proposes a method based on generative adversarial networks (GANs) The new dehazing algorithm combines residual learning, channel attention mechanism, and skip connections to effectively eliminate haze and restore clear image details, We designed a generator network containing multiple residual blocks and a discriminator network based on PatchGAN framework, These networks were trained using a mixed loss function that combines adversarial loss. pixel level loss, and perceptual loss, Experimental results show that the proposed method achieves notable improvements in the quantitative metrics of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM). When compared with traditional haze removal techniques such as DCP and ASM, the proposed method realizes an average PSNR increase of 13.5% and an average SSIM improvement of 11.0%, which demonstrates its effectiveness in processing images with varying haze densities.