A Single Image Dehazing Algorithm Based on Cycle-GAN

Abstract

Due to the effect of atmospheric light scattering, the quality of images taken under haze weather conditions will be seriously degraded. These characteristics affect the judgment and extraction of image features and reduce the application value of images. Image dehazing is therefore a necessary step in many computer vision tasks. In this paper, an end-to-end image dehazing algorithm Dehaze-GAN based on Cycle-GAN is proposed. In order to ensure that the structure of the images before and after dehazing is basically the same, the algorithm adds structure consistency loss on the basis of Cycle-GAN. The input of Dehaze-GAN is a hazy image and the output is a clean image. Dehaze-GAN is trained by a large number of hazy images and their corresponding clean images. The experimental results show that Dehaze-GAN is superior to other dehazing algorithms in both PSNR and SSIM dehazing performance indicators.