Patch-based Generative Adversarial Network for Single Image Haze Removal
Qianli Jia, Zhikang Ma · 2020 International Conference on Computer Engineering and Application (ICCEA) · 2020
Image haze removal is widely used in various fields and is an important link to improve the ability of image recognition under special circumstances. In order to improve the image haze removal level, this paper proposes a single image haze removal method based on network image translation (Pix2Pix). Relying on the encoder-decoder model and the Markov Discriminator (PatchGAN), it is possible to maintain a high resolution and texture structure of the image while defogging. Finally, experiments on synthetic prove that this algorithm is better than others in objective indicators of PSNR and SSIM.