An Image Dehazing Algorithm Based on the Improved CGAN
Xuelian Li, Zhibin Chen, Wenqian Zheng, Ying Chang · IOP Conference Series Materials Science and Engineering · 2020
Abstract In order to improve the haze removal effect of image, a modified Conditional Generative Adversarial Nets (CGAN) based algorithm is proposed. In the new algorithm, the pre-trained visual geometry group (VGG) model is adopted, the DenseNet instead of the traditional U-net as the network structure of the generator, the Patch-GAN as the network structure of the discriminator, and the loss function is modified by the total variation regularization gradient. The defogged image can be obtained without estimating the projection map and the related defogging feature. The experiments indicate that our new algorithm effectively reduces the halo phenomenon and haze residue problem caused by the traditional dehazing method, and can preserve more details of the image, the structural similarity is improved from 75.9% to 92.6%.