Single image dehazing algorithm based on generative adversarial network

This paper proposes a kind of generative adversarial network which is used to remove the haze for single image. In this paper, the generator uses U-Net as the backbone, and in order to effectively fuse the feature of different scales between the non-adjacent layers of the generator, a dense linking module which based on back-projection is used in the generator. In this paper, a kind of enhancement strategy which based on boosting strategy is used to improve the effectiveness of skip connection between the encoder and the decoder in the generator model. In order to evaluate the effect of haze removing, the proposed model is trained on the RESIDE and evaluated on the SOTS. The experiment proves that our method has advantages in both qualitative comparison and quantitative assessment.