An end-to-end single image dehazing network based on U-net

Single image haze removal is always significant for computer advanced vision tasks, while it is also a challenging problem. In this paper, inspired of the recent methods, we proposed an end-to-end network with encoding–decoding structure and jumping layers for single image dehazing. The network combined the advantages of VGG16 and the U-net and adopted different jumping layers to retain most of the image feature information. In order to clarify the image features like contrast and color distribution, the scale-invariant loss function and the proposed histogram loss function were used. We compared the algorithm with the several state-of-the-art algorithms qualitatively and quantitatively. Experimental results demonstrated that the proposed algorithm has achieved favorable dehazing results on both indoor and outdoor synthetic hazy testing set and real-world set. In particular, it obtained the better dehazing results for the slight hazy conditions than other density of haze.