Image Defogging Based on Improved Residual Block Feature Fusion Network

In this paper, we propose an image defog network based on improved residual block feature fusion, which aims to reconstruct fog-free images. Conventional methods usually cannot handle different fog density features differentially, and are limited to only focus on processing local features. To address this, we design a new residual block to obtain the depth information of the foggy images. In addition, our model can make use of complementary features with different weights, which increases the receptive field, makes better use of global information, speeds up network convergence, and achieves more accurate results. Extensive experimental results on the RESTED dataset have shown the effectiveness of our proposed method for image defog.