Image Dehazing Base on Dual Attention Generative Adversarial Network
Bohong Liu, Jie Yuan · 2024
In the field of image dehazing based on deep learning, the lack of clarity and color distortion in defogged images has been an ongoing problem. To address the above problems this paper proposes a novel generative adversarial network defogging algorithm. Its generator is composed of an encoder-decoder network, and the shortcoming of insufficient feature connectivity is compensated by adding feature extraction group (FEG) and dual attention module (DAM) between adjacent layers. FEG is used to extract shallow feature information at each scale, and DAM performs feature weighting on feature maps at different scales to enable it to perceive color changes in different regions and enhance the ability of the generator to express features. Comparative experiments were conducted on RESIDE public dataset to compare several mainstream image dehazing algorithms, and the experimental results show that the algorithm in this paper has a greater advantage over other algorithms in color restoration and detail recovery.