Progressive Image Defogging Network with Depth Information
Ting Liu, Mengyu Dong, Xiayang Huang, Kai Zheng, Kun Wang · 2025
Under complex non-uniform haze conditions, Conventional defogging networks cannot achieve the desired defogging effect. To address this problem, this paper proposes a progressive image defogging network that incorporates depth information, which is designed to improve defogging in non-uniform haze conditions. Specifically, a progressive image defogging network architecture is designed to gradually remove the fog in the image by iteratively defogging the unit. The defogging unit consists of a feature memory module, an encoder and a decoder. The feature memory module is designed to effectively enhance the retention and updating of key feature information during the defogging process. In the encoder stage, a depth information branch and a visible light image branch are introduced, and the extraction capability of image features is improved through the local and global fusion modules in the design, achieving an effective fusion of depth information and visible light information. In the decoder stage, a multi-scale feature enhancement module is designed to improve the recovery of image details and textures. The experimental results show that the designed network has significantly improved the performance of dehazing on both synthetic and real datasets.