A robust edge-guided and color-guided network for single image dehazing. (October 2022)

• Record Type: Journal Article
• Authors: Xu, Yujie; Zhang, Yongjun; Cui, Jingrong; Cui, Zhongwei; Yang, Yitong
• Is Part Of: Computers & electrical engineering. Volume 103(2022)
• Publication Date: 2022-10
• Subjects: Image dehazing -- Convolutional neural network -- Attention mechanism -- Encoder–decoder architecture; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2022.108301
• Languages: English
• ISSNs: 0045-7906

Abstract

Abstract: Most current dehazing methods tend to cause color distortion and blurred edge details. In this work, we design a robust edge-guided and color-guided dehazing network (ECGDN) to mitigate these issues. ECGDN consists of a cross-aggregation dual U-Net (CADU), a color feature extractor (CFE), a prior attention module(PAM), and a prior fusion group (PFG). Specifically, the CADU can take advantage of the complementarity between edge and texture features in a cross-aggregation manner. The CFE focuses on Hue- and Saturation-channel features to solve color distortion under the constraint of color correction loss. Finally, the PAM and the PFG can handle different priors(i.e., edge, texture, color) with attention mechanism and fusion strategy, respectively, to guide the feature learning procedures. To demonstrate the effectiveness of the proposed method, we perform extensive experiments on synthetic and real-world hazy images. The experimental results show that our method outperforms existing state-of-the-art methods in terms of accuracy and robustness. Graphical abstract: Highlights: A robust dehazing network based on CNN is proposed. An efficient algorithm with the guidance of edge and color priors. Using the attention mechanism and the fusion strategy improves the performance.