Article
Mathematics, Interdisciplinary Applications
Qingliang Jiao, Ming Liu, Bu Ning, Fengfeng Zhao, Liquan Dong, Lingqin Kong, Mei Hui, Yuejin Zhao
Summary: The study introduces a novel dehazing model using fractional derivative and data-driven regularization terms to enhance dehazing quality. The proposed model is solved using half-quadratic splitting and a dual-stream network based on CNN and Transformer is introduced for data-driven regularization. Experimental results demonstrate that the proposed method outperforms state-of-the-art dehazing methods on both synthetic and real-world images.
FRACTAL AND FRACTIONAL
(2022)
Article
Computer Science, Artificial Intelligence
Usman Ali, Jeongdan Choi, KyoungWook Min, Young-Kyu Choi, Muhammad Tariq Mahmood
Summary: For single image dehazing, regularization-based schemes improve the initial transmission map iteratively using a guidance map as a structural prior. However, these methods do not constrain the transmission map to its valid range, which affects the robustness and quality of the recovered image. To address this, we propose a robust regularization scheme that constrains the transmission map during enhancement by leveraging mutual structural information and solving a nonconvex problem.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Information Systems
Usman Ali, Ik Hyun Lee, Muhammad Tariq Mahmood
Summary: This paper proposes an improved image dehazing method by optimizing a nonconvex energy function that leverages structural information from the transmission map and guidance. The proposed method provides robust regularization and achieves high-quality haze-free images.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Yan Li, De Cheng, Dingwen Zhang, Nannan Wang, Xinbo Gao, Jiande Sun
Summary: This paper proposes a single image dehazing method with an independent Detail Recovery Network (DRN) that captures details from the input image and integrates them into the dehazed image. The method consists of two independent networks, DRN and the dehazing network. DRN aims to recover the details of the dehazed image through the joint efforts of the local branch and the global branch. Experiments demonstrate the effectiveness of the proposed method and show its superiority over existing dehazing methods.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Nanfeng Jiang, Kejian Hu, Ting Zhang, Weiling Chen, Yiwen Xu, Tiesong Zhao
Summary: Deep learning technologies have been successfully applied in Single Image Dehazing (SID) tasks, but most algorithms neglect the refinement of image details during dehazing, leading to detail-loss regions in the dehazed results. To address this issue, a deep hybrid network is designed to improve dehazing performance and remedy the loss of details. The network consists of two sub-networks with a multi-term loss function, one for effective haze removal using a haze residual attention sub-network, and the other for detail refinement via multi-scale contextual information aggregation. Joint training of the two sub-networks achieves clear haze removal and well-preserved image details. The detail refinement sub-network can also be applied to other dehazing methods to enhance their model performances. Extensive experiments demonstrate the superiority of the proposed network over state-of-the-art methods.
PATTERN RECOGNITION
(2023)
Article
Engineering, Electrical & Electronic
Mingye Ju, Can Ding, Wenqi Ren, Yi Yang
Summary: In this research, a new image dehazing technique called IDBP is developed using a robust and promising atmospheric scattering model. It overcomes the limitations of existing techniques by incorporating multiple priors and minimal information loss principle. The proposed technique consists of two modules, the atmospheric light estimation module and the multiple prior constraint module, and outperforms the state-of-the-art alternates according to numerous experiments.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Environmental Sciences
Jing Qin, Liang Chen, Jian Xu, Wenqi Ren
Summary: In this paper, a novel method based on sparse representation is proposed for haze removal in single input image. Contextual regularization tool is used to reduce block artifacts and halos, and a dictionary is utilized to smooth the image and generate sharp dehazed result. Experimental results show that the proposed method outperforms state-of-the-art dehazing techniques and produces high-quality dehazed and vivid color results.
Article
Engineering, Civil
Guisik Kim, Junseok Kwon
Summary: A novel dehazing framework is proposed in this paper for real-world images containing hazy and low-light areas, which unifies dehazing and low-light enhancements using an illumination map estimated by a convolutional neural network. Experimental results show that the method outperforms state-of-the-art algorithms in real-world image dehazing in both quantitative and qualitative terms.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Pulkit Dwivedi, Soumendu Chakraborty
Summary: Image dehazing is an important technique that aims to remove haze from atmospheric conditions to improve the visual quality of images. The dark channel prior methodology has been recognized as an effective approach, but it has limitations in accurately removing haze due to its reliance on a single color channel. To overcome this limitation, we propose a method that computes the dark channel from all three channels of an image, resulting in better performance on synthetic and real-world hazy images.
IMAGE AND VISION COMPUTING
(2023)
Article
Chemistry, Analytical
Bingnan Yan, Zhaozhao Yang, Huizhu Sun, Conghui Wang
Summary: The preservation of image details in the defogging process is a key challenge in deep learning. In this paper, we propose a detail enhanced image CycleGAN to retain the detail information during the process of defogging. The algorithm combines the CycleGAN network with the U-Net network's idea to extract visual information features and introduces Dep residual blocks to learn deeper feature information. It also introduces a multi-head attention mechanism to enhance the expressive ability of features and balance the deviation produced by the same attention mechanism. Experimental results on the D-Hazy dataset show that compared with the CycleGAN network, the network structure in this paper improves the SSIM and PSNR of the image dehazing effect by 12.2% and 8.1% respectively, and can retain image dehazing details.
Article
Engineering, Electrical & Electronic
Jinyuan Liu, Jingjie Shang, Risheng Liu, Xin Fan
Summary: Recent advances in deep learning networks have shown impressive progress in multi-exposure image fusion. However, restoring realistic texture details while correcting color distortion remains a challenging problem. In this paper, an attention-guided global-local adversarial learning network is proposed to address these issues. Experimental results demonstrate the superiority of this method in visual inspection and objective analysis.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Yong Liu, Xiaorong Hou
Summary: In this paper, a local multi-scale feature aggregation network (LMFA-Net) is proposed for real-time dehazing. LMFA-Net can directly restore the haze-free image by learning the local mapping relationship between the clean value of a haze image at a certain point and its surrounding local region. LMFA-Net achieves fast and efficient dehazing with its lightweight model structure.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Yuda Song, Zhuqing He, Hui Qian, Xin Du
Summary: Image dehazing is a low-level vision task to estimate haze-free images. Convolutional neural network methods dominate this task, but vision Transformers haven't made a breakthrough. This study introduces DehazeFormer, an improved version of Swin Transformer, with modified normalization layer, activation function, and spatial information aggregation. Multiple variants of DehazeFormer were trained and shown to be effective.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Robotics
Kun Wang, Liquan Shen, Yufei Lin, Mengyao Li, Qijie Zhao
Summary: This study proposes a novel underwater image enhancement method by jointly optimizing the results of color correction and dehazing. It first uses a triplet-based color correction module to obtain color-balanced images, followed by a recurrent dehazing module to address the haze effect. Finally, an iterative mechanism is proposed to jointly optimize color correction and dehazing.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2021)
Article
Environmental Sciences
Bo Jiang, Guanting Chen, Jinshuai Wang, Hang Ma, Lin Wang, Yuxuan Wang, Xiaoxuan Chen
Summary: This study proposed a single remote sensing image dehazing method based on the encoder-decoder architecture, which combines wavelet transform and deep learning technology to address the non-uniform haze in remote sensing images. Extensive experiments showed that the proposed method qualitatively outperformed typical traditional methods and deep learning methods, with improved peak-signal-to-noise ratio (PSNR) and structural similarity (SSIM) values compared to the average of four typical deep learning methods. This comprehensively verified the effectiveness of the proposed method for the problem of remote sensing non-uniform dehazing.
