Unsupervised Dark-Channel Attention-Guided CycleGAN for Single-Image Dehazing

Abstract

In this paper, we propose a new unsupervised attention-based cycle generative adversarial network to solve the problem of single-image dehazing. The proposed method adds an attention mechanism that can dehaze different areas on the basis of the previous generative adversarial network (GAN) dehazing method. This mechanism not only avoids the need to change the haze-free area due to the overall style migration of traditional GANs, but also pays attention to the different degrees of haze concentrations that need to be changed, while retaining the details of the original image. To more accurately and quickly label the concentrations and areas of haze, we innovatively use training-enhanced dark channels as attention maps, combining the advantages of prior algorithms and deep learning. The proposed method does not require paired datasets, and it can adequately generate high-resolution images. Experiments demonstrate that our algorithm is superior to previous algorithms in various scenarios. The proposed algorithm can effectively process very hazy images, misty images, and haze-free images, which is of great significance for dehazing in complex scenes.

Haze concentration adaptive network for image dehazing

“…However, these methods require additional labeled data and might be inaccurate due to the complexity of practical scenes [12]. 2) Density-awareness via extracting density features directly: Research works [15,17,26] directly learn haze density information without estimating a T-map. Deng et al [17] design a Haze-Aware Representation Distillation (HARD) module to extract global brightness and a haze-aware map.…”

“…Deng et al [17] design a Haze-Aware Representation Distillation (HARD) module to extract global brightness and a haze-aware map. Chen et al [26] propose an attention mechanism based on dark channel prior to describe haze concentration. However, not estimating the T-map would result in a lack of a comparator to measure density.…”

Visibility Enhancement and Fog Detection: Solutions Presented in Recent Scientific Papers with Potential for Application to Mobile Systems

“…It seems that learning-based methods are becoming more and more popular since many studies focused on this approach. The paper [ 63 ] proposes an unsupervised attention-based cycle generative adversarial network to resolve the problem of single-image dehazing. The novelty of the paper consists of an attention mechanism that can be used to dehaze different areas based on the previous generative adversarial network dehazing method.…”

RRG-GAN Restoring Network for Simple Lens Imaging System

“…The generative adversarial network (GAN), proposed by Goodfellow [ 23 ], became one of the most attractive schemes in deep learning in recent years. GAN utilize a discriminator network as a special robust loss function instead of the traditional hand-crafted loss function, which improves the performance and robustness of deep networks [ 24 , 25 ]. Specifically, we adopt the recursive residual groups (RRG) network [ 26 ] as the generative model G, thus our proposed generative adversarial network is named as RRG-GAN.…”