Abstract:Images captured in hazy outdoor conditions often suffer from colour distortion, low contrast, and loss of detail, which impair high-level vision tasks. Single image dehazing is essential for applications such as autonomous driving and surveillance, with the aim of restoring image clarity. In this work, we propose WTCL-Dehaze an enhanced semi-supervised dehazing network that integrates Contrastive Loss and Discrete Wavelet Transform (DWT). We incorporate contrastive regularization to enhance feature representation by contrasting hazy and clear image pairs. Additionally, we utilize DWT for multi-scale feature extraction, effectively capturing high-frequency details and global structures. Our approach leverages both labelled and unlabelled data to mitigate the domain gap and improve generalization. The model is trained on a combination of synthetic and real-world datasets, ensuring robust performance across different scenarios. Extensive experiments demonstrate that our proposed algorithm achieves superior performance and improved robustness compared to state-of-the-art single image dehazing methods on both benchmark datasets and real-world images.

What problem does this paper attempt to address?

This paper attempts to solve the problems such as color distortion, low contrast and detail loss in images taken under hazy weather conditions. These problems will seriously affect the performance of high - level vision tasks. Specifically, single - image dehazing is a fundamental problem in computer vision, aiming to restore the clarity of images obscured by haze. This is crucial for applications such as autonomous driving, surveillance and remote sensing. ### Core problems of the paper 1. **Color distortion, low contrast and detail loss**: These problems are caused by haze, which degrades the image quality and affects subsequent high - level vision tasks. 2. **Domain Gap**: Existing methods usually rely on synthetic data for training, but perform poorly in real - world scenarios because there are differences between real - world data and synthetic data. 3. **Insufficient feature representation**: Traditional dehazing methods have limitations in feature extraction and are difficult to effectively capture multi - scale information. ### Solutions To solve the above problems, the authors propose WTCL - Dehaze, an enhanced semi - supervised dehazing network that combines contrastive loss and discrete wavelet transform (DWT). Specific improvements include: 1. **Contrastive Regularization**: By comparing pairs of hazy and clear images, the feature representation is enhanced, enabling the model to better learn useful features. 2. **Multi - scale feature extraction**: Use DWT for multi - scale feature extraction to effectively capture high - frequency details and global structures. 3. **Semi - supervised learning**: Utilize labeled and unlabeled data to alleviate the domain gap and improve the generalization ability of the model. ### Model architecture The architecture of WTCL - Dehaze includes two branches: - **Supervised branch**: Use labeled synthetic data for training to minimize the reconstruction loss between the generated image and the real image. - **Unsupervised branch**: Use unlabeled real data for training to ensure that the model can generate high - quality dehazed images without real labels. ### Loss functions To ensure the effective training of the model, the authors introduce multiple loss functions, including: - **Supervised losses**: Such as mean squared loss, perceptual loss and adversarial loss. - **Unsupervised losses**: Such as total variation loss and dark channel loss. - **Contrastive regularization loss**: Used to enhance the feature representation, ensure that the restored image is closer to the clear image and far from the hazy image. ### Experimental results Experiments show that WTCL - Dehaze performs well on both standard benchmark datasets and real - world images, outperforming existing dehazing methods. In summary, this paper effectively solves the key problems in single - image dehazing by introducing contrastive loss and discrete wavelet transform, combined with semi - supervised learning, and improves the performance and robustness of the model.

WTCL-Dehaze: Rethinking Real-world Image Dehazing via Wavelet Transform and Contrastive Learning

Vision Transformers for Single Image Dehazing

From Coarse to Fine: A Two-Stage Network for Dense Haze removal

UCL-Dehaze: Toward Real-World Image Dehazing via Unsupervised Contrastive Learning

UCL-Dehaze: Towards Real-world Image Dehazing via Unsupervised Contrastive Learning

Restoring vision in hazy weather with hierarchical contrastive learning

SCL-Dehaze: Toward Real-World Image Dehazing via Semi-Supervised Codebook Learning

From Synthetic to Real: Image Dehazing Collaborating with Unlabeled Real Data

DehazeDCT: Towards Effective Non-Homogeneous Dehazing via Deformable Convolutional Transformer

Single image dehazing via cycle-consistent adversarial networks with a multi-scale hybrid encoder-decoder and global correlation loss

Multi-Scale Attention Feature Enhancement Network for Single Image Dehazing

Contrastive Learning in Wavelet Domain for Image Dehazing

Contrastive Multiscale Transformer for Image Dehazing

CTHD-Net: CNN-Transformer hybrid dehazing network via residual global attention and gated boosting strategy

Visual-quality-driven unsupervised image dehazing

Contrastive adaptive frequency decomposition network guided by haze discrimination for real-world image dehazing

Advancing Real-World Image Dehazing: Perspective, Modules, and Training

DW-GAN: A Discrete Wavelet Transform GAN for NonHomogeneous Dehazing

Adaptive discrete wavelet transform and optimized residual-based deep CNN for image dehazing with a new meta-heuristic algorithm

WaveDH: Wavelet Sub-bands Guided ConvNet for Efficient Image Dehazing