GUSL-Dehaze: A Green U-Shaped Learning Approach to Image Dehazing

TL;DR

The paper tackles single-image dehazing with a focus on resource efficiency by cancelling reliance on heavy neural networks. It introduces GUSL-Dehaze, a Green Learning-based U-shaped pipeline that starts from a Modified Dark Channel Prior initialization and proceeds through PixelHop-based representation learning, discriminative feature selection with RFT, LNT-based feature generation, and hierarchical decision learning via two XGBoost regressors per level, all without backpropagation. Key contributions include learning to adapt the DCP parameter omega via a random forest, a multiscale, interpretable feature extraction framework, and a multilevel regression strategy that yields competitive PSNR/SSIM with about 1.72M parameters on RESIDE. This approach enables edge-device deployment with low latency and memory usage while maintaining transparency and performance comparable to state-of-the-art deep learning models $I(x)=J(x)t(x)+A(1-t(x))$ and $t(x)=e^{-eta d(x)}$ under varying haze densities.

Abstract

Image dehazing is a restoration task that aims to recover a clear image from a single hazy input. Traditional approaches rely on statistical priors and the physics-based atmospheric scattering model to reconstruct the haze-free image. While recent state-of-the-art methods are predominantly based on deep learning architectures, these models often involve high computational costs and large parameter sizes, making them unsuitable for resource-constrained devices. In this work, we propose GUSL-Dehaze, a Green U-Shaped Learning approach to image dehazing. Our method integrates a physics-based model with a green learning (GL) framework, offering a lightweight, transparent alternative to conventional deep learning techniques. Unlike neural network-based solutions, GUSL-Dehaze completely avoids deep learning. Instead, we begin with an initial dehazing step using a modified Dark Channel Prior (DCP), which is followed by a green learning pipeline implemented through a U-shaped architecture. This architecture employs unsupervised representation learning for effective feature extraction, together with feature-engineering techniques such as the Relevant Feature Test (RFT) and the Least-Squares Normal Transform (LNT) to maintain a compact model size. Finally, the dehazed image is obtained via a transparent supervised learning strategy. GUSL-Dehaze significantly reduces parameter count while ensuring mathematical interpretability and achieving performance on par with state-of-the-art deep learning models.

Figures (2)

• Figure 1: An overview of the proposed pipeline in three modules. Module 1 represents the preliminary dehazing using a modified DCP method. Module 2 learns robust representations via the Saab transform, selects the most relevant raw features, and then generates secondary features through the LNT. Module 3 employs two XGBoost regressors at each level to estimate the residuals based on the initial predictions of the preceding (coarser) level.
• Figure 2: Ranked features based on MSE loss using RFT feature selection.