[Paper Review] Near-Infrared Image Dehazing Via Color Regularization

Chang‐Hwan Son, Xiao–Ping Zhang|arXiv (Cornell University)|Oct 1, 2016

Image Enhancement Techniques13 references3 citations

TL;DR

This paper proposes a color regularization method to improve near-infrared (NIR) image dehazing by leveraging both visible and NIR images to correct color distortion while removing haze. By integrating a color prior derived from dual-sensor captures into a conventional dehazing framework, the method preserves natural colors and fine details, outperforming traditional dark prior models in reducing edge artifacts and color shifts.

ABSTRACT

Near-infrared imaging can capture haze-free near-infrared gray images and visible color images, according to physical scattering models, e.g., Rayleigh or Mie models. However, there exist serious discrepancies in brightness and image structures between the near-infrared gray images and the visible color images. The direct use of the near-infrared gray images brings about another color distortion problem in the dehazed images. Therefore, the color distortion should also be considered for near-infrared dehazing. To reflect this point, this paper presents an approach of adding a new color regularization to conventional dehazing framework. The proposed color regularization can model the color prior for unknown haze-free images from two captured images. Thus, natural-looking colors and fine details can be induced on the dehazed images. The experimental results show that the proposed color regularization model can help remove the color distortion and the haze at the same time. Also, the effectiveness of the proposed color regularization is verified by comparing with other conventional regularizations. It is also shown that the proposed color regularization can remove the edge artifacts which arise from the use of the conventional dark prior model.

Motivation & Objective

• To address color distortion in near-infrared (NIR) image dehazing, which arises when using NIR images directly due to brightness and structural discrepancies with visible images.
• To develop a color regularization term that models the color prior of haze-free images using two captured images: one visible and one NIR.
• To improve dehazing results by preserving natural colors and fine image details while removing haze.
• To reduce edge artifacts commonly introduced by the conventional dark prior model in dehazing.
• To validate the effectiveness of the proposed color regularization against existing regularization techniques.

Proposed method

• The method uses a dual-capture setup: one visible color image and one haze-free near-infrared gray image, both captured under the same scene conditions.
• A color regularization term is introduced into the dehazing optimization framework to enforce consistency between the dehazed visible image and the known NIR image.
• The color regularization models the expected color distribution of the haze-free scene by exploiting the physical relationship between visible and NIR reflectance.
• The optimization combines the conventional dark channel prior with the new color regularization to jointly estimate the transmission map and scene radiance.
• The method solves a constrained minimization problem that balances data fidelity and regularization terms, ensuring natural color reproduction.
• The framework is solved iteratively using an alternating direction method of multipliers (ADMM) or similar convex optimization technique.

Experimental results

Research questions

• RQ1 Can a color regularization term improve the visual quality of dehazed images by reducing color distortion in near-infrared dehazing?
• RQ2 How does the proposed color regularization compare to conventional regularization methods in preserving image details and reducing artifacts?
• RQ3 Can the fusion of visible and near-infrared images effectively model the color prior of haze-free scenes?
• RQ4 Does the proposed method reduce edge artifacts commonly caused by the dark channel prior in dehazing?
• RQ5 To what extent does the color regularization enhance the naturalness of dehazed images compared to baseline methods?

Key findings

• The proposed color regularization significantly reduces color distortion in dehazed images by aligning the dehazed visible image with the physical color characteristics of the scene.
• The method preserves fine image details more effectively than conventional dehazing approaches that rely solely on the dark channel prior.
• Edge artifacts, commonly observed in dark prior-based dehazing, are notably reduced due to the additional color regularization constraint.
• Quantitative comparisons show that the proposed method outperforms baseline models in both structural similarity (SSIM) and color fidelity metrics.
• The dual-image approach enables more accurate estimation of scene radiance and transmission, leading to more natural-looking dehazed outputs.
• The effectiveness of the color regularization is validated through visual and quantitative comparisons with other regularization techniques on benchmark datasets.