Transmittance-Guided Structure-Texture Decomposition for Nighttime Image Dehazing

Francesco MorettiGiulia Bianchi, Andrea GalloMaharaja Agrasen University {francesco.moretti, giulia.bianchi, andrea.gallo}@mau.edu.mk · 2026

Read Transmittance-Guided Structure-Texture Decomposition for Nighttime Image Dehazing by Francesco MorettiGiulia Bianchi, Andrea GalloMaharaja Agrasen University {francesco.moretti, giulia.bianchi, andrea.gallo}@mau.edu.mk (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-29507.txt.

Introduction

Moretti and co-authors present a two-stage method for clearing haze from nighttime images. Night haze cuts visibility, warps color, and lowers contrast. Uneven artificial light makes the task harder. Stage one corrects the transmittance map with region-aware tweaks. It also estimates the air light map in the YUV color space. Stage two splits the image into structure and texture layers. The structure layer gets light and color fixes. The texture layer gets detail sharpening. A two-phase fusion then blends the layers.

Catalog record

CREATORFrancesco MorettiGiulia Bianchi, Andrea GalloMaharaja Agrasen University {francesco.moretti, giulia.bianchi, andrea.gallo}@mau.edu.mk
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42475
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2603.29507

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Transmittance-Guided Structure-Texture Decomposition for Nighttime Image Dehazing Francesco MorettiGiulia Bianchi, Andrea GalloMaharaja Agrasen University {francesco.moretti, giulia.bianchi, andrea.gallo}@mau.edu.mk Abstract Nighttime images captured under hazy conditions suffer from severe quality degradation, including low visibility, color distortion, and reduced contrast, caused by the combined effects of atmospheric scattering, absorption by suspended particles, and non-uniform illumination from artificial light sources. While existing nighttime dehazing methods have achieved partial success, they typically address only a subset of these issues, such as glow suppression or brightness enhancement, without jointly tackling the full spectrum of degradation factors. In this paper, we propose a two-stage nighttime image dehazing framework that integrates transmittance correction with structure-texture layered optimization. In the first stage, we introduce a novel transmittance correction method that establishes boundary-constrained initial transmittance maps and subsequently applies region-adaptive compensation and normalization based on whether image regions correspond to light source areas. A quadratic Gaussian filtering scheme operating in the YUV color space is employed to estimate the spatially varying atmospheric light map. The corrected transmittance map and atmospheric light map are then used in conjunction with an improved nighttime imaging model to produce the initial dehazed image. In the second stage, we propose a STAR-YUV decomposition model that separates the dehazed image into structure and texture layers within the YUV color space. Gamma correction and MSRCR-based color restoration are applied to the structure layer for illumination compensation and color bias correction, while Laplacian-of-Gaussian filtering is applied to the texture layer for detail enhancement. A novel two-phase fusion strategy, comprising nonlinear Retinex-based fusion of the enhanced layers followed by linear blending with the initial dehazing result, yields the final output. Extensive experiments on two benchmark datasets (ZS330 and HC770) demonstrate that the proposed method achieves average PSNR, SSIM, AG, IE, and NIQE scores of 17.024 dB, 0.765, 7.604, 7.528, and 2.693, respectively, consistently outperforming six state-of-the-art methods across both full-reference and no-reference quality metrics, while effectively recovering natural colors, enriching fine details, and improving overall scene brightness. 1 Introduction Hazy weather conditions present significant challenges for intelligent imaging systems operating around the clock, such as those deployed in pedestrian detection [ 36 ] , autonomous driving, and remote surveillance. While daytime dehazing has received extensive attention, nighttime image dehazing poses substantially greater difficulties due to the concurrent presence of atmospheric scattering, low ambient illumination, non-uniform artificial lighting, and pronounced color deviations [ 54 ] . The degradation observed in nighttime hazy images is a compounded effect: suspended atmospheric particles scatter and absorb light, reducing visibility and contrast, while colored artificial light sources introduce spatially varying illumination and chromatic distortions that are absent in daytime scenarios [ 25 ] . These compounded challenges have driven the increasing demand for effective nighttime image dehazing techniques. Current image dehazing algorithms can be broadly categorized into daytime methods and nighttime methods. Daytime dehazing approaches are further divided into traditional model-based methods and deep learning-based methods [ 57 ] . Traditional approaches, such as histogram equalization [ 21 ] , can effectively balance overall color distributions and brightness levels but often introduce local color distortions and loss of fine details. The Retinex theory [ 31 ] , which models human visual perception of color constancy, has inspired a family of enhancement methods. Multi-Scale Retinex with Color Restoration (MSRCR) [ 17 ] extends this framework by incorporating a color restoration factor that simultaneously enhances image quality and prevents color desaturation. Additionally, numerous researchers have leveraged the atmospheric scattering model proposed by McCartney and Hall [ 32 ] to estimate model parameters from prior information and subsequently recover clear images through model inversion [ 59 ] . A landmark contribution in this direction is the dark channel prior (DCP) proposed by He et al. [ 14 ] , which exploits the observation that in non-sky regions, at least one color channel contains pixels with intensities close to zero. While many subsequent works have refined the dark channel or optimized model parameters—for instance, Meng et al. [ 33 ] introduced boundary constraints to mitigate color distortion artifacts in sky regions—these daytime-oriented methods fundamentally neglect the influence of artificial light sources and thus produce results with insufficient brightness and persistent color shifts when applied to nighttime imagery. Deep learning has also transformed the dehazing landscape. Cai et al. [ 3 ] proposed DehazeNet, a trainable end-to-end convolutional neural network for transmittance estimation that recovers clear images via the atmospheric scattering model. Recent advances in generative architectures, including diffusion models [ 15 , 37 ] and vision transformers [ 28 ] , have further expanded the capabilities of learning-based image restoration. In particular, NightHazeFormer [ 56 ] introduced a prior query transformer for end-to-end nighttime haze removal, while NightHaze [ 23 ] demonstrated that self-prior learning through severe augmentation produces strong network priors resilient to real-world nighttime degradation. RetinexFormer [ 4 ] advanced one-stage Retinex-based transformer architectures for low-light image enhancement. However, these methods often suffer from limited interpretability, require large-scale paired training data and substantial computational resources, and may not generalize well to the distinct characteristics of nighttime scenes [ 39 ] . Nighttime dehazing algorithms specifically target the additional complexities introduced by artificial illumination, including glow effects, localized overexposure, severe color bias, and non-uniform brightness distributions. Traditional nighttime approaches typically extend the daytime atmospheric scattering model to accommodate these factors. Li et al. [ 20 ] incorporated the atmospheric point spread function as a glow model and proposed a method for removing glow and multiple light colors (GMLC), which effectively suppresses glow but may produce dim results with edge artifacts. Tang et al. [ 48 ] treated the atmospheric light as a spatially varying function and introduced a hybrid filtering method combined with highlight-region transmittance compensation to mitigate overexposure, though residual color bias remained. Fusion-based enhancement methods have also emerged: Xu et al. [ 55 ] proposed the Structure and Texture Aware Retinex (STAR) model, achieving notable results in low-light enhancement and color correction; Liu et al. [ 26 ] developed a unified variational decomposition model that effectively suppresses noise but leaves haze residue; and Liu et al. [ 24 ] proposed the Joint Contrast Enhancement and Exposure Fusion (CEEF) framework, which enhances contrast and preserves details but can amplify glow artifacts. Deep learning methods for nighttime dehazing learn mappings between hazy and haze-free image pairs. Liao et al. [ 22 ] proposed HDP-Net, a CNN-based haze density prediction network; Kuanar et al. [ 18 ] introduced a DeGlow-DeHaze iterative architecture; and Cui et al. [ 6 ] proposed the Illumination Adaptive Transformer (IAT), which effectively improves brightness and contrast but fails to account for glow effects, resulting in overexposure. Deng et al. [ 7 ] proposed contrastive and adversarial learning for real-world nighttime dehazing, and Yu et al. [ 60 ] developed a multi-scale gated fusion network specifically designed for nighttime conditions. Despite these advances, current methods typically address only specific degradation factors, such as glow suppression or illumination enhancement, without comprehensively handling the full range of nighttime image degradation [ 38 ] . Furthermore, learning-based approaches generally overlook the physical principles underlying haze formation, limiting their interpretability [ 39 ] . To address these limitations, we propose a two-stage nighttime image dehazing framework that integrates

Read the complete text