Light Pollution Reduction in Nighttime Photography

Chang Liu, Xiaolin WuarXiv 2021 · OpenReview

Abstract

Nighttime photographers are often troubled by light pollution of unwanted artificial lights. Artificial lights, after scattered by aerosols in the atmosphere, can inundate the starlight and degrade the quality of nighttime images, by reducing contrast and dynamic range and causing hazes. In this paper we develop a physically-based light pollution reduction (LPR) algorithm that can substantially alleviate the aforementioned degradations of perceptual quality and restore the pristine state of night sky. The key to the success of the proposed LPR algorithm is an inverse method to estimate the spatial radiance distribution and spectral signature of ground artificial lights. Extensive experiments are carried out to evaluate the efficacy and limitations of the LPR algorithm.

The Challenge of Urban Night Skies

For nighttime photographers, urbanization presents a significant technical hurdle. Artificial lights from cities scatter off atmospheric aerosols—such as dust, water droplets, and pollutants—creating a pervasive "glow" that obscures the faint light of stars. This light pollution reduces contrast, shrinks dynamic range, and introduces unnatural brownish or grayish hazes that diminish the aesthetic appeal of long-exposure photography.

While image dehazing is a well-studied field, light pollution is a distinct phenomenon. Unlike daytime haze, where the sun provides a relatively uniform, white light source, light pollution is spatially non-uniform, varies in spectral signature (color), and originates from the ground. Existing enhancement techniques often fail because they do not account for the specific physics of how artificial light attenuates and scatters at varying altitudes and distances.

Physically-Based Methodology

The researchers frame the task as an inverse problem: if a polluted image $\hat{I}$ is the sum of a pristine image $I$ and a "jamming" pollution image $J$, the goal is to recover $I$.

Physical model of light pollution scattering in the atmosphere.

Physical Modeling and Adaptive Estimation

The algorithm is built upon a model derived from Bouguer’s exponential law and the inverse-square law, describing how ground light reaches atmospheric points and scatters toward the camera. Recognizing that city lights are non-uniform, the authors developed an Adaptive Method to estimate the horizontal radiance profile of the pollution by comparing input images to "calibration images"—known pristine night sky photos.

Advanced Filtering and Depth Awareness

To prevent celestial bodies like stars and the moon from interfering with the estimation of the background glow, the team utilized a novel Quasi-Quartile Filter. This isolates the smooth background light pollution from sharp points of light.

For urban scenes, the volume of polluted air between the camera and a subject is critical. The authors integrated deep-learning-based single-image depth estimation to determine the appropriate amount of pollution to subtract at different distances. Edge-guided filtering was then applied to refine these depth maps, preventing artifacts like "halos" around buildings.

Key Results and Insights

The LPR algorithm was rigorously compared against eight other methods, including CLAHE, LIME, and state-of-the-art dehazing networks like GridDehazeNet.

Comparison between original light-polluted images (top/bottom left) and results from the LPR algorithm (top/bottom right).
  • Natural Landscapes: LPR successfully reduced background glow, revealing hidden star details and cloud textures that other methods either obscured or distorted.
  • Urban Scenes: In cityscapes, the algorithm significantly increased dynamic range and enhanced building surface details by removing the "veil" of nearby streetlights.
  • Ablation Study: The results confirmed that the "Adaptive" approach to non-uniform ground lights is superior to baseline models that assume uniform lighting, which often leave visible intensity ramps in the sky.

Conclusion and Future Work

The LPR algorithm demonstrates that treating light pollution as a volumetric, physically-grounded effect allows for superior restoration compared to general-purpose dehazing. While the method currently relies on calibration priors and the accuracy of depth estimation models, it represents a major step forward for nighttime image enhancement. Future research may involve more sophisticated spectral analysis of specific lamp types, such as Sodium-vapor versus LED, to further improve color accuracy in complex urban environments.

Summary updated

Abstract

Nighttime photographers are often troubled by light pollution of unwanted artificial lights. Artificial lights, after scattered by aerosols in the atmosphere, can inundate the starlight and degrade the quality of nighttime images, by reducing contrast and dynamic range and causing hazes. In this paper we develop a physically-based light pollution reduction (LPR) algorithm that can substantially alleviate the aforementioned degradations of perceptual quality and restore the pristine state of night sky. The key to the success of the proposed LPR algorithm is an inverse method to estimate the spatial radiance distribution and spectral signature of ground artificial lights. Extensive experiments are carried out to evaluate the efficacy and limitations of the LPR algorithm.
BibTeX
@misc{liu2021light,
  author = {Chang Liu and Xiaolin Wu},
  title = {{Light Pollution Reduction in Nighttime Photography}},
  year = {2021},
  eprint = {2106.10046},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2106.10046},
}

Loading related content...

Figure