IllumiCurveNet: Low-Light Image Enhancement of Lunar Permanently Shadowed Regions Using a Self-Guided Loss Framework

Saksham Jain, Sparsh Jain, Ashish Prajapati, Garvit Singh, Dinesh Kumar Vishwakarma

Abstract: Lunar Permanently Shadowed Regions (PSRs) are areas near the Moon’s poles that remain in perpetual darkness due to its axial tilt. Obtaining clear and high-quality images of these regions are crucial for exploring lunar surface and detecting valuable minerals. However, due to the absence of illumination, PSR images often suffer from low visibility, poor contrast, and elevated noise levels, making their enhancement a significant challenge. To overcome these challenges, this paper introduces IllumiCurveNet, a novel framework leveraging an encoder-decoder architecture with spatial attention, dilated convolutions, and adaptive gamma correction for illuminance optimization. It uses the proposed Self-Guided Loss Framework that integrates the novel texture preservation and contrast enhancement losses, along with exposure control, spatial consistency, color consistency, and total variation losses, enabling robust enhancement without paired training data. IllumiCurveNet achieves state-of-the-art performance on PSR images with no-reference image quality metrics, surpassing other zero-shot methods. The results highlight IllumiCurveNet’s potential for applications in lunar mapping, rover navigation, and resource analysis, advancing visual perception in unlit extraterrestrial environments.

External IDs:dblp:conf/ijcnn/JainJPSV25