RL-AWB: Deep Reinforcement Learning for Auto White Balance Correction in Low-Light Night-time Scenes
Image generated by Gemini AI
The RL-AWB framework addresses nighttime color constancy in computational photography by merging statistical methods with deep reinforcement learning. It utilizes a tailored statistical algorithm for gray pixel detection and illumination estimation, then optimizes parameters dynamically, emulating expert tuning. A new multi-sensor nighttime dataset supports cross-sensor evaluation, showing improved performance in varying light conditions. More details can be found on the project page.
Novel Framework Utilizes Deep Reinforcement Learning for Nighttime White Balance Correction
Researchers have unveiled RL-AWB, a framework designed to enhance color constancy in low-light nighttime photography, addressing challenges related to low-light noise and complex illumination conditions.
RL-AWB integrates statistical methods with deep reinforcement learning (DRL) to improve white balance. It uses a statistical algorithm for nighttime scenes, including gray pixel detection and illumination estimation, addressing unique challenges posed by night-time environments.
Notably, this is the first application of deep reinforcement learning to color constancy, optimizing parameters for individual images to enhance color reproduction quality in low-light settings.
Cross-Sensor Evaluation and Dataset Introduction
The research team has introduced the first multi-sensor nighttime dataset for comprehensive cross-sensor evaluation, serving as a valuable resource for future studies in nighttime photography.
Experimental results indicate that RL-AWB excels in low-light conditions and shows superior generalization across various lighting scenarios, critical for real-world applications.
Related Topics:
📰 Original Source: https://arxiv.org/abs/2601.05249v1
All rights and credit belong to the original publisher.