LUCK: Lighting Up Colors in the Dark
Yaping Zhao, Edmund Y. M. Lam · 2024
Low-light imaging is challenging, especially in scenarios like nighttime or dim indoor environments, where images often suffer from color distortion and noise. Traditional RGB cameras with Bayer filters face limitations such as low photon capture rates and quantum efficiency, leading to darker images. When compensated by longer exposure times or higher sensitivity settings, issues such as motion blur and noise amplification arise. Enhancements for RGB cameras are limited by these hardware constraints and do not address the fundamental problem of insufficient photon reception. This paper focuses on the dual-camera system that combines an RGB camera with a monochrome camera to improve low-light imaging. This system uses the color processing capabilities of the RGB camera alongside the higher photon capture rate of the monochrome camera. Specifically, we design a comprehensive computational imaging framework with feature extraction, alignment, and fusion modules to process and synthesize images from both cameras into a single high-quality output. Experimental results confirm the effectiveness of our approach, significantly enhancing image quality and achieving over a 2dB increase in PSNR compared to state-of-the-art methods. Our research demonstrates the potential of dualcamera systems in low-light settings and indicates a promising direction for future advancements in photography technology. Code is available at: https://github.com/IndigoPurple/LUCK.