Computational multi-spectral fusion of visible to near-infrared bands for enhanced low-light imaging

Achieving high-quality imaging under extremely low-light conditions is crucial for applications such as autonomous driving and night surveillance. Traditional RGB and visible (VIS) fusion methods struggle with noise reduction due to limited input and significant noise interference, while near-infrared (NIR) and VIS fusion face color interpretation differences. This study proposes a computational framework for multi-spectral fused imaging, integrating information across NIR and VIS ranges while considering human perception of color. Introduce NVFNet for noise suppression by referring and fusing rich optical information from VIS and NIR spectrums. A novel, to our knowledge, attention fusion module systematically combines spectral and spatial features, enabling multi-source cross-referencing for noise suppression and high color fidelity imaging. A pilot optical system was developed to capture a real-scene multi-spectral-RGB dataset under illuminance conditions below 0.001 lx per spectrum. Experimental results demonstrate that the proposed method significantly outperforms traditional RGB imaging techniques, achieving an average improvement of over 8.58 dB in peak signal-to-noise ratio (PSNR) and 0.25 in structural similarity index (SSIM).