Innovative grayscale-assisted method converts near-infrared images to vibrant RGB visuals

Researchers at Beijing Institute of Technology (BIT) have developed an innovative method to convert near-infrared (NIR) images into visible RGB images. The study, titled "Grayscale-Assisted RGB Image Conversion from Near-Infrared Images," recently published in Tsinghua Science and Technology, offers a robust solution for enhancing NIR image quality, which typically suffers from poor luminance and lack of chromatic detail.

"NIR imaging provides critical advantages like atmospheric penetration and robust anti-interference capability, crucial for fields like assisted driving and surveillance systems," explained Professor Ying Fu, the corresponding author of the study. "However, these images inherently lack luminance and chrominance, limiting their usability. Our approach effectively resolves this issue by breaking down the conversion process into two more manageable phases."

In the first phase, the NIR images are converted into grayscale images, significantly simplifying the luminance restoration process. This step addresses the inherent luminance deficiency typically seen in NIR images. In the second phase, the grayscale images undergo colorization using extensive datasets, such as ImageNet, to accurately restore chrominance. The researchers further enhanced this method by integrating Frequency Domain Learning (FDL), which leverages Fast Fourier Convolution to improve the preservation and recovery of image details, textures, and edges.

“Our method demonstrates remarkable improvements over previous direct NIR-to-RGB conversions,” said Yunyi Gao, the first author and master's student at BIT. "By using grayscale as an intermediary step, we harness large-scale grayscale-RGB datasets that significantly enhance the conversion quality."

Extensive tests on widely recognized datasets, including ICVL and TokyoTech, showed the superiority of the new method. The experimental results demonstrate higher Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and reduced color errors (Delta-E), validating its effectiveness in producing more visually appealing and detailed RGB images.

"This grayscale-assisted method has practical implications for enhancing image quality in systems that rely heavily on NIR imaging, such as assisted vehicles and security monitoring," commented Qiankun Liu, a co-author of the paper. "By significantly improving the visual clarity and color accuracy, our approach can lead to more reliable automated image interpretation and decision-making processes."

The researchers plan to further refine this method, particularly focusing on domain adaptation techniques to enhance compatibility between NIR and grayscale images. Future research aims to extend the applicability of this technology to more advanced imaging systems and practical scenarios.

The work was supported by the National Natural Science Foundation of China (Grant Nos. 62331006, 62171038, and 62088101) and the Fundamental Research Funds for the Central Universities.

About the Authors:

Yunyi Gao is a master's student at the School of Computer Science & Technology, Beijing Institute of Technology. His research primarily focuses on near-infrared to visible light image conversion.

Qiankun Liu received his B.S. degree in information engineering from Xidian University in 2017, and Ph.D. degree in information and communication engineering from University of Science and Technology of China in 2022. He finished his Postdoc research in Beijing Institute of Technology in 2024. He is currently a Special-Term Associate Professor in University of Science and Technology Beijing. His research interests include object detection and tracking, image synthesis, and 3D reconstruction and synthesis.

Lin Gu completed his Ph.D. at the Australian National University and NICTA in 2014. Currently, he is a research scientist at RIKEN AIP, Japan, and a special researcher at the University of Tokyo. His research interests include hyperspectral imaging and color science.

Ying Fu (corresponding author) received her BEng degree from Xidian University, China, in 2009, her MEng degree from Tsinghua University, China, in 2012, and her PhD degree from The University of Tokyo, Japan, in 2015. She is currently a professor at the School of Computer Science & Technology, Beijing Institute of Technology, China. Her research interests include computational videography, computer vision, machine learning, and multimedia image video analysis. More information can be found at her homepage: https://ying-fu.github.io/.

[1] Y. Gao, Q. Liu, L. Gu and Y. Fu, "Grayscale-Assisted RGB Image Conversion from Near-Infrared Images," in Tsinghua Science and Technology, vol. 30, no. 5, pp. 2215-2226, October 2025, doi: 10.26599/TST.2024.9010115.