Human sight is limited when it comes to seeing in the dark. This is because humans perceive light in the visible spectrum (400-700 nm), and to effectively see in the dark unaided requires the ability to see on the infrared scale.
Image Credit: Browne, A., et al., (20220) Deep learning to enable color vision in the dark. PLOS ONE, [online] 17(4), p.e0265185. Available at: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0265185
However, some night vision systems are able to help humans see better in nighttime/dark conditions as they harness infrared light and project a monochromatic image of the surroundings to a display.
These systems typically shine infrared light onto objects and capture the signals that bounce back. These signals are then converted into visible light, and while infrared technology aids human vision in a number of applications, researchers have been keen to advance the technology to render full-color videos or images in places with no light.
Now, a team of researchers at the University of California Irvine has made some progress towards achieving this goal with the application of an imaging algorithm that was optimized by deep learning architectures. Published in the journal PLOS ONE, the team describes how they have applied deep learning to achieve color vision in dark conditions.
Advancing Deep Learning
Deep learning has already proven effective in certain applications when it comes to adding color to monochromatic images and data. By teaching the system’s neural network what colors should be in an image or scene, deep learning techniques are able to turn black and white images into color.
In order to ‘teach’ the program, the neural network is presented with a large number of samples, i.e., input. Thus, training a model requires the input of historical data (images), which may be comprised of observations or examples with elements that describe the appropriate conditions – for seeing color in the dark – and an element that is able to translate what the observations mean.
We investigated if a combination of infrared illuminants in the red and nearinfrared (NIR) spectrum could be processed using deep learning to recompose an image with the same appearance as if it were visualized with visible spectrum light.
Andrew W. Browne, Assistant Professor of Ophthalmology at UC Irvine
The UC Irvine team has taken things a little further by applying some pretty sophisticated and advanced techniques to better train the deep learning neural network. By printing images of color palettes and faces, they were able to create a monochromatic camera that can be set to capture images at extremely specific wavelengths.
The team used a variety of monochromatic light sources across different wavelengths in the visible and near-infrared spectrums to train the system by taking numerous pictures of people’s faces to familiarize the AI system with the colors that make up human faces. They then applied the same approach to take a sequence of photographs in the dark.
The results showed that the system had the capability to make accurate estimations to color the pictures which were then transmitted to an external display. What this demonstrated was that by building upon the foundations of the existing research, the advanced deep learning architecture was able to reproduce full-color images from infrared data.
Future Applications
While this proof-of-concept is presently limited in scope – as the researchers admit the technology is only trained on images of human faces – it does mark an exciting leap for infrared-based night vision applications.
Advanced night vision has a wide range of applications, including security, military and surveillance operations, as well as a research tool for other fields such as monitoring the habits of nocturnal animals. The technology could also be used to study artifacts that are sensitive to light and require low- or no-light conditions.
Additionally, such a tool could have a positive impact in the field of medicine.
Studying the lightsensitive retinal tissue, for example, may require processing the sample in darkness to avoid altering its biochemistry and function. Likewise, performing eye surgery benefits from low light exposure to avoid retinal damage.
Andrew W. Browne, Assistant Professor of Ophthalmology at UC Irvine
While this technology will not be rolled out immediately, it does show that full-color night vision is a very real possibility as the technology evolves and becomes even more sophisticated.
The UC Irvine team will continue training the deep learning technology to develop infrared visualization systems that would benefit applications where there is an absence of visible light and help humans ‘see’ in color in the dark.
References and Further Reading
Browne, A., et al., (20220) Deep learning to enable color vision in the dark. PLOS ONE, [online] 17(4), p.e0265185. Available at: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0265185
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.