Tech News

Goodbye green night vision: They develop a camera that captures images as if they were daytime

When you talk about night vision everyone imagines that the displayed image will be in green with black, however this is about to change. Researchers have found a way to cameras capture, even at night, a color imageas if it had been taken in the day.


This April 6 the magazine Pls One, published an American article where researchers present the finding an optimized algorithm with architecture of deep learning what does it achieve transform the visible spectrum of a night scene to how a person could see it in the day.

During the night, people cannot see the colors and contrasts due to the lack of light, for this they need to illuminate the area or use night vision goggles, the latter give a greenish image. By solving the monochrome of the viewfinders, it will be possible for everyone to see and take photos that look as if it were daytime, which will be of great help in tactical military reconnaissance work, among others.


To achieve this, the researchers used a monochrome camera sensitive to visible and infrared light to acquire the database of an image or printed images of faces under multispectral illumination covering the standard visible eye.

They later optimized a convolutional neural network (U-Net) to predict visible spectrum images from near infrared images. Its algorithm is powered by deep learning by spectral light.

“We set up a controlled visual context with limited pigments to test our hypothesis that deep learning can render scenes visible to humans using NIR illumination [Infrarrojo Cercano] which is otherwise invisible to the human eye.


To learn the spectrum of spectral reflectance of cyan, magenta, and yellow inks, they printed the Rainboy color palette to record their wavelengths. They then printed several images and placed them under multispectral illumination with a monochrome (black and white) camera, mounting on a dissecting microscope focused on the image.

In total they printed a library of more than 200 human faces available in the publicationLabeled Faces in the Wild”, with a Canon printer and MCYK paint. The images were put under different wavelengths and then used in the training of the machine learning (machine learning) focused on predict color (RGB) images from single or combined single wavelength illuminated images.

“To predict RGB color images from single wavelength illuminations or combinations, we evaluated the performance of the following architectures: a baseline linear regression, a CNN [Red Neuronal Convolucional) inspirada en U-Net (UNet) y una U-Net aumentada con pérdida adversaria (UNet-GAN )”.

Para todos los experimentos, siguieron el modelo práctico de machine learning: dividieron la base de datos en 3 partes, reservando 140 imágenes para entrenamiento, 40 para validación y 20 para pruebas. Para comparar el rendimiento entre diferentes modelos evaluaron varias métricas para la reconstrucción de la imagen.


Los investigadores señalaron que este estudio sirve como un paso para la predicción de escenas del espectro visible humano a partir de una iluminación infrarroja cercana imperceptible.

Aseguraron que “sugiere que la predicción de imágenes de alta resolución depende más del contexto de entrenamiento [de la Máquina] that of the spectroscopic signatures of each ink” and that this work should be a step for night vision videos, for which the number of frames it processes per second will depend.


Back to top button