Color imaging provides more information about the captured scene, but under conditions of low illumination or scattering, this color information is degraded or lost. It is necessary to use longer wavelengths, such as infrared (IR), to pass through the medium and capture the scene image. Although these images obtained through IR are monochromatic, in some applications, the color information provides details that can help us to discriminate objects or make decisions.
We proposed using a vision system active illumination single-pixel imaging 2D color in the near-infrared (NIR) at wavelength 850-1550 nm, with a resolution of 32 x 32 image size. For the generation of 2D color imaging, the spatial multi-spectral mask method was proposed in combination with deep learning using a NoGAN approach. To evaluate the reconstructed color image, we make laboratory tests, where we will measure the performance of the color quality of the reconstructed image through No-reference image quality assessment (NR-IQA) methods and processing time. To determine the feasibility of our system, we will make tests in real scenarios where the identification of scene objects' colors would be necessary, such as in rescue or surveillance conditions.
