Seeing in Extra Darkness Using a Deep-Red Flash

Jinhui Xiong*, Jian Wang*, Wolfgang Heidrich, Shree Nayar
CVPR 2021 (Oral)

Top left: Human vision uses cones and rods for the perception of light. Photopic vision is associated with cones, occurring at bright-light conditions (over 3 cd/m^2). Scotopic vision is associated with rods, occurring at dim-light conditions (below 10^{-3} cd/m^2). At intermediate light levels, both rods and cones are active, which is called mesopic vision. Bottom left: We propose to use deep-red (e.g. 660 nm) light as flash for low-light imaging in mesopic light levels. This new flash can be introduced into smartphones with a minor adjustment. Middle: The eye spectral sensitivity in a dimly lit environment (0.01 cd/m^2) and the relative responses of R, G and B color channels of the camera we used, as well as the emissions spectrum of the red LED flash. Under dim lighting, rod vision dominates, yet the rods are nearly insensible to deep-red light. Meanwhile, our LED flash can be sensed by the camera especially in the red and green channels. Right: Inputs to our videography pipeline are a sequence of no-flash and flash frames, and the outputs are denoised and would yield temporally stable videos with no frame rate loss.


We propose a new flash technique for low-light imaging, using deep-red light as an illuminating source. Our main observation is that in a dim environment, the human eye mainly uses rods for the perception of light, which are not sensitive to wavelengths longer than 620 nm, yet the camera sensor still has spectral response. We propose a novel modulation strategy when training a modern CNN model for guided image filtering, fusing a noisy RGB frame and a flash frame. This fusion network is further extended for video reconstruction. We have built a prototype with minor hardware adjustments and tested the new flash technique on a variety of static and dynamic scenes. The experimental results demonstrate that our method produces compelling reconstructions, even in extra dim conditions.


paper [SeeingInExtraDarkness.pdf (12.2MB)]
code  []