Spectral response characteristic-driven ICMOS color night vision imaging system and color reconstruction network

Area of Science:

• Optics and Photonics
• Computer Vision
• Image Processing

Background:

• Color night vision devices face challenges with nonlinear color shifts and noise under low-light conditions.
• Existing low-light datasets lack the quality and paired data needed for effective training.
• Spectral response mismatch significantly impacts color accuracy in low-light imaging.

Purpose of the Study:

• To develop an intensified complementary metal-oxide semiconductor (ICMOS) system for high-fidelity color night vision.
• To address severe color deviations and noise in extremely low-light imaging.
• To create a robust method for generating high-quality, paired low-light color datasets.

Main Methods:

• Implemented an ICMOS-based low-light color imaging system utilizing time-division imaging and spectral matching.
• Developed a physics-based synthetic data generation method using real multispectral and environmental light data.
• Designed a dual-branch color reconstruction network for image enhancement.

Main Results:

• Successfully corrected severe color deviations caused by spectral response mismatch.
• Significantly improved the signal-to-noise ratio and color fidelity of low-light images.
• Achieved high-quality, natural-looking color night vision at illuminance levels of 10^-3-10^-4 lx.

Conclusions:

• The proposed ICMOS system and reconstruction network effectively overcome limitations in extreme low-light color imaging.
• The physics-based synthetic dataset generation method addresses the scarcity of high-quality training data.
• This approach enables natural-looking color night vision imaging under challenging conditions.