IPLNet: a neural network for intensity-polarization imaging in low light

Area of Science:

• Computer Vision
• Optical Imaging
• Artificial Intelligence

Background:

• Low-light imaging presents significant challenges across various applications.
• Polarization imaging offers advantages over conventional intensity imaging but faces quality limitations.
• Integrating intensity and polarization information is difficult due to differing operator characteristics.

Purpose of the Study:

• To address the limitations of low-light imaging by enhancing both intensity and polarization images simultaneously.
• To introduce the first polarimetric imaging dataset specifically collected in low-light conditions.
• To develop a specialized neural network for improved low-light polarimetric imaging.

Main Methods:

• Collection of the first polarimetric imaging dataset in low-light environments.
• Design and implementation of a specialized neural network architecture.
• Simultaneous enhancement of intensity and polarization image qualities using the proposed network.

Main Results:

• Demonstrated effectiveness of the neural network in enhancing both intensity and polarization images.
• Superior performance compared to conventional methods in low-light conditions.
• Validation through both indoor and outdoor experimental setups.

Conclusions:

• The developed neural network effectively enhances low-light polarimetric images.
• This approach offers a significant improvement for vision in photon-starved environments.
• Potential applications include object detection and enhanced imaging in challenging conditions.