Single photon compressive imaging with enhanced quality using an untrained neural network.

Area of Science:

• Photonics
• Computational Imaging
• Artificial Intelligence

Background:

• Traditional single photon compressive imaging suffers from poor image quality.
• Deep learning methods improve imaging but require extensive training datasets, posing a significant challenge.

Purpose of the Study:

• To address the limitations of traditional and deep learning-based single photon compressive imaging.
• To introduce an untrained neural network approach for improved imaging quality and reduced data requirements.

Main Methods:

• Development and implementation of an imaging system utilizing an untrained neural network.
• Conducting simulation studies using the Monte Carlo method to validate the approach.
• Analyzing the impact of input images, imaging types, and anti-noise capabilities on Convolutional Neural Networks (CNNs).

Main Results:

• The proposed untrained neural network method significantly improved image quality in single photon compressive imaging.
• The approach successfully eliminated the need for large training datasets, addressing a key challenge.
• The study demonstrated the network's ability to maintain imaging speed and alter system sensitivity to photon numbers.

Conclusions:

• Untrained neural networks offer a viable solution for enhancing single photon compressive imaging quality and overcoming data limitations.
• The findings support the potential of CNNs for natural image processing and provide a foundation for future research in this domain.
• This work paves the way for advancements in single photon compressive imaging and the application of untrained neural networks.