Photon-counting statistics-based support vector machine with multi-mode photon illumination for quantum imaging.

Area of Science:

• Quantum Imaging
• Photon Statistics
• Machine Learning

Background:

• Background photon noise degrades image quality in quantum imaging systems.
• Distinguishing signal photons from noise is crucial for high-fidelity imaging.
• Practical quantum imaging often involves low photon counts and noise challenges.

Purpose of the Study:

• To develop a photon-counting statistics-based method for eliminating background noise in quantum imaging.
• To improve image quality by distinguishing signal photons from background noise.
• To apply machine learning for classifying photon-counting distributions.

Main Methods:

• Utilizing photon-counting statistics derived from the multi-mode Bose-Einstein distribution.
• Employing a machine learning technique, specifically a support vector machine (SVM), for classification.
• Developing a photon-counting statistics-based support vector machine (PSSVM) to learn pixel-level photon distributions.

Main Results:

• The PSSVM effectively distinguishes between signal photons and background noise based on photon-counting distributions.
• Demonstrated quantum imaging of a binary object using spontaneous parametric down-conversion (SPDC).
• Achieved a 2.89 dB gain in peak signal-to-noise ratio (PSNR) and a 27.7% gain in structural similarity index measure (SSIM).

Conclusions:

• The PSSVM method successfully enhances quantum image quality by mitigating background noise.
• Photon-counting statistics offer a viable approach for noise reduction in low-cost quantum imaging.
• The proposed technique provides significant improvements over conventional direct single-photon imaging.