Photophysical image analysis: Unsupervised probabilistic thresholding for images from electron-multiplying...

Area of Science:

• Image analysis
• Computational imaging
• Photonics

Background:

• Electron-multiplying charge-coupled device (EMCCD) cameras are widely used for low-light imaging.
• Existing image thresholding methods often lack a priori error estimation.
• Accurate noise modeling is crucial for reliable image analysis.

Purpose of the Study:

• To introduce a novel unsupervised probabilistic image thresholding pipeline for EMCCD cameras.
• To develop a method for a priori determination of misclassified pixels.
• To provide a robust framework for photophysical image analysis (PIA).

Main Methods:

• Developed a closed-form analytic expression for the characteristic function of EMCCD image counts.
• Incorporated photon arrival stochasticity and camera detection noise.
• Estimated background photon statistics (λbg) using a truncated fit procedure.
• Introduced a probabilistic thresholding method with a priori error estimation.
• Validated using synthetic images and compared against the Otsu method.

Main Results:

• The proposed method accurately estimates background photon statistics and camera noise parameters.
• Probabilistic thresholding allows for a priori determination of misclassification fractions.
• Demonstrated superior performance compared to the Otsu method in synthetic image benchmarks.
• Successfully applied to segment synthetic and experimental images.

Conclusions:

• The developed PIA pipeline offers automated, unsupervised, and probabilistic image thresholding for EMCCD cameras.
• This approach enables precise error quantification in image analysis.
• Publicly available software facilitates advanced photophysical image analysis.