A Comparison of Thresholding Methods for Forensic Reconstruction Studies Using Fluorescent Powder Proxies for Trace...

Area of Science:

• Forensic Science
• Image Analysis
• Computational Imaging

Background:

• Quantitative image analysis relies on accurate image segmentation.
• A standardized methodology for segmenting fluorescent proxy images for trace evidence is currently unavailable.
• Manual segmentation and thresholding methods exhibit variability and limitations in reproducibility.

Purpose of the Study:

• To evaluate the reproducibility of manual image segmentation among examiners.
• To assess the accuracy and reproducibility of manual thresholding.
• To compare algorithmic global thresholding methods against manual approaches for fluorescent proxy image segmentation.

Main Methods:

• Manual segmentation reproducibility was assessed with three participants tracing three images repeatedly.
• Manual thresholding accuracy and reproducibility were evaluated by 20 examiners segmenting 10 images.
• Sixteen global thresholding algorithms were tested for accuracy, reproducibility, and efficiency.

Main Results:

• Statistically significant differences were observed in manual segmentation outputs between examiners.
• Manual thresholding demonstrated good average accuracy (within ±1%) but poor reproducibility with outliers.
• Three algorithms (Yen, MaxEntropy, RenyiEntropy) achieved comparable accuracy with superior reproducibility and efficiency.

Conclusions:

• Algorithmic thresholding provides a more robust and reproducible alternative to manual methods for fluorescent proxy image segmentation.
• The tested algorithms (Yen, MaxEntropy, RenyiEntropy) can enhance the reliability of quantitative image analysis in trace evidence reconstruction studies.
• Implementing standardized algorithmic approaches is crucial for consistent and accurate trace evidence analysis.