Enhancing automated shoeprint comparison via synthetic data generation and deep segmentation.

Area of Science:

• Forensic Science
• Computer Vision
• Pattern Recognition

Background:

• Forensic footwear impressions are crucial for criminal investigations.
• Automated shoeprint analysis faces challenges like background clutter, occlusion, and print degradation.
• Limited availability of labeled data hinders algorithm development.

Purpose of the Study:

• To develop and evaluate a framework for improving algorithmic shoeprint comparison.
• To address challenges in automated analysis using controllable synthetic data and deep segmentation.
• To provide a reproducible evaluation pipeline for shoeprint comparison methods.

Main Methods:

• Generated a synthetic dataset with systematic variations in background, noise, occlusion, and degradation.
• Trained a deep segmentation model to isolate shoeprint regions from cluttered backgrounds.
• Evaluated the impact of segmentation-based preprocessing on shoeprint similarity scoring and verification performance.

Main Results:

• Segmentation-based preprocessing significantly improved verification performance compared to non-segmented baselines.
• The framework demonstrated improved shoeprint comparison under challenging conditions.
• Synthetic data generation allowed for controlled evaluation of algorithm robustness.

Conclusions:

• The proposed framework offers a robust method for evaluating algorithmic shoeprint comparison.
• Deep segmentation preprocessing enhances the accuracy of forensic footwear analysis.
• This approach facilitates reproducible and reliable assessment of shoeprint comparison technologies.