Building a ground-truth fingerprint dataset for proficiency testing and research.

Area of Science:

• Forensic Science
• Biometrics
• Image Analysis

Background:

• Collecting realistic fingermark evidence is crucial for forensic investigations.
• Standardized datasets are essential for training and validating fingerprint identification systems.
• The Royal Canadian Mounted Police (RCMP) required a robust dataset for research and proficiency testing.

Purpose of the Study:

• To outline the methodology for creating a controlled fingermark image dataset.
• To ensure the dataset accurately simulates crime scene evidence.
• To detail the process of identifying and including close non-matches within the dataset.

Main Methods:

• Fingermarks were collected from individuals on various substrates using diverse development techniques and residues.
• Close non-matches were identified using an Automated Fingerprint Identification System (AFIS).
• An efficient workflow was established for image capture, encoding, and storage, resulting in 3011 images from 62 individuals.

Main Results:

• A comprehensive fingermark dataset was successfully created, mimicking crime scene evidence.
• The dataset includes fingermarks and carefully selected close non-matches.
• An efficient capture and storage workflow was implemented by the RCMP.

Conclusions:

• The developed dataset provides a foundation for future research and proficiency testing in fingerprint analysis.
• Lessons learned emphasize the importance of object cleanliness and rigorous quality control, including fingerprint comparison verification.
• The dataset is currently restricted to internal RCMP use for proficiency testing.