Statistics of fingerprint minutiae frequency and distribution based on automatic minutiae detection method

Area of Science:

• Forensic Science
• Biometrics
• Computer Vision

Background:

• The scientific validity of fingerprint evidence has been debated since the Daubert case.
• Traditional fingerprint identification methods face ongoing scrutiny.
• Quantitative, statistics-based approaches to fingerprint minutiae analysis are emerging.

Purpose of the Study:

• To develop an automatic algorithm for classifying fingerprint minutiae.
• To statistically evaluate the occurrence frequencies of different minutiae types.
• To provide data for developing probabilistic fingerprint identification models.

Main Methods:

• Utilized the deep convolution neural network YOLOv5 for automatic minutiae detection and classification.
• Analyzed a large dataset of 619,297 fingerprint images.
• Calculated frequency ranges for six types of minutiae: ridge endings, bifurcations, independent ridges, spurs, lakes, and crossovers.

Main Results:

• Established frequency ranges for six minutiae types per finger (e.g., ridge endings [68.49, 70.81]%, bifurcations [26.37, 27.26]%).
• Observed variations in minutiae distribution across ten finger positions and four finger patterns (arch, left loop, right loop, whorl).
• Quantified the number and frequency of minutiae, differentiating their evaluative significance.

Conclusions:

• The study provides crucial quantitative data on fingerprint minutiae frequencies.
• Results support the development of more robust, probability-based fingerprint identification systems.
• This research addresses the need for statistically grounded evidence in forensic biometrics.