Latent Fingerprint Matching: Performance Gain via Feedback from Exemplar Prints

Area of Science:

• Forensic Science
• Computer Vision
• Biometrics

Background:

• Latent fingerprints are crucial forensic evidence.
• Matching latent prints to exemplars is challenging due to poor quality and noise.
• High accuracy in automatic latent fingerprint matching is vital for legal applications.

Purpose of the Study:

• To improve the accuracy of automatic latent fingerprint matching.
• To develop a feedback mechanism using exemplar information for latent feature refinement.
• To determine the conditions under which feedback is necessary for improved matching.

Main Methods:

• Incorporating top-down information (feedback) from exemplar prints to refine latent features.
• Using refined features (ridge orientation, frequency) for re-matching.
• Developing a feedback paradigm adaptable to existing latent matchers.
• Evaluating the necessity of feedback for accuracy improvement.

Main Results:

• The proposed feedback paradigm improved identification accuracy by 0.5-3.5% on NIST SD27 and WVU databases.
• The system was tested against a background database of 100,000 exemplars.
• The feedback approach enhanced the performance of a state-of-the-art latent matcher.

Conclusions:

• The feedback mechanism effectively refines latent features, boosting matching accuracy.
• This approach offers a generalizable method to enhance existing latent fingerprint identification systems.
• The research provides a systematic way to leverage exemplar data for better forensic analysis.