Latent Fingerprint Quality Assessment for Criminal Investigations: A Benchmark Dataset and Method

Area of Science:

• Biometrics
• Forensic Science
• Computer Vision

Background:

• Automated fingerprint recognition is vital for criminal investigations but is hindered by low-quality crime scene fingerprints.
• Effective Fingerprint Quality Assessment (FQA) is crucial for improving the usability and accuracy of biometric identification systems.

Purpose of the Study:

• To introduce the Crime Scene Fingerprints quality assessment Dataset (CSFD-10k), the largest dataset of its kind for forensic fingerprint analysis.
• To propose a novel deep neural network-based Dual-Branch FQA (DB-FQA) framework for enhanced fingerprint quality assessment.

Main Methods:

• Established the CSFD-10k dataset with 11,500 crime scene fingerprint images, labeled by police officers.
• Developed the DB-FQA framework integrating image-level and edge-level features using a Logical/Linear operator for edge map transformation.
• Utilized a Multi-scale Adaptive Cross feature Fusion (MACF) module to fuse features and highlight quality-related regions.

Main Results:

• The proposed DB-FQA method demonstrated robustness and superiority in extensive experiments.
• The framework effectively enhances ridge details and improves identification accuracy for low-quality fingerprints.
• The CSFD-10k dataset and DB-FQA framework provide significant support for forensic fingerprint biometrics.

Conclusions:

• The developed DB-FQA framework offers a significant advancement in assessing fingerprint quality from challenging crime scene conditions.
• The CSFD-10k dataset facilitates further research and development in forensic fingerprint analysis.
• This work enhances the reliability and accuracy of fingerprint biometrics in real-world forensic applications.