Hair-MAP: a prototype automated system for forensic hair comparison and analysis.

Area of Science:

• Forensic Science
• Computer Science
• Artificial Intelligence

Background:

• Forensic hair analysis is crucial for criminal investigations.
• Manual analysis is time-consuming and subjective.
• Automation can improve efficiency and objectivity.

Purpose of the Study:

• To demonstrate the feasibility of automating forensic hair analysis and comparison.
• To develop a reliable and understandable automated system using neural networks.
• To assess the accuracy of an automated hair matching system.

Main Methods:

• Utilized neural network explanation systems (NNESs) for hair analysis.
• Employed image processing techniques, including wavelet analysis and Haralick texture algorithm.
• Implemented neural networks for feature classification and statistical tests for match determination.
• Developed decision trees to explain neural network behavior.

Main Results:

• Achieved 83% accuracy in hair match classification.
• Successfully compressed large image data using pre-processing techniques.
• Demonstrated the system's ability to classify features from microscopic hair images.
• The system used 5 out of 21 morphological characteristics for analysis.

Conclusions:

• Automated forensic hair analysis using NNESs is feasible.
• The developed system shows potential to assist forensic experts.
• Further development could lead to a more comprehensive automated analysis tool.
• The system provides a means to pre-process data, easing the expert's workload.