Forensic gender and stature identification from footprint images using machine learning

Area of Science:

• Forensic Science
• Biometrics
• Computer Vision

Background:

• Traditional footprint analysis for forensic investigation is time-consuming and requires expert judgment.
• Automated methods using image analysis and machine learning offer potential for efficiency and accuracy.

Purpose of the Study:

• To develop and benchmark traditional machine learning (ML) models for automated gender classification and stature estimation from footprint images.
• To assess the feasibility of traditional ML methods for forensic gait analysis given modest data requirements.

Main Methods:

• Employed image pre-processing for Region of Interest extraction and segmentation of footprints.
• Utilized a dataset of 396 footprints from 33 participants.
• Benchmarked Logistic Regression (LR), Gaussian Naive Bayes (GNB), K-Nearest Neighbor (KNN), Decision Tree Classifier (DTC), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) models.

Main Results:

• K-Nearest Neighbor (KNN) achieved the highest accuracy (0.91) for gender classification.
• Extreme Gradient Boosting (XGBoost) demonstrated superior performance for stature estimation with a Mean Absolute Error (MAE) of 4.10 cm and Root Mean Square Error (RMSE) of 5.42 cm.
• Observed varying strengths and weaknesses among classifiers for both tasks.

Conclusions:

• Traditional ML methods show promise as a feasible solution for automated gender and stature estimation from footprints in forensic contexts.
• Expanding datasets with greater diversity and varying quality is crucial for developing a robust end-to-end solution.
• Future work may explore advanced deep learning methods with larger, more diverse datasets.