Vein pattern visualisation for biometric identification with cGAN on a New Zealand dataset

Area of Science:

• Forensic Science
• Biometrics
• Computer Vision

Background:

• Vein pattern recognition for forensic identification is challenged by low visibility in standard color (RGB) images.
• Existing computational methods like neural networks and optical vein disclosure show limitations in reliability compared to Near-Infrared (NIR) imaging.
• A scarcity of synchronized RGB and NIR image datasets for body limbs hinders research progress.

Purpose of the Study:

• To introduce a novel, ethically approved dataset of 602 synchronized NIR and RGB forearm images from a diverse population.
• To propose a conditional Generative Adversarial Network (cGAN) model for translating RGB images into their NIR equivalents.
• To evaluate the effectiveness of the cGAN model in enhancing vein pattern visibility and accuracy for forensic identification.

Main Methods:

• Collection of a new dataset of 602 paired NIR and RGB forearm images.
• Development and implementation of a conditional Generative Adversarial Network (cGAN) model.
• Evaluation of the translated vein patterns based on matching accuracy, vein length measurements, and contrast quality.

Main Results:

• The proposed cGAN model successfully translates RGB images into NIR-like vein patterns.
• Translated vein patterns demonstrate high similarity to actual NIR counterparts.
• Evaluations show promising results in matching accuracy, vein length consistency, and improved contrast quality.

Conclusions:

• The developed dataset and cGAN model offer a significant advancement for research in vein pattern-based forensic identification.
• The ability to generate NIR-equivalent images from RGB data enhances the potential for reliable identification using standard imaging.
• This approach holds promise for improving the accuracy and accessibility of forensic identification techniques.