Research publication

Research that helps face-recognition systems reject fake inputs

Published deep-learning research on distinguishing live faces from presentation attacks such as photos, videos, and masks.

Fact 1
80,000 frames
Fact 2
Multi-modal data
Fact 3
Published by IEEE
Period
2022

The problem

Face-recognition systems can be fooled by presentation attacks. Security-critical deployments need a reliable way to distinguish a live person from a fake input.

What I built

Working with a research team, I developed and evaluated deep-learning methods on a multi-modal dataset of 80,000 frames, including data focused on South Asian ethnicity.

The result

The work achieved high detection accuracy and was published through IEEE, contributing practical evidence to biometric-security research.

Technical details

For technical reviewers, the delivery used the following tools and platforms. The business problem and result remain the primary scope.

  • Deep learning
  • Computer vision
  • Transformer models
  • RGB and thermal data