Biometric Authentification with Timeless Learner

IDIAP proposes a multispectral sensor array, backed by a modular recognition system, for detecting face presentation attacks (PAs). The proposed system consists of three stages. Biometric data (such as the images shown in the figure) are first captured by a set of sensors, (e.g., RGB camera, NIR camera, depth sensor, etc.). Data from the sensors are processed by an array of PA detectors. A detector may process data from multiple sensors. Scores from the individual detectors are combined into a final binary face PAD decision in the fusion stage. IDIAP explored already several approaches for face? PAD, including techniques involving video based liveness detection, analysis of image artifacts and image quality as well as image-texture properties. So far we have focused on testing the efficacies of specific hand crafted features in detecting specific kinds of PAs. In this project we plan to explore the use of deep learning tools to automatically extract features specifically adapted to the problem of PAD.

Biometric Person Recognition
Hochschule Darmstadt, Idiap Research Institute
Intelligence Advanced Research Project Activity
Mar 01, 2017
Feb 28, 2021