Experiments show promise for real-time fingerprint liveness detection

Researchers say they have devised lightweight software that can perform real-time fingerprint biometric liveness detection.

A team of scientists working in China, Portugal and Canada say their method requires shorter training and fewer parameters than previous approaches.

The researchers reportedly apply a broad learning system (BLS) to fingerprint liveness detection, something they say has not been done before, and something that does not required GPU training.

A BLS boosts the performance and applicability of detection algorithms on mobile devices, they claim. Authentication can follow a positive liveness detection.

A paper about their work describes a three-step process.

In the first step, regions of interest are extracted from a print and noise is stripped. Then, distinguishable texture features are built using uniform local binary pattern, or ULBP, descriptors as BLS input.

The descriptors minimize the variety of binary patterns in the submitted fingerprint biometrics features while preserving information critical to liveness detection.

Last, extracted features go to the BLS for training. The BLS is a flat network that places the mapped-feature original input in feature notes. That generalizes the structure in augmentation nodes, according to the paper.

Competing deep-learning methods, say the researchers, have shortcomings that are addressed in their approach.

Mobile devices cannot operate the complex neural network algorithms needed due to storage and power constraints. And Convolutional neural network models result in too many parameters for efficient training.

Article Topics

biometric liveness detection  |  biometrics  |  biometrics research  |  fingerprint biometrics  |  real-time biometrics