Fingerprint Presentation Attack Detection: Multispectral imaging with a narrow-band camera using Bag of Features

Biometric recognition is ever-increasingly utilized in our daily lives due to its advantages in convenience, speed and security. Nevertheless, biometric systems can be vulnerable to attacks at many points, presentation attacks being a common issue. Thus, accurate Presentation Attack Detection methods need to be studied and evaluated to overcome these vulnerabilities. In this paper, a narrow-band camera with 10nm increments in wavelength will be used to observe real fingerprints and artefacts (Play-Doh, latex and transparent nail polish) and study the different classification accuracies depending on the wavelength. A total of 9,646 images were captured in 36 wavelengths, and an APCER of 1.98% at 50% training, 50% testing was obtained. All results are calculated in accordance to the requirements of the standard ISO/IEC 30107-3.