Classification of Fingerprints Captured Using Optical Coherence Tomography

Abstract

We propose a technique for analysis of fingerprints scanned free-air (not pressed against a glass) with Optical Coherence Tomography (OCT). Fingerprints from the surface and subdermal parts of the finger are extracted from a 2GB volumetric scan in cca. 2 s using our specialized technique and GPU acceleration on GeForce GTX 980. The technique provides fingerprints that perform with promising error rates that demonstrate the potential of the OCT for improved fingerprint identification, as well as its potential for prevention of biometric spoofing (PAD).

Robust Identification and Segmentation of the Outer Skin Layers in Volumetric Fingerprint Data

“…This problem was later addressed at NTNU, where Sousedik et al. developed a novel edge detection algorithm that can be very efficiently executed on GPUs [ 32 ]. We therefore considered the BSI and NBL research on OCT fingerprinting as the starting point for our own work.…”

“…Inspired by the work of Sousedik et al [ 32 ], we tried to identify the fingerprints at the boundaries between the different layers in the scan observed along the scan lines. For this, we applied a custom 1D edge detection filter with additional low-pass characteristics for noise suppression individually to each scan line.…”

“…To address this problem, we propose the use of a multi-resolution volume pyramid for noise reduction, created by downsampling the scan volume multiple times in all three dimensions. Sousedik et al used a similar approach, but performed the downsampling only in the horizontal plane and left the depth unchanged [ 32 ]. After implementing and comparing both approaches, we concluded that downsampling in all dimensions as the better alternative because it does not deform the finger by changing the aspect ratio.…”