The ability to identify individuals fast and accurately is a cornerstone of many security operations. High performance, universal applicability, and unique matching make fingerprint recognition the most commonly used and widely accepted biometric technique. However, artificial finger dummies with embedded fingerprints can be made using just $10 worth of household products and can easily spoof common identification systems.¹ If these systems are to be foolproof, they must be improved.

Significant advances, reported in recent years by several scientific groups, include more robust fingerprint readers based on surface topology recognition. Suzuki and colleagues have described a smart card holder authentication system, for instance, that links fingerprint verification to a personal identification number (PIN) by applying a double random phase-encoding scheme.² However, such improvements concentrate on decreasing the FAR (false accept rate) and FRR (false reject rate) or shortening scanning time. They do not prevent the use of artificial fingerprints.

Recently, we demonstrated that optical coherence tomography (OCT), a high-resolution tissue-imaging technique,³ can successfully identify artificial materials commonly used to spoof optical fingerprint scanning systems (see Figure 1).⁴ We fabricated dummies with readily available products, including a plasticene (Dixon Ticonderoga), household cement (ITW Devcon), and liquid silicon rubber (GE Silicones).

Figure 1(a) shows a typical OCT image of finger skin beneath a dummy layer with artificial fingerprints. Three layers of human skin (the stratum corneum or pigment layer, epidermis, and dermis) are clearly visible. The corresponding 1D OCT signal is shown in Figure 1(b). The dummy layer is a homogeneous media as illustrated by the OCT signal curve, and it has a significantly lower scattering profile than that of the skin.

We tested our fingerprint dummies with the Microsoft Fingerprint Reader (model 1033) and an OCT system. When the artificial dummies were applied to the reader, FARs ranged from 80 to 100%. Each dummy was tested at least 10 times.

By contrast, the OCT system invariably detected the dummies, both visually in 2D images and corresponding OCT signal curves, and also after processing with autocorrelation analysis (see Figure 2). Autocorrelation is commonly used in signal processing to analyze functions and series. We applied it here to test whether OCT images recorded from artificial materials and real skin can be distinguished, depending on their speckle-noise pattern.

We demonstrated that high-resolution OCT could be successfully applied to identify artificial materials commonly used to make fake fingerprints. We also showed that autocorrelation analysis has potential for use with automatic fingerprint recognition systems.