Article Preview
Top1. Introduction
The robustness of a fingerprint verification system under fraudulent attempts through fake fingers is an important issue. In fact, Matsumoto et al. (2002) and Sandstrom (2005) showed that many fingerprint sensors can be deceived by submitting a “gummy” finger. Acquired image is processed as well as a “genuine” image.
In order to prevent these fraudulent attempts, some solutions have been proposed. The majority of them is based on the use of hardware, embedded in the sensor, which can detect the vitality, or liveness, of the finger, e.g., through a blood pressure measurement.
Another class of approaches is based on the extraction of features which can discriminate between live and fake fingerprints by processing images acquired by the sensor. These software- based solutions are obviously less intrusive and cheaper than the hardware-based ones. Recently, two integrated systems (hardware and software methods) have been proposed (Russo, 2007; Diaz-Santana & Parziale, 2006).
According to Coli et al. (2007a), the majority of approaches are aimed to characterize physiological features of the skin by image processing algorithms, e.g., physiological features such as the perspiration (Derakshani et al., 2003; Parthasaradhi et al., 2005; Marcialis et al., 2010), and the elastic deformation of fingerprints (Antonelli et al., 2008; Chen & Jain, 2005). Morphological algorithms and wavelet transformations are employed to this aim (Coli et al., 2008, Moon et al., 2005; Tan & Schuckers, 2006, 2008; Abhyankar & Schuckers, 2006; Drahansky & Lodrova, 2008; Yau et al., 2007; Choi et al., 2007). These features are related to characteristics of live fingers which can be detected by image processing. In the following, we will refer to them as vitality-based features (or live-based features).
It is worth noting that no work so far paid attention to features based on fake finger characteristics (in the following, fake-based features). However, a visual analysis of two example images (Figure 1) may help in detecting some differences among images obtained from live and fake fingers. In particular, Figure 1 shows in the left side a live fingerprint image, whilst the correspondent fake fingerprint image is shown in the right side. At the center, we zoomed on the rectangle pointed out in both images. It is possible to observe three kind of peculiarities: absence or reduction of pores (recently exploited in Marcialis et al., 2010), alteration of pores width, general smoothing of several details, presence of artifacts (the scratch at the center is present only in the fake fingerprint). Such differences are mainly due to the stamp fabrication process of fake fingers which causes an alteration of the frequency details between ridges and valleys. But state-of-the-art vitality-based features have not been conceived to detect and exploit such differences for liveness detection purposes.
Figure 1. An example of live and fake fingerprint images (left and right side, respectively) corresponding to the same subject. At the center, we zoomed on the rectangle pointed out in both images, in order to point out differences mainly due to the fabrication process of the fake fingerprint.