Synthetic Fingerprint Generation

Significant efforts are continuously being made in designing new fingerprint recognition algorithms both in academic and industrial institutions. However, the accuracy of each algorithm is usually evaluated on relatively small databases. An evaluation on small databases makes the accuracy estimates highly data dependent; as a result, they do not generalize well on fingerprint images captured in different applications and different environments. Furthermore, when the databases are proprietary, the accuracy of various matching algorithms cannot be compared directly. A sharable large database of fingerprints (thousands or tens of thousands of images) is required to evaluate and compare various fingerprint recognition algorithms due to the very small error rates that have to be estimated. Unfortunately, collecting large databases of fingerprint images is: (i) expensive both in terms of money and time; and (ii) tedious for both the data collection technicians and for the subjects providing the data. Even if one is able to collect such a large fingerprint database, it is difficult to share it with others due to privacy legislations that often protect such personal data. Finally, publicly available databases of real fingerprints, such as those used in FVC technology evaluations (Maio et al., 2002a, b, 2004), Cappelli et al. (2006), do not constitute lasting solutions for evaluating and comparing different algorithms because they expire once “used,” and new databases have to be collected for future evaluations. In other words, once an evaluation database is released, algorithm developers can “train” their algorithm to perform well on this specific database.

A potential alternative to collecting large fingerprint databases is fingerprint sample synthesis, i.e., generating images similar to human fingerprints, through parametric models that encode the salient characteristics of such images and their modes of variation. This chapter describes SFinGe, a synthetic fingerprint generation approach developed by Cappelli, Maio, and Maltoni (2000a, b, 2004); see also Maltoni (2004). SFinGe can be used to automatically create large databases of fingerprints, thus allowing fingerprint recognition algorithms to be effectively trained, tested, optimized, and compared. The artificial fingerprints emulate images acquired with electronic fingerprint scanners, because most commercial applications require on-line acquisition. It is also possible to generate impressions similar to those acquired by the traditional “ink-technique” with relatively minor changes in the algorithm.