Empirical tests on enhancement techniques for a hybrid fingerprint matcher based on minutiae and texture

[abstract] This paper focuses on the use of enhancement techniques for improving the performance of a hybrid fingerprint matcher based on the fusion of image-based fingerprint matchers and a minutiae-based matcher. A review of the existing literature is provided, and several methods are compared on all four FVC2004 databases. Through extensive testing, we find that the best performing system is obtained by an ensemble of image-based matchers with features extracted by local phase quantization and Tico's minutiae matchers. Performance of our fusion approaches are compared to reported results of competing commercial methods using FVC2004.

Contributions of this study include a fair comparison of different preprocessing techniques using the different matchers. We also study their fusion for improving the performance of stand-alone preprocessing techniques. In this way, we demonstrate that different enhancement methods can be used for building a multi-matcher method. We also propose a genetic optimization approach to improve the enhancement step using different optimization functions. Finally, we contribute all the source code used in our experiments (available at bias.csr.unibo.it/nanni/finger.rar). Providing a free Matlab toolbox that contains functions for minutiae extraction, enhancement, image-based matching, and minutiae-based matching (a system we show compares relatively well to commercial matchers) could form the basis for future work by other researchers in this and similar areas.

Keywords: fingerprint identification; texture descriptors; minutiae; local phase quantization; multi-matcher ensemble.