Feature space fusion and feature selection for an enhanced robustness of the fingerprint forgery detection for printed artificial sweat

The recognition of forged fingerprints at crime scenes is a very old challenge. Various forgery techniques can be applied to produce such traces. However, the detection of such forgeries usually involves a thorough manual inspection. For the example of fingerprints printed using ink-jet printers and artificial sweat, first pattern recognition approaches are proposed in prior work employing two different feature spaces. In this paper we investigate the potential for a fusion of the two prior feature spaces towards the robustness of the classifier's decision. Furthermore, we perform a feature selection to reduce the dimensionality of the feature space. We use independent test and training sets, each consisting of 3000 samples captured by a Confocal Laser Scanning Microscope. The results of the StirTrace-based benchmarking of the recognition of printed fingerprints indicate improved detection accuracies for several simulated influences.