In this paper, we address the problem of assessing the overall quality of forgery detection approaches for artificial sweat printed latent fingerprints placed at crime scenes. It is very important to have reliable detection mechanisms tested on manifold characteristics caused for example by different surfaces, printers and during acquisition, avoiding misleading crime scene investigations. Today only a limited number of detection methods exist in the literature and test sets are still limited in size and quality covering all different conditions (influence factors). Based on the recently introduced publicly available StirTrace tool, we enhance the functionality to simulate complex and realistic test sets and discuss how detection approaches can be tuned by further preprocessing and feature selection.
Our contributions here are twofold. First, we suggest a benchmarking design in 16-bit domain working in full bit-depth of today’s nanometer sensory and propose enhancements for further simulations of sensor and substrate characteristics as well as single and combined scan artifacts (simulated, novel experimental data set of in sum 1.254.000 samples). Second, we benchmark exemplarily two known feature sets on nonsimulated and simulated data and compare findings with additional preprocessing and feature selection. Finally, we summarize lessons learned how good today’s detection works and which challenges exist for achieving a high reliability. For the community we provide a tool, which can be used as fundamental basis to simulate influence factors allowing a systematic comparison and benchmarking of results. We also want to motivate further research in the design and tuning of forgery detection approaches.