StirTraceV3.0 and printed fingerprint detection: Simulation of acquisition condition tilting and its impact to latent fingerprint detection feature spaces for crime scene forgeries

Based on the existing StirTraceV2.0 framework including 13 single artifact simulations for benchmarking artificial sweat printed fingerprint detection to identify crime scene forgeries, we propose and investigate the tilting of the sample as a further acquisition condition for Confocal Laser Scanning Microscopes (CLSM). We study Benford's law, edge- and circle-based feature detection spaces on intensity (int) and on topography (topo) image data separately. Tilting artifact reduction pre-processing is proposed as Best Fit Plane Subtraction (subp, using the known least squares method) to improve detection results. An evaluation with seven different tilting parameters with and without the proposed Best Fit Plane Subtraction is performed and discussed. To support benchmarking, StirTrace is enhanced with so-called StirTrace Evaluation Modes to perform different benchmarking tasks, such as the "printedFP" mode offering 10 edge-based features and 67 circle-based as well 9 Benford's law based detection features. The experimental data consists of 3000 printed and 3000 real fingerprint samples acquired by a CLSM. Based on different tilting parameters 21000 samples are created using StirTrace. We observe that tilting has a higher impact on the detection of forgeries using intensity data and that the proposed corrections with the Best Fit Plane Subtraction can be recommended to stabilize the detection performance. Furthermore, we analyze the impact of this pre-processing on the distribution of the most significant digits within noise data relevant for Benford's law based detection feature space.