Forensic sensor quality evaluation: Towards objective visual quality indices for contact-less fingerprint and fiber traces

Testability and error rates are fundamental requirements for forensic investigations methods. For digitized forensics (e.g. fingerprint or fiber traces) signal processing and pattern recognition concepts are used. The quality of the acquisition process is sparsely researched. In our article we outline a first approach to address this and to open the discussion about the general challenge. We propose using Visual Quality Indices (VQI) as objective measure to compare scanned (digitized) traces for 3 different 3D contactless sensors (in-tensity + topography data) and 1 sensor with spectrometer-based wavelength output. We propose a fitness matrix containing VQI results for trace data acquired for each sensor and suggest using the number of outliers (non-occupied elements of the main diagonal) as fitness measure. We compare 182 data sets using 4 contactless sensors for 2 consecutive scans of 13 traces (3 fingerprints + 10 fibers, with 78 intensity and 78 topography data sets for the three 3D sensors and 26 for the spectography-based wavelength output). We compare the 2 consecutive scans vice versa by computing VQIs for all 7 outputs (intensity, topography and wavelength output) per trace, yielding 196 (14×14) single comparisons per trace and VQI (in all 356720 comparisons). We research the best performing VQIs to select intra-sensor, intra-trace data from intra- and inter-sensor data, giving a starting point for further inter-sensor investigations. Our research shows the Luminance Similarity Score (LSS) performing best in intra-sensor, intra-trace scenarios for fingerprint traces and Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR) and PSNR in the YUV Color Space (PSNRY) equally performing best for synthetic fiber traces as a first result for further discussion.