A First Public Research Collection of HighResolution Latent Fingerprint Time Series for Short- and Long-Term Print Age Estimation

Abstract– The creation of publicly available image databases for the signal processing community is a very time-consuming, yet immensely valuable task, enabling scientific progress by providing the opportunity of an objective comparison and reproduction of results. This paper presents for the first time a public research
collection of high-resolution latent fingerprint time series for age estimation, captured from a pool of 116 different test subjects. It comprises ten different sets with a total of 2,618 time series (117,384 scans), varying between capturing devices (CWL and CLSM), data types (intensity versus topography), aging periods (short-term aging: 24 h, long-term aging: 0.5 – 3 years) and resolutions (1,270 – 180,142 ppi). Most series are annotated with donor information (age and gender) and capturing conditions (scan parameters, ambient temperature, and humidity). The data are anonymized (using partial prints only) and an organizational revocation mechanism is included to assure non-identifiability of donors in the future. Baseline results for age estimation on all ten sets are provided in the
form of correlation coefficients and machine-learning based age estimation (kappa), using 19 features from prior feature spaces as well as new ones (Tamura contrast, Benford’s law, and improved dust feature). Classification results exhibit kappa values between 0.51 and 0.85, highlighting the progress made in this very challenging area in recent years and also emphasizing the need of future studies on the issue.