Application of stirtrace benchmarking for the evaluation of latent fingerprint age estimation robustness

Abstract

The long-researched challenge of latent print age estimation has recently been re-visited using non-invasive capturing devices and time series. While this approach was able to provide an objective performance measure for the first time, the robustness of the scheme towards different contextual conditions is barely known. Also, StirTrace has recently been adapted from the well-known StirMark framework to the benchmarking of latent fingerprints. In this paper, the image manipulation techniques of StirTrace are transferred to the age estimation challenge and estimation performance is benchmarked for 500 short-term aging time series under varying contextual conditions. Results show that time series preprocessing is sensitive to pixel-depth reduction as well as different types of noise. Furthermore, a decreased image size has a stronger impact on the age estimation performance than a decreased resolution. The addition of texture for substrate simulation and median cut filtering for smudged print simulation are found to not be optimal in the aging context for representing this kind of distortions.