Latent fingerprint persistence: A new temporal feature space for forensic trace evidence analysis

In forensic applications, traces are often hard to detect and segment from challenging substrates at crime scenes. In this paper, we propose to use the temporal domain of forensic signals as a novel feature space to provide additional information about a trace. In particular we introduce a degree of persistence measure and a protocol for its computation, allowing for a flexible extraction of time domain information based on different features and approximation techniques. At the example of latent fingerprints on semi-/porous surfaces and a CWL sensor, we show the potential of such approach to achieve an increased performance for the challenge of separating prints from background. Based on 36 earlier introduced spectral texture features, we achieve an increased separation performance (0.01 ≤ Δκ ≤ 0.13, respective 0.6% to 6.7%) when using the time domain signal instead of spatial segmentation. The test set consists of 60 different prints on photographic-, catalogue- and copy paper, acquired in a sequence of ten times. We observe a dependency on the used surface as well as the number of consecutive images and identify the accuracy and reproducibility of the capturing device as the main limitation, proposing additional steps for even higher performances in future work.