Separation of contactless captured high-resolution overlapped latent fingerprints: Parameter optimisation and evaluation

Although overlapped latent fingerprints often occur in crime scenes, they usually can not directly be used as forensic evidence as it is difficult for the state-of-the-art fingerprint matchers to separate and match them. There exist separation techniques based on chemical imaging, but they require invasive fingerprint acquisition (e.g. applying chemicals) methods which not only compromise the integrity of potential evidence but also eliminate the possibility of analysing the fingerprints using other means (e.g. spectroscopy). Addressing this issue, in this work we use scanned samples of two overlapped latent fingerprints acquired noninvasively by a Chromatic White Light (CWL) sensor. To properly apply our previous separation approach adapted from Chen et al. on those samples, in this paper we conduct a thorough parameter optimisation. A test set of 20 authentic overlapped fingerprints is used. With a cross examination with 4 templates of 2 different kinds using NIST Biometric Image Software (NBIS), we obtain 160 samples from the matching test, which shows a clear increasing number of positive matching after the separation of the fingerprints and leads to an overall Equal Error Rate (EER) of 8.3%, which is lower than the results from Chen et al. where simulated samples are used.