Statistical latent fingerprint residue recognition in contact-less scans to support fingerprint segmentation

The search, acquisition and analysis of latent fingerprints are performed in forensics for over a century. The acquisition methods have evolved during the decades but they are still primarily contact-based and alter the trace. Contactless acquisition systems are subject to ongoing research, allowing for a non-destructive acquisition of latent fingerprints. Those techniques pose opportunities and challenges for forensic investigations. In particular the visibility enhancement of the fingerprint needs to be performed digitally in contrast to the contact-based methods. In this paper we propose and evaluate a pattern recognition based fingerprint residue detection using 16 statistical features, whereas 10 are motivated in [1], [2] and 6 newly introduced, as well as 9 features based on Benford's law [3]. Our goal is to recognize the fingerprint residue as digital visibility enhancement as a foundation for fingerprint segmentation and subsequent biometric comparison of trace evidence from crime scenes. Our evaluation is performed for three different surfaces (white furniture surface, veneered plywood, brushed stainless steel) that are usually found at crime scenes. The test set contains 30 untreated contact-less captured latent fingerprints with additional labeling information as ground truth gathered from differential scans. Our evaluation is split into two parts: in the first evaluation a fingerprint residue recognition accuracy of up to 92.7% is achieved on a cooperative surface. In the second evaluation, based on biometric matching after our residue recognition, we can outperform the matching performance of the ground truth from the differential scans using J48 decision tree classifier on the cooperative white furniture surface, achieving 6 instead of 5 successful matches with an exemplar fingerprint.