Biometric Recognition of the Hand in Uncontrolled Images

Publications

Overview

Incidents of abuse material found online have massively increased in the last decade, presenting an epidemic that law enforcement agencies are struggling to keep up with. In image or video documented crime perpetrators often take steps to maintain anonymity, including hiding their faces; biometric analysis of this type of content is one way to identify the people involved. Hands and forearms are more often visible and contain many unique features such as hand geometry, palm and knuckle prints, under-skin vein patterns, androgenic hair patterns, and marks such as scars, freckles, and tattoos. The key aim of this project is to investigate the usability of these features for offender identification, and develop effective methods of extracting them.

There are research gaps in the application of hand biometric feature extraction in uncontrolled images, where there is a solid basis of research in controlled images. A method designed for a forensic investigation context has several extra requirements: detection of features before extraction, invariance to different image conditions, a matching process that evaluates within-source and between-source variance, and being explainable to a layperson – a requirement of providing the algorithm’s output as evidence. Uncontrolled hand image data is scarce so creation of further datasets will be necessary to develop these methods. Requirements for this data will be determined by an initial analysis of pre-existing data and the available features.

The project will consider the history of more established biometrics used in forensic investigation (fingerprints, DNA, facial recognition) with regards to data privacy, use and mis-use by law enforcement, and admissibility of evidence. This will be done with the aim of better designing methods from the start, avoiding the same pitfalls encountered with biometric evidence in the past.

Research Questions

What methods can be used to extract hand biometric information consistently and reliably, from the uncontrolled conditions often found in criminal material?
How accurate will recognition be when performed on these extracted features?
Which of these features, or combinations of these feature, are best for automatic recognition?

References

1. N. A. Spaun and R. W. V. Bruegge, 2008, "Forensic Identification of People from Images and Video," in 2008 IEEE Second International Conference on Biometrics: Theory, Applications and Systems, pp. 1-4, doi: 10.1109/BTAS.2008.4699363.

2. A. Slot, Z. J. Geradts, 2014, “The possibilities and limitations of forensic hand comparison”, in Journal of Forensic Sciences, 59(6), pp. 1559-1567.

3. D. Meuwly, 2006, “Forensic individualisation from biometric data”, in Science & Justice, 46(4), pp.205-213