Burton, I (2023) Post-mortem and ante-mortem 2D and 3D facial comparison for forensic identification. Doctoral thesis, Liverpool John Moores University.

	Text 2023BurtonPhDpublic.pdf - Published Version Restricted to Repository staff only until 1 July 2024. Available under License Creative Commons Attribution Non-commercial. Download (5MB)
	Text 2023BurtonPhDinternal.pdf - Submitted Version Restricted to Repository staff only Download (40MB)

Abstract

This research explores the application of manual facial comparison methods and offline (semi-) automated face recognition algorithms to human identification of recently deceased individuals, using 3D (N=3) and 2D (N=6) data. Current methods are generally designed for and applied to the living, are not commonly used in combination or hierarchical order, and are mostly tested under controlled, lab-based conditions. The aforementioned makes their applicability to real-life settings questionable, and research on their application to the deceased is scarce, although desperately needed. Further issues arise from the lack of uniformity and standardisation in methodological approaches, as well as feature descriptions and terminology.
This study investigates the applicability of combined, manual face comparison methods in hierarchical order (preliminary feature-based analysis, facial superimposition, detailed morphological comparison) and two (semi-) automated face recognition algorithms (MATLAB® and Picasa) to the recently deceased. Pilot and ancillary studies explore pretend-dead faces as a data source and evaluate geometry vs. texture in 3D face models. 3D data was obtained using the Artec handheld laser scanner.
Key findings indicate that human face matching ability is superior and more resilient to PM facial changes, non-standardised AM data, and limited data availability, compared to the automated methods tested. Results further suggest that a hierarchical approach to manual comparison is highly beneficial. MATLAB®’s algorithm is unreliable even as a filtering tool and Picasa struggled to detect PM faces in images; an issue not encountered with pretend-dead data. Both automated approaches utilised here are not suitable for application on similar datasets or in casework settings, unlike the manual approach – the latter requiring further validation. Limitations arise primarily from the small sample size and non-quantifiable approaches to manual facial superimposition.
In forensic casework and disaster victim identification scenarios, comparative AM data for primary and even secondary methods of identification are often lacking. However, facial photographs are almost always attainable and should be considered an important resource for post-mortem identification. The interdisciplinary nature of this field requires collaborative efforts to address remaining challenges in the future.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Facial Comparison; Identification; Postmortem; Morphological
Subjects:	Q Science > Q Science (General) T Technology > T Technology (General) T Technology > TR Photography
Divisions:	Art & Design
SWORD Depositor:	A Symplectic
Date Deposited:	18 Jul 2023 15:05
Last Modified:	18 Jul 2023 15:05
DOI or ID number:	10.24377/LJMU.t.00020271
Supervisors:	Wilkinson, C, Roberts, J and Sudirman, S
URI:	https://researchonline.ljmu.ac.uk/id/eprint/20271

View Item