forensic face recognition from characteristic descriptors to strength of evidence
In this dissertation, FISWG characteristic descriptors are the main object of study. FISWG is an organisation in which several forensic institutes participate, notably the FBI and NFI. Characteristic descriptors are facial features that can be used by a forensic face examiner during forensic evidence evaluation of trace and reference images. The trace image often captures a crime scene and is most of the time taken under uncontrolled conditions. The reference image is a photograph of a suspect and is taken under controlled conditions. During this forensic evidence evaluation, the forensic face examiner pays attention to these characteristic descriptors, mostly shape like and potentially highly discriminating features, and computes the strength of evidence that can be used in a court of law.
The mere fact that the characteristic descriptors are documented does not automatically imply their suitability, in particular for their intended use under forensically relevant conditions. Actually, little research is done on this topic. Also, in light of in the recent adoption of the Daubert rule (“a trial judge must ensure that any and all scientific testimony or evidence admitted is not only relevant, but reliable”), there should be more insight into this matter.
This dissertation addresses two major research questions.
The first research question deals with the suitability of FISWG characteristic descriptors as a means to discriminate, taking (a) human, (b) classifier, (c) feature, and (d) forensic aspects into account. A classifier is a computational structure that uses characteristic descriptors, or more generally features, to compute a value that can be either interpreted as or converted to strength of evidence.
There does not exist a single simple answer to this research question. On one hand, one can argue that from a biometric point of view that is primarily interested in general performance, the results of (b) are both somewhat positive, but mostly negative when we consider the forensic circumstances. The former result is obtained in a limited and less representative setting, whereas the latter result is designed to be representative of various forensic use cases (d) and considers a large subset of FISWG characteristic descriptors. Also, the results of (c) show that the characteristic descriptors are difficult to measure in representative forensic use cases (d), although almost any feature in general would be useless under those circumstances. Results of (a) and (b) seem to indicate that there is little added value in using FISWG characteristic descriptors.
The second research question deals with the suitability of a subject based approach in forensic evidence evaluation, taking (e) empirical results from specific applications, (f) theoretical results, and (g) a framework approach into account.
The framework (g) itself integrates the design of the feature, the biometric score and its forensic use, and the evaluation of performance characteristics, with a specific emphasis on a subject based approach. A theoretical construction (f), the results of the facial marks study (e) and the two applications (e) contained in (g) clearly show that large differences between subject based and general performance exists. Also, the effect of using subject data in classifiers (e) is clearly seen. These results confirm the suitability of a subject based approach in forensic evidence evaluation; the presented framework can be used as a tool for such a subject based approach.
Overall, we conclude that from a general biometric perspective, FISWG characteristic descriptors are not suitable as a means to discriminate. However, if we also consider them from a biometric perspective that includes the use of subject based data and subject based performance, then in limited cases a FISWG characteristic descriptor can be used as a biometric feature to discriminate a particular subject from a group of subjects. More generally, subject based performance evaluation provides insight into the contribution and limits of FISWG characteristic descriptors.
Recommendations of this dissertation include the conduction of more studies with respect to the use of FISWG characteristic descriptors by forensic face examiners, the collection or creation of more forensically relevant data and information, and the incorporation of a subject based evaluation in method validation.
As a closing remark, this dissertation has systematically considered FISWG characteristic descriptors, both directly and indirectly, starting from a human approach and zooming out to a general framework. It is a contribution that serves the scientific approach as meant by the Daubert rule.
