Facial Image Comparison: Morphological and Holistic Methods

Forensic facial comparison reached UK courts in the 1980s and 1990s, largely through the work of practitioners who had backgrounds in medical illustration and physical anthropology. The term Facial Mapping entered legal vocabulary to describe the process, and early cases relied on a relatively informal methodology that varied between practitioners. The reliability of the field was contested, and courts struggled with how to weigh evidence that lacked the population-frequency databases underpinning fingerprint and DNA work.

In Australia, similar debates played out, with the Bluey Clark case in the 1990s among the early examples of facial image evidence being tested in court. In the United States, law-enforcement interest was driven partly by the expansion of photographic surveillance and partly by post-9/11 investment in biometric identification. The FISWG was established in 2009, bringing together facial comparison practitioners, cognitive psychologists, and computer scientists. Its published guidelines in 2012 and subsequent updates established a methodology built around documentation, sequential unmasking, and a structured conclusion vocabulary.

Morphological facial comparison is systematic: the examiner records the characteristics of a defined set of anatomical regions and assesses whether the pattern of correspondences and differences between two images is consistent with a single identity or two distinct individuals. The regions examined follow anatomical logic.

• Orbital region: eye shape, palpebral fissure inclination, medial and lateral canthal angles, epicanthal fold presence, lid crease morphology, and eyebrow form and position.
• Nasal region: dorsal profile (straight, convex, concave), nasal tip shape, tip projection, nostril shape and orientation, columella visibility, and alar base width relative to facial width.
• Auricular region: helix shape and attachment, antihelix bifurcation, lobule form (attached, free, intermediate), tragus size and shape, concha depth, and overall ear size and projection.
• Mandibular and chin region: chin form (pointed, rounded, squared), chin projection, jaw angle definition, gonial angle, and the overall lower-face shape in frontal view.
• Soft-tissue mouth region: lip vermilion border shape, philtrum column definition, philtrum width, commissure position, and the relative proportions of upper and lower lips.

The examiner records whether features show correspondence, difference, or are indeterminate (due to pose, resolution, lighting, or age difference between images). A difference in a single feature does not automatically mean exclusion, because the same person photographed years apart, at different angles, or under different lighting conditions will show apparent variation. The comparison matrix as a whole drives the conclusion, not any single outlier.

Face recognition is one of the most heavily studied areas of human perception. A substantial body of research from cognitive psychology shows that people, including trained examiners, process faces holistically: the brain encodes a face as a unified pattern, not as a list of features assembled at runtime. This is why inverting a face dramatically impairs recognition, and why a feature that looks right in isolation can look wrong once the rest of the face changes around it.

In forensic facial comparison, holistic assessment serves as a check on the morphological feature matrix. After working through the anatomy, an examiner also forms an overall impression: does the general facial gestalt fit with the two images being the same person? This step can catch things the feature-by-feature pass misses, particularly when subtle proportional relationships across the face carry identity signal that no single feature captures alone. But it is also where examiner subjectivity is highest, which is why FISWG guidelines require that the holistic assessment supplements rather than replaces the documented feature analysis.

Research by David White, Richard Kemp, and colleagues at UNSW and other institutions has repeatedly shown that even trained facial comparison examiners make errors when working with poor-quality images. Accuracy drops significantly when the two images to be compared differ in pose, lighting, or age. This research has been influential in pushing the field toward formalised methodology and quality controls, and in reinforcing that facial comparison evidence should be presented with explicit statements of its limitations.

The external ear, or pinna, has a complex three-dimensional structure made up of cartilaginous ridges and hollows that are largely set by genetics and remain relatively stable through adult life. The helix, antihelix, tragus, antitragus, concha, and lobule all show variation across individuals, and the overall shape and size of the ear can be assessed from a two-dimensional image if the angle of view is favourable.

Ear comparison entered forensic literature in the late 1990s, and Dutch police officer and forensic instructor Cor van der Lugt was among the early systematic researchers. In the UK, earprint evidence was central to the Dallagher case: Mark Dallagher was convicted of murder in 1998 partly on earprint evidence, but his conviction was quashed by the Court of Appeal in 2002 and he was acquitted at retrial in 2004 after DNA analysis failed to link the earprint to him. The evidentiary value of ear comparison on its own remains debated, and the field acknowledges that the population-frequency data for ear morphology are thin compared with fingerprint or DNA databases. In practice, ear comparison is most useful when the face is fully obscured or when the ear happens to be captured clearly on surveillance footage that shows the face at an unhelpful angle.

Photo-anthropometric overlay is a technique that places two images in alignment and measures the proportional relationships between anatomical landmarks: the intercanthal distance, the alar width, the mouth width, the distance between pupil centres, and several other fixed-point pairs. The examiner converts these measurements to ratios that are independent of absolute image scale, then compares the ratio profiles between the reference and questioned images.

The approach derives from craniofacial anthropometry. In forensic contexts the numerical output carries apparent objectivity, but that appearance requires careful qualification. But the technique carries important caveats. Accurate landmark localisation requires adequate image resolution, and even small errors in placing a landmark can shift ratio values substantially. More critically, all metric comparisons are sensitive to camera geometry: a face photographed with a telephoto lens at a distance will have systematically different apparent proportions than the same face photographed with a wide-angle lens at close range. Correcting for this requires knowing or estimating the camera and its focal length, which is rarely possible from crime-scene footage.

When the geometry can be controlled, for instance when measurements are taken from calibrated photography of a suspect alongside the crime-scene image, anthropometric overlay adds a useful quantitative dimension to the comparison. When the geometry is uncertain, the numerical output can create a false impression of precision, and examiners are expected to state clearly which conditions applied.

One of FISWG's most practically important contributions was a standardised conclusion vocabulary. Before this standardisation, different examiners used phrases such as consistent with, cannot be excluded, and highly probable in ways that courts interpreted inconsistently. FISWG proposed a scale anchored at both ends by strong conclusions and with graduated intermediate levels, structured to communicate the weight of the evidence in a way that does not overstate certainty.

1. Exclusion
   The examiner concludes the two images depict different individuals. Reserved for cases where there are clear, unambiguous morphological differences that cannot be explained by image variation, pose, or age change.
2. Supports exclusion
   The evidence is more consistent with the images depicting different people than the same person, but the examiner cannot exclude the possibility of one identity.
3. Inconclusive
   The evidence is insufficient to support either conclusion. This can reflect poor image quality, unfavourable pose, extensive occlusion, or a genuine balance of corresponding and differing features.
4. Supports identification
   The evidence is more consistent with the images depicting the same person than different people, but the examiner stops short of positive identification.
5. Identification
   The examiner concludes the two images depict the same individual. This is the strongest positive conclusion and requires a high level of correspondence across multiple independent features with no unaccounted differences.

The scale is not a probability scale in the statistical sense: there are no likelihood ratios attached to each level, and the verbal labels are not mapped to fixed numerical ranges. What the scale does do is force an examiner to commit to a position rather than hiding behind deliberately vague language, and it gives courts a framework to understand the relative weight of positive and negative conclusions. FISWG guidelines also require that any conclusion be accompanied by an explanation of what features drove it and what limitations affected the analysis.