Forensic Science International: Reports (Nov 2021)
Scientific estimation of the subject-to-camera distance from facial photographs for craniofacial superimposition
Abstract
Subject-to-camera distance (SCD) is widely acknowledged to be a critical variable in craniofacial superimposition, but in practice it has been notoriously difficult to estimate from facial photographs. The recently proposed PerspectiveX algorithm offers the first potential capability to objectively undertake this estimation using quantified data. In this study, we test PerspectiveX’s legitimacy first by determining if the palpebral fissure length that it uses, falls in the relative-parallel-zone of the face (RPZ) where perspective distortion effects are minimal, thereby justifying its use for craniofacial superimposition (Study 1). Secondly, we validation test the PerspectiveX algorithm by using facial photographs acquired at known, whole meter ground truth distances between 1 and 10 m (Study 2). For study 1, we used 3D face scans to quantitatively map the relative parallel zone (RPZ) in 30 upright living subjects. For study 2, we used 2D facial photographs of up to 50 subjects at known 1 m interval distances. Results show the RPZ to cover a very thin vertical area of the face, but one that includes the palpebral fissure and the eyelids. This justifies the use of the palpebral fissure length in PerspectiveX. When tested using photographs taken at known distances, PerspectiveX performed with a signed mean and mean absolute error of 3.8% and 6.8% respectively for two prime lenses (85 and 100 mm). For a zoom lens operated at its extremes of 24 and 105 mm focal-lengths, PerspectiveX performed with a signed mean and mean absolute error of 4.4% and 11.0%, respectively. These errors are small in the superimposition context, translating into less than 1% difference in vertical face height. This supports the use of PerspectiveX in superimposition to objectively set the skull-to-camera distance, such that it replicates the face-to-camera distance to provide optically valid 1:1 comparisons of skulls and faces.
