Facial anthropometry as an evidential tool in forensic image comparison

Kleinberg, Krista F. (2008) Facial anthropometry as an evidential tool in forensic image comparison. PhD thesis, University of Glasgow.

Full text available as:
PDF (edited version, 3rd party copyright removed)
Download (5MB) | Preview
Printed Thesis Information: https://eleanor.lib.gla.ac.uk/record=b2636999


Anthropometry can be used in certain circumstances to facilitate comparison of a photograph of a suspect with the potential offender portrayed in video surveillance crime footage. Anthropometry does not have the same success rate in identification as DNA or fingerprinting. However, these types of evidence are not always left at crime scenes. Sometimes the only evidence available relating to an offence is from surveillance videos and research was needed to lend credence to anthropometry as a viable method of identification. An alternative method of detecting individuals from surveillance video, morphology, was also investigated to determine its accuracy in confirming the identity of individuals based on facial descriptions and for use as a comparison tool in forensic identification.
Pilot Laboratory Study: A number of different techniques are employed in facial image comparison of living persons. In this study, the effect of rotation on angles and proportions between selected facial landmarks is evaluated as a first step to assess whether facial anthropometry could be usefully applied to facial image comparison. The faces of five volunteers were photographed in the Frankfort plane at different angles of rotation from 0º (frontal) to 90º (side view), rotating every 10º both clockwise and counter-clockwise. Four landmarks were used: right and left ectocanthions, nasion, and stomion. The proportions of the measurements between these landmarks were calculated as well as the angles created by the lines connecting the same landmarks. The results show a consistent and predictable variation between the five subjects. With rotation, the greatest variation is seen where horizontal landmark connecting lines are combined with the ectocanthion/stomion or nasion/stomion lines. There is less variation in the proportions for vertical and diagonally orientated landmark connecting lines. In principle, the data from these empirical measurements could also be used to develop a photogrammetric model of the face which, if calibrated, could be used to correct anthropometric measurements for distortions caused by a camera angle which differs from the one specified in a protocol for facial comparison. The purpose of developing such a model would lie in its use to calculate correction factors to convert observed proportions and angles back to the full-face orientation values, which could then, for example, be used to search a database of the proportions.

Investigation of Uncertainty of Anthropometric Measurements: The objective of this study was to estimate the uncertainty in the measurements of the chosen facial proportions caused by landmark placement and by operators taking photographs, including the uncertainty contributions resulting from different people performing these tasks. The aim of this was to simulate effects found in the real world, as there would be different operators both placing landmarks and taking suspect photographs in various police departments. In addition, this study was completed in order to address variables encountered in the Pilot Laboratory Study that occurred as a result of the experimental set up. The first section of the study reviewed the errors involved in measuring facial proportions as a result of variations in landmark placement. Intra and inter-operator studies in landmark placement were conducted and as expected the average and range of coefficients of variation for the set of proportions were larger in the inter-operator error than that obtained in the intra-operator error. The second section of the study reviewed the errors in measuring facial proportions as a result of the process of taking photographs. The lowest variation in facial measurements was seen in the series of photographs taken of a single subject by a single operator and in general, the lowest variation in facial measurements was seen at 45° and the highest variation at 20°. The contributions of errors from landmark placement and photography were determined to produce an overall estimated uncertainty of 5%. When a comparison of 2D images is conducted in this manner this estimation of uncertainty should be taken into account.

Anthropometry Study: An existing database of video and photograhic images was examined, which had previously been used in a psychological research project with the aim to test the hypothesis: “Using a comparison of anthropometric facial proportions, it is possible to discriminate between individuals of two samples.” Material avaliable consisted of 80 video (Sample 1) and 119 photograhic (Sample 2) images and were of high resolution, though taken with different cameras. A set of 37 anthropometric landmarks were placed measuring 59 proportions to conduct within sample and between sample comparisons using the following calculations; mean absolute value between proportions, Eulcidean distance and Cosine θ distance between proportions. First, the statistics of the two samples were examined to determine which calculation best ascertained if there were any differences between faces which fall under the same conditions. Subsequent to a between sample, the removal of up to 50% of the lowest variant proportions along with the determination of a subsample of faces requiring human verification were tested. Relative frequency distribution histograms were created from the data and the normal histogram curves of true positive and true negative faces were superimposed to determine their separation rate and how likley it may be to mix up the two categories of faces. Presented results showed that the Cosine θ distance equation using Z-normalized values was the preferred equation because it achieved the largest separation between true positive and true negative faces. Results also indicated that there was no benefit to removing up to 50% of the lowest variant proportions in the comparison of Sample 1 against Sample 2. Finally, applying the Cosine θ distance equation allowed a decrease to five database images to be verified by a human in approximately 75% of the cases tested.

Morphology study: A morphological analysis was conducted on high resolution images and although highly relevant to the process of facial identification did not contribute to the continuity of the thesis and thus was included as an Appendix. The morphological analysis was performed on a total of 199 images: 119 photographs and 80 images from video using a checklist of 20 facial characteristics. Each facial characteristic had numerous choices in which it could be described. Once the analysis on all 199 images was carried out, a comparison was conducted between each video (unknown) image and the database of 199 (known) images. In the research conducted, only 2.5% of the comparisons showed a true positive match between video and photograph with zero false positives in the group. Subsequent to analysis it was determined not possible to differentiate between individuals, however, when looking directly at the individuals’ photographs, it is clear that there were differentiating characteristics amongst them.

Conclusions: After embarking upon a series of anthropometrical investigations using high resolution images to compare video images with photographic images, it was concluded that anthropometry, when used as a comparison method of identification, does not generate the results necessary for use as evidence in a court of law. Identifying individuals based on a morphological analysis of a check list of features alone also did not result in clear consistent identifications. If descriptions of facial characteristics are to be fully utilized, a side by side comparison is likely to be less subjective. This outcome was as expected and provides additional insights into forensic morphological research.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Additional Information:
Keywords: forensic science, forensic anthropometry, facial identification, image comparison, identification
Subjects: G Geography. Anthropology. Recreation > GN Anthropology
R Medicine > R Medicine (General)
T Technology > T Technology (General)
Colleges/Schools: College of Medical Veterinary and Life Sciences > Institute of Cancer Sciences
Supervisor's Name: Anderson, Dr. Robert A. and Siebert, Dr. Jan P. and Peter, Prof. Vanezis
Date of Award: 2008
Depositing User: Miss Krista Kleinberg
Unique ID: glathesis:2008-245
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 21 Aug 2008
Last Modified: 10 Dec 2012 13:17
URI: http://theses.gla.ac.uk/id/eprint/245

Actions (login required)

View Item View Item


Downloads per month over past year