Forensic facial reconstruction using 3-D computer graphics: evaluation and improvement of its reliability in identification

Vanezis, Maria (2008) Forensic facial reconstruction using 3-D computer graphics: evaluation and improvement of its reliability in identification. PhD thesis, University of Glasgow.

Abstract

This thesis is concerned with computerised forensic 3-D facial reconstruction as a means of identification and involves the restoration of the face on the skull in an attempt to achieve a close likeness of the individual when alive. The reconstruction process begins with the biological identification of the skeletal remains, (age, sex, ancestry and build). Facial reconstruction is then carried out and essentially works by building the “face” up from the skull using soft tissue thicknesses at specific locations from existing data. However, it is used as a last resort on skeletonised, badly decomposed or mutilated corpses, when no other information is available; even then it is only accepted as corroborative evidence in court. It is performed in the hope that it may stimulate recognition, and consequently narrow the field of identification, allowing other tests to be carried out, such as radiographic and/or dental comparisons, DNA analysis or other means, to establish positive identification.
The advantages of the computerised method over the manual clay reconstruction are speed, rapid editing capability, production of images that can be stored and reconstructions repeated at any time if required.
Furthermore, in many cases, the original skull instead of a cast or model may be used for reconstruction because the 3-D computerised procedure is rapid and non invasive. However, the most significant advantage of this technique with regard to the aims and objectives of the thesis is that a number of alternative reconstructions may be produced sequentially for the same skull by using different facial templates from the database that meet the anthropological/biological criteria of the skull.
The issues addressed by the study and therefore its main aims are:
a) evaluation and b) improvement of the reliability of facial reconstruction using 3-D computer graphics. The methodology involved initially digitizing a skull using a low-power laser scanner and a video camera interfaced to a computer. From a database of previously scanned faces, ten facial templates were selected that matched the anthropological criteria of each of the skulls, i.e. age, sex, ancestry and build. Landmarks with their corresponding soft tissue thicknesses were then located and placed on the skull and the equivalent ones on the face. The 3-D computer graphics then reconstructed the face by morphing (warping) the facial template over the skull by matching the corresponding landmarks on the skull and face with the appropriate soft tissue thicknesses at those landmark locations. The soft tissue thicknesses used at their specific landmark locations also matched the anthropological criteria of the skulls, since soft tissue depths are dependent on age, sex, ancestry and build.
One of the major problems with any reconstruction which affect its reliability for identification is the uncertainty of the shape of some of the individual characteristics of soft tissue structures such as shape of lips, ears and nose/nasal tip since there is not direct information on the skull regarding the shape of some of these features. In addition, with the laser scanning system, the faces within the database all have closed eyes, because of the potential laser hazard to the eyes. Thus it is necessary to add “opened” eyes, head and facial hair (where appropriate) to give a realistic appearance to the face. The software provides the facility to export a 2-D view in a TIFF or JPEG format from the 3-D reconstructed image. The file can then be imported into a police identi-kit system such as E-FIT™, which allows the addition of features.
In this study five skulls of known individuals were used for reconstruction in the manner explained. Ten facial templates which fulfilled the anthropological criteria (age, sex, ancestry and build) for each skull were used for the rebuilding process, thus totalling fifty reconstructions.
The study employed a psychological resemblance test (experiment 1) where 20 different assessors, were asked to select in each case study, the best three matches of the ten reconstructions with the ante-mortem photograph of the individual during life. The results from these tests were correlated with a mathematical shape analysis assessment using Procrustes Analysis in which, the skull was compared in turn with each of the ten facial templates of each case study (experiment 2).The ranking of the assessors’ reconstruction choice was correlated with the ranking of the Procrustes Analysis by using Spearman’s Rank Order Correlation. The results indicate that although not statistically significant, it would seem however, that in some of the case studies, the mathematical approach using Procrustes Analysis does seems to capture some perceptual similarity in human observers.
Experiment 3, similar to experiment 1, was a further psychological resemblance test, which involved implementing E-Fit features on four of the ten reconstructed images per case study. Assessors were asked to select the closest E-Fit image match with the ante-mortem photograph. Again, results indicated that, although not statistically significant, adding E-Fit feature to the images appears to improve perceptual similarity in human observers, provided, the limitations of adding these characteristics are addressed. Furthermore, there also appears to be good agreement in most of the case studies between the two psychological resemblance tests using the two different sets of assessors in experiment 1 and 3 (reconstruction choice and E-Fit choice, respectively).
Further work involving anthropometric comparisons and using two methods of assessment (landmark line matching between images and proportion indices) was also carried out (experiment 4). It was found that matching landmark lines between images appeared to be only of limited value due to the images not being aligned at exactly the same viewpoint and magnification. It should be appreciated that because the thesis was based on recognition and was not an anthropometric study, precise alignment of viewpoints was not a requirement. Hence using the same data from the study, although images were in the frontal view, they were not aligned to the accuracy acceptable for an anthropometric study as there was no requirement to so. It would appear that, although there was some correspondence between the discrepant distances and the first and second ranked reconstructions, no firm conclusions could be drawn from this technique and therefore does not assist in understanding the way observers made their choices. Further tests would need to be carried out (beyond the scope of the thesis) to reach any firm conclusions.
Undoubtedly, given the complex nature of the recognition process, it would have been desirable to use reconstructions of persons known to the assessors rather than asking them to assess unfamiliar persons, since it is well established that familiar faces are easier to recognize than those that are unfamiliar to observers. It should be appreciated however, that, although the study was designed in this way for practical and ethical reasons, it nevertheless does not truly reflect the real operational forensic scenario.
Furthermore, recognition/matching is a much more complex process and even a reconstructed face which may be generally morphologically similar to the person in life may not capture perceptual similarity in human observers, especially in an unfamiliar scenario. It is not certain that identification will always occur even when the facial reconstruction bears good resemblance to the target individual.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Keywords: Forensic cranio-facial identification, 3-D computer graphics, facial reconstruction, psychology of recognition
• Subjects: R Medicine > RA Public aspects of medicine > RA1001 Forensic Medicine. Medical jurisprudence. Legal medicine; Q Science > QM Human anatomy; B Philosophy. Psychology. Religion > BF Psychology
• Colleges/Schools: College of Medical Veterinary and Life Sciences > School of Cancer Sciences
• Supervisor's Name: Burton, Professor Mike
• Date of Award: 2008
• Depositing User: Mrs Maria Vanezis
• Unique ID: glathesis:2008-375
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 18 Nov 2008
• Last Modified: 10 Dec 2012 13:18
• URI: https://theses.gla.ac.uk/id/eprint/375