Introduction
In issue 1 of assemblage we presented a feature article on our research in computerised three-dimensional facial reconstruction (Evison 1996). Readers are referred to that publication, and to Prag and Neave (1997) and Tyrrell, et al. (1997), for introductions to facial reconstruction and computerisation. Here we would like to give an update on our research in progress and introduce some of the 3-D prototype facial reconstructions we have developed.

The aim of our project is to develop a rapid, flexible and highly repeatable means of generating a resemblance of an individual from her or his skull. The primary purpose here is forensic -- our research is funded by the UK Home Office -- and the end product will be used to help identify missing persons and homicide victims. Facial reconstruction can be the last resort in cases were no other means of identification is available, and it does seem to work in a substantial number of such cases. Fortunately for us in the UK, few skeletonised human remains are discovered -- only about eight per year -- and most of these are identified by other means such as possessions, clothing, etc. Internationally, the situation is more serious. In the United States, for example, the homicide rate hugely exceeds that of the UK (women in Detroit are more deadly than men in London) and skeletonised remains appear all too frequently, deposited on the outskirts of major cities, particularly those adjacent to desert terrain. Even more distressing are the mass graves resulting from civil war and 'ethnic cleansing' in countries such as Rwanda and Bosnia. Here again is a considerable potential for bodies to be recovered which cannot be identified by any other means than facial reconstruction. To provide an international service for facial reconstruction which can also be used in human rights abuse cases in the field, we must add accessibility and portability to the requirements of our system.

We should recall that facial reconstruction is really a creative or artistic process with only a limited scientific basis. At best, facial reconstruction will generate a resemblance to the individual during life. Its value in identification is that a resemblance can lead the public to put forward a number of names which can be reduced by process of elimination until a positive identification can be established from DNA analysis or dental records. The limited fidelity of facial reconstruction is less of a problem in archaeology, where a definite likeness is not strictly required. Here the main objective is to use facial reconstructions to create empathy with those whose lives we hope to re-create, and considerable 'artistic licence' can be encouraged.

Improving the scientific basis

Before moving on to our experimental prototypes, we would like to describe some of the difficulties (or 'opportunities', as they are described in information technology-speak) we are facing. Currently, facial reconstructions are based on the application of average tissue depths at 20 to 30 landmark sites on the skull collected from cadavers or, more recently, from living individuals using ultrasound. We would like to dispense with this small number of data points in favour of a matrix of up to 10,000 measurements collected automatically from magnetic resonance (MR) imaging equipment. The apparently complex images depicted in MR scans (e.g. Figure 1) are represented digitally as a stream of grey-scale values (i.e. from black to white with a range of values in between) built up as a series of dots arranged in lines, much like the way the picture is built up on your television screen.

Figure 1. Magnetic resonance (MR) 'slice' from the scan of a head

In theory, tissue depths can be 'stripped out' from such images by a cunning algorithm which can recognise the outside of the head (i.e. when the black background values suddenly jump to the white values of the soft tissues of the face) and the inside (i.e. when the white values of the soft tissues of the face suddenly jump to the grey or black value of the bone and cartilage of the skull). Developing such a computer program, and applying it to collect tissue-depth data from volunteers of both sexes and of various ages, builds, and ethnic groups is likely to be a major long-term project. Whilst MR imaging should make data collection easier, it suffers from one disadvantage in comparison to ultrasound -- MR scanning is carried out with the subject lying down. Further mathematical transformation of the data may be required to convert a lying face (so to speak) to a sitting one.

Accessibility and flexibility in forensic facial reconstruction

Figure 2. 40×40 co-ordinate VRML image of a forensic facial reconstruction (appropriate browser plug-in or external viewer required)

Figure 3. 151×151 co-ordinate VRML image of an archaeological facial reconstruction with a point light source added (appropriate browser plug-in or external viewer required)

Figure 4. 151×151 co-ordinate VRML image of an archaeological facial reconstruction with the point light source in a different position (appropriate browser plug-in or external viewer required)

The Internet is the obvious medium within which to develop an internationally accessible service for forensic facial reconstruction. Virtual reality modelling language or VRML (Ames, et al. 1997) -- the recognised standard for 3-D on the Internet -- is the appropriate file format. We have been able to convert scanned 'plastic' reconstructions to VRML images for display on the Internet. To view VRML images you will require a VRML browser 'plug-in'. An Internet search will reveal a variety of such products. We encourage the reader to download the latest version of Cosmo Player® from Silicon Graphics. (Instructions are provided.)

When you have downloaded your VRML browser you will be able to view a variety of facial reconstructions. Figures 2 and 3 illustrate the difference between images generated at 40×40 and 151×151 co-ordinate matrix resolutions, respectively. Figure 3 also demonstrates the results of incorporating a point light source into the scene by means of a 3-D VRML editor and Figure 4 shows the light source at a different position.

Our most productive recent research has been directed at methods for interactively changing the facial appearance to correspond to the processes of fattening and ageing of the face, and to begin to address the difficult problem of facial appearance and ethnicity. Simulation of ageing and fattening of the face has been achieved by using standard plastic reconstructions of the young and old and thin and fat extremes. Once captured in digital form using a 3-D scanner (see Evison 1996), the files can be translated into VRML format. The coordinateInterpolator node of VRML permits a VRML scene subsequently to be constructed, in which the image of the face interpolates or 'morphs' between the young and old (Figure 5) or thin and fat (Figure 6) extremes. A colour approximating to 'white' skin was rendered using a VRML editor and the degree of 'faceting' was similarly reduced (cf. Figure 2).

Figure 5. Using the coordinateInterpolator node of VRML to morph between young and old reconstructions (appropriate browser plug-in or external viewer required)

Figure 6. Using the coordinateInterpolator node of VRML to morph between 'thin' and 'fat' reconstructions (appropriate browser plug-in or external viewer required)

The existence and nature of any relationship between ethnic identity and biological characteristics has been the most controversial and ill-used topic in anthropology. The issues involved are worthy of detailed consideration beyond the scope of this article. Suffice to say, we reject the existence of biological races in the human species. Race is a social construct and, since the term is entirely misleading, it is one which anthropologists should take the lead in abandoning. We believe, however, that a certain small number of biological characteristics are sometimes more frequent in some ethnic groups than others, and that it may therefore sometimes be possible to estimate which ethnic group a deceased individual is more likely to have identified with or been identified as.

We also believe that human groups have historically, and almost universally, undergone a process of genetic admixture and that this process continues apace in the present. People are not stereotypical biologically, and many people exist outside the main ethnic categories. Debate continues in the black community as to whether a person of mixed ancestry has a 'black' identity or a 'mixed' identity -- or, conceivably, a 'white' identity.

How should one approach facial reconstruction from a skull which exhibits a balance of features some of which are more common in black and some in white populations? Such a skull would not be unexpected in groups identifying themselves as 'black', 'mixed', or 'white'. Which tissue depth data set should we use and with what complexion should we render the finished reconstruction? Computerisation may help us here in that it enables us to interpolate or morph between the various options. Figure 7 illustrates one such example. An archaeological skull exhibiting features common in both white and black populations was classified by physical anthropologists as '?Negroid'.

Figure 7. Using the coordinateInterpolator and colorInterpolator nodes of VRML to morph between 'black' and 'white' reconstruction (appropriate browser plug-in or external viewer required)

Historical records associated with the site indicated that one 'foreign' and one 'Costa Rican' individual may have been present in the cemetery from which the skull was recovered. The selection of possible ethnic identities attributable to this individual is large and might include 'African', 'British', 'White', 'Latin', 'Caribbean', etc. There is no reason to assume that the individual was particularly light-skinned or dark-skinned either. In this example, casts were taken from the archaeological skull and reconstructions carried out using published 'white' and 'black' tissue depths. The faces were scanned into the computer and a VRML file constructed, in which the face appears to morph between the white and black means. At the same time, the skin surface interpolates between light and dark colourings. Whilst there is no reason to assume a simple correlation between tissue depths, skin colour, and ethnic identity, such a model offers a range of likely outcomes, some of which will offer a greater resemblance to the individual during life. Ethnicity apart, there is scope for considerable further work on the computer simulation of skin colour and texture in facial reconstruction.

Issues in archaeology

Figure 8. VRML image of a plastic reconstruction of the priestess from Anemospilia, constructed by Richard Neave (appropriate browser plug-in or external viewer required)

Whilst our emphasis is on the application of computerised 3-D facial reconstruction in the forensic field, the method will clearly be amenable to applications within archaeology. Although public interest in facial reconstruction is strong, facial reconstruction is almost totally ignored in academic archaeology. A number of factors may contribute to this situation. The misuse of anthropometry by racial anthropologists is an issue easily raised in criticism of facial reconstruction, especially in contexts where an estimation of ethnicity is being attempted. The importation of 'science wars' from US campuses may also contribute to suspicion of another scientific technique being applied in archaeology.

Ironically, facial reconstruction specialists are probably the only artists in archaeology who are keen to emphasise that their work does have a scientific basis! The novelty and technological complexity of computerised 3-D facial reconstruction may also seem daunting to some. Fortunately, public interest in facial reconstruction is a factor recognised by archaeological program makers and cultural resource managers. The ability of a thoughtfully presented facial reconstruction to create empathy with someone who may have lived many thousands of years ago would seem to be a tool of unique value in archaeology, even, or perhaps especially, in 'social' archaeology -- an observation which may have passed most social archaeologists by. It will be clear from the VRML images we have presented here, however, that considerable refinement will be required before computers can be used to present images as attractive as those produced by the sculptor (Figure 8, courtesy of Mr. Richard Neave).

Acknowledgements
We thank Andrew Chamberlain for making skeletal material available for analysis and Mandy Holmes for work on facial reconstruction and the fattening of the face. We thank Richard Neave for permission to reproduce the reconstruction of the priestess from Anemospilia in VRML. This project is supported by the UK Home Office.

References

Copyright © M. Evison, O.M. Finegan, and T.C. Blythe 1998

Copyright © assemblage 1998