鈥淒O YOU know this person?鈥� runs the headline above an eerie photograph of a
person鈥檚 face. It鈥檚 eerie because the image is not of a real face but a
reconstruction. It was created from the skull of a person who died with nothing
else in the world to identify them but their bones.
Such reconstructions are published in newspapers as a last resort, when a
skeleton is found and the police cannot put a name to it. That they are often
recognised is a testament to the skill of the people who create them. Yet they
are not proof positive of identity. 鈥淲e wouldn鈥檛 ever suggest that a face we
create is a portrait of a person,鈥� says Caroline Wilkinson, a medical artist at
the University of Manchester. 鈥淏ut it should be recognisable as that
辫别谤蝉辞苍.鈥�
Facial reconstruction was once the sole preserve of artists like Wilkinson,
who builds up a face by painstakingly moulding clay onto a cast of a skull.
Today, however, computer artists are gradually gaining ground, wrapping
cyberflesh round a three-dimensional image of a skull. This movement from the
tactile to the virtual is, in part, an attempt to produce a faster, more
flexible way of reconstructing a face. There is also the hope that digital
technologies will erode uncertainties inherent in the process. Researchers
believe that their attempts to improve the scanning of faces and skulls and to
find better correlations between features on the face and those of the skull
will raise the art of reconstruction to a higher plane.
Advertisement
In her work, Wilkinson first places pegs of different lengths at a few dozen
鈥渓andmarks鈥� around the skull to indicate the depth of flesh at each point. These
tissue measurements are averages taken from tables collated from postmortem or
ultrasound studies, and are listed by sex, age, build and race. Guided by these
limits, Wilkinson crafts a face using her anatomical knowledge first to sculpt
the larger facial muscles, followed by the finer musculature.
Virtual clay
At the Human Identification Centre of the University of Glasgow, forensic
pathologist Peter Vanezis leads a team that has replaced modelling knives with
computers. Here, the researchers work with a 3D laser scan of a skull and apply
virtual pegs to fix the tissue depths. Their 鈥渃lay鈥� is a facial template which
they wrap round the prepared skull, and the job of moulding the face between the
pegs is left to a computer algorithm. The team has about 200 templates to choose
from: digitised faces of men and women of different ages, builds and racial
groups.
This method is faster than modelling in clay and makes it easier to modify a
face. Weight can be added or removed from a face simply by recalibrating the
lengths of the pegs, and individual features can easily be changed. The drawback
is finding a template that is a good match for the scanned skull. For example,
you get imperfections in the final image if gross features of the template
differ from those of the skull. 鈥淚f you have a face that doesn鈥檛 match very well
with a skull,鈥� says Vanezis, 鈥渢he face will have to stretch quite a bit to
produce a final image, so you would get more distortion of the facial features
than you really want.鈥�
Despite this drawback, the Glasgow team has had some significant successes.
Vanezis points to the murder in London of a homeless woman who had been buried
anonymously after an investigation failed to uncover her identity. Later, when
two people confessed to the murder, the police reopened the case and exhumed the
body, and Vanezis began to reconstruct the face.
The researchers scanned and prepared the skull and wrapped a template round
it. Then they brought other information about the remains to bear. Hair still
remained on the victim鈥檚 head, giving them an idea of hair colour and length at
the time of death. Later, they drew in the likeness of a pair of glasses found
near the body. And a key distinguishing feature came from a dental colleague.
鈥淲e showed the jaws to him and he said they were smoothed down and [this
indicated] it wasn鈥檛 her practice to wear dentures,鈥� says Vanezis. 鈥淪o we
[created] a face which had a slightly sunken-in mouth to take that into account.
It actually wrapped around [the skull] quite nicely.鈥�
Within days of the reconstructed face appearing in London鈥檚 Evening
Standard, two people came forward with a name. Forensic scientists
contacted the family and compared a DNA sample from a female member with one
taken from the corpse. The samples matched, and finally they were able to give
the dead woman a name.
Shape, sex and race
The centre鈥檚 work is not just confined to Britain. Its reconstructions have
been used to identify individuals in a mass grave in Chile and to recreate the
faces of several Sri Lankans killed in a string of related murders. In all these
cases, as with the unidentified body in London, the reconstructions relied
heavily on other evidence from the grave.
The skull alone reveals only so much about how a person once looked. Features
such as high cheek bones or a jutting jaw help to define the shape of the face.
Often, sex and racial origin can also be gleaned from combinations of
measurements, such as proportional variations in brow and cheek size, says Alf
Linney, reader in medical physics at University College London and an expert on
facial variation. But these measurements are not foolproof, he adds. As in so
many other areas of biology, race in particular is not a clear cut
definition.
Beyond the skull, details such as whether or not a person wore glasses can
profoundly change their appearance. Then there is the rest of the skeleton.
Deterioration of joints, for example, can help with a person鈥檚 weight and age.
鈥淚f you are looking at a historical person the weight of that person might be
reflected in the state of their knee joints,鈥� says Linney. 鈥淲ith more modern
bodies you may simply look at the clothing. You often find that clothing does
not deteriorate very fast. Nylon, for example will outlive soft tissue. So if
nylon or even cotton underwear survives, it will give an idea of the girth of an
颈苍诲颈惫颈诲耻补濒.鈥�
Finer details about a person鈥檚 face are even harder to come by. Cartilage
decomposes quickly, so ears are particularly difficult to recreate. So too is
the nose, though the nasal spine, a bony outcrop just above the teeth, gives at
least some idea of how it might have looked. Hardest of all to recreate are the
lips. 鈥淪o what you end up with is a facial shape with a reasonable
representation of the nose, no proper representation of the ears and possibly a
vague representation of the lips,鈥� says Linney.
Wilkinson tackles this problem in her clay models by moulding features in
sympathy with the rest of the face. Vanezis has a different solution. He would
like to use maths to eliminate the uncertainties, so his team is preparing to
compile a cephalometric index of skulls and faces.
Their aim is to collect 3D images of the people鈥檚 faces and their
skulls鈥攆rom CT, MRI or ultrasound scans. From these pairs of images, the
researchers intend to locate landmarks on both skull and face and calculate
their relative positions. This, says Vanezis, will allow his team to find a
facial template for an unidentified skull based on calculated values rather than
visual comparison.
It could also bring added refinement when wrapping cyberskin round a
digitised skull. It may be, for example, that a particular arrangement of
landmarks on the skull around the cheek or nose will correspond to puffed cheeks
or a turned-up nose. Such relationships would remove much of the guesswork from
reconstructing facial features.
鈥淓ars are going to be difficult to sort out whatever you do,鈥� says Vanezis.
But the technique should firm up choices over the shape of mouth and nose. The
teeth and the way the jaw is positioned鈥攚hether it has an overbite, an
underbite or a jutting jaw, for example鈥攚ould give a good framework on
which to build the mouth. Likewise, pinning down the precise trajectory of the
nasal spine should help with the nose. Vanezis hopes that the size of the nasal
aperture in the skull may also give a clue to the size of the nostrils. 鈥淭here
has been work done on these relationships, but we need to explore them further,鈥�
he says.
Another key to improving facial reconstruction will be increasing the number
of templates well beyond the 200 mark. The more facial shapes that are stored,
the more chance there is that one of them will match a skull under
investigation. This should ease the burden on the software. 鈥淲hat you are trying
to do is minimise that distortion so the computer just鈥oes the
fine-tuning,鈥� says Vanezis.
One other area being tackled by the Glasgow researchers is creating better
optical images of skulls and faces. At present, they place a skull or person on
a turntable in a darkened room. A laser produces a single, thin stripe of light,
down the face or skull. As the subject rotates, a 鈥渟tereo鈥� pair of widely spaced
cameras records the changing contours. A program then uses triangulation to turn
these into a 3D digital image.
The problem with this approach is that the scan takes about 30 seconds and
people tend to move during this time, which can impair resolution. The image
created by the software also has a plain-coloured synthetic texture, which can
make it difficult to locate landmarks, such as where different bones meet.
To overcome these obstacles, Vanezis鈥檚 group is working with the Turing
Institute, a 3D imaging company controlled by Glasgow University. Researchers
there have dispensed with the laser. Instead, new software builds 3D images from
stereo pairs of photographs of the whole face taken in one go. With exposure
times cut to about 10 milliseconds, errors caused by movement are greatly
reduced.
The institute has also added a third, colour camera which allows a
鈥減hotographic render鈥� to be added to the image, says Tim Niblett, chief
executive of the Turing Institute. This gives the image the same colour and
texture as the original subject. 鈥淲hat this will enable is easier capture of an
image and more precision in landmark placement,鈥� says Vanezis.
All this effort should make those eerie images in newspapers more
recognisable. But there will always be some imponderables. Expression on a
person鈥檚 face, for example, is crucial to identification, says Vanezis. And this
can create a real dilemma: you have a man鈥檚 skull, you create a face, but what
expression do you give him? 鈥淚f the guy was miserable all his life and we put a
smile on his face no one would recognise him,鈥� he says.
With all the technological improvements, then, is there a chance that facial
reconstruction could ever be used as proof positive of who a person was? Here,
everyone in the field agrees. No. 鈥淭his is a tool for recognition,鈥� says
Wilkinson, 鈥渘ot identification.鈥�
IT鈥橲 an intriguing idea. Could a smidgen of DNA from a corpse reveal what a
person once looked like? If so, you could abandon clay and computer-based facial
reconstruction. 鈥淵ou wouldn鈥檛 need the skull and you wouldn鈥檛 need any
reconstruction software because the [genetic] code would determine the face,鈥�
says Robin Richards, a senior research fellow who works with Alf Linney at
University College London.
One of Linney鈥檚 interests is whether faces can be used for identification in
the same way that fingerprints already are. He takes 3D scans of people, and
marks the images at points such as the corners of the eyes and mouth. He can
then describe the faces in terms of distances or angles between these marks. He
hopes that one set of these values will turn out to be unique for every
person.
Also at UCL, David Hopkinson, director of the Medical Research Council鈥檚
Human Biochemical Genetics Unit, has other plans for Linney鈥檚 measurements. He
wants to see if close correlations exist between any of these numbers and the
occurrence of specific genes.
But with all the thousands of human genes, where do you start? Genes of
obvious interest, such as hair and eye colour, are slowly emerging. Beyond
these, Hopkinson looks to a clutch of rare genetic disorders. People with
Treacher Collins syndrome and Crouzon鈥檚 syndrome have common facial features,
and genes for these disorders have recently been pinpointed. 鈥淭hese lessons
coming from the study of abnormality may be applicable to the study of normality
or normal variation,鈥� says Hopkinson. 鈥淏ut how directly applicable they are I
don鈥檛 think we really know.鈥�
The researchers face a huge task if not an impossible one. Too much of our
appearance is dictated by influences that are not reducible to genes, say critics
(Picture of a killer, New Scientist supplement,
4 October 1997, p4).
But until someone looks, it鈥檚 not clear who is right. Richards for one, is
undaunted. 鈥淚 am not saying the route from one end to the other is easy,鈥� he
says. 鈥渂ut the information in principle is there, all you need to know is how to
read it.鈥�