The purpose of forensic facial reconstruction is to produce an image from a skull which offers a sufficient likeness of the living individual that it will facilitate identification of skeletal remains when there are no other means available. Although facial reconstruction had begun in the nineteenth century, the method gained notoriety with the work of Gerasimov (1968), depicted on film in Gorky Park. These traditional 'plastic' methods use modelling clay or plasticine to build up the depth of tissue on the skull (or a cast of the skull) to that of a living individual. Tissue depths are known for 'landmark' sites on the skull; the depths elsewhere are interpolated between these points (Figure 1) and then into the interstices (Figure 2). The shape of the eyes, nose and mouth cannot be confidently predicted and are largely guesswork (Figure 3). Even for skilled practitioners, plastic reconstructions take one or two days. The results obtained will differ between reconstructions and between practitioners.

Figure 1 Figure 2 Figure 3

Left to right: Figure1.Establishment of tissue depths at landmark sites on the skull (in white) and the interpolation between these sites. Figure 2. Interpolation of tissue depths into the interstices. Figure 3. Completed "plastic" reconstruction. The shape of the eyes, nose and mouth are guesswork.

The tissue depth measurements used tend to be those collected from cadavers in the early part of the twentieth century, or before. These measurements are biased because they come from small samples, because a dead person's tissues are not the same as in life, and because they take only limited account of the average differences known to occur between people of different age, build and sex, and between the major human diversity aggregates. For over a century, forensic artists and scientists have been attempting to improve the quality of facial reconstructions from the skull, efforts which have met with very limited success. Most recently, computerised methods for 3D facial reconstruction have been developed. These methods employ computer programs to transform laser-scanned 3D skull images into faces. Although the results are more reproducible than sculpted reconstructions, some subjectivity can remain in the 'pegging' of a composite facial image onto the digitised skull matrix. The use of such a standardised image will reduce the influence of the individual shape of each skull, which is after all fundamental to the person's appearance. Computerised methods may be repeatable, fast and precise, but as long as they employ the old data, the quality of the reconstruction will be undermined.

Figure 4. The Sheffield facial reconstruction suite: laser scanner, PC server and Industrial graphics workstation.

Facial reconstructions now employ color laser scanners and silicon graphics computer to capture 3D images of the skull (Figure 4). As the platform rotates (Figure 5) a 'wireframe' matrix is generated (Figure 6). Computed tomography (CT) scanning permits more accurate measurement of tissue depths (Figure 7). Large samples of tissue depth measurements can be collected, with associated attributes of age, sex, build and, where appropriate, ethnic group. A pilot study on the collection of tissue depth measurements from CT scans has been carried out by one of our team .Digitised images of facial features not predicted by the skull contours (nose, eyes and mouth) must be added by separate means to generate a wireframe face (Figure 8), onto which colour and texture can subsequently be rendered (Figure 9). If necessary, a skull can be reconstructed 'virtually' from the separately scanned parts (Figure 10).

Left to right: Figure 6. A wireframe matrix of a skull shown at 32 x 32 resolution. Figure 7.Computed tomography (CT) scan of the head of a living individual. Figure 8. A wireframe matrix of a face shown at 32 x 32 resolution. Figure 9. Rendered image of a facial reconstruction. Figure 10. Computerized reconstruction of a skull from the separately scanned fragments.

Facial reconstruction techniques use C++ and Silicon Graphics' programming languages to provide a coherent Windows-based medium for scanning, facial reconstruction and display. They build features into their models for the selective application of tissue depth datasets defined by parameters such as age, sex, build or ethnic origin. Another component allows the selection of facial features from computerised image libraries. The graphic and animation facilities of Open Inventor allow the display of a moving facial image cycling through a range of versions in a variety of lighting conditions - with the aim of maximising the likelihood of recognition. The animated 3D reconstruction could be downloaded to videotape or, as the Open Inventor.iv file format is the bas