During the Renaissance period, people began to take an interest in the anatomy of the human body. Artists such as Verrocchio, Michelangelo, Versalius are known to have used wax models to document their works. Towards the end of the 18th century, scientific art became popularised and the artists of that time have been considered to be the first sculptors to realise that the skeleton is the ideal frame upon which to build the musculature and the body (Wilkinson, 2004). The credit for developing the theory behind facial reconstruction can be thus credited to these artists. In the 19th century, work was done to obtain the facial tissue depth measurements from cadavers.
Facial reconstructions were earlier achieved through the collaboration of scientists and artists. Anatomists depended upon the sculptors to depict their data, as can be seen in the cases of His and Sefner (1895), Kollmann and Buchly(1898) (Vermeulen, 2012). In 1895, the German anatomist His made the first scientific endeavour in this field and worked with the artist Sefner to reconstruct a plaster skull cast of Johann Sebastian Bach using the skin depth measurements at nine midline facial points and six lateral points of twenty-four males and four female white cadavers in Leipzig. He further authenticated the reconstruction by comparing it with available portraits of Bach.
A few years later, Kollmann and Buchly also made a facial approximation of Dante in 1898 (Snow et al., 1970; Rynn et al., 2012) from the tissue depth measurements taken at ten midline and eight lateral points of 21 males and four female cadavers to His’s total, thus producing mean measurements for 45 male and eight female European White cadavers. The subjects were ranged between 17 and 72 years of age and were all described as moderately well-nourished (Snow et al, 1970). Kollman then went on to reconstruct the face of a stone-age woman from France with the help of tissue depth measurements taken from hundreds of women around that area and produced technical drawings, which were then brought to life by Buchly (Vérze, 2009).
Various anatomists and anthropologists produced many further reconstructions of early hominoids such as Neanderthal and Pithecanthropus, and others of the stone age such as that of a well preserved Neanderthal skull found in the Chapelleoux Saint, in France, in 1908; the head of an old Neanderthal male from the cave of Le Moustier, France. In 1913 anthropologists Martin and Von Heggeling at the Anatomy Department of Jena University produced different reconstructions of a Neanderthal face from the same skull (Tyrrell et al., 1997).
With the advent of the 20th century, facial reconstructions began to be used in museums and also the various manual reconstruction techniques began to spring up. Also, in 1989, the first ever three-dimensional computer-assisted reconstruction technique for forensic identification was developed by Vanezis et al. (1989). The method utilised a low-power laser scanner and a video camera interfaced to a computer. The current trend is to move towards computer-assisted techniques that are considered to be less subjective and more rapid.
THREE-DIMENSIONAL FACIAL RECONSTRUCTION TECHNIQUES
Traditional Manual Reconstruction
Sculpting the face over the unknown skull with clay, wax or plasticine is one of the most common forms of three-dimensional reconstruction. There are a number of ways to go about it. Nonetheless, in any reconstruction technique, the first step is to examine the skull to determine biological characteristics such as age, gender, race, etc., after which a replica of the skull is obtained to be worked on. One of the most influential and significant pioneer in early facial reconstruction work was the Russian anthropologist, Mikhael Gerasimov (1971) who developed a manual technique, that relied upon anatomical knowledge and sculptural skills. He developed the Russian method, also known as the anatomical method, in which the development of the musculature of the skull and neck is regarded as being of fundamental importance. Firstly, three main muscles of mastication namely the temporalis, masseter and the buccinators are modelled over the skull while taking extra care not to exaggerate the bulk of those muscles. Secondly, the circular muscles around the mouth and eyes were built up. Other tissues such as the parotid glands and any fatty deposits were then added, if required. Next, a layer of skin which can be textured is applied as a last step in the process.
Some experts (Snow et al., 1970; Cherry and Angel, 1977; Gatliffe and Snow, 1979; Helmer, 1984; Krogman and Iscan, 1986) perform the reconstructions only based on the soft tissue thickness of the face. Such method utilises tissue depth markers such as dowels or rubber of different lengths depending on the tissue thickness at various anatomical points on the face. The dowels are cut according to the required thickness and glued to the skull. This method is also known as the American or anthropometrical method and was developed from the works of Krogman by the forensic artist Betty Pat Gatliff and the physical anthropologist, Clyde Snow. Soft tissues are added in bulk and strips without any regard for the underlying anatomy, all the while making sure that the clay does not exceed the length of the markers at any point. Others (Neav, 1997; Wilkinson, 2004) prefer a combination to the two methods above, which employs the detailed traces of muscle insertion on the skull to ascertain facial detail and form, and relies on tissue thickness data, as in the American method, to model soft tissue depth. This is known as the Manchester or the combination method.
Facial features such as the eyes, nose, ears and mouth cannot be directly determined from the skull and these are added based on some existing guidelines obtained from various studies ( Gerasimov, 1971; Gatliff, 1984; Krogman and Iscan, 1986). However such guidelines have been found to be inaccurate (Suk, 1935) to some extent and does not give an exact representation but only an approximate one. Although various successful cases have been reported in the past, the traditional technique is subjective and controversial due to the presence of an artistic aspect. Also, the obtained results always tend to differ between practitioners and also between reconstructions. This point was clearly illustrated in the Green River serial killer cases, in which multiple facial reconstructions of several victims were created by different practitioners. The results were highly variable from practitioner to practitioner and met with little success (Davy et al., 2005).
Contemporary Computer-based Reconstruction
A computer, when compared with a human expert, was found to be consistent and objective. Knowing all the modelling assumptions and given the same input data, a computer always generated the same output data. Furthermore, it was possible to generate many faces with little variations from the same skull (Claes et al., 2010). The first introduction of the computer based technique was done by Vanezis and his colleagues (1989). Like every computer-based reconstruction technique, the skull is first digitised. This was done using a laser scanner and a video camera interfaced to a computer, forming a fully shaded 3D surface. An advantage of this system, thus, was that the reconstruction work could be carried out on the image of the actual skull, rather than a replica as in manual reconstructions. Various markers indicating the tissue depths are then placed on different selected sites he skull. Then a face from a databank of previously scanned face of live subjects is placed over the skull in the form of a mask, allowing it to specifically conform to the skull based on the different tissue depths at various points. Vanezis et al. (1989) performed both the traditional clay as well as the computer-based technique to make a comparison. After their studies, they concluded that although the manual reconstruction has a number of limitations, yet it can prove to be quite effective when done the right way. The computer based technique provided greater speed and flexibility but it was still far from perfect. They later improved the technique by upgrading the computer software (Vanezis et. al., 2000) and added the same tissue markers from the skull on to the face as well. However, live subjects were scanned with their eyes closed and hence, in the last step, eyes and other facial features were added using the police identikit systems to humanise the face.
Subsequently, Evenhouse et al (1992) proposed the idea that the individual identity of a face is the direct result of the scale, position, and the ratio of facial features relative to one another. since the relative positions of facial features are dependent on the skull, Evenhouse put forward a hypothesis that an individual’s face could actually be formed by substituting an average face onto the skull and formulated an image warping algorithm to create an average face. He merged the facial features of five females to create the average face and with the help of 37 tissue markers, mapped the face onto the skull. He believed that with further research on tissue depth measurements, and a more detailed averaging of the facial features, this technique could be a good tool for the identification process.
A further technique was proposed by Quatrehomme et al. in 1997 to produce facial reconstructions based on deformable models, instead of tissue depth measurements. They used a CT scanner to obtain the digitised 3D models of two pairs of skulls along with their facial data. The first pair was used as a reference and the second to validate their method. A global parametric transformation algorithm was then used to turn the reference skull into the skull to be reconstructed. This was based on what is known as the crest lines, which are lines of absolute maxima of the largest principal curvature of the skulls. The crest lines of the reference skull were matched to the crest lines of the skull to be reconstructed. The algorithm was then applied to the reference face to obtain the reconstructed unknown face. (Vanezis et al., 2000; Nelson and Michael, 1998).
Nelson and Michael (1998) believed that the lack of fully understanding the relationship between the soft tissues and the underlying skull was one of the main limitations of every reconstruction technique. Therefore they introduced a new approach through volume deformation. In the first stage, the unknown skull and a number of reference heads similar to the unknown skull in terms of age, sex, and race are selected which are digitised using a CT scanner. In the next stage, a set of control points are then placed at specific anatomical positions on the the heads and the skull. Then a single head is selected for deformation by calculating and comparing the spatial distribution of the control points. Lastly, the selected head is then deformed to the shape of the skull with the help of control points and any adjustments such as the addition of facial features, facial expressions and tissue depth variations are also done by manipulating the control points. This method has one major advantage over surface deformation methods in that all the data representing the facial soft tissues are deformed and not just the surface, thus the face is not merely a mask suspended on a restricted number of reference points.
Several regional methods have recently been proposed, where the face and skull are segmented into regions, and the relationship of each region is then learned independently. After which, the facial regions for a given skull are estimated and finally glued together to generate a face (Qingqiang Deng et al., 2015).
All Current computer-based techniques share the same general model-based work flow which have been summarised by Claes et al. (2010). They are basically a virtual mimicking of the traditional sculpting techniques where anthropological examination of the unknown skull is conducted to create a biological profile such as age, gender and race, etc. Then a virtual copy of the skull to be reconstructed is obtained by digitising the real skull through various techniques like CT scan, MRI, Ultrasound and the like. The core of every technique lies within the craniofacial model being used, which is considered to be the equivalent of the expert performing a manual reconstruction. A reconstruction is then performed by finding the geometrical relationship between the craniofacial model and the unknown skull based on an appropriate skull representation such as tissue depth markers, that may be manually or automatically placed. In the final stage, the reconstructed facial shape can be textured and visualised in order to generate images for further distribution and recognition triggering.
The current inclination in facial reconstruction is towards the computerisation of the methodology. But with the ever changing and advancing technology, it is too soon to designate the best approach or make comparison between methods. On the occasion of introducing the computer system into the facial reconstruction method, the most important matter may be to establish reliable standard databases of facial components (Miyasaka et al., 1995). However, while the computer technology is advancing at an astounding rate, one of the major problems remain the lack of tissue depth data upon which reconstructions are based on and also the fact that such data are seen as unreliable, and is still the subject of ongoing research..
TECHNIQUES FOR TISSUE DEPTH MEASUREMENT
A number of studies have gone into quantifying the relationship between the soft facial tissues and the underlying skull, for the purpose of reconstruction. The first of its work, however, can be credited to Welcker (1883), a German physiologist and anatomist in 1883, who documented the average tissue depth thickness of cadavers, by inserting a small surgical blade into various anthropometric landmarks on the face and then measuring the depth of penetration. In the late 1880’s and early 1890’s, His and Kollman built upon Welcker’s work, slightly modifying it by inserting a thin sharp needle which had a small piece of rubber on its tip, instead of using wider blade, resulting in a more accurate data. Subsequently, the anthropologist Kollman along with the sculptor Buchly, used a needle that was covered in soot and considered the clean part of the needle to be the tissue depth.
However, with the advancement in technology, tissue depth measurements began to be taken using radiographs, MRI images and CT data, but more recent reliable measurements were performed using ultrasound, as the test subjects could be seated in an upright position, with the skin elasticity and muscle tone in life adding bulk to the face (Vandermeulen et al., 2012). Nonetheless, ultrasonic scanning had the additional problems of being difficult to measure and reproduce due to the imperceptibility of the bone underlying the tissue (Lee et al., 2012). All computerised reconstruction techniques up to this day use CT scanners to digitise the unknown skull. One of the major advantages of using a CT scanner to acquire samples for the database is the possibility to have both the reference skull and face surface information obtained simultaneously and in the correct spatial relationship to each other. However, there is the disadvantage of the level of radiation absorbed by the live subjects and the fact that CT scans are obtained in a horizontal position, due to which facial shapes extracted from CT images will differ from the typical facial shape as viewed in an upright position, which is the normal pose for viewing, and hence recognising faces. Lastly, CT scans are sensitive to high density material, for example, the dental restorations used in patients such as amalgam (Vandermeulen et al., 2012). Another alternative to CT scanners were the use of MRI scanners for soft- tissue imaging. In contrast to CT, MRI is considered to not be harmful and soft-tissue depth measurements can be acquired, including differentiation between different types of soft tissue (muscle, fat). However, the subjects were also scanned in the supine position as in the CT scanners. Furthermore, because of the poor hard-tissue visualisation in MRI estimation of the hard tissue boundary is complicated and no robust solution for the complete facial region is yet available.
During the twentieth century, various anatomists studied and collected the measurements of tissue depths in different racial and ethnic groups. Birkner (1904) studied Chinese cadavers; Fischer (1903) studied Papuan cadavers; Von Eggeling (1909) studied Herero (Namibian) cadavers; Suzuki (1948) studied Japanese cadavers; Rhine and Campbell (1980) studied American Black cadavers; Rhine et al (1982) studied American White cadavers; and Rhine (1983) studied Southwestern Indian cadavers (Wilkinson, 2015).
Later on, Cone-beam CT (CBCT) scanners were developed and being used in medical practice. In contrast to conventional CT, CBCT scanners typically acquired images of subjects in an upright position. Furthermore, CBCT images could be obtained with a lower absorbed dose (Loubele et al., 2009) at the cost of a slightly reduced image quality (Loubele et al., 2008). Hwang et al. (2011) tested the reproducibility of CBCT scanners by measuring the soft tissue thickness at 31 landmarks on 20 subjects. They then found that thickness was measured with high reproducibility, while also suggesting that certain landmarks should be redefined.