See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/51789914

An Accuracy Assessment of Forensic Computerized Facial Reconstruction

Employing Cone-Beam Computed Tomography from Live Subjects

Article in Journal of Forensic Sciences · November 2011

DOI: 10.1111/j.1556-4029.2011.01971.x · Source: PubMed

CITATIONS READS

110 2,457

3 authors:

Won Joon Lee Caroline Wilkinson

National Forensic Service Liverpool John Moores University
47 PUBLICATIONS 581 CITATIONS 113 PUBLICATIONS 2,893 CITATIONS

SEE PROFILE SEE PROFILE

Hyeon-Shik Hwang
Chonnam National University
87 PUBLICATIONS 2,477 CITATIONS

SEE PROFILE

All content following this page was uploaded by Won Joon Lee on 16 June 2019.

The user has requested enhancement of the downloaded file.

 J Forensic Sci, March 2012, Vol. 57, No. 2
doi: 10.1111/j.1556-4029.2011.01971.x
PAPER Available online at: onlinelibrary.wiley.com

ANTHROPOLOGY

Won-Joon Lee,1 M.Sc.; Caroline M. Wilkinson,1 Ph.D.; and Hyeon-Shik Hwang,2 Ph.D.

An Accuracy Assessment of Forensic

Computerized Facial Reconstruction
Employing Cone-Beam Computed
Tomography from Live Subjects

ABSTRACT: The utilization of 3D computerized systems has allowed more effective procedures for forensic facial reconstruction. Three 3D
computerized facial reconstructions were produced using skull models from live adult Korean subjects to assess facial morphology prediction accu-
racy. The 3D skeletal and facial data were recorded from the subjects in an upright position using a cone-beam CT scanner. Shell-to-shell deviation
maps were created using 3D surface comparison software, and the deviation errors between the reconstructed and target faces were measured. Results
showed that 54%, 65%, and 77% of the three facial reconstruction surfaces had <2.5 mm of error when compared to the relevant target face. The
average error for each reconstruction was )0.46 mm (SD = 2.81) for A, )0.31 mm (SD = 2.40) for B, and )0.49 mm (SD = 2.16) for C. The facial
features of the reconstructions demonstrated good levels of accuracy compared to the target faces.

KEYWORDS: forensic science, forensic facial reconstruction, accuracy, 3D computerized modeling method, cone-beam CT, Korean

Forensic (cranio-) facial reconstruction (FFR) is a technique remains (6) and were ultimately developed as a useful forensic
based on both scientific standards and artistic skill to rebuild a face identification tool.
onto a skull to recreate the antemortem appearance of the individ- Currently, two basic techniques are utilized in FFR: two-dimen-
ual in order to recognize and identify the decedent. It is also known sional (2D) and three-dimensional (3D) facial reconstruction. Each
as ‘‘forensic facial approximation’’ (1,2). FFR may be very useful technique might be also subdivided into two methods: manual and
in forensic investigations where other approaches are not possible, computerized methods (9,10). The 3D manual methods that have
or few clues are remaining to aid the identification of human been mainly employed in forensic or archaeological facial recon-
remains. As the public are fascinated with the facial appearance of struction cases include the anatomical (Russian), anthropometrical
our ancestors or historically important figures, FFR is also utilized (American), and combination (Manchester or British) methods,
in archaeological research particularly where there are no portraits which were developed by Gerasimov, Krogman, and Neave,
or sculptures available (3). respectively (5,11–13). Whatever the techniques applied to the
It is widely accepted that the first scientific facial reconstruction FFR, these 3D manual methods share the common principle of the
was attempted by the Swiss-born German anatomist Wilhelm His relationship between the facial skeleton structure and the overlying
in 1895 (4,5). He applied average tissue depth thicknesses mea- soft tissue.
sured from German cadavers to reconstruct the face of the com- It is important for forensic investigation and academic theory to
poser Johann Sebastian Bach. At the turn of the 20th century and establish the accuracy of the process of facial reconstruction in rela-
for the next few decades, the early techniques of facial reconstruc- tion to facial morphology prediction, as facial resemblance is
tion were mainly applied to archaeological investigations (6). In the related to recognition. A number of accuracy studies using tradi-
former Soviet Union, the renowned archaeologist and anthropolo- tional 3D manual methods have demonstrated good levels of like-
gist Mikhail Gerasimov was a pioneer in the field of FFR, and he ness to the target faces (5,14–18). However, some researchers have
reconstructed numerous faces for research purposes and forensic criticized the 3D manual methods as being highly subjective, time-
identifications (7). In the United States, it was not until 1946 that consuming, requiring artistic intuition, and producing a single-facial
facial reconstruction was studied, when the anthropologist Wilton estimation (19,20). In an attempt to overcome these limitations,
Krogman became interested in the procedure and performed automated FFR methods have been developed (21,22). The concept
research into the accuracy (8). The techniques for reconstructing of the automated, computerized FFR originated from a system for
faces were then employed for the identification of unknown human the simultaneous visualization of soft and hard tissues in the field
of maxillofacial surgery. Moss et al. (23,24) at the University
1
Centre for Anatomy and Human Identification, College of Life Sciences, College London developed a computerized system using a laser
University of Dundee, CAHID, MSI Building, Dow Street, Dundee, DD1 line scanner.
5EH, U.K.
2
Department of Orthodontics, School of Dentistry, Chonnam National
Automated 3D computerized FFR commonly fits an ‘‘average’’
University, Gwangju 500-757, Republic of Korea. face template to a target skull using either a ‘‘sparse approach’’
Received 9 Sept. 2010; and in revised form 16 Dec. 2010; accepted 30 based on a set of anatomical landmarks placed on the face template
Dec. 2010. or a ‘‘dense approach’’ based on a spatial volumetric template from
318  2011 American Academy of Forensic Sciences
 LEE ET AL. • AN ACCURACY ASSESSMENT OF FORENSIC COMPUTERIZED FACIAL RECONSTRUCTION 319

both the face and skull (25). The sparse approach was first devel- reconstruction and participants were then asked to match the recon-
oped by Vanezis et al. (21), modifying the 3D visualization system structed face to images in a face pool. The facial reconstruction of
of Moss et al. (23,24). Since their pioneering work, the sparse the female face scored 26% correct matching, whereas the male
method has been developed using a geometry-based muscle model- face scored 68%.
ing integrating with the facial deformation technique (26) or a flex- Vanezis et al. (21) attempted a comparison of the FFRs produced
ible statistical model to reduce the facial template bias (27). from the manual and computerized methods in a single-blind test.
The dense approach was first introduced by Quatrehomme et al. The results showed that both methods could be employed as a
(19) and has been modified and improved by other researchers useful tool for identification. Helmer et al. (16) performed a dou-
using control data sets (28), a dense placement of soft tissue depth ble-blind accuracy test using 12 skulls reconstructed independently
(29,30), a statistical shape model of both the skull and the face employing a manual method by two practitioners, and three
instead of a sole extrapolation of the deformation field (31), or assessors determined the degree of resemblance using five-point rat-
computed tomography (CT)-derived implicit surface representations ing scale from one (great) to five (no resemblance). The result
(20). showed 50% approximate resemblance (scale of three) as the mean
Automated, 3D computerized FFR has exhibited some advanta- rating.
ges over the traditional manual methods, such as increased effi- Research into the reliability of FFRs has continued in the 21st
ciency, producing many variations for a face, and partial removal century. Stephan and Henneberg (11) investigated facial morphol-
of practitioner subjectivity. However, the new approach has ogy prediction using 16 reconstructed faces and 37 assessors. They
exposed significant disadvantages as the reconstructed faces are stated that ‘‘facial approximation’’ should be considered highly
biased by the template or generated faces, because the template inaccurate and unreliable, as only one facial reconstruction showed
used in a facial reconstruction is usually derived from a limited true positive identification at above chance rate, and 403 incorrect
database; therefore, the facial appearance of the reconstructed face identifications were made in 592 identification scenarios. On the
will always ultimately resemble the facial template (2,20). In an contrary, Wilkinson and Whittaker (17) demonstrated more optimis-
attempt to reduce this limitation, researchers have collected large- tic results for facial morphology prediction. Employing five juve-
scale data with different ages, sexes, and body builds (32,33) or nile female forensic cases (from the West serial murder
developed a statistical average model acquired from the analysis of investigation in Gloucester, U.K.) facial reconstructions were pro-
a whole database (27). Another shortcoming is the paucity of duced manually and then were compared with a face pool by 50
published accuracy and reliability studies for the computerized volunteers. The participants were asked to select a face from the
systems. face pool that most resembled each reconstruction. The results
Another approach to the computer-generated FFR has been showed that the mean hit rate was 44%, and all hit rates were
developed by Wilkinson (34) who introduced a ‘‘virtual sculpture’’ above chance (10%). The authors concluded that facial reconstruc-
method utilizing a 3D modeling system (FreeForm Modelling tion can produce a good resemblance to an individual.
Plus; Sensable Technologies, Wilmington, MA) with haptic feed- More recently, Quatrehomme et al. (18) performed an accuracy
back (Phantom Desktop Haptic Device; Sensable Technologies). study using 25 3D FFRs produced from a manual method employ-
This method attempts to mimic the 3D manual methods enabling ing skulls donated by the decedents. The subjects were divided into
practitioner-led facial reconstruction. The skull is imported into three groups, and then, each group was reconstructed by either of
FreeForm Plus as a stereolithography (STL) file, converted origi- the two scientists who have a different degree of experiences in the
nally from Digital Imaging and Communication in Medicine (DI- field of FFR. During reconstructing the faces, the scientists were
COM) data, or an object file translated from laser scan data. The given discriminative information on the forensic analysis of the
operator is therefore able to feel the surface in detail during the skulls. The results suggested that resemblance rates were improved
process of facial reconstruction. in proportion to the experience of practitioners and the amount of
The first recorded accuracy study related to facial morphology available information from the analysis of the skulls.
prediction was produced by Von Eggeling (35) who reported little Facial morphology prediction studies for computer-generated
success. He employed a death mask and facial soft tissue measure- facial reconstructions have been performed recently by researchers
ments recorded from a cadaver. Two plaster casts of the skull of and presented promising results. Wilkinson et al. (39) produced
the cadaver were provided along with the tissue depth measure- FFRs from two CT scanned skulls by utilizing a 3D computerized
ments, and two sculptors reproduced the face independently. Von modeling system. Two face pools composed of the five facial CT
Eggeling concluded that there was no resemblance between either scanned images including the target individual were prepared for a
of the two reconstructed faces or the death mask. Stadtmller (36) matching task. The assessors were asked to choose the face from
reconstructed two faces from the skulls of an elderly man and a the face pool that most resembled each reconstruction. The results
young man, using the Kollmann and Bchly (37) facial soft tissue demonstrated that the combined hit rate was 50% above the level
data, and then reported that no similarities were found between the recorded by chance (20%). The researchers also reported another
reconstructed faces and the corresponding facial photographs from pioneering study in the same article into a quantitative comparison
the corpses. Suk (38) concluded that facial reconstruction from the of the facial morphology between the facial reconstruction and the
skull must resort to fantasy. facial scan of the identified individual using reverse modeling soft-
In contrast, Gerasimov (15) demonstrated high accuracy for ware. The results showed that the computer-generated facial recon-
facial morphology prediction. He performed a series of experiments structions also have good levels of quantitative accuracy in relation
using 12 skulls from cadavers and asserted that all 12 reconstruc- to facial morphology prediction.
tions were recognizable with strong similarity when compared to Claes et al. (27) attempted to evaluate the accuracy and identifi-
photographs of the deceased. He also reported that all of the 140 cation success rate from facial reconstructions issuing from auto-
facial reconstructions attempted in his laboratory were successfully mated computerized systems employing combined statistical
identified. Snow et al. (14) carried out an accuracy study using models. Results showed an average of 1.14-mm deviation between
antemortem photographs of the subjects instead of death masks or the scanned target and the computer-generated face, and the identi-
photographs of cadavers. They produced a male and female facial fication rate was 100% (based on Euclidean distance matrix
320 JOURNAL OF FORENSIC SCIENCES

signatures of the facial surfaces in the database of 118 individuals). with a neutral and relaxed facial expression (Figs 1 and 3Ac–Cc).
Identification success rate was demonstrated to be 81% when Immediately after the completion of CT scanning, the faces of each
employing a face pool comparison with 28 participants. subject were photographed from frontal, three-quarter, and profile
Although some researchers have maintained that the prediction views (Fig. 3Aa–Ca). To differentiate two layers of soft and hard
ability of FFR is unreliable, much of the previous research has tissues, two scales of Hounsfield units (HUs) were applied during
shown an acceptable prediction level for recognition or identifica- head scanning: one for a hard tissue image with +150 to +200
tion. Also a number of forensic cases have shown that the technique HUs and another for a soft tissue image with )500 to )550 HUs.
of FFR can be used to assist in the identification of individuals from Hence, all sliced images created from the CT scanning were thres-
unknown skulls (3,5,12–14,40–44). According to these studies and holded using HU limits to be segmented into soft or hard tissue uti-
forensic case reports, it could be suggested that FFR can be lizing an available visualization computer software. These images
employed as an effective forensic tool especially when other means were stored as a format of DICOM files and then were transmitted
of identification have failed for the identification of human remains. electronically to the CAHID at the University of Dundee with the
Average facial tissue depth data are a major contributor to the facial photographs of the subjects. The DICOM data of the heads
accuracy and reliability of FFR. Until present, lateral cephalometric were converted to STL image files using a 3D visualization com-
radiographs (45,46), CT (20,25,47), magnetic resonance imaging puter program (Amira version 5.2.2; Visage Imaging, San Diego,
(48,49), and ultrasound (32,50–53) have been employed to measure CA) by an independent CAHID staff member. A practitioner
facial tissue depths with more accurate noninvasive analysis. Utiliz- (W-JL), with one and half years training in FFR, produced the
ing these techniques, a large amount of facial tissue depth data has facial reconstructions. The 3D facial scan images and the photo-
been collected from various ethnic groups relating to sex, age, and graphs of the subjects were not exposed to the practitioner until the
body mass index and has been applied practically to FFR. Owing facial reconstructions had been completed.
to further progression of 3D medical diagnosis technology, new
equipment has been modified for the purpose of collecting more Preparation of the Skull Models
accurate tissue depths. Of those, recently developed cone-beam CT
(CBCT) scanner has been introduced to the study of tissue depth A 3D modeling system (FreeForm Modelling Plus software)
measurements. The CBCT enables obtaining the head image of soft was utilized for the facial reconstruction process. The system
and hard tissue from the subject in an upright position with less radi- involves Phantom Desktop Haptic Device enabling the use of
ation doses than the typical multislice CT system (54). The 3D the sense of touch to handle and feel digitalized 3D data
image of a face taken from the CBCT may also allow a comparison developed by Wilkinson (34). The three skull models (skulls A,
between the face of a subject and the facial reconstruction without B, and C) were imported into FreeForm Modelling Plus as STL
the soft tissue distortion caused by gravity and body position. files (Figs 1 and 2A). As the biological information relating to the
This study aims to assess the accuracy of facial reconstructions three live subjects was unknown to the practitioner (except ances-
generated from a computerized 3D modeling system, to explore the try group—all were northeastern Asian), anthropological assess-
availability of comparison between the facial reconstruction and the ments for the skulls were initially carried out regarding sex and
corresponding face of a live subject scanned from CBCT, and to age. Skull A (Fig. 1A) was determined as a man and aged
investigate the validity of facial guidelines for the FFR of Korean 20–30 years; skull B (Fig. 1B) was determined as female and
adults. aged 20–30 years; skull C (Fig. 1C) was determined as a male
and aged 20–30 years. The actual ages of the subjects were pro-
vided by the CNU after completion of the facial reconstructions
Materials and Methods as 30.4 years for subject A, 27.3 years for subject B, and
For the purpose of a single-blind accuracy test for FFRs, this 27.7 years for subject C.
research was carried out by two study groups in different countries,
United Kingdom and Republic of Korea. The Department of Ortho- Facial Reconstruction
dontics at the School of Dentistry of the Chonnam National Uni-
The faces of the subjects were reconstructed according to the
versity (CNU) in Korea undertook the recruitment of participants
combination method (5,13) (Fig. 2). Facial soft tissue depths data
and the collection of facial scan data from live subjects. The facial
for living Korean adults was utilized (Lebedinskaya et al. [50],
reconstructions and comparisons between the reconstructed faces
who employed Russian–Korean residents in a province of central
and the scanned facial surfaces were performed by a practitioner
Russia). The tissue depth pegs were placed onto the surface of the
(W-JL) in the Centre for Anatomy and Human Identification
skull at the corresponding anatomical sites using modeling clay and
(CAHID) at the University of Dundee in the United Kingdom.
scale tools to adjust the exact lengths of each peg in the FreeForm
software (Fig. 2B). Each individual facial muscle was rebuilt as
Acquisition of Facial Scan Data accurately as possible following anatomical guidelines. A data bank
of premodeled facial muscles containing 15 major facial muscles
Three volunteers (subjects A, B, and C) were recruited from stu-
and the parotid glands was utilized in the anatomical stage. Each
dents at the CNU in Gwangju, Korea. All volunteers had neither
muscle was imported and positioned onto the skull according to the
any experiences of orthodontic treatment or facial plastic surgery
analysis of those origins and insertions. The shape and size of the
nor facial deformities. Informed consents were obtained from all
muscles were altered utilizing 3D deformation tools to customize
subjects. This study was approved by the Institutional Review
the muscle to the target skull (Fig. 2C). A number of guidelines
Board for the Medical Science at the CNU Hospital, Gwangju,
were employed to predict facial components: eyes, nose, mouth,
Korea. Three-dimensional images for the skulls and head surfaces
and ears including:
were obtained using a CBCT scanner (Alphard Vega; Asahi
Roentgen Co., Kyoto, Japan) with a voxel size of 0.39 mm and • 25-mm-diameter eyeballs (55) were placed in the orbits and
field of view (FOV) of 200 · 179 mm. The subjects were scanned positioned so that the eyeball and pupil were centrally located
to acquire 3D skull and facial images in the seated upright position within the orbits (4,56).
 LEE ET AL. • AN ACCURACY ASSESSMENT OF FORENSIC COMPUTERIZED FACIAL RECONSTRUCTION 321

FIG. 1—The skull models from the scans of subjects A, B, and C.

FIG. 2—3D computerized facial reconstruction procedure following the combination method.

• The eyeballs were positioned in the orbit so that a tangent taken facial anatomy and musculature by utilizing transparency tools in
from the superior to the inferior mid-orbital margins touched the FreeForm software. Finally, the FFRs were completed and
the iris (57). displayed (Figs 2E and 3Ab–Cb).
• The inner canthus was placed 2 mm lateral to the lacrimal crest
at its middle, and outer canthus can be placed 3–4 mm medial
to the malar tubercle (58). Where the malar tubercle was absent, Comparison of Facial Reconstruction and Target Face
the outer canthus was positioned 10 mm below the line of the
The accuracy of the reconstructed faces was assessed using 3D
zygomatico-frontal suture and 5–7 mm from the orbital margin
morphometric surface comparison between each reconstruction and
(58).
the relevant subject face. First, the FFR and corresponding CT
• The maximum width of the soft nose was estimated by that the
scanned face of the subject were aligned manually in FreeForm
bony nasal aperture at its widest point as three-fifths of the
using the embedded skull in the reconstructed face and the CT
overall width of the soft nose (59), which has been confirmed
scanned head for alignment (Fig. 3Ad–Cd) so that the two skull
by a CT study of living subjects regardless of ethnic group (60).
models were positioned identically in dimension and orientation.
The end of the soft nose was predicted as the point where a line
Therefore, the most reliable 3D discrepancy between the facial
following the projection of the portion of the nasal bones
reconstruction and the subject could be obtained. As there was a rel-
crosses a line following the direction of the nasal spine (59).
atively large defect area on the occipital region of the subject scan,
The shape and size of the alae and profile of the nose were
because of the limitation of FOV, the back of the head including
determined by the nasal aperture in profile. These prediction
ears and below of jaw line of both the facial reconstruction and the
methods for the nose have also been confirmed using a blind
scanned head were removed, so that only the facial region was com-
study by Rynn and Wilkinson (61).
pared (Fig. 3Ac–Cc). Second, the comparison model was imported
• The corners of the mouth were estimated by the maxillary
as STL files to the Geomagic Qualify software (Geomagic Qual-
canine and first premolar teeth (4,6), and sets of regression
ify Version 10; Geomagic, Morrisville, NC) to quantitatively assess
formulae derived from the positive correlation between lip
surface morphology discrepancy between the facial reconstruction
thicknesses and the enamel heights of upper and lower incisors
and the subject. This reverse modeling software provides nine differ-
were utilized for the estimation of lip thickness (62).
ent 3D work activities and together allows high-quality polygon
• The broad length of the ears was predicted by the length of the
meshes, accurate freeform Non-Uniform Rationale B-Spline surfaces
nose (63), and the ear canal was positioned using the external
and geometrically perfect solid models to be created. Geomagic
auditory meatus (4). The angle of ear was set as parallel to the
Qualify generates data as absolute mean shell deviations, SD of the
jaw line, and the earlobe adherence was predicted using the
errors during shell overlaps, maximum and minimum range maps,
direction of the mastoid processes (59).
histogram plots, and finally color maps. Within the software, a shell-
In this way, each facial component was rebuilt onto the skull to-shell deviation map may be computed and automatically
(Fig. 2D). For the final stage of facial reconstruction, a skin layer produced. The results include the maximum and minimum range of
was added over the muscle and skull structure referring to the shell deviations, the average distance between the two shells, and
322 JOURNAL OF FORENSIC SCIENCES

FIG. 3—From the top row: facial photographs of subjects A, B, and C (Aa, Ba, and Ca), reconstructed faces (Ab, Bb, and Cb), scanned facial surfaces
(Ac, Bc, and Cc), alignments each of the facial reconstruction, and corresponding scanned face (Ad, Bd, and Cd; gold-colored for the scanned faces, orange-
colored for the reconstructions).

the SD. This function was used to statistically analyze the differ- the errors: ‘‘green’’ representing the deviation of within €1.0 mm;
ences between each facial reconstruction and the facial scan of its ‘‘yellow to red’’ representing from above +1.0 to +10 mm; and
target. ‘‘darkening blue’’ representing from below )1.0 mm to )10 mm.
The ‘‘+’’ (the areas of the yellow and red) implies that the skin sur-
face of the reconstruction is more prominent than the subject face,
Results
and the ‘‘)’’ (the areas of the bluish color) implies that the skin
Shell-to-shell deviation maps for three comparisons between each surface of the reconstruction is less prominent than the subject face.
reconstructed face and the subject face were created (Figs 4–6) and The deviation map for subject A (Fig. 4) showed that 45% of the
the percentage distributions for the deviations are presented in overall surface of the reconstructed face deviated within an error
Tables 1 and 2. The discrepancies between the two shells (the €2.0 mm in the alignment with the scanned face (Table 1), and the
errors) were computed as the minimum limit of deviation error percentage was increased to 54% of the overall surface of the recon-
defined within either €2 or €2.5 mm. The Geometric deviation structed face when the error deviation was broadened to within
maps (Figs 4–6) and tables for percentage distributions of devia- €2.5 mm (Table 2). Eighty-eight percent of the facial surface was
tions (Tables 1 and 2) were created applying €2.5 mm of the mini- within an error €5.0 mm (Tables 1 and 2). The averages of the error
mum limit of deviation. In the figures, the colors on the spectrum and SD were )0.46 and 2.81 mm, respectively. The most accurate
bars and the faces of the reconstructions indicate the distribution of areas (errors between €1.0 mm; green-colored areas, occupied by
 LEE ET AL. • AN ACCURACY ASSESSMENT OF FORENSIC COMPUTERIZED FACIAL RECONSTRUCTION 323

FIG. 4—Deviation map for subject A created from morphometric compar-

ison between the facial reconstruction and the scanned face.

26% of the overall surface of the facial reconstruction) were found FIG. 5—Deviation map for subject B created from morphometric compar-
at the frontal forehead, parts of lateral forehead, eyes, both infra- ison between the facial reconstruction and the scanned face.
orbits, nasal bridge, lateral nose, parts of the medial cheeks, phil-
trum, some parts of the lips, and majority of the chin. Both lateral
foreheads, parts of the nasal bridge, alae, both sides of the nose, lat-
eral and inferior portions of the nose and partial upper and lower
lips, and a minor part of the chin were between +1.0 and +2.5 mm
(yellow-colored areas) more prominent than the subject face. The
areas around the orbits, middle of the lower forehead, both partial
medial cheeks, both sides of the lips, and lateral portions of the chin
were between )1.0 and )2.5 mm (light blue-colored areas) less
prominent than the subject face. The largest areas of error (‡+4 and
£)4 mm) occurred at the minor parts of the both lateral foreheads
and a small part of the nose (orange-colored areas; more prominent
than the subject face) and at the majority of both cheeks (dark blue-
colored areas; less prominent than the subject face).
The deviation map for subject B (Fig. 5) presented that 50% of
the reconstructed facial surface aligned with the scanned face
within an error €2.0 mm (Table 1), and the percentage was
increased in 65% when the error deviation was broadened to within
€2.5 mm (Table 2). Ninety-seven percent of the whole surface of
the facial reconstruction was within an error €5.0 mm. The aver-
ages of the error deviation and SD were )0.31 and 2.40 mm,
respectively. The most accurate areas (errors between €1 mm;
green-colored areas, occupied by 25% of the overall surface of the
facial reconstruction) were at the lower forehead, eyes, eyelids,
parts of the supra-orbits, infra-orbits, majority of the dorsal nose, FIG. 6—Deviation map for subject C created from morphometric compar-
both lateral portions of the nose, philtrum, mouth corners, and lat- ison between the facial reconstruction and the scanned face.
eral chin. The majority of the frontal and lateral foreheads, parts of
the nasal bridge, tip of the nose, alae, lips, and medial chin were
between +1.0 and +2.5 mm (yellow-colored areas) more prominent The deviation map for subject C (Fig. 6) revealed that 66% of
than the subject face. Some parts of the orbits, medial and lower the overall surface of the reconstructed face was aligned to the tar-
cheeks, and both temples and lower chin were between )1.0 and get face within an error €2.0 mm (Table 1), and the proportion of
)2.5 mm (light blue-colored areas) less prominent than the subject the surface was increased in 77% when the error deviation was
face. The largest areas of error (‡+4 and £)4 mm) were found at extended to within €2.5 mm (Table 2). Ninety-five percent of the
the minor portions of the lateral foreheads (orange-colored areas; overall surface of the facial reconstruction deviated within an error
more prominent than the subject face) and at the both endocanthi €5.0 mm in the alignment with the scanned face. The averages of
and small parts of the both lateral and lower cheeks (dark blue-col- the error deviation and SD were )0.49 and 2.16 mm, respectively.
ored areas; less prominent than the subject face). The most accurate areas (errors between €1 mm; green-colored
324 JOURNAL OF FORENSIC SCIENCES

TABLE 1—Distribution (%) of the deviation error between the surfaces of the reconstruction and the subject within each defined error range (minimum
range within €2 mm)

Deviation (X: mm, minimum range within €2 mm)

)10.0 £ X < )5.0 )5.0 £ X < )2.0 )2.0 £ X £ 2.0 2.0 < X £ 5.0 5.0 < X £ 10.0 Total (%)
Subject A 10.12 33.25 44.63 10.39 1.61 100
88.27
Subject B 2.24 32.63 50.43 14.12 0.58 100
97.18
Subject C 4.59 24.93 66.18 4.21 0.09 100
95.32

TABLE 2—Distribution (%) of the deviation error between the surfaces of the reconstruction and the subject within each defined range (minimum range
within €2 mm).

Deviation (X: mm, minimum range within €2.5 mm)

)10.0 £ X < )5.0 )5.0 £ X < )2.5 )2.5 £ X £ 2.5 2.5 < X £ 5.0 5.0 < X £ 10.0 Total (%)
Subject A 10.12 26.53 54.30 7.44 1.61 100
88.27
Subject B 2.04 25.58 64.55 7.25 0.58 100
97.18
Subject C 4.59 16.41 76.74 2.17 0.09 100
95.32

areas, occupied by 35.28% of the overall surface of the facial software enabled the comparison of a facial reconstruction with
reconstruction) were found at the lower and lateral foreheads, both the 3D model of the corresponding subject face or another
lateral orbits, infra-orbits, lateral and inferior nose, nasal tip, phil- facial reconstruction. Two 3D facial models can be aligned so
trum, upper cheeks, around the mouth, and lateral chin. The frontal that the differences in surface contours between two models can
forehead, nasal bridge, alae, lower tip of the nose, lips, and medial be computed numerically. Many current FFR studies have
and lateral chins were between +1.0 and +2.5 mm (yellow-colored acquired facial soft- and hardtissue data from CT scans, but
areas) more prominent than the subject face. The eyes, orbits, only a few studies have used this approach to assess the accu-
medial and lower cheeks, and parts of the temples were between racy of the facial reconstruction.
)1.0 and )2.5 mm (light blue-colored areas) less prominent than In a study by Wilkinson et al. (39), CT scanned skull models
the subject face. The largest areas of error (‡+4 and £)4 mm) from two live individuals (white North American male and female)
occurred at the both endocanthi, left side of the upper eyelid, and were used for facial reconstructions. The researchers employed a
partial temples (dark blue-colored areas; less prominent than the 3D computer modeling system (the same one utilized in this study)
subject face). There was no area of the reconstructed face more and facial soft tissue depth data for white North American male
prominent than the subject face with the error deviation above and female (51) to reconstruct the faces. The accuracy of the recon-
+4.0 mm. structions was assessed quantitatively using reverse modeling soft-
ware (Rapidform; Rapidform, Seoul, Korea). The results for the
white man demonstrated that 60% of the reconstructed face devi-
Discussion
ated with the facial scan no more than €2.5 mm and that the most
There are a number of methods in use to evaluate the accuracy accurate areas of the reconstruction were at the nose, chin, mouth,
of FFRs, yet the methods might be divided into two main groups: eyes, and left forehead. The right temple, upper cheek, ears, the
qualitative and quantitative assessments. In the early period of the nasal tip, and parts of the neck were more than 5 mm less or more
accuracy studies, death masks or photographs of the deceased were prominent than the facial scan. The results for the white woman
compared with the reconstructed faces to determine the level of demonstrated that 52% of the reconstructed face deviated with the
resemblance (35,36,64). These attempts can be considered as primi- facial scan no more than €2.5 mm and that the most accurate areas
tive qualitative studies for assessing the accuracy of facial recon- of the reconstruction were at the nose, chin, upper mouth, upper
structions. Demands for more realistic in vivo experimentation lead cheeks, and cranium. The nasal alae, lower cheeks, upper lip, and
to the one-to-one comparisons of the facial reconstruction with a ears were between 3 and 8 mm or more than 8 mm more promi-
photograph of the target individual during life (5,15,41,42). For nent than the facial scan.
more accurate qualitative evaluation, some researchers tested the The results of this study are comparable to those of Wilkinson
accuracy using face pool matching (14,17,65,66). In the face pool et al. (39) whose research demonstrated that 60% for the male face
test, the assessors were asked to match an image of the facial and 52% for the female face were in the deviation within
reconstruction to a face pool of images including the antemortem €2.5 mm, and 90% for the male and 75% of the female were in
target face. the deviation between €5.0 mm. Considering the overall deviation
The one-to-one comparison and face pool matching are very useful errors between the reconstructions and the scanned faces from this
methods for the assessment of the accuracy of facial reconstruc- study with the results from the previous research by Wilkinson
tion, but further objective methods were required to evaluate the et al. (39), we conclude that the 3D computerized facial reconstruc-
accuracy quantitatively. Recently developed 3D anthropometrical tion system is capable of reliable facial prediction. However, it
 LEE ET AL. • AN ACCURACY ASSESSMENT OF FORENSIC COMPUTERIZED FACIAL RECONSTRUCTION 325

must be acknowledged that this study is small, and even when tissue components, and a number of guidelines have been intro-
combined with the previous study (39), the total number of subjects duced. However, as there is a paucity of research into determining
(n = 5) is not sufficient to reach any firm conclusions regarding facial features of Koreans or other northeastern Asians, this
reliability. research applied guidelines derived from white European or black
The average errors and SD between the surfaces of the recon- African populations. Figures 3–6 show that all the facial features
structions and subjects were )0.46 and 2.81 mm for subject A, demonstrated very low error deviations and a good likeness to the
)0.31 and 2.40 mm for subject B, and )0.49 and 2.16 mm for sub- subject except the mouth position for subject A. From these results,
ject C. Claes et al. (27) carried out a series of accuracy tests for it can be concluded that the preexisting guidelines are applicable to
the 118 facial reconstructions generated from an automated com- reconstruct the face of northeastern Asians, but also further studies
puterized FFR system using a flexible statistical model. The are demanded to produce more accurate prediction methods from
researcher reported a result that the overall average absolute recon- diverse ethnic groups.
struction error and SD between the reconstructed skin surface and Currently CBCT scanning has been used in the field of orthodon-
the real test skin surface are 1.14 and 1.04 mm, respectively. This tics and maxillofacial surgery (72–74). The CBCT enables to collect
study showed lower errors than the result from the study of Claes image data from the subjects in an upright position rather than
et al. (27) and may suggest that the 3D computerized modeling supine position, and it is reported that radiation dose in CBCT scan
method for FFR enables to reconstruct the face more accurately is much less than multislice CT (75–77). These advantages over the
than the automated system. Moreover, contrasting with the recently conventional spiral CT and other diagnostic imaging devices can be
updated data in the research of Claes et al. (27), the average facial modified for the purpose of measuring facial skin thicknesses and
soft tissue depths for adult Koreans used in this project (50) was scanning facial surface. The results from this study present good
obtained from the Korean population group who had emigrated examples to demonstrate the usefulness of the CBCT in accuracy
from Korea to central Russia 80 years ago and have resided in the tests of the facial reconstruction. In addition, the CBCT has a great
region with little interchange of people between the group in Russia potentiality in tissue depth measurements as once the scanned
and the majority in Korea. Even though the population has main- images are obtained, the measurements can be repeated, and an
tained a homogeneous ethnicity after the emigration, different cli- unlimited number of measurements can be added according to the
mates, diet, and lifestyle might affect facial tissues. Indeed, the research requirements. Because of the advances in the technology of
cheeks of all the target faces were more prominent than the recon- diagnostic imaging software integrating with CBCT, scanned images
structions (Figs 3Ad–Cd and 4–6). The results may imply that the can contribute to allow displaying soft and hard tissue images simul-
tissue depth data from the study of Lebedinskaya et al. (50) are not taneously on a computer screen. Some researchers have started to
wholly appropriate for current Korean adults and that further use the CBCT in FFR (55), and it is expected that further research
research to update the data using contemporary population group is employing the CBCT will be carried out in the domain of FFR.
required. The under- or overestimation of the tissue depths may The 3D computer modeling system (FreeForm Modelling Plus)
also be a reflection of variation related to occlusion and facial type with haptic feedback (Phantom Desktop) has involved actively
(51) and highlights one of the problems associated with a small not only in computer-generated FFRs but also in forensic investiga-
sample size of three subjects. Despite the differences at the cheeks, tions to simulate an incidence with engaging our tactile senses to
the errors recorded for the reconstructions were similar to those shape and manipulate digitized 3D models in a noninvasive manner
recorded in previous research (39). (78). Some of the strengths of the system for the FFR are that
There are other considerations when comparing this research to muscles, layer by layer, and skin can be visualized as separate
the previous study. Wilkinson et al. (39) utilized a conventional units, with the ‘‘transparency’’ tool—a process not available in the
spiral CT scanner in which the subject is required to be in supine manual method, and the simultaneous visibility of multilayers
position. The position may cause gravity effects on the cheek and might allow alignment of two reference objects (the skull images
mouth areas of the face to sag downward, whereas they employ in this study) to assess the accuracy. This method is also reproduc-
the facial tissue depth data taken from upright subjects (52). This ible, quick, and provides little or no damage to the original speci-
positional difference may explain the fuller cheeks, more prominent men. The additions of skin texture, eyelid position, and hairstyle,
upper lips and ear pattern errors seen for the reconstructions when as well as altering degrees of facial tissue depth are quicker and
compared to the facial scans. The current study employed CBCT easier to integrate with animation or other computerized programs.
to acquire the soft and hard tissue scan data for the subjects, which Because of the merits, demands for computerized modeling have
allows the face to be in upright position eliminating the possible increased, especially in historical or archaeological facial recon-
distortions of the facial soft tissue caused by the gravity with lower structions where realistic facial depictions are required (3,79,80).
radiation doses (55). Consequently, this computer facial reconstruction system can be
However, there is a limit to the FOV in CBCT (the FOV was applied widely and reliably to forensic identification investigation.
200 · 179 mm in this study) resulting in an inability to scan the
whole head and an unnecessary artifact on the back of the scanned
Conclusions
head. Nevertheless, it is thought that the excluded portions would
not influence the reconstruction errors because the adult tissue The results from this study demonstrate that this computerized
depths at these areas show a constant thickness regardless of age, 3D modeling method is capable of producing reliable facial recon-
sex, and ancestry (67–71) and are relatively unproblematic to structions with acceptable levels of resemblance employing the
rebuild. Therefore, it is possible that the reconstruction error would combination method and the images scanned from CBCT. This
decrease rather than to increase if the rest of the head were study also suggests that previously published guidelines for the
included and compared. prediction of facial features are applicable to the reconstruction of
The prediction of the location, size, and morphology of facial adult Korean faces. However, further research is recommended to
features—eyes, nose, mouth, and ears—is critical to the level of update Korean tissue depth data and to examine feature prediction
the accuracy of FFR. The greater part of the feature prediction is standards in relation to northeastern Asians in order to increase the
based on analysis of the relation between skull structures and soft accuracy of craniofacial reconstruction in this demographic.
326 JOURNAL OF FORENSIC SCIENCES

References 29. Turner WD, Brown REB, Kelliher TP, Tu PH, Taister MA, Miller
KWP. A novel method of automated skull registration for forensic facial
1. Stephan CN. Anthropological facial ‘reconstruction’—recognizing the approximation. Forensic Sci Int 2005;154(2–3):149–58.
fallacies, ‘unembracing’ the errors, and realizing method limits. Sci Jus- 30. Pei Y, Zha H, Yuan Z. The craniofacial reconstruction from the local
tice 2003;43(4):193–200. structural diversity of skulls. Comput Graph Forum 2008;27(7):1711–8.
2. Wilkinson CM. Computerized forensic facial reconstruction: a review of 31. Berar M, Desvignes M, Bailly G, Payan Y. 3D semi-landmarks-based
current systems. Forensic Sci Med Pathol 2005;1(3):173–7. statistical face reconstruction. J Comp Info Technol 2006;14(1):31–43.
3. Wilkinson CM. Facial reconstruction—anatomical art or artistic anat- 32. De Greef S, Claes P, Mollemans W, Loubele M, Vandermeulen D, Sue-
omy? J Anat 2010;216(2):235–50. tens P, et al. Semi-automated ultrasound facial soft tissue depth registra-
4. Krogman WM, _Is¸ can MY. The human skeleton in forensic medicine, tion: method and validation. J Forensic Sci 2005;50(6):1282–8.
2nd edn. Springfield, IL: Charles C. Thomas, 1986. 33. De Greef S, Claes P, Vandermeulen D, Mollemans W, Suetens P,
5. Prag J, Neave RAH. Making faces. London, UK: British Museum Press, Willems G. Large-scale in-vivo Caucasian facial soft tissue thickness
1997. database for craniofacial reconstruction. Forensic Sci Int 2006;159(1):
6. George RM. Anatomical and artistic guidelines for forensic facial recon- S126–46.
struction. In: _Is¸ can MY, Helmer RP, editors. Forensic analysis of the 34. Wilkinson CM. Virtual sculpture as a method of computerized facial
skull. New York, NY: Wiley-Liss, 1993;215–27. reconstruction. Proceedings of the 1st International Conference on Recon-
7. Kunstkamera. Face of our ancestors: an exhibition on the occasion of struction of Soft Facial Parts; 2003 March 17–18; Potsdam, Germany.
Mikhail Gerasimov’s centenary, 1998, http://www.kunstkamera.ru/en/ Remagen, Germany: Kreartive Konzepte, Beate Surek, 2003;59–63.
temporary_exhibitions/virtual/gerasimov/03/the_predecessors (accessed 35. Von Eggeling H. Die leistungsfahigkeit physiognomischerrekon-struk-
April 12, 2010). tionsversuche auf grundlagc des schdels. Arch Anthropol 1913;12:44–7.
8. Wilkinson CM, Neave RAH. The reconstruction of a face showing a 36. Stadtmller F. Zur beurteilung der plastischen rekonstruktionsmethode
healed wound. J Arch Sci 2003;30(10):1343–8. der physiognomie auf dem schdel. J Morphol Anthropol 1922;22:337–
9. Ubelaker DH, O’ Donnell G. Computer-assisted facial reproduction. 72.
J Forensic Sci 1992;37(1):155–62. 37. Kollmann J, Bchly W. Die persistenz der rassen und die reconstruction
10. Wilkinson CM. Facial anthropology and reconstruction. In: Thompson der physiognomie prahistorischer schdels. Arch Anthropol 1898;25:
TJU, Black SM, editors. Forensic human identification: an introduction. 329–59.
Boca Raton, FL: CRC Press Inc., 2007;231–56. 38. Suk V. Fallacies of anthropological identifications. Publications de la
11. Stephan CN, Henneberg M. Building faces from dry skulls: are they rec- Facultae des sciences de l’Universitae Masaryk 1935;207:3–18.
ognized above chance rates? J Forensic Sci 2001;46(3):432–40. 39. Wilkinson C, Rynn C, Peters H, Taister M, Kau CH, Richmond S.
12. Taylor KT. Forensic art and illustration. Boca Raton, FL: CRC Press, 2001. A blind accuracy assessment of computer-modeled forensic facial recon-
13. Wilkinson CM. Forensic facial reconstruction. Cambridge, UK: Cambridge struction using computed tomography data from live subjects. Forensic
University Press, 2004. Sci Med Pathol 2006;2(3):179–88.
14. Snow CC, Gatliff BP, McWilliams KR. Reconstruction of facial features 40. Suzuki T. Reconstitution of a skull. Int Crim Police Rev 1973;264:76–80.
from the skull: an evaluation of its usefulness in forensic anthropology. 41. Gatliff BP, Snow CC. From skull to visage. J Biocommun 1979;6(2):
Am J Phys Anthropol 1970;33(2):221–8. 27–30.
15. Gerasimov MM. The face finder. New York, NY: Lippincott, 1971. 42. Haglund WD, Reay DT. Use of facial approximation techniques in iden-
16. Helmer RP, Rçhricht S, Petersen D, Mçhr F. Assessment of the reliabil- tification of Green River Serial Murder victims. Am J Forensic Med
ity of facial reconstruction. In: _Is¸ can MY, Helmer RP, editors. Forensic Pathol 1991;12(2):132–42.
analysis of the skull: craniofacial analysis, reconstruction, and identifica- 43. Vanezis P, Vanezis M, McCombe G., Niblett T. Facial reconstruction
tion. New York, NY: Wiley Liss Publishers, 1993;75–83. using 3-D computer graphics. Forensic Sci Int 2000;108(2):81–95.
17. Wilkinson CM, Whittaker DK. Juvenile forensic facial reconstruction: a 44. Wilkinson C, Neave RAH. Skull re-assembly and the implications for
detailed accuracy study. Proceedings of the 10th Conference of the forensic facial reconstruction. Sci Justice 2001;41(3):233–4.
International Association of Craniofacial Identification; 2002 Sept 11– 45. George RM. The lateral craniographic method of facial reconstruction.
14. Bari, Italy: University of Bari Press, 2002;11–4. J Forensic Sci 1987;32(5):1305–30.
18. Quatrehomme G, Balaguer T, Staccini P, Alunni-Perret V. Assessment 46. Utsuno H, Kageyama T, Deguchi T, Umemura Y, Yoshino M, Nakam-
of the accuracy of three-dimensional manual craniofacial reconstruction: ura H, et al. Facial soft tissue thickness in skeletal type I Japanese chil-
a series of 25 controlled cases. Int J Legal Med 2007;121(6):469–75. dren. Forensic Sci Int 2007;172(2–3):137–43.
19. Quatrehomme G, Cotin S, Subsol G, Delingette H, Garidel Y, Grvin G, 47. Kim KD, Ruprecht A, Wang G, Lee JB, Dawson DV, Vannier MW.
et al. A fully three-dimensional method for facial reconstruction based Accuracy of facial soft tissue thickness measurements in personal com-
on deformable models. J Forensic Sci 1997;42(4):649–52. puter-based multiplanar reconstructed computed tomographic images.
20. Vandermeulen D, Claes P, Loeckx D, De Greef S, Willems G, Suetens Forensic Sci Int 2005;155(1):28–34.
P. Computerized craniofacial reconstruction using CT-derived implicit 48. Sahni D, Jit I, Gupta M, Singh P, Suri S, Sanjeev S. Preliminary study
surface representations. Forensic Sci Int 2006;159(1):164–74. on facial soft tissue thickness by magnetic resonance imaging in north-
21. Vanezis P, Blowes RW, Linney AD, Tan AC, Richards R, Neave R. west Indians. Forensic Sci Commun 2002;4:1–7.
Application of 3-D computer graphics for facial reconstruction and com- 49. Sahni D, Sanjeev S, Singh G, Jit I, Singh P. Facial soft tissue thickness
parison with sculpting techniques. Forensic Sci Int 1989;42(1–2):69–84. in northwest Indian adults. Forensic Sci Int 2008;176(2–3):137–46.
22. Evenhouse R, Rasmussen M, Sadler L. Computer-aided forensic facial 50. Lebedinskaya GV, Balueva TS, Veselovaskaya VS. Principles of facial
reconstruction. J Biocommun 1992;19(2):22–8. reconstruction. In: _Is¸ can MY, Helmer RP, editors. Forensic analysis of
23. Moss JP, Linney AD, Grindrod SR, Arridge SR, Clifton JS. Three- the skull: craniofacial analysis, reconstruction and identification. New
dimensional visualization of the face and skull using computerized York, NY: Wiley, 1993;183–98.
tomography and laser scanning techniques. Eur J Orthod 1987;9(4):247– 51. Utsuno H, Kageyama T, Uchida K, Yoshino M, Miyazawa M, Inoue K.
53. Facial soft tissue thickness in Japanese children. Forensic Sci Int
24. Moss JP, Linney AD, Grindrod SR, Mosse CA. A laser scanning system 2010;199(1):109e1–e6.
for the measurement of facial surface morphology. Opt Lasers Eng 52. Manhein MH, Listi GA, Barsley RE, Musselman R, Barrow NE, Ubela-
1989;10(3–4):179–90. ker DH. In vivo facial tissue depth measurements for children and
25. Tilotta F, Richard F, Glauns J, Berar M, Gey S, Verdeille S, et al. Con- adults. J Forensic Sci 2000;45(1):48–60.
struction and analysis of a head CT-scan database for craniofacial recon- 53. Wilkinson CM. In vivo facial tissue depth measurements for white Brit-
struction. Forensic Sci Int 2009;191(1–3):112.e1–e12. ish children. J Forensic Sci 2002;47(3):459–65.
26. Khler K, Haber J, Seidel H-P. Reanimating the dead: reconstruction of 54. Smith SL, Throckmorton GS. A new technique for three-dimensional
expressive faces from skull data. ACM Trans Graph 2003;22(3):554–61. ultrasound scanning of facial tissues. J Forensic Sci 2004;49(3):451–7.
27. Claes P, Vandermeulen D, De Greef S, Willems G, Suetens P. Cranio- 55. Fourie Z, Damstra J, Gerrits PO, Ren Y. Accuracy and reliability of
facial reconstruction using a combined statistical model of face shape facial soft tissue depth measurements using cone beam computer tomog-
and soft tissue depths: methodology and validation. Forensic Sci Int raphy. Forensic Sci Int 2010;199(1–3):9–14.
2006;159(Suppl):S147–58. 56. Tian S, Nishida Y, Isberg B, Lennerstrand G. MRI measurements of
28. Nelson LA, Michael SD. The application of volume deformation to normal extraocular muscles and other orbital structures. Graefes Arch
three-dimensional facial reconstruction: a comparison with previous Clin Exp Ophthalmol 2000;238(5):393–404.
techniques. Forensic Sci Int 1998;94(3):167–81.
 LEE ET AL. • AN ACCURACY ASSESSMENT OF FORENSIC COMPUTERIZED FACIAL RECONSTRUCTION 327

57. Gatliff BP. Facial sculpture on the skull for identification. Am J Foren- 72. Kau CH, Richmond S, Palomo JM, Hans MG. Three-dimensional cone
sic Med Pathol 1984;5(4):327–32. beam computerized tomography in orthodontics. J Orthod 2005;32(4):
58. Wilkinson CM, Mautner SA. Measurement of eyeball protrusion and its 282–93.
application in facial reconstruction. J Forensic Sci 2003;48(1):12–6. 73. Hechler SL. Cone-beam CT: applications in orthodontics. Dent Clin
59. Gerasimov MM. Facial reconstruction from the cranium. Translator North Am 2008;52(4):809–23.
Tshernezky W, Moscow, Russia: Izdat. Akademii Nauk SSSR, 1955. 74 Cattaneo PM, Melsen B. The use of cone-beam computed tomography
60. Angel JL. Restoration of head and face for identification. Proceedings of in an orthodontic department in between research and daily clinic.
the 30th Annual Meeting of the American Academy of Forensic Sci- World J Orthod 2008;9(3):269–82.
ences; 1987 Feb 20-25; St. Louis, MO. Colorado Springs, CO: American 75. Mah JK, Danforth RA, Bumann A, Hatcher D. Radiation absorbed in
Academy of Forensic Sciences, 1987. maxillofacial imaging with a new dental computed tomography device.
61. Rynn C, Wilkinson CM. Appraisal of traditional and recently proposed Oral Surg Oral Med Oral Pathol Oral Radiol Endod 2003;96(4):508–13.
relationships between the hard and soft dimensions of the nose in pro- 76. Tsiklakis K, Donta C, Gavala S, Karayianni K, Kamenopoulou V,
file. Am J Phys Anthropol 2006;130(3):364–73. Hourdakis CJ. Dose reduction in maxillofacial imaging using low dose
62. Rynn C, Wilkinson CM, Peters HL. Prediction of nasal morphology cone beam CT. Eur J Radiol 2005;56(3):413–7.
from the skull. Forensic Sci Med Pathol 2009;6(1):1–15. 77. Swennen GR, Schutyser F. Three-dimensional cephalometry: spiral
63. Wilkinson CM, Motwani M, Chiang E. The relationship between the multi-slice vs cone-beam computed tomography. Am J Orthod Dento-
soft tissues and the skeletal detail of the mouth. J Forensic Sci facial Orthop 2006;130(3):410–6.
2003;48(4):728–32. 78. Buck U, Naether S, Braun M, Thali M. Haptics in forensics: the possi-
64. Fedosyutkin BA, Nainys JV. The relationship of skull morphology to bilities and advantages in using the haptic device for reconstruction
facial features. In: _Is¸ can MY, Helmer RP, editors. Forensic analysis of approaches in forensic science. Forensic Sci Int 2008;180(2–3):86–92.
the skull. New York, NY: Wiley-Liss, 1993;199–213. 79. Benazzi S, Fantini M, De Crescenzio F, Mallegni G, Mallegni F, Persi-
65. Wilder HH, Wentwoith B. Personal identification. Boston, MA: Gormon ani F, et al. The face of the poet Dante Alighieri reconstructed by virtual
Press, 1918. modelling and forensic anthropology techniques. J Arch Sci 2009;36(2):
66. Stephan CN, Henneberg M. Recognition by forensic facial approxima- 278–83.
tion: case specific examples and empirical tests. Forensic Sci Int 80. Benazzi S, Stansfield E, Milani C, Gruppioni G. Geometric morphomet-
2005;156(2–3):182–91. ric methods for three-dimensional virtual reconstruction of a fragmented
67. Rhine JS, Campbell HR. Thickness of facial tissues in American blacks. cranium: the case of Angelo Poliziano. Int J Legal Med 2009;123(4):
J Forensic Sci 1980;25(4):847–58. 333–44.
68. Helmer R. Schdelidentifizierung durch elektronische bildmischung.
Heidelberg, Germany: Kriminalistik Verlag GmbH, 1984. Additional information and reprint requests:
69. Phillips VM, Smuts NA. Facial reconstruction: utilization of computer- Caroline M. Wilkinson, Ph.D.
ized tomography to measure facial tissue thickness in a mixed racial Centre for Anatomy and Human Identification
population. Forensic Sci Int 1996;83(1):51–9. College of Life Sciences
70. Stephan CN, Simpson EK. Facial soft tissue depths in craniofacial University of Dundee
identification (part I): an analytical review of the published adult data. MSI ⁄ WTB Complex
J Forensic Sci 2008;53(6):1257–72. Dow Street
71. Tedeschi-Oliveira SV, Melani RFH, de Almeida NH, de Paiva LAS. Dundee DD1 5EH
Facial soft tissue thickness of Brazilian adults. Forensic Sci Int U.K.
2009;193(1–3):127e1–e7. E-mail: [email protected]

View publication stats