Abstracts from the 9th Intntl. Craniofacial ID Conference--Part 4 (Forensic Science Communications, October 2000)

U.S. Department of Justice

Federal Bureau of Investigation

October 2000   Volume 2   Number 4

Presentations at the
9th Biennial Scientific Meeting of the
International Association for
Craniofacial Identification
Washington, DC
July 24–28, 2000

Part 4

The following abstracts of the presentations are ordered alphabetically by authors' last names.

Face Reconstruction Using Flesh Deformation Models

P. Tu, R. Hartley, W. Lorensen, M. Allyassin, R. Gupta, and L. Heier
General Electric Company
Niskayuna, New York

A system for reconstructing a face from its skeletal remains using principle component analysis (PCA) is described. Given a CT scan of a person's head, 3D surfaces of the face and skull can be extracted. Using a large database of CT head scans, a statistical flesh-depth model is developed to define the relationship between the surface of the face and the surface of the skull. The flesh-depth modes are composed of a set of principal components or global deformation models that are associated with soft tissue variation. These models can be tailored to the skull of an unidentified subject to define a range of probable 3D reconstructions. Geometric editing tools are used to change specific features such as the eyes, nose, and lips. Lastly, a texture map is fitted to the 3D model to provide a natural appearance.

The process of generating a flesh-depth model for a particular skull requires several steps. First, a model of the subject's skull is generated using a 3D scanner. Second, an image-like representation of the model is generated on the basis of a canonical cylindrical coordinate system. At this point, 2D operators such as smoothing and morphing can be used. Third, a set of fiducial points on the subject's skull is identified. These points have been previously defined for all the skulls in the CT database. Fourth, each skull face surface pair in the database is morphed so that all the skulls assume the shape of the subject's skull. This process transforms the skin surfaces as well so that each skin point is registered with the skull in a radial sense. However, there is still variation in skull-to-skin (flesh) depths. This is due to differences in fat and muscle composition between individuals in the database. Finally, the covariance matrix associated with these depths is measured and the eigenvectors (which are independent global deformation modes) are extracted. The operator is presented with an initial reconstruction based on average flesh depths. Each deformation mode can be adjusted to provide a continuum of probable reconstructions.

Additional modifications are achieved using a face-editing tool that changes the shape of specific facial features such as the lips, nose, and eyes. For example, the operator can select a specific nose from a parts library and fuse the nose with the current 3D model. Once the 3D model has been completed, a texture map is fitted by specifying certain fiducial points. The texture map is created from a representative population, where averaging is used to suppress any unsupported details. The combination of the 3D geometric model and a texture map results in a lifelike 3D reconstruction.

A History of Smithsonian–FBI Collaboration in
Forensic Anthropology, Especially in Regard to
Facial Imagery

D. H. Ubelaker
National Museum of Natural History, Smithsonian Institution
Washington, DC

Since the 1930s, physical anthropologists at the Smithsonian Institution's National Museum of Natural History have consulted with FBI Headquarters on hundreds of cases. Increasingly, this consultation has involved techniques of facial reproduction and photographic superimposition. This presentation reviews the history of that collaboration, documenting the introduction and evolution of these specialized techniques.

FBI consultation in forensic anthropology at the Smithsonian began with Aleš Hrdli…ka (1869–1943) in the 1930s and has continued primarily through the work of T. D. Stewart (1901–1997), J. Lawrence Angel (1915–1986), and Douglas H. Ubelaker.

Since 1977, Ubelaker has reported on approximately 404 cases for the FBI. Of these, about 3 percent have involved photographic superimposition and 9 percent have involved facial reproduction. The cases involving photographic superimposition date from about 1990, when the necessary equipment became available, to 1996. Cases involving facial reproduction date from throughout this period, but generally show a temporal increase. Artistic techniques have varied with individual preferences but consistently have involved collaboration of the artist and anthropologist.

Preliminary Study on the Measurement of
Facial Tissue Thickness in Japanese Children

H. Utsuno and K. Minaguchi
Tokyo Dental College
Tokyo, Japan

H. Miyazawa
Matsumoto Dental University
Matsumoto, Japan

M. Yoshino
National Research Institute of Police Science
Kashiwa, Japan

The facial reconstruction technique used in forensic anthropology is based on soft tissue thickness measurements. Many data sets on facial tissue thickness that account for the race and ethnicity of adults have been published; however, there are no statistics on facial tissue thickness in children other than American children. For this reason, the authors attempted to compile information on facial tissue thickness in contemporary Japanese children, which is necessary to produce an accurate facial likeness and to evaluate a match of skull photo superimposition image.

Twenty-four X-ray cephalographs of Japanese children (15 male and 9 female) ranging in age from 4 to 13 years were taken at the Department of Pediatric Dentistry at Matsumoto Dental University and used as samples in this study. The film-tube distance was 165 cm. The profiles of the skull and face were traced from the lateral X-ray cephalograph, and then the following measurement points were plotted: glabella, nasion, rhinion, subnasale, labrale superius, stomion, labrale inferius, labiomentale, pogonion, and gnathion. The facial tissue thickness of these points was measured at a right angle to the skull surface. The actual measurement was calculated from the diameter of the ear rod.

Table 1 shows the means, standard deviations, and ranges of facial tissue thickness. The tissue thickness in the upper facial part of Japanese children was thinner than that of American children. However, in the middle and lower facial parts, the facial tissue thickness was thicker in Japanese children than in American children. More samples are being collected to establish reliable data.

Application of the Facial Transformation Program for the
Creation of Interethnic Images for Studies in
Applied Psychology

M. Vanezis and P. Vanezis
Department of Forensic Medicine and Science
University of Glasgow
Glasgow, Scotland

H. Minnis, M. Gillies, and S. Smith
Department of Child and Family Psychiatry
Royal Hospital for Sick Children, Yorkhill
Glasgow, Scotland

The facial transformation program, an integral part of computerized 3D facial reconstruction systems, was used in studies of racial stereotyping to create interethnic images assessing the response of various groups to facial appearance.

A 3D facial image from a young black (Negroid) male volunteer was acquired using an optical laser-scanning system. His face was used as a template over a Caucasian skull to produce one reconstruction—the first using facial criteria applicable to white males.

Caucasian criteria were applied to obtain the resulting face. Appropriate soft tissue thicknesses were used for all landmarks on the face, and the location of the nasal alare and thickness of lips on the skull were adopted for a Caucasian face. This produced a face with some Caucasian characteristics, but the skin tone was still more Negroid than Caucasian. A commercially available electronic identikit system, E-FIT™, was used to reduce this effect. A Caucasian hairstyle was added to the reconstruction. In addition, the contrast and lighting on the face were reduced to give the resulting face a paler appearance. This was a relatively straightforward process involving the use of grayscale images rather than color.

The male black scanned face was treated with E-Fit™, and a black image was thus made available for testing. The resulting images may be applied to psychological-testing studies designed to assess the response of groups of individuals, such as potential employers, doctors, and law enforcement personnel, to these images as a means of evaluating the prevalence of racial bias and stereotyping among individuals in these groups.

New Advances in 3D Facial Reconstruction
Using Computer Models

J-N. Vignal, A. Brejeon, and Y. Schuliar
lnstitut de Recherche Criminelle
Rosny-sous-Bois, France

In 1995 a facial reconstruction technique using a 2D computer image deformation was developed. Soft tissue thicknesses, which can be determined by linear regression functions according to sex, age, and stoutness, were determined using computerized axial tomography scans. Soft tissue thicknesses of a group of 34 Caucasian women and 41 Caucasian men were measured. This technique allowed the identification of skeletal remains.

To improve this method, a 3D computer model was used. Because detailed intracranial information and thousands of thickness points were unavailable, 3D data using photographic images of the sagittal and coronal planes were acquired. These images provided coordinates of individual points in the sagittal and coronal planes from which 3D spatial coordinates could be determined and, subsequently, reconstructed and warped. This approach allowed a facial reconstruction from only a few points. The results were demonstrated first with a 2D method and then with a 3D model.

Identifying Racially Distinct Facial Morphologies in
Subadult Skeletal Remains Using
Geometric Morphometric Techniques

U. Strand Viðarsdóttir
University of Durham
Durham, United Kingdom

P. O'Higgins
University of Central Lancashire
Preston, United Kingdom

It has long been established that the adult facial skeleton can be used to reliably assign modern human individuals to major racial groups or populations and, consequently, aid forensic identification of skeletal remains.

Similar identification has hitherto proved difficult in subadults as the degree of allometric changes in the craniofacial skeleton during postnatal growth is greater than the extent of the possible population-specific morphologies. However, recent advances in analytical techniques have allowed the study of inter- and intra-population differences in facial form during growth and the possibility of developing models of population-specific morphologies at all ages.

This study examines morphological separation between age series of ten modern human groups, some with known morphological differences in adult facial form. These groups include populations from Africa, Europe, North America, Melanesia, Australia, and Polynesia. In addition, they include closely related populations such as Alaskan Inupiaq Eskimo and Aleutians. The facial growth trajectories of the groups are modeled using principal components analysis of Procrustes' registered 3D landmark data from the facial skeleton. This method allows the isolation of those shape differences that are due solely to allometric growth and, subsequently, a comparison of the groups irrespective of the age or sex of the individuals within them.

The results show that all the populations can be statistically separated on the basis of some aspect of facial shape, irrespective of age and sex. This indicates that population-specific morphologies are established early in development and carried through and accentuated by allometric changes during growth. The shape relationships between the populations based on the age series are significantly correlated with those based on an adult sample from the same populations. Of greater importance, an average of 70.74 percent of the individuals can be assigned to the correct population on the basis of the first 90 percent of total variance. The number of individuals correctly assigned to each group is dependent on the number of individuals and, in particular, subadult individuals in each particular group.

A separate analysis of three populations—African Americans, Caucasians, and North American Indians—gives yet stronger results, with individuals being assigned to their correct populations with 91, 88, and 97 percent certainty, respectively. The greater success of this analysis as opposed to the previous one, is due, on the one hand, to the relatively large number of individuals in the samples concerned, and on the other, to the smaller number of populations included.

It is concluded that the findings of this study, given data sets of adequate size, could be used to develop a facility that would assign unknown subadult individuals to population groups with considerable certainty.

Identification of Individuals Through
Photographic Facial Comparisons

R. W. Vorder Bruegge and T. Musheno
Federal Bureau of Investigation
Washington, DC

Examiners in the Forensic Audio, Video, and Image Analysis Unit (FAVIAU) of the FBI Laboratory conduct photographic facial comparisons as a means of identifying or eliminating known individuals as the persons depicted in questioned images. These questioned images most frequently originate from bank surveillance film or videotape images, as well as surveillance images from other sources, such as convenience stores. The known images usually take the form of arrest photographs of suspects.

The process of photographic facial comparison is based on the same principle used in the analysis and comparison of fingerprints, footwear and tire tread impressions, firearms, and toolmarks. This "Principle of Individualization" states that

The individualization of an impression [or other piece of physical evidence] is established by finding agreement of corresponding individual characteristics of such number and significance to preclude the possibility (or probability) of their having occurred by mere coincidence, and establishing that there are no differences that cannot be accounted for (Tuthill 1994).

Two types of characteristics are considered in a facial comparison: class and individual identifying characteristics. Class characteristics include such characteristics as the overall shape of the face, nose, mouth, chin, eyes, and ears. Individual identifying characteristics include such features as moles, scars, freckle patterns, chipped teeth, and the detailed configuration of the ears. Known individuals may be identified or eliminated as subjects depicted in questioned images on the basis of the overall correspondence or dissimilarity of these characteristics. Examples of comparisons based on these features will be discussed.

A variety of factors must be considered when conducting this type of analysis. One of the most crucial factors involves the 3D nature of the subject and the effects of perspective. In addition to the obvious differences that result from a change in the orientation of a subject's head relative to the camera (e.g., profile view versus full-frontal view), further variations in the apparent morphology and relative position of facial features are observed when the distance between the subject and camera is varied. Also, differences in the resolution of the camera systems used to acquire the questioned and known images (e.g., film versus video) can result in apparent differences in feature dimensions when no such differences exist (Vorder Bruegge and Musheno 1996). In most cases, these factors prevent an exact, one-to-one comparison of specific features. Failure to consider these factors, particularly when attempting video superimposition or biometric analysis, can lead to incorrect conclusions.

References

Tuthill, H. Individualization: Principles and Procedures in Criminalistics. Lightning Powder Company, Salem, Massachusetts, 1994.

Vorder Bruegge, R. W. and Musheno, T. M. Some Cautions Regarding the Application of Biometric Analysis and Computer-Aided Facial Recognition in Law Enforcement. Presented at the American Defense Preparedness Association's 12th Annual Joint Government–Industry Security Technology Symposium and Exhibition, Williamsburg, Virginia, June 17–20, 1996.

Skull Reassembly and Modeling and the Effect Upon
3D Facial Reconstruction

C. Wilkinson and R. Neave
University of Manchester
Manchester, United Kingdom

One of the tasks of forensic anthropology is to identify the skeletal remains of unknown individuals. In cases where a general description has not elicited a response from the authorities or the public and there are no other clues, an attempt may be made to reconstruct the face from the skull. Many archaeological specimens may also be studied using a facial reconstruction technique. One of the methods used to reconstruct the face onto the skull is the 3D method.

In order to conduct any facial reconstruction, the skull must ideally be whole and intact. The more detail the skull shows, the more accurate the reconstruction will be. Unfortunately, forensic cases often include skulls that are fragmented, damaged, or incomplete, and it is not always possible to work from the ideal specimen. In addition, many archaeological specimens are damaged by the passage of time or movement of earth, and the specimens are fragile and incomplete.

Frequently the skull fragments may require reassembly and missing portions may require modeling. These tasks have traditionally been left to the expertise of the forensic anthropologist or, more recently, the medical artist. This paper discusses the techniques of skull reassembly and remodeling.

The accuracy of the technique is established using the following two examples of forensic facial reconstruction:

• An incomplete, fire-damaged skull that resulted in identification of the individual (Staffordshire, United Kingdom, Police, Major Incident Squad, 1997), and
  
  
• An incomplete skull of a murder victim that established a cause of death (Lady of the Lake incident, Coniston, The Lake District, United Kingdom, 1998).

Computer-Assisted Facial Image Identification System

M. Yoshino, H. Matsuda, S. Kubota, K. Imaizumi, and S. Miyasaka
National Research Institute of Police Science
Kashiwa, Chiba, Japan

This system consists of a 3D physiognomic range finder and a computer-assisted facial image superimposition unit. The 3D range finder is composed of a detector for measuring facial surface and its control computer. The detector has two sinusoidal grating projection devices and two CCD cameras. The computer-assisted facial image superimposition unit consists of a host computer including proprietary software, a flat surface color display, and a color image scanner for inputting 2D facial images of a criminal.

The 3D facial shape and texture of a suspect are obtained by using the range finder based on the sinusoidal grating projection with phase shift method at 2.5 seconds, with an accuracy on the order of 0.16 mm. To compare the 3D facial image and the 2D facial image, the 3D facial image is first reproduced on a display of the host computer from a MO disk; then the 2D facial image is taken with the color image scanner and reproduced on the display. After the 3D facial image is adjusted exactly to match the orientation and size of the 2D facial image under the fine framework mode, the fine framework mode of the 3D facial image is converted to the fine texture image. The shape and positional relationships of facial components between the 3D and 2D facial images are examined by the fade-out or wipe image mode. In this system, 18 points were plotted on the 3D and 2D facial images to evaluate the anthropometrical data and the reciprocal point matches between the images. The distance between the two selected points and the angle among the three points selected on the 3D and 2D facial images are automatically measured. The reciprocal point-to-point differences between both images are also calculated.

The face-to-face superimposition was experimentally performed to assess the reliability of the facial image comparison with this system. To evaluate the match of 3D and 2D facial images in the same person, the 3D facial images obtained from 25 examinees were compared to the 2D oblique facial images of the same examinees times ten, yielding 250 superimpositions. In the case of the different person, the 3D facial images of 25 examinees were each compared to the 2D facial images of the other 24 examinees, yielding 600 superimpositions. The average distance obtained from the reciprocal point-to-point differences on 16 anthropometrical points of the 3D and 2D images was used as a matching criterion. The results indicated that the measuring system for the reciprocal point-to-point differences on the superimposition image was reproducible. The ranges of the average distance were 1.4–3.3 for the same person and 2.6–7.0 for the different person, respectively. The average distance and percentage error at the false positive/false negative crossover point were 3.1 and 4.2 percent, respectively.

A model case in which the 2D facial image of one examinee is identified from the 3D facial images of 25 examinees was experimentally investigated. It was suggested that the facial image comparison using the matching of reciprocal points was reliable when the threshold of the average distance was 2.5.

In conclusion, this facial image identification system, which includes morphological comparison, anthropometrical analysis, and reciprocal points matching, provides accurate and reliable identification.

Back to the index

FORENSIC SCIENCE COMMUNICATIONS     OCTOBER 2000   VOLUME 2   NUMBER 4