Individual identification of disguised faces by morphometrical matching

The reliability of a morphometrical matching method for identifying disguised faces was examined experimentally using a computer-assisted facial image identification system. The 2D right oblique facial images of three target persons disguised with sunglasses, cap and gauze mask were each compared with each of the 3D facial images of 100 subjects, yielding 900 face-to-face superimpositions. The average perpendicular distance between the facial outlines and the average point-to-point distance of the corresponding landmarks in the 2D image of the disguised face and the 3D facial image, were calculated. As a matching criterion, the sum of the values of the average perpendicular difference of the facial outlines and the average point-to-point difference between the corresponding landmarks was used (abbreviation: average difference).The range of the average difference was 2.3-2.8mm for the same person (a match) and 4.0-14.6mm for different (non-matching) people, respectively. The ranges for matching and non-matching faces did not overlap. Even the 3D facial images of the non-matching person showing the closest value of average difference to the average difference for the matching person could be excluded easily. It was concluded that the morphometrical matching method can reliably identify disguised faces and the results produced by this method could be easily understandable by a court of law.     

2D/3D image (facial) comparison using camera matching

A problem in forensic facial comparison of images of perpetrators and suspects is that distances between fixed anatomical points in the face, which form a good starting point for objective, anthropometric comparison, vary strongly according to the position and orientation of the camera. In case of a cooperating suspect, a 3D image may be taken using e.g. a laser scanning device. By projecting the 3D image onto a 2D image with the suspect's head in the same pose as that of the perpetrator, using the same focal length and pixel aspect ratio, numerical comparison of (ratios of) distances between fixed points becomes feasible. An experiment was performed in which, starting from two 3D scans and one 2D image of two colleagues, male and female, and using seven fixed anatomical locations in the face, comparisons were made for the matching and non-matching case. Using this method, the non-matching pair cannot be distinguished from the matching pair of faces. Facial expression and resolution of images were all more or less optimal, and the results of the study are not encouraging for the use of anthropometric arguments in the identification process. More research needs to be done though on larger sets of facial comparisons.     

Application of superimposition-based personal identification using skull computed tomography images

Superimposition has been applied to skulls of unidentified skeletonized corpses as a personal identification method. The current method involves layering of a skull and a facial image of a suspected person and thus requires a real skeletonized skull. In this study, we scanned skulls of skeletonized corpses by computed tomography (CT), reconstructed three-dimensional (3D) images of skulls from the CT images, and superimposed the 3D images with facial images of the corresponding persons taken in their lives. Superimposition using 3D-reconstructed skull images demonstrated, as did superimposition using real skulls, an adequate degree of morphological consistency between the 3D-reconstructed skulls and persons in the facial images. Three-dimensional skull images reconstructed from CT images can be saved as data files and the use of these images in superimposition is effective for personal identification of unidentified bodies.     

Computer-assisted facial image identification system using a 3-D physiognomic range finder

This system consists of a 3-D physiognomic range finder and a computer-assisted facial image superimposition unit. The 3-D 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 a proprietary software, a flat surface color display and a color image scanner for inputting 2-D facial images of a criminal. The 3-D facial shape and texture of a suspect is obtained by using the range finder. To make the comparison between the 3-D facial image and the 2-D facial image, the 3-D facial image is first reproduced on a display of the host computer from a MO disk and then the 2-D facial image is taken with the color image scanner and reproduced on the display. The 3-D facial image is exactly adjusted to match the orientation and size of the 2-D facial image under the fine framework mode, and then the fine framework mode of 3-D facial image is converted to the fine texture image. The shape and positional relationships of facial components between the 3-D and 2-D facial images are examined by the fade-out or wipe image mode. The distance between the selected two points and angle among the selected three points on the 3-D and 2-D facial images are automatically measured for the assessment of anthropometrical data between both images. For evaluating the fit between the anthropometrical points on the 3-D and 2-D facial images, the reciprocal point-to-point difference between both images is compared.     

Computerized craniofacial reconstruction using CT-derived implicit surface representations

In forensic craniofacial reconstruction, facial features of an unknown individual are estimated from an unidentified skull, based on a mixture of experimentally obtained guidelines on the relationship between soft tissues and the underlying skeleton. In this paper, we investigate the possibility of using full 3D cross-sectional CT images for establishing a reference database of densely sampled distances between the external surfaces of the skull and head for automated craniofacial reconstruction. For each CT image in the reference database, the hard tissue (skull) and soft tissue (head) volumes are automatically segmented and transformed into signed distance transform (sDT) images, representing for each voxel in this image the Euclidean distance to the closest point on the skull and head surface, respectively, distances being positive (negative) for voxels inside (outside) the skull/head. Multiple craniofacial reconstructions are obtained by first warping each reference skull sDT maps to the target skull sDT using a B-spline based free form deformation algorithm and subsequently applying these warps to the reference head sDT maps. A single reconstruction of the target head surface is defined as the zero level set of the arithmetic average of all warped reference head sDT maps, but other reconstructions are possible, biasing the result to subject specific attributes (age, BMI, gender). Both qualitative and quantitative tests (measuring the similarity between the 3D reconstructed and corresponding original head surface) on a small (N = 20) database are presented to proof the validity of the concept.     

Pistol image retrieval by shape representation

Databases have been used to record data in forensic science, such as fingerprints, shoeprints, and photos. In traditional databases, we use “text” as the “keyword” for retrieving data (text-keyword retrieval); however, in some applications, “text” is not proper to describe the searching target, and in this case, “image” plays an important role. In this paper, we use “image features” as the “keywords”, and show its potential for building up a prototype of pistol image databases.In current firearm databases of forensic science, the retrieval method is still by text-keyword retrieval. For experienced forensic examiners, this kind of databases may satisfy their requirement. However, for people who have little gun knowledge, how can they find the possible gun candidates or similar guns in the databases, if there are not any available words or marks on the gun? In this paper, we try to retrieve similar pistol images by the pistol shape instead of “text-keyword”. This method can narrow down the searching range while identifying pistols by firearm databases.There are more than 300 pistol images in our pistol image database. From the experimental results, we can retrieve the similar pistol images in top five candidates for each pistol image.     

Shoeprint image retrieval and crime scene shoeprint image linking by using convolutional neural network and normalized cross correlation

A shoeprint image retrieval process aims to identify and match images of shoeprints found at crime scenes with shoeprint images from a known reference database. It is a challenging problem in the forensic discipline of footwear analysis because a shoeprint found at the crime scene is often imperfect. Recovered shoeprints may be partial, distorted, left on surfaces that do not mark easily, or perhaps come from shoes that do not transfer marks easily. In this study, we present a shoeprint retrieval method by using a convolutional neural network (CNN) and normalized cross-correlation (NCC). A pre-trained CNN was used to extract features from the pre-processed shoeprint images. We then employed NCC to compute a similarity score based on the extracted image features. We achieved a retrieval accuracy of 82% in our experiments, where a “successful” retrieval means that the ground truth image was returned in the top 1% of returned images. We also extend our shoeprint retrieval method to the problem of linking shoeprints recovered from crime scenes. This new method can provide a linkage between two crime scenes if the two recovered shoeprints originated from the same shoe. This new method achieved a retrieval accuracy of 88.99% in the top 20% of returned images.     

Investigation into the use of photoanthropometry in facial image comparison

Photoanthropometry is a metric based facial image comparison technique. Measurements of the face are taken from an image using predetermined facial landmarks. Measurements are then converted to proportionality indices (PIs) and compared to PIs from another facial image. Photoanthropometry has been presented as a facial image comparison technique in UK courts for over 15 years. It is generally accepted that extrinsic factors (e.g. orientation of the head, camera angle and distance from the camera) can cause discrepancies in anthropometric measurements of the face from photographs. However there has been limited empirical research into quantifying the influence of such variables. The aim of this study was to determine the reliability of photoanthropometric measurements between different images of the same individual taken with different angulations of the camera. The study examined the facial measurements of 25 individuals from high resolution photographs, taken at different horizontal and vertical camera angles in a controlled environment. Results show that the degree of variability in facial measurements of the same individual due to variations in camera angle can be as great as the variability of facial measurements between different individuals. Results suggest that photoanthropometric facial comparison, as it is currently practiced, is unsuitable for elimination purposes. Preliminary investigations into the effects of distance from camera and image resolution in poor quality images suggest that such images are not an accurate representation of an individuals face, however further work is required.     

The Effect of Image Quality and Forensic Expertise in Facial Image Comparisons

Images of perpetrators in surveillance video footage are often used as evidence in court. In this study, identification accuracy was compared for forensic experts and untrained persons in facial image comparisons as well as the impact of image quality. Participants viewed thirty image pairs and were asked to rate the level of support garnered from their observations for concluding whether or not the two images showed the same person. Forensic experts reached their conclusions with significantly fewer errors than did untrained participants. They were also better than novices at determining when two high‐quality images depicted the same person. Notably, lower image quality led to more careful conclusions by experts, but not for untrained participants. In summary, the untrained participants had more false negatives and false positives than experts, which in the latter case could lead to a higher risk of an innocent person being convicted for an untrained witness.     

A symmetry perceiving adaptive neural network and facial image recognition

The paper deals with the forensic problem of comparing nearly from view and facial images for personal identification. The human recognition process for such problems, is primarily based on both holistic as well as feature-wise symmetry perception aided by subjective analysis for detecting ill-defined features. It has been attempted to approach the modelling of such a process by designing a robust symmetry perceiving adaptive neural network. The pair of images to be compared should be presented to the proposed neural network (NN) as source (input) and target images. The NN learns about the symmetry between the pair of images by analysing examples of associated feature pairs belonging to the source and the target images. In order to prepare a paired example of associated features for training purpose, when we select one particular feature on the source image as a unique pixel, we must associate it with the corresponding feature on the target image also. But, in practice, it is not always possible to fix the latter feature also as a unique pixel due to pictorial ambiguity. The robust or fault tolerant NN takes care of such a situation and allows fixing the associated target feature as a rectangular array of pixels, rather than fixing it as a unique pixel, which is pretty difficult to be done with certainty. From such a pair of sets of associated features, the NN searches out proper locations of the target features from the sets of ambiguous target features by a fuzzy analysis during its learning. If any of target features, searched out by the NN, lies outside the prespecified zone, the training of the NN is unsuccessful. This amounts to non-existence of symmetry between the pair of images and confirms non-identity. In case of a successful training, the NN gets adapted with appropriate symmetry relation between the pair of images and when the source image is input to the trained NN, it responds by outputting a processed source image which is superimposable over the target images and identity may subsequently be established by examining detailed matching in machine-made superimposed/composite images which are also suitable for presentation before the court. The performance of the proposed NN has been tested with various cases including simulated ones and it is hoped to serve as a working tool of forensic anthropologists.     

Accuracy of facial soft tissue thickness measurements in personal computer-based multiplanar reconstructed computed tomographic images

The purpose of this study was to determine the precision and accuracy of facial soft tissue measurement using personal computer (PC)-based multiplanar reconstructed (MPR) computed tomography (CT) images and to evaluate the effect of the various CT scanning protocols on the facial soft tissue thickness measurement. Thirteen different CT imaging protocols were used to image a cadaver head. MPR reformations and three-dimensional (3D) reconstructions viewed on a laptop PC were used to make measurements at six specific sites on each set of images. These measurements were compared to physical measurements at the same sites. Increasing the slice thickness resulted in decreased image quality. Within the same slice thickness, increasing the pitch ratio in the spiral mode, resulted in decreasing image quality. The image quality of conventional CT scanning was relatively poorer than that of the spiral CT scanning. However, the mean deviation from the physical measurement was within 0.43 mm in every instance. This mean deviation was quite small and clinically acceptable for measuring the soft tissue thickness of the facial area. PC-based MPR CT images of the face using routine scanning CT protocols can be used to accurately measure soft tissue thickness in the facial region. However, for more fine and accurate data collection, scanning protocols with slice thicknesses less than 5mm, and a spiral/helical mode pitch less than 2:1 are recommended.     

A new experimental approach to computer-aided face/skull identification in forensic anthropology

The present study introduces a new approach to computer-assisted face/skull matching used for personal identification purposes in forensic anthropology.In this experiment, the authors formulated an algorithm able to identify the face of a person suspected to have disappeared, by comparing the respective person's facial image with the skull radiograph. A total of 14 subjects were selected for the study, from which a facial photograph and skull radiograph were taken and ultimately compiled into a database, saved to the hard drive of a computer. The photographs of the faces and corresponding skull radiographs were then drafted using common photographic software, taking caution not to alter the informational content of the images. Once computer generated, the facial images and menu were displayed on a color monitor. In the first phase, a few anatomic points of each photograph were selected and marked with a cross to facilitate and more accurately match the face with its corresponding skull. In the second phase, the above mentioned cross grid was superimposed on the radiographic image of the skull and brought to scale. In the third phase, the crosses were transferred to the cranial points of the radiograph. In the fourth phase, the algorithm calculated the distance of each transferred cross and the corresponding average. The smaller the mean value, the greater the index of similarity between the face and skull.A total of 196 cross-comparisons were conducted, with positive identification resulting in each case. Hence, the algorithm matched a facial photograph to the correct skull in 100% of the cases.     

Large-scale in-vivo Caucasian facial soft tissue thickness database for craniofacial reconstruction

A large-scale study of facial soft tissue depths of Caucasian adults was conducted. Over a 2-years period, 967 Caucasian subjects of both sexes, varying age and varying body mass index (BMI) were studied. A user-friendly and mobile ultrasound-based system was used to measure, in about 20min per subject, the soft tissue thickness at 52 facial landmarks including most of the landmarks used in previous studies. This system was previously validated on repeatability and accuracy [S. De Greef, P. Claes, W. Mollemans, M. Loubele, D. Vandermeulen, P. Suetens, G. Willems, Semi-automated ultrasound facial soft tissue depth registration: method and validation. J. Forensic Sci. 50 (2005)]. The data of 510 women and 457 men were analyzed in order to update facial soft tissue depth charts of the contemporary Caucasian adult. Tables with the average thickness values for each landmark as well as the standard deviation and range, tabulated according to gender, age and BMI are reported. In addition, for each landmark and for both sexes separately, a multiple linear regression of thickness versus age and BMI is calculated. The lateral asymmetry of the face was analysed on an initial subset of 588 subjects showing negligible differences and thus warranting the unilateral measurements of the remaining subjects. The new dataset was statistically compared to three datasets for the Caucasian adults: the traditional datasets of Rhine and Moore [J.S. Rhine, C.E. Moore, Tables of facial tissue thickness of American Caucasoids in forensic anthropology. Maxwell Museum Technical series 1 (1984)] and Helmer [R. Helmer, Sch?delidentifizierung durch elektronische bildmischung, Kriminalistik Verlag GmbH, Heidelberg, 1984] together with the most recent in vivo study by Manhein et al. [M.H. Manhein, G.A. Listi, R.E. Barsley, R. Musselman, N.E. Barrow, D.H. Ubelbaker, In vivo facial tissue depth measurements for children and adults. J. Forensic Sci. 45 (2000) 48-60]. The large-scale database presented in this paper offers a denser sampling of the facial soft tissue depths of a more representative subset of the actual Caucasian population over the different age and body posture subcategories. This database can be used as an updated chart for manual and computer-based craniofacial approximation and allows more refined analyses of the possible factors affecting facial soft tissue depth.     

An accuracy assessment of forensic computerized facial reconstruction employing cone-beam computed tomography from live subjects

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 accuracy. 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.     

Facial Age Synthesis Using Sparse Partial Least Squares (The Case of Ben Needham)

Automatic facial age progression (AFAP) has been an active area of research in recent years. This is due to its numerous applications which include searching for missing. This study presents a new method of AFAP. Here, we use an active appearance model (AAM) to extract facial features from available images. An aging function is then modelled using sparse partial least squares regression (sPLS). Thereafter, the aging function is used to render new faces at different ages. To test the accuracy of our algorithm, extensive evaluation is conducted using a database of 500 face images with known ages. Furthermore, the algorithm is used to progress Ben Needham's facial image that was taken when he was 21 months old to the ages of 6, 14, and 22 years. The algorithm presented in this study could potentially be used to enhance the search for missing people worldwide.     

Configural and featural information in facial-composite images

Eyewitnesses are often invited to construct a facial composite, an image created of the person they saw commit a crime that is used by law enforcement to locate criminal suspects. In the current paper, the effectiveness of composite images was investigated from traditional feature systems (E-FIT and PRO-fit), where participants (face constructors) selected individual features to build the face, and a more recent holistic system (EvoFIT), where they ‘evolved’ a composite by repeatedly selecting from arrays of complete faces. Further participants attempted to name these composites when seen as an unaltered image, or when blurred, rotated, linearly stretched or converted to a photographic negative. All of the manipulations tested reduced correct naming of the composites overall except (i) for a low level of blur, for which naming improved for holistic composites but reduced for feature composites, and (ii) for 100% linear stretch, for which a substantial naming advantage was observed. Results also indicated that both featural (facial elements) and configural (feature spacing) information were useful for recognition in both types of composite system, but highly-detailed information was more accurate in the feature-based than in the holistic method. The naming advantage of linear stretch was replicated using a forensically more-practical procedure with observers viewing an unaltered composite sideways. The work is valuable to police practitioners and designers of facial-composite systems.     

Computer-assisted skull identification system using video superimposition

This system consists of two main units, namely a video superimposition system and a computer-assisted skull identification system. The video superimposition system is comprised of the following five parts: a skull-positioning box having a monochrome CCD camera, a photo-stand having a color CCD camera, a video image mixing device, a TV monitor and a videotape recorder. The computer-assisted skull identification system is composed of a host computer including our original application software, a film recorder and a color printer. After the determination of the orientation and size of the skull to those of the facial photograph using the video superimposition system, the skull and facial photograph images are digitized and stored within the computer, and then both digitized images are superimposed on the monitor. For the assessment of anatomical consistency between the digitized skull and face, the distance between the landmarks and the thickness of soft tissue of the anthropometrical points are semi-automatically measured on the monitor. The wipe images facilitates the comparison of positional relationships between the digitized skull and face. The software includes the polynomial functions and Fourier harmonic analysis for evaluating the match of the outline such as the forehead and mandibular line in both the digitized images.     

A new technique for three-dimensional ultrasound scanning of facial tissues

We report the development of an ultrasonic facial scanning technique that allows for the visualization of continuous contours without deforming surface tissues. Adhesive markers are placed on the face to enable measurement of facial tissue thicknesses at specific landmarks. The subject immerses the face in a clear plastic box filled with water for about 20 seconds while the researcher moves the transducer along the bottom of the box, guiding transducer movement by watching the facial image in a mirror placed below. 3D Echotech software (1) builds the images from sequentially acquired 2D frames. Reliability of repeat measurements at landmarks is good, and individual tissues (skin, subcutaneous, muscle) can be distinguished. The method is simple, reliable, less expensive and less time consuming than alternatives such as magnetic resonance imaging (MRI). It is applicable in both research and clinical contexts.     

The Use of Scale‐Invariance Feature Transform Approach to Recognize and Retrieve Incomplete Shoeprints

Shoeprints left at the crime scene provide valuable information in criminal investigation due to the distinctive patterns in the sole. Those shoeprints are often incomplete and noisy. In this study, scale‐invariance feature transform is proposed and evaluated for recognition and retrieval of partial and noisy shoeprint images. The proposed method first constructs different scale spaces to detect local extrema in the underlying shoeprint images. Those local extrema are considered as useful key points in the image. Next, the features of those key points are extracted to represent their local patterns around key points. Then, the system computes the cross‐correlation between the query image and each shoeprint image in the database. Experimental results show that full‐size prints and prints from the toe area perform best among all shoeprints. Furthermore, this system also demonstrates its robustness against noise because there is a very slight difference in comparison between original shoeprints and noisy shoeprints.