Automated human face recognition is a computer vision problem of considerable practical significance. Existing two dimensional (2D) face recognition techniques perform poorly for faces with uncontrolled poses, lighting and facial expressions. Face recognition technology based on three dimensional (3D) facial models is now emerging. Geometric facial models can be easily corrected for pose variations. They are illumination invariant, and provide structural information about the facial surface. Algorithms for 3D face recognition exist, however the area is far from being a matured technology. In this dissertation we address a number of open questions in the area of 3D human face recognition. Firstly, we make available to qualified researchers in the field, at no cost, a large Texas 3D Face Recognition Database, which was acquired as a part of this research work. This database contains 1149 2D and 3D images of 118 subjects. We also provide 25 manually located facial fiducial points on each face in this database. Our next contribution is the development of a completely automatic novel 3D face recognition algorithm, which employs discriminatory anthropometric distances between carefully selected local facial features. This algorithm neither uses general purpose pattern recognition approaches, nor does it directly extend 2D face recognition techniques to the 3D domain. Instead, it is based on an understanding of the structurally diverse characteristics of human faces, which we isolate from the scientific discipline of facial anthropometry. We demonstrate the effectiveness and superior performance of the proposed algorithm, relative to existing benchmark 3D face recognition algorithms. A related contribution is the development of highly accurate and reliable 2D+3D algorithms for automatically detecting 10 anthropometric facial fiducial points. While developing these algorithms, we identify unique structural/textural properties associated with the facial fiducial points. Furthermore, unlike previous algorithms for detecting facial fiducial points, we systematically evaluate our algorithms against manually located facial fiducial points on a large database of images. Our third contribution is the development of an effective algorithm for computing the structural dissimilarity of 3D facial surfaces, which uses a recently developed image similarity index called the complex-wavelet structural similarity index. This algorithm is unique in that unlike existing approaches, it does not require that the facial surfaces be finely registered before they are compared. Furthermore, it is nearly an order of magnitude more accurate than existing facial surface matching based approaches. Finally, we propose a simple method to combine the two new 3D face recognition algorithms that we developed, resulting in a 3D face recognition algorithm that is competitive with the existing state-of-the-art algorithms. / text

<p>This thesis explores a data driven machine learning based solution for Facial reconstruction from three dimensional (3D) skull shape for recognizing or identifying unknown subjects during forensic investigation. With over 8000 unidentified bodies during the past 3 decades, facial reconstruction of disintegrated bodies in helping with identification has been a critical issue for forensic practitioners. Historically, clay modelling has been used for facial reconstruction that not only requires an expert in the field but also demands a substantial amount of time for modelling, even after acquiring the skull model. Such manual reconstruction typically takes from a month to over 3 months of time and effort. The solution presented in this thesis uses 3D Cone Beam Computed Tomography (CBCT) data collected from many people to build a model of the relationship of facial skin to skull bone over a dense set of locations on the face. It then uses this skin-to-bone relationship model learned from the data to reconstruct the predicted face model from a skull shape of an unknown subject. The thesis also extends the algorithm in a way that could help modify the reconstructed face model interactively to account for the effects of age or weight. This uses the predicted face model as a starting point and creates different hypotheses of the facial appearances for different physical attributes. Attributes like age and body mass index (BMI) are used to show the physical facial appearance changes with the help of a tool we constructed. This could improve the identification process. The thesis also presents a methods designed for testing and validating the facial reconstruction algorithm. <br></p>

Understanding of the shape and size of different features of human body from the scanned data is necessary for automated design and evaluation of product ergonomics. The traditional method of finding required body dimensions by manual measurements (Anthropometry) has many sociological, logistical and technical drawbacks such as prolonged time, skilled researcher for consistency and accuracy of measurements, undesirable physical contact between the subject and the researcher, required presence of people from different demographic categories or travel of researcher with equipments. If these di- mensions are extracted from the stored digital human models, above drawbacks can be eliminated. With the emergence of laser based 3d scanners, it is now possible generate a large database of surface models of humans from different demographic backgrounds but the automatic processing of 3d meshes is under development. Though some commercial packages are available for extraction of a limited number of dimensions from full body scans, mostly belonging to topologically separable body parts like hands and legs, the dimensions associated with head and face are particularly not available in public domain. The processing of surface models of head and face from the automatic measurement point of view is also not discussed in literature though this type of data has many practical applications like ergonomic design of close-fitting products like respiratory masks,ophthalmic frames (spectacles), helmets and similar head-mounted devices; Creation of a facial feature database for face modeling coding and reconstruction and for use in forensic sciences; Automated anthropological surveys and Medical growth analysis and aesthetic surgery planning. Hence, in this thesis, a computational framework is developed for automatic detection, recognition and measurement of important facial features namely eyes, eyebrows, nose, mouth and moustache (if applicable) from scanned head and shoulder polyhedral models. After preprocessing the scanned mesh manually to fill holes and remove singular vertices, discrete differential geometric operators were implemented to compute surface normals and curvatures. Mean curvature magnitude was used as the primary metric to segment the mesh using morphological watershed algorithms which treat the mesh as a height map and separate the regions according to the water catchment basins. After visualization it was hypothesized that the important facial features consist of relatively high curvature regions and based on this hypothesis a much faster approach was then employed based on mathematical morphology to group the high curvature vertices into regions based on adjacency. The important feature regions isolated this way were then identified and labeled to be belonging to different facial features by a decision tree based on their relative spatial disposition. Adaptive selection of parameters was incorporated later to ensure robustness of this algorithm. Critical points of these identified features are recognized as the standard landmarks associated with those primary facial features. A number of clinically identified landmarks lie on the facial mid-line. An efficient algorithm is proposed for detection and processing of the mid-line using a point sampling technique which is fast and has immunity to noise in the data. An algorithm to find shortest path between two vertices while traveling along the edges is implemented to measure on-surface distances and to isolate the nose. Complete program comprising of curvature and surface normal computations, seg- mentation and identification of 6 important features, facial mid-line processing, detection of total 17 landmarks and shortest path computations to separate nose takes about 2 minutes to work including visualization on a full resolution mesh of typically 2,15,521 Vertices and 4,30,560 Faces. The algorithm was tested successfully on more than 40 faces with minor exceptions. The results match human perception. The computed measurements were also compared with the physical measurements for a few subjects, the measurements were found to be in good agreement and satisfactory for its usage in product ergonomics and clinical applications.

The main purpose of this study is to conduct a contrastive analysis on a corpus of Swedish and Spanish family conversations with respect to two adjacency pairs: opinion-agreement/disagreement (OADs) and offer-acceptance/rejection (OARs). On one hand, from a structural perspective, based on the methodology of Conversation Analysis, one of the objectives is to observe how (dis)preferred turns of the OADs and OARs are managed by the interlocutors. On the other hand, from a functional perspective, based on the methodology of Sociocultural Pragmatics, the intention is to study how face is constructed and how politeness is managed by the family members when expressing OADs and OARs. The structural analysis of OADs and OARs shows that the majority of agreements and acceptances follow the rules for preferred turns proposed by orthodox conversation analysts, i.e. they appear directly after the first part of the adjacency pair (opinion or offer), and they are brief and unambiguous. However, the structural analysis also reveals that 70% (Swedish corpus) and 72% (Spanish corpus) of the disagreements as well as 64% (Swedish corpus) and 70% (Spanish corpus) of the rejections have a tendency to not follow the proposed rules for dispreferred turns, i.e. they are not delayed or accompanied by hesitations, justifications, etc. and nor are they evaluated as dispreferred by the participants. This indicates that social perspective, especially face, has to be considered when deciding what is considered (dis)preferred. The functional analysis of the OADs indicates that the majority of the disagreements in both Swedish (68%) and Spanish (79%) corpus are not mitigated, but rather are expressed in a fairly direct manner. Swedes tend to avoid disagreements, and therefore we expected to find a major difference between the two groups. One explanation could be that family members enjoy close relationships, and therefore the Swedes feel free to express their disagreements. As for the impact on the family members face, in both groups, it is both autonomy face and affiliation face that are influenced when OADs are expressed. As for agreement, for example, it is usually autonomy face that is affected. We interpret this as a way for the participants to show that both speakers and listeners have valuable opinions that deserve to be both voiced and commented on. This reveals the more discursive (rather than ritual) nature of OADs. In addition, the functional study of OARs shows that acceptances and rejections in both corpora are expressed using both ritual and attenuating politeness according to the norms required by the situation. Concerning the impact on face, autonomy face has different requirements in the two cultures: in the Swedish conversations, it is important to offer food without insisting several times, and in the Spanish corpus, it is important to offer food more than one or two times, and there is also a tendency to refuse the offer several times before accepting it. Therefore, according to one’s situational role, one has to know how to both give and receive offers, which points to the more ritual nature of OARs. Finally, we want to emphasize that by adding a social perspective to the structural one, we can interpret the meaning of the conversations in a way that provides a broader understanding of what is being said as participants express OADs and OARs.

Baigiamajame magistro darbe nagrinėjamos technologijos naudojamos 3D veido atkūrime. Pasirinkus 3D veido atkūrimą iš nuotraukos yra nagrinėjami šio proceso metu naudojami 3D bendriniai modeliai, jų sandara, bei juos naudojantys metodai. Taip pat nagrinėjamas veido savybių taškų aptikimas naudojant aktyvios formos modelius, bendrinio modelio interpoliavimas naudojant plonųjų plokštumų splainų metodą, bei ortogonalus tekstūros projektavimas modeliui. Taip pat pateikiamas naujas siūlomas veido formos modelis aprašantis papildomas savybes. Praktinėje dalyje yra pateikiama sukurta programa, kuri pateikus vieną veido nuotrauką automatiškai atkuria 3D veidą su tekstūra. Gautas 3D veidas gali būti eksportuotas į OBJ formato bylą. Gauti rezultatai yra įvertinami apskaičiuojant paklaidą tarp realių ir atkurtų veidų, naudojant lazeriu nuskaitytų 3D veidų duomenų bazę. / The aim of this master thesis is to examine the technologies that are being used for 3D face reconstruction. 3D generic face models, their structure, and methods they are being used with, are being analyzed after selecting reconstruction from single photo technology. Also facial feature points detection while using active shape models, 3D generic face interpolation, and orthogonal texture mapping for model are being analyzed. Also a new facial shape model is defined. Automated 3D face reconstruction application was developed based on previous analysis. Application is capable of automatically reconstructing full 3D face model with texture from single frontal face picture. Resulting face model can be exported to another software by using OBJ file format. The results of application are being evaluated while calculating error between real and reconstructed faces, while using laser scanned 3D face database.

M.Sc. (Computer Science) / A video conference is an interactive meeting between two or more locations, facilitated by simultaneous two-way video and audio transmissions. People in a video conference, also known as participants, join these video conferences for business and recreational purposes. In a typical video conference, we should properly identify and authenticate every participant in the video conference, if information discussed during the video conference is confidential. This prevents unauthorized and unwanted people from being part of the conference and exposing any confidential information during the video conference. Present existing video conferencing systems however, have problems in this area, resulting in some risks. These risks relate precisely to the lack of facilities to properly identify and authenticate participants, making it possible for unwanted/unauthorised participants to join the conference or masquerade as another participant. It is especially a problem, when facilitators or organisers are the only participants that know the authorised participants, or participants allowed in a video conference. In this dissertation, we review the risks that are present in video conferencing, and create a security system, (called BioVid) that mitigates the identification and authentication risks in video conferences. BioVid uses a Speeded-Up Robust Features or SURF-based face recognition approach, to identify and authenticate any participant in a video conference. BioVid continuously monitors the participants to check if masquerading has occurred and when it does detect an unauthorised participant, it informs the Service Provider. The Service Provider can then deal with the problem by either kicking the participant or asking the other participants to vote the unauthorised participant out of the video conference.

Face recognition that is invariant to pose and illumination is a problem solved effortlessly by the human brain, but the computational details that underlie such efficient recognition are still far from clear. This thesis draws on research from psychology and neuroscience about face and object recognition and the visual system in order to develop a novel computational method for face detection, feature selection and representation, and memory structure for recall. A biologically plausible framework for developing a face recognition system will be presented. This framework can be divided into four parts: 1) A face detection system. This is an improved version of a biologically inspired feedforward neural network that has modifiable connections and reflects the hierarchical and elastic structure of the visual system. The face detection system can detect if a face is present in an input image, and determine the region which contains that face. The system is also capable of detecting the pose of the face. 2) A face region selection mechanism. This mechanism is used to determine the Gabor-style features corresponding to the detected face, i.e., the features from the region of interest. This region of interest is selected using a feedback mechanism that connects the higher level layer of the feedforward neural network where ultimately the face is detected to an intermediate level where the Gabor style features are detected. 3) A face recognition system which is based on the binary encoding of the Gabor style features selected to represent a face. Two alternative coding schemes are presented, using 2 and 4 bits to represent a winning orientation at each location. The effectiveness of the Gabor-style features and the different coding schemes in discriminating faces from different classes is evaluated using the Yale B Face Database. The results from this evaluation show that this representation is close to other results on the same database. 4) A theoretical approach for a memory system capable of memorising sequences of poses. A basic network for memorisation and recall of sequences of labels have been implemented, and from this it is extrapolated a memory model that could use the ability of this model to memorise and recall sequences, to assist in the recognition of faces by memorising sequences of poses. Finally, the capabilities of the detection and recognition parts of the system are demonstrated using a demo application that can learn and recognise faces from a webcam.

The aim of this work is to study and test current methods for face detection on veiled faces and evaluate the results. In the first chapter, five selected methods are theoretically analyzed and in the second chapter the individual methods are evaluated, both for the Wider Face file and for the actual set of photos with veiled faces. Subsequently, the Dlib CNN method is improved for better detection of veiled faces and reprogrammed to detect the degree of veil from the tested image

There has been a significant increase in the interest for the task of classifying demographic profiles i.e. race and ethnicity. Ethnicity is a significant human characteristic and applying facial image data for the discrimination of ethnicity is integral to face-related biometric systems. Given the diversity in the application of ethnicity-specific information such as face recognition and iris recognition, and the availability of image datasets for more commonly available human populations, i.e. Caucasian, African-American, Asians, and South-Asian Indians. A gap has been identified for the development of a system which analyses the full-face and its individual feature-components (eyes, nose and mouth), for the Pakistani ethnic group. An efficient system is proposed for the verification of the Pakistani ethnicity, which incorporates a two-tier (computer vs human) approach. Firstly, hand-crafted features were used to ascertain the descriptive nature of a frontal-image and facial profile, for the Pakistani ethnicity. A total of 26 facial landmarks were selected (16 frontal and 10 for the profile) and by incorporating 2 models for redundant information removal, and a linear classifier for the binary task. The experimental results concluded that the facial profile image of a Pakistani face is distinct amongst other ethnicities. However, the methodology consisted of limitations for example, low performance accuracy, the laborious nature of manual data i.e. facial landmark, annotation, and the small facial image dataset. To make the system more accurate and robust, Deep Learning models are employed for ethnicity classification. Various state-of-the-art Deep models are trained on a range of facial image conditions, i.e. full face and partial-face images, plus standalone feature components such as the nose and mouth. Since ethnicity is pertinent to the research, a novel facial image database entitled Pakistani Face Database (PFDB), was created using a criterion-specific selection process, to ensure assurance in each of the assigned class-memberships, i.e. Pakistani and Non-Pakistani. Comparative analysis between 6 Deep Learning models was carried out on augmented image datasets, and the analysis demonstrates that Deep Learning yields better performance accuracy compared to low-level features. The human phase of the ethnicity classification framework tested the discrimination ability of novice Pakistani and Non-Pakistani participants, using a computerised ethnicity task. The results suggest that humans are better at discriminating between Pakistani and Non-Pakistani full face images, relative to individual face-feature components (eyes, nose, mouth), struggling the most with the nose, when making judgements of ethnicity. To understand the effects of display conditions on ethnicity discrimination accuracy, two conditions were tested; (i) Two-Alternative Forced Choice (2-AFC) and (ii) Single image procedure. The results concluded that participants perform significantly better in trials where the target (Pakistani) image is shown alongside a distractor (Non-Pakistani) image. To conclude the proposed framework, directions for future study are suggested to advance the current understanding of image based ethnicity verification. / Acumé Forensic

Call number: LD2668 .T4 1965 V145 / Master of Science