Recent advances in 4D data acquisition systems in the field of Computer Vision have opened up many exciting new possibilities for the interpretation of complex moving surfaces. However, a fundamental problem is that this has also led to a huge increase in the volume of data to be handled. Attempting to make sense of this wealth of information is then a core issue to be addressed if such data can be applied to more complex tasks. Similar problems have been historically encountered in the analysis of 3D static surfaces, leading to the extraction of higher-level features based on analysis of the differential geometry. Our central hypothesis is that there exists a compact set of similarly useful descriptors for the analysis of dynamic 4D surfaces. The primary advantages in considering localised changes are that they provide a naturally useful set of invariant characteristics. We seek a constrained set of terms - a vocabulary - for describing all types of deformation. By using this, we show how to describe what the surface is doing more effectively; and thereby enable better characterisation, and consequently more effective visualisation and comparison. This thesis investigates this claim. We adopt a bottom-up approach of the problem, in which we acquire raw data from a newly constructed commercial 4D data capture system developed by our industrial partners. A crucial first step resolves the temporal non-linear registration between instances of the captured surface. We employ a combined optical/range flow to guide a conformation over a sequence. By extending the use of aligned colour information alongside the depth data we improve this estimation in the case of local surface motion ambiguities. By employing a KLT/thin-plate-spline method we also seek to preserve global deformation for regions with no estimate. We then extend aspects of differential geometry theory for existing static surface analysis to the temporal domain. Our initial formulation considers the possible intrinsic transitions from the set of shapes defined by the variations in the magnitudes of the principal curvatures. This gives rise to a total of 15 basic types of deformation. The change in the combined magnitudes also gives an indication of the extent of change. We then extend this to surface characteristics associated with expanding, rotating and shearing; to derive a full set of differential features. Our experimental results include qualitative assessment of deformations for short episodic registered sequences of both synthetic and real data. The higher-level distinctions extracted are furthermore a useful first step for parsimonious feature extraction, which we then proceed to demonstrate can be used as a basis for further analysis. We ultimately evaluate this approach by considering shape transition features occurring within the human face, and the applicability for identification and expression analysis tasks.

Automatically determining the temporal characteristics of facial expressions has extensive application domains such as human-machine interfaces for emotion recognition, face identification, as well as medical analysis. However, many papers in the literature have not addressed the step of determining when such expressions occur. This dissertation is focused on the problem of automatically segmenting macro- and micro-expressions frames (or retrieving the expression intervals) in video sequences, without the need for training a model on a specific subset of such expressions. The proposed method exploits the non-rigid facial motion that occurs during facial expressions by modeling the strain observed during the elastic deformation of facial skin tissue. The method is capable of spotting both macro expressions which are typically associated with emotions such as happiness, sadness, anger, disgust, and surprise, and rapid micro- expressions which are typically, but not always, associated with semi-suppressed macro-expressions. Additionally, we have used this method to automatically retrieve strain maps generated from peak expressions for human identification. This dissertation also contributes a novel 3-D surface strain estimation algorithm using commodity 3-D sensors aligned with an HD camera. We demonstrate the feasibility of the method, as well as the improvements gained when using 3-D, by providing empirical and quantitative comparisons between 2-D and 3-D strain estimations.

The video-based computational analyses of human face and gesture signals encompass a myriad of challenging research problems involving computer vision, machine learning and human computer interaction. In this thesis, we focus on the following challenges: a) the classification of hand and body gestures along with the temporal localization of their occurrence in a continuous stream, b) the recognition of facial expressivity levels in people with Parkinson's Disease using multimodal feature representations, c) the prediction of student learning outcomes in intelligent tutoring systems using affect signals, and d) the personalization of machine learning models, which can adapt to subject and group-specific nuances in facial and gestural behavior. Specifically, we first conduct a quantitative comparison of two approaches to the problem of segmenting and classifying gestures on two benchmark gesture datasets: a method that simultaneously segments and classifies gestures versus a cascaded method that performs the tasks sequentially. Second, we introduce a framework that computationally predicts an accurate score for facial expressivity and validate it on a dataset of interview videos of people with Parkinson's disease. Third, based on a unique dataset of videos of students interacting with MathSpring, an intelligent tutoring system, collected by our collaborative research team, we build models to predict learning outcomes from their facial affect signals. Finally, we propose a novel solution to a relatively unexplored area in automatic face and gesture analysis research: personalization of models to individuals and groups. We develop hierarchical Bayesian neural networks to overcome the challenges posed by group or subject-specific variations in face and gesture signals. We successfully validate our formulation on the problems of personalized subject-specific gesture classification, context-specific facial expressivity recognition and student-specific learning outcome prediction. We demonstrate the flexibility of our hierarchical framework by validating the utility of both fully connected and recurrent neural architectures.

No / In this paper we discuss a novel method of mathematically modelling facial action units for accurate representation of human facial expressions in 3- dimensions. Our method utilizes the approach of Facial Action Coding System (FACS). It is based on a boundary-value approach, which utilizes a solution to a fourth order elliptic Partial Differential Equation (PDE) subject to a suitable set of boundary conditions. Here the PDE surface generation method for human facial expressions is utilized in order to generate a wide variety of facial expressions in an efficient and realistic way. For this purpose, we identify a set of boundary curves corresponding to the key features of the face which in turn define a given facial expression in 3-dimensions. The action units (AUs) relating to the FACS are then efficiently represented in terms of Fourier coefficients relating to the boundary curves which enables us to store both the face and the facial expressions in an efficient way.

Leveraging task relatedness has been proven to be beneficial in many machine learning tasks. Extensive researches has been done to exploit task relatedness in various forms. A common assumption for the tasks is that they are intrinsically similar to each other. Based on this assumption, joint learning algorithms are usually implemented via some forms of information sharing. Various forms of information sharing have been proposed, such as shared hidden units of neural networks, common prior distribution in hierarchical Bayesian model, shared weak learners of a boosting classifier, distance metrics and a shared low rank structure for multiple tasks. However, another very common and important task relationship, i.e., task competition, has been largely overlooked. Task competition means that tasks are competing with each other if there are competitions or conflicts between their goals. Considering that tasks with competition relationship are universal, this dissertation is to accommodate this intuition from an algorithmic perspectives and apply the algorithms to various visual recognition problems. Focus on exploiting the task competition relationships in visual recognition, the dissertation presents three types of algorithms and applied them to different visual recognition tasks. First, hypothesis competition has been exploited in a boosting framework. The proposed algorithm CompBoost jointly model the target and auxiliary tasks with a generalized additive regression model regularized by competition constraints. This model treats the feature selection as the weak learner (\ie, base functions) selection problem, and thus provides a mechanism to improve feature filtering guided by task competition. More specifically, following a stepwise optimization scheme, we iteratively add a new weak learner that balances between the gain for the target task and the inhibition on the auxiliary ones. We call the proposed algorithm CompBoost, since it shares similar structures with the popular AdaBoost algorithm. In this dissertation, we use two test beds for evaluation of CompBoost: (1) content-independent writer identification by exploiting competing tasks of handwriting recognition, and (2) actor-independent facial expression recognition by exploiting competing tasks of face recognition. In the experiments for both applications, the approach demonstrates promising performance gains by exploiting the between-task competition relationship. Second, feature competition has been instantiated through an alternating coordinate gradient algorithm. Sharing the same feature pool, two tasks are modeled together in a joint loss framework, with feature interaction encouraged via an orthogonal regularization over feature importance vectors. Then, an alternating greedy coordinate descent learning algorithm (AGCD) is derived to estimate the model. The algorithm effectively excludes distracting features in a fine-grained level for improving face verification. In other words, the proposed algorithm does not forbid feature sharing between competing tasks in a macro level; it instead selectively inhibits distracting features while preserving discriminative ones. For evaluation, the proposed algorithm is applied to two widely tested face-aging benchmark datasets: FG-Net and MORPH. On both datasets, our algorithm achieves very promising performances and outperforms all previously reported results. These experiments, together with detailed experimental analysis, show clearly the benefit of coordinating conflicting tasks for improving visual recognition. Third, two ad-hoc feature competition algorithms have been proposed to apply to visual privacy protection problems. Visual privacy protection problem is a practical case of competition factors in real world application. Algorithms are specially designed to achieve best balance between competing factors in visual privacy protection based on different modeling frameworks. Two algorithms are developed to apply to two applications, license plate de-identification and face de-identification. / Computer and Information Science

No / Explains how to implement computational techniques for human smile analysis Shares insights into the human personality traits hidden in a smile Enriches the understanding of human emotions through examples of face analysis Includes key examples of the practical use of computer based smile analysis.

No / How many times have you smiled today? How many times have you frowned today? Ever thought of being in a state of self-consciousness to be able to relate your own mood with your facial emotional expressions? Perhaps with our present-day busy lives, we may not consider these as crucial questions. However, as researchers uncover more and more about the human emotional landscape they are learning the importance of understanding our emotions.

L’analyse de visages 3D y compris la reconnaissance des visages et des expressions faciales 3D est devenue un domaine actif de recherche ces dernières années. Plusieurs méthodes ont été développées en utilisant des images 2D pour traiter ces problèmes. Cependant, ces méthodes présentent un certain nombre de limitations dépendantes à l’orientation du visage, à l’éclairage, à l’expression faciale, et aux occultations. Récemment, le développement des capteurs d’acquisition 3D a fait que les données 3D deviennent de plus en plus disponibles. Ces données 3D sont relativement invariables à l’illumination et à la pose, mais elles restent sensibles à la variation de l’expression. L’objectif principal de cette thèse est de proposer de nouvelles techniques de reconnaissance/vérification de visages et de reconnaissance d’expressions faciales 3D. Tout d’abord, une méthode de reconnaissance de visages en utilisant des matrices de covariance comme des descripteurs de régions de visages est proposée. Notre méthode comprend les étapes suivantes : le prétraitement et l’alignement de visages, un échantillonnage uniforme est ensuite appliqué sur la surface faciale pour localiser un ensemble de points de caractéristiques. Autours de chaque point, nous extrayons une matrice de covariance comme un descripteur de région du visage. Deux méthodes d’appariement sont ainsi proposées, et différentes distances (géodésiques / non-géodésique) sont appliquées pour comparer les visages. La méthode proposée est évaluée sur troisbases de visages GAVAB, FRGCv2 et BU-3DFE. Une description hiérarchique en utilisant trois niveaux de covariances est ensuite proposée et validée. La deuxième partie de cette thèse porte sur la reconnaissance des expressions faciales 3D. Pour ce faire, nous avons proposé d’utiliser les matrices de covariances avec les méthodes noyau. Dans cette contribution, nous avons appliqué le noyau de Gauss pour transformer les matrices de covariances en espace d’Hilbert. Cela permet d’utiliser les algorithmes qui sont déjà implémentés pour l’espace Euclidean (i.e. SVM) dans cet espace non-linéaire. Des expérimentations sont alors entreprises sur deux bases d’expressions faciales 3D (BU-3DFE et Bosphorus) pour reconnaître les six expressions faciales prototypiques. / 3D face analysis including 3D face recognition and 3D Facial expression recognition has become a very active area of research in recent years. Various methods using 2D image analysis have been presented to tackle these problems. 2D image-based methods are inherently limited by variability in imaging factors such as illumination and pose. The recent development of 3D acquisition sensors has made 3D data more and more available. Such data is relatively invariant to illumination and pose, but it is still sensitive to expression variation. The principal objective of this thesis is to propose efficient methods for 3D face recognition/verification and 3D facial expression recognition. First, a new covariance based method for 3D face recognition is presented. Our method includes the following steps : first 3D facial surface is preprocessed and aligned. A uniform sampling is then applied to localize a set of feature points, around each point, we extract a matrix as local region descriptor. Two matching strategies are then proposed, and various distances (geodesic and non-geodesic) are applied to compare faces. The proposed method is assessed on three datasetsincluding GAVAB, FRGCv2 and BU-3DFE. A hierarchical description using three levels of covariances is then proposed and validated. In the second part of this thesis, we present an efficient approach for 3D facial expression recognition using kernel methods with covariance matrices. In this contribution, we propose to use Gaussian kernel which maps covariance matrices into a high dimensional Hilbert space. This enables to use conventional algorithms developed for Euclidean valued data such as SVM on such non-linear valued data. The proposed method have been assessed on two known datasets including BU-3DFE and Bosphorus datasets to recognize the six prototypical expressions.

L'analyse automatique des expressions faciales est une étape clef pour le développement d'interfaces intelligentes ou l'analyse de comportements. Toutefois, celle-ci est rendue difficile par un grand nombre de facteurs, pouvant être d'ordre morphologiques, liés à l'orientation du visage ou à la présence d'occultations. Nous proposons des adaptations des Random Forest permettant d' adresser ces problématiques:- Le développement des Pairwise Conditional Random Forest, consistant en l'apprentissage de modèles à partir de paires d'images expressives. Les arbres sont de plus conditionnés par rapport à l'expression de la première image afin de réduire la variabilité des transitions. De plus, il est possible de conditionner les arbres en rapport avec une estimation de la pose du visage afin de permettre la reconnaissance quel que soit le point de vue considéré.- L'utilisation de réseaux de neurones auto-associatifs pour modéliser localement l'apparence du visage. Ces réseaux fournissent une mesure de confiance qui peut être utilisée dans le but de pondérer des Random Forests définies sur des sous-espaces locaux du visage. Ce faisant, il est possible de fournir une prédiction d'expression robuste aux occultations partielles du visage.- Des améliorations du récemment proposé algorithme des Neural Decision Forests, lesquelles consistent en une procédure d'apprentissage simplifiée, ainsi qu'en une évaluation "greedy" permettant une évaluation plus rapide, avec des applications liées à l'apprentissage en ligne de représentations profondes pour la reconnaissance des expressions, ainsi que l'alignement de points caractéristiques. / Automatic face analysis is a key to the development of intelligent human-computer interaction systems and behavior understanding. However, there exist a number of factors that makes face analysis a difficult problem. This include morphological differences between different persons, head pose variations as well as the possibility of partial occlusions. In this PhD, we propose a number of adaptations of the so-called Random Forest algorithm to specifically adress those problems. Mainly, those improvements consist in:– The development of a Pairwise Conditional Random Forest framework, that consists in training Random Forests upon pairs of expressive images. Pairwise trees are conditionned on the expression label of the first frame of a pair to reduce the ongoing expression transition variability. Additionnally, trees can be conditionned upon a head pose estimate to peform facial expression recognition from an arbitrary viewpoint.– The design of a hierarchical autoencoder network to model the local face texture patterns. The reconstruction error of this network provides a confidence measurement that can be used to weight Randomized decision trees trained on spatially-defined local subspace of the face. Thus, we can provide an expression prediction that is robust to partial occlusions.– Improvements over the very recent Neural Decision Forests framework, that include both a simplified training procedure as well as a new greedy evaluation procedure, that allows to dramatically improve the evaluation runtime, with applications for online learning and, deep learning convolutional neural network-based features for facial expression recognition as well as feature point alignement.

No description available.