No / Recently, automatic age progression has gained popularity due to its nu-merous applications. Among these is the search for missing people, in the UK alone up to 300,000 people are reported missing every year. Although many algorithms have been proposed, most of the methods are affected by image noise, illumination variations, and most importantly facial expres-sions. To this end we propose to build an age progression framework that utilizes image de-noising and expression normalizing capabilities of kernel principal component analysis (Kernel PCA). Here, Kernel PCA a nonlinear form of PCA that explores higher order correlations between input varia-bles, is used to build a model that captures the shape and texture variations of the human face. The extracted facial features are then used to perform age progression via a regression procedure. To evaluate the performance of the framework, rigorous tests are conducted on the FGNET ageing data-base. Furthermore, the proposed algorithm is used to progress images of Mary Boyle; a six-year-old that went missing over 39 years ago, she is considered Ireland’s youngest missing person. The algorithm presented in this paper could potentially aid, among other applications, the search for missing people worldwide.

Yes / 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 ageing function is then modelled using Sparse Partial Least Squares Regression (sPLS). Thereafter, the ageing 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 paper could potentially be used to enhance the search for missing people worldwide.

In this thesis we study Particle Filter for visual tracking applications. The sequential Monte Carlo methods or Particle Filter provides approximate solutions when the tracking problem involves non-linear and/or non-Gaussian state space models. Also in this study, in order to make the visual tracker robust against change in target appearance and unexpected target motion, an adaptive target appearance model and a first order motion estimator are embedded in particle filtering. Additionally, since pixels that don&rsquo / t belong to target makes the motion estimation biased, the algorithm includes robust estimators to make the tracker reliable. Within the scope of this thesis the visual tracker proposed in [5] is implemented and the same problem is solved by proposing a Rao-Blackwellized Particle Filter. To deal with problems encountered during the implementation phase of the algorithm some improvements are made such as utilizing learning rate for the computation of adaptive velocity estimation. Moreover, some precautions are taken such as checking the velocity estimations to validate them. Finally, we have done several experiments both in indoor and outdoor environments to demonstrate the effectiveness and robustness of the implemented algorithm. Experimental results show that most of the time the visual tracker performs well. On the other hand the drawbacks of the implemented tracker are indicated and we explain how to eliminate them.

This dissertation presents an efficient multilayer background modeling approach to distinguish among midground objects, the objects whose existence occurs over varying time scales between the extremes of short-term ephemeral appearances (foreground) and long-term stationary persistences (background). Traditional background modeling separates a given scene into foreground and background regions. However, the real world can be much more complex than this simple classification, and object appearance events often occur over varying time scales. There are situations in which objects appear on the scene at different points in time and become stationary; these objects can get occluded by one another, and can change positions or be removed from the scene. Inability to deal with such scenarios involving midground objects results in errors, such as ghost objects, miss-detection of occluding objects, aliasing caused by the objects that have left the scene but are not removed from the model, and new objects’ detection when existing objects are displaced. Modeling temporal layers of multiple objects allows us to overcome these errors, and enables the surveillance and summarization of scenes containing multiple midground objects.

Ces dernières années ont vu l'émergence de la capture des modèles spatio-temporels (modélisation 4D) à partir d'images réelles, avec de nombreuses applications dans les domaines de post-production pour le cinéma, la science des sports, les études sociales, le divertissement, l'industrie de la publicité. A partir de plusieurs séquences vidéos, enregistrées à partir de points de vue variés, la modélisation 4D à partir de vidéos utilise des modèles spatio-temporels pour extraire des informations sur la géométrie et l'apparence de scènes réelles, permettant de les enregistrer et de les reproduire. Cette thèse traite du problème de la modélisation d'apparence.La disponibilité des donnée d'images offre de grands potentiels pour les reconstructions haute fidélité, mais nécessite des méthodes plus élaborées. En outre, les applications du monde réel nécessitent des rendus rapides et des flux réduits de données. Mais l'obtention de représentations d'apparence compactes, indépendantes du point de vue, et à grande résolution est toujours un problème ouvert.Pour obtenir ces caractéristiques, nous exprimons l'information visuelle de l'objet capturé dans un espace de texture commun. Les observations multi-caméra sont considérées comme des réalisations de l'apparence commune et un modèle linéaire est introduit pour matérialiser cette relation. Le modèle linéaire d'apparence proposé permet une première étude du problème de l'estimation d'apparence dans le cas multi-vue et expose les sources variées de bruit et les limitations intrinsèques du modèle.Basé sur ces observations, et afin d'exploiter l'information visuelle de la manière la plus efficace, nous améliorons la méthode en y intégrant un modèle de super-résolution 2D. Le modèle simule le procédé de capture d'image avec une concaténation d'opérations linéaires, générant les observation d'image des différents points de vue et permettant d'exploiter la redondance. Le problème de super-résolution multi-vue résultant est résolu par inférence bayésienne et une représentation haute-résolution d'apparence est fournie permettant de reproduire la texture de l'objet capturé avec grand détail.La composante temporelle est intégrée par la suite au modèle pour permettre d'y recouper l'information visuelle commune sous-jacente. En considérant des petits intervalles de temps ou l'apparence de l'objet ne change pas drastiquement, une représentation super-résolue cohérente temporellement est introduite. Elle explique l'ensemble des images de l'objet capturé dans cet intervalle. Grâce à l'inférence statistique Bayésienne, l'apparence construite permet des rendus avec une grande précision à partir de point de vue nouveau et à des instants différent dans l'intervalle de temps prédéfini.Pour améliorer l'estimation d'apparence d'avantage, l'inter-dépendance de la géométrie et de la photométrie est étudiée et exploitée. Les modélisations de la géométrie et de l'apparence sont unifiées dans le framework de super-résolution permettant une amélioration géométrique globale, ce qui donne à son tour une amélioration importante de l'apparence.Finalement pour encoder la variabilité de l'apparence dynamique des objets subissant plusieurs mouvements, une représentation indépendante du point de vue s'appuyant sur l'analyse en composantes principales est introduite. Cette représentation décompose la variabilité sous-jacente d'apparence en texture propres et déformations propres. La méthode proposée permet de reproduire les apparences de manière précise avec des représentation compactes. Il permet également l'interpolation et la complétion des apparences.Cette étude montre que la représentation compacte, indépendante du point de vue, et super-résolue proposée permet de confronter les nouvelles réalités du problème de modélisation d'apparence. Elle représente un contribution vers des représentations d'apparence 4D haute-qualité et ouvre de nouvelles directions de recherche dans ce domaine. / Capturing spatio-temporal models (4D modelling) from real world imagery has received a growing interest during the last years urged by the increasing demands of real-world applications and the tremendous amount of easily accessible image data. The general objective is to produce realistic representations of the world from captured video sequences. Although geometric modelling has already reached a high level of maturity, the appearance aspect has not been fully explored. The current thesis addresses the problem of appearance modelling for realistic spatio-temporal representations. We propose a view-independent, high resolution appearance representation that successfully encodes the high visual variability of objects under various movements.First, we introduce a common appearance space to express all the available visual information from the captured images. In this space we define the representation of the global appearance of the subject. We then introduce a linear image formation model to simulate the capturing process and to express the multi-camera observations as different realizations of the common appearance. Identifying that the principle of Super-Resolution technique governs also our multi-view scenario, we extend the image generative model to accommodate it. In our work, we use Bayesian inference to solve for the super-resolved common appearance.Second, we propose a temporally coherent appearance representation. We extend the image formation model to generateimages of the subject captured in a small time interval. Our starting point is the observation thatthe appearance of the subject does not change dramatically in a predefined small time interval and the visual information from each view and each frame corresponds to the same appearance representation.We use Bayesian inference to exploit the visual redundant as well as the hidden non-redundant information across time, in order to obtain an appearance representation with fine details.Third, we leverage the interdependency of geometry and photometry and use it toestimate appearance and geometry in a joint manner. We show that by jointly estimating both, we are able to enhance the geometry globally that in turn leads to a significant appearance improvement.Finally, to further encode the dynamic appearance variability of objects that undergo several movements, we cast the appearance modelling as a dimensionality reduction problem. We propose a view-independent representation which builds on PCA and decomposesthe underlying appearance variability into Eigen textures and Eigen warps. The proposed framework is shown to accurately reproduce appearances with compact representations and to resolve appearance interpolation and completion tasks.

This thesis attempts to create an application to determine head rotation angle in video recorded from a security camera. The application consists of three parts: face detection, facial landmarks detection and determination of person's head rotation. The face detection has been implemented using Viola-Jones and HOG algorithms. Facial landmarks detection has been done using algorithm based on active shape model. Two methods to calculate the head rotation angles have been used: the first method works with anthropometric head features. The second method uses Perspective-n-Point algorithm to find the right rotation angles. Finally, all algorithms implemented have been tested and the proper parameters have been determined.

This biomedical engineering thesis explores the opportunities and challenges of 2D+t contrast echocardiography for left ventricle functional analysis, both clinically and within a computer vision atlas-based deformable template model framework. A database was created for the experiments in this thesis, with 21 studies of contrast Dobutamine Stress Echo, in all 4 principal planes. The database includes clinical variables, human expert hand-traced myocardial contours and visual scoring. First the problem is studied from a clinical perspective. Quantification of endocardial global and local function using standard measures shows expected values and agreement with human expert visual scoring, but the results are less reliable for myocardial thickening. Next, the problem of segmenting the endocardium with a computer is posed in a standard landmark and atlas-based deformable template model framework. The underlying assumption is that these models can emulate human experts in terms of integrating previous knowledge about the anatomy and physiology with three sources of information from the image: texture, geometry and kinetics. Probabilistic atlases of contrast echocardiography are computed, while noting from histograms at selected anatomical locations that modelling texture with just mean intensity values may be too naive. Intensity analysis together with the clinical results above suggest that lack of external boundary definition may preclude this imaging technique for appropriate measuring of myocardial thickening, while endocardial boundary definition is appropriate for evaluation of wall motion. Geometry is presented in a Principal Component Analysis (PCA) context, highlighting issues about Gaussianity, the correlation and covariance matrices with respect to physiology, and analysing different measures of dimensionality. A popular extension of deformable models ---Active Appearance Models (AAMs)--- is then studied in depth. Contrary to common wisdom, it is contended that using a PCA texture space instead of a fixed atlas is detrimental to segmentation, and that PCA models are not convenient for texture modelling. To integrate kinetics, a novel spatio-temporal model of cardiac contours is proposed. The new explicit model does not require frame interpolation, and it is compared to previous implicit models in terms of approximation error when the shape vector changes from frame to frame or remains constant throughout the cardiac cycle. Finally, the 2D+t atlas-based deformable model segmentation problem is formulated and solved with a gradient descent approach. Experiments using the similarity transformation suggest that segmentation of the whole cardiac volume outperforms segmentation of individual frames. A relatively new approach ---the inverse compositional algorithm--- is shown to decrease running times of the classic Lucas-Kanade algorithm by a factor of 20 to 25, to values that are within real-time processing reach.