Dissertations / Theses on the topic '2D/3D'

The area of this work is a combination of draping and printing. It strives towards the technique that dazzles the eye with illusions of more than one dimension. As a viewer you will believe that the prints are real drapings while they are flat surfaces. Today prints in fashion are categorized as placed prints or all-over prints, and generally created as a flat surface to decorate the garments. In this work the idea is to manipulate and challenge the boundaries of print and give it life through the body shapes and in the movement. Potentially this work could be an introduction to a new way of working with prints in fashion. This work could poosibly be presented as a new technique where placed- print and all-over prints comes together- called placed all-over prints. Also, it could develop into further techniques in using two-dimensional flatness and save fabric in using photography as an option to the real three-dimensional drapings. Furthermore could it mean savings in material as a conscious choice in the process ? This investigation explores two particular kinds of techniques, - print and draping, that are merged into one expression. The aims of this work is to find new ways of using print in combination with draping in dress and explore the possibilities to find a new technique to create interesting womenswear. To unite dimensions like two-dimensional and three-dimensional as a method of finding new forms and expressions. Through experiments with striped textiles the focus is to investigate the possibilities of greater visual effects on two-dimensional prints. For a depth and to exaggerate the directions in the fabric before translating it into a flat surface the striped textile can be a tool for further design. The striped textile has the potential to help the eye to understand the directions in the print and can be used in more than one dimension and color. To explore how to create 3D effect on 2D in print design through draping in dress is the aim of this work.
Program: Modedesignutbildningen

3D modeling of real objects has an increasing importance in numerous areas. Although many methods and solutions are already proposed for 3D data acquisition, research continuing in this area is still intense. However, a crucial drawback about 3D data extraction algorithms is their testing and validation difficulty. Additionally, obtaining calibrated 2D and 3D imaging systems is troublesome due to their high effort demand for calibration and high cost. In this thesis, a 2D/3D Imaging Simulator is proposed in order to ease development and testing of 3D data interpretations of different methods and also to generate synthetic images for miscellaneous use. Furthermore, an example application on FRGC database is explained in detail.

While interaction with computers used to be dominated by mice and keyboards, new types of sensors now allow users to interact through touch, speech, or using their whole body in 3D space. These new interaction modalities are often referred to as "natural user interfaces" or "NUIs." While 2D NUIs have experienced major success on billions of mobile touch devices sold, 3D NUI systems have so far been unable to deliver a mobile form factor, mainly due to their use of cameras. The fact that cameras require a certain distance from the capture volume has prevented 3D NUI systems from reaching the flat form factor mobile users expect. In this dissertation, we address this issue by sensing 3D input using flat 2D sensors. The systems we present observe the input from 3D objects as 2D imprints upon physical contact. By sampling these imprints at very high resolutions, we obtain the objects' textures. In some cases, a texture uniquely identifies a biometric feature, such as the user's fingerprint. In other cases, an imprint stems from the user's clothing, such as when walking on multitouch floors. By analyzing from which part of the 3D object the 2D imprint results, we reconstruct the object's pose in 3D space. While our main contribution is a general approach to sensing 3D input on 2D sensors upon physical contact, we also demonstrate three applications of our approach. (1) We present high-accuracy touch devices that allow users to reliably touch targets that are a third of the size of those on current touch devices. We show that different users and 3D finger poses systematically affect touch sensing, which current devices perceive as random input noise. We introduce a model for touch that compensates for this systematic effect by deriving the 3D finger pose and the user's identity from each touch imprint. We then investigate this systematic effect in detail and explore how users conceptually touch targets. Our findings indicate that users aim by aligning visual features of their fingers with the target. We present a visual model for touch input that eliminates virtually all systematic effects on touch accuracy. (2) From each touch, we identify users biometrically by analyzing their fingerprints. Our prototype Fiberio integrates fingerprint scanning and a display into the same flat surface, solving a long-standing problem in human-computer interaction: secure authentication on touchscreens. Sensing 3D input and authenticating users upon touch allows Fiberio to implement a variety of applications that traditionally require the bulky setups of current 3D NUI systems. (3) To demonstrate the versatility of 3D reconstruction on larger touch surfaces, we present a high-resolution pressure-sensitive floor that resolves the texture of objects upon touch. Using the same principles as before, our system GravitySpace analyzes all imprints and identifies users based on their shoe soles, detects furniture, and enables accurate touch input using feet. By classifying all imprints, GravitySpace detects the users' body parts that are in contact with the floor and then reconstructs their 3D body poses using inverse kinematics. GravitySpace thus enables a range of applications for future 3D NUI systems based on a flat sensor, such as smart rooms in future homes. We conclude this dissertation by projecting into the future of mobile devices. Focusing on the mobility aspect of our work, we explore how NUI devices may one day augment users directly in the form of implanted devices.
Die Interaktion mit Computern war in den letzten vierzig Jahren stark von Tastatur und Maus geprägt. Neue Arten von Sensoren ermöglichen Computern nun, Eingaben durch Berührungs-, Sprach- oder 3D-Gestensensoren zu erkennen. Solch neuartige Formen der Interaktion werden häufig unter dem Begriff "natürliche Benutzungsschnittstellen" bzw. "NUIs" (englisch natural user interfaces) zusammengefasst. 2D-NUIs ist vor allem auf Mobilgeräten ein Durchbruch gelungen; über eine Milliarde solcher Geräte lassen sich durch Berührungseingaben bedienen. 3D-NUIs haben sich jedoch bisher nicht auf mobilen Plattformen durchsetzen können, da sie Nutzereingaben vorrangig mit Kameras aufzeichnen. Da Kameras Bilder jedoch erst ab einem gewissen Abstand auflösen können, eignen sie sich nicht als Sensor in einer mobilen Plattform. In dieser Arbeit lösen wir dieses Problem mit Hilfe von 2D-Sensoren, von deren Eingaben wir 3D-Informationen rekonstruieren. Unsere Prototypen zeichnen dabei die 2D-Abdrücke der Objekte, die den Sensor berühren, mit hoher Auflösung auf. Aus diesen Abdrücken leiten sie dann die Textur der Objekte ab. Anhand der Stelle der Objektoberfläche, die den Sensor berührt, rekonstruieren unsere Prototypen schließlich die 3D-Ausrichtung des jeweiligen Objektes. Neben unserem Hauptbeitrag der 3D-Rekonstruktion stellen wir drei Anwendungen unserer Methode vor. (1) Wir präsentieren Geräte, die Berührungseingaben dreimal genauer als existierende Geräte messen und damit Nutzern ermöglichen, dreimal kleinere Ziele zuverlässig mit dem Finger auszuwählen. Wir zeigen dabei, dass sowohl die Haltung des Fingers als auch der Benutzer selbst einen systematischen Einfluss auf die vom Sensor gemessene Position ausübt. Da existierende Geräte weder die Haltung des Fingers noch den Benutzer erkennen, nehmen sie solche Variationen als Eingabeungenauigkeit wahr. Wir stellen ein Modell für Berührungseingabe vor, das diese beiden Faktoren integriert, um damit die gemessenen Eingabepositionen zu präzisieren. Anschließend untersuchen wir, welches mentale Modell Nutzer beim Berühren kleiner Ziele mit dem Finger anwenden. Unsere Ergebnisse deuten auf ein visuelles Modell hin, demzufolge Benutzer Merkmale auf der Oberfläche ihres Fingers an einem Ziel ausrichten. Bei der Analyse von Berührungseingaben mit diesem Modell verschwinden nahezu alle zuvor von uns beobachteten systematischen Effekte. (2) Unsere Prototypen identifizieren Nutzer anhand der biometrischen Merkmale von Fingerabdrücken. Unser Prototyp Fiberio integriert dabei einen Fingerabdruckscanner und einen Bildschirm in die selbe Oberfläche und löst somit das seit Langem bestehende Problem der sicheren Authentifizierung auf Berührungsbildschirmen. Gemeinsam mit der 3D-Rekonstruktion von Eingaben ermöglicht diese Fähigkeit Fiberio, eine Reihe von Anwendungen zu implementieren, die bisher den sperrigen Aufbau aktueller 3D-NUI-Systeme voraussetzten. (3) Um die Flexibilität unserer Methode zu zeigen, implementieren wir sie auf einem großen, berührungsempfindlichen Fußboden, der Objekttexturen bei der Eingabe ebenfalls mit hoher Auflösung aufzeichnet. Ähnlich wie zuvor analysiert unser System GravitySpace diese Abdrücke, um Nutzer anhand ihrer Schuhsolen zu identifizieren, Möbelstücke auf dem Boden zu erkennen und Nutzern präzise Eingaben mittels ihrer Schuhe zu ermöglichen. Indem GravitySpace alle Abdrücke klassifiziert, erkennt das System die Körperteile der Benutzer, die sich in Kontakt mit dem Boden befinden. Aus der Anordnung dieser Kontakte schließt GravitySpace dann auf die Körperhaltungen aller Benutzer in 3D. GravitySpace hat daher das Potenzial, Anwendungen für zukünftige 3D-NUI-Systeme auf einer flachen Oberfläche zu implementieren, wie zum Beispiel in zukünftigen intelligenten Wohnungen. Wie schließen diese Arbeit mit einem Ausblick auf zukünftige interaktive Geräte. Dabei konzentrieren wir uns auf den Mobilitätsaspekt aktueller Entwicklungen und beleuchten, wie zukünftige mobile NUI-Geräte Nutzer in Form implantierter Geräte direkt unterstützen können.

This thesis evaluates and tests different methods utilized to produce a 2D aesthetic within a 3D pipeline, as in, converting 3D geometry to an aesthetic that is similar to hand-drawn classic films such as Snow White And The Seven Dwarfs. This thesis explores methods to produce both exterior and interior lines that indicate shape and form of 3D models, the conclusion from the tested methods leaves with the statement that it is unlikely the human factor will be ever entirely replaced by automated solutions, and instead a mixed approach with shader relied solutions and involvement of texturing techniques which provides artistic controls where necessary, is deemed to be the most effective way of preserving the hand-drawn 2D aesthetic within a 3D-pipeline.

The field of computer vision studies the computational tools and methods required for computers to be able to process visual information, for example images and video. Shape descriptors are one of the tools commonly used in image processing applications. Shape descriptors are mathematical functions which are applied to an image and produce numerical values which are representative of a particular characteristic of the image. These numerical values can then be processed in order to provide some information about the image. For example, these values can be fed to a classifier in order to assign a class label to the image. There are a number of shape descriptors already existing in the literature for 2D and 3D images. The aim of this thesis is to develop additional shape descriptors which provide an improvement over (or an alternative to) those already existing in the literature. A large majority of the existing 2D shape descriptors use surface information to produce a measure. However, in some applications surface information is not present and only partially extracted contours are available. In such cases, boundary based shape descriptors must be used. A new boundary based shape descriptor called Linearity is introduced. This measure can be applied to open or closed curve segments. In general the availability of 3D images is comparatively smaller than that of 2D images. As a consequence, the number of existing 3D shape descriptors is also relatively smaller. However, there is an increasing interest in the development of 3D descriptors. In this thesis we present two basic 3D measures which afterwards are modified to produce a range of new shape descriptors. All of these descriptors are similar in their behaviour, however they can be combined and applied in different image processing applications such as image retrieval and classification. This simple fact is demonstrated through several examples.

COORDENAÇÃO DE APERFEIÇOAMENTO DO PESSOAL DE ENSINO SUPERIOR
FINANCIADORA DE ESTUDOS E PROJETOS
Esta dissertação discute a aplicação de efeitos de colorização 3D a animações 2D produzidas pela técnica quadro-a-quadro. Utilizando algoritmos de processamento de imagens, desenhos 2D são preparados para receber técnicas de sombreamento evitando a transformação da cena para uma geometria 3D. Esta preparação se dá através da obtenção de mapas de normais que aproximam a geometria do desenho. O sombreamento é obtido através de um conjunto de técnicas de renderização foto-realistas e não-foto-realistas, que podem ser adaptadas para utilização de normais aproximadas. Visando amenizar o trabalho exaustivo de colorir cada desenho, um método baseado no relacionamento entre imagens é apresentado para colorir automaticamente cada quadro numa seqüência de desenhos. Este processo de colorização considera a necessidade de possíveis intervenções humanas para garantir a qualidade final de cada imagem da animação. Um estudo sobre aproximação de normais, técnicas de sombreamento, segmentação de imagens e rastreamento de objetos é amplamente discutido nesta dissertação.
This dissertation discusses the 3D colorization effects usage over a 2d animation, which has been produced through frame-by-frame techniques. Normal vector maps approximates the drawing geometry and provide the ability to perform shading effects by applying digital image processing algorithms, avoiding 3D geometry scene transformation. A set of photorealistic and non-photorealistic renderization techniques, which can be adapted to normal approximation usage, is proposed in the colorization process. Also, a method based on interframe dependence is presented, aiming to reduce the thoroughgoing effort of colorizing each individual frame within an animation. This colorization process considers possible human interventions to ensure image´s result quality. Finally, this dissertation provides a comprehensive study regarding several topics, such as normal approximations, shading techniques, image segmentation and object tracking.

Face verification is a challenging pattern recognition problem. The face is a biometric that, we as humans, know can be recognised. However, the face is highly deformable and its appearance alters significantly when the pose, illumination or expression changes. These changes in appearance are most notable for texture images, or two-dimensional (2D) data. But the underlying structure of the face, or three dimensional (3D) data, is not changed by pose or illumination variations. Over the past five years methods have been investigated to combine 2D and 3D face data to improve the accuracy and robustness of face verification. Much of this research has examined the fusion of a 2D verification system and a 3D verification system, known as multi-modal classifier score fusion. These verification systems usually compare two feature vectors (two image representations), a and b, using distance or angular-based similarity measures. However, this does not provide the most complete description of the features being compared as the distances describe at best the covariance of the data, or the second order statistics (for instance Mahalanobis based measures). A more complete description would be obtained by describing the distribution of the feature vectors. However, feature distribution modelling is rarely applied to face verification because a large number of observations is required to train the models. This amount of data is usually unavailable and so this research examines two methods for overcoming this data limitation: 1. the use of holistic difference vectors of the face, and 2. by dividing the 3D face into Free-Parts. The permutations of the holistic difference vectors is formed so that more observations are obtained from a set of holistic features. On the other hand, by dividing the face into parts and considering each part separately many observations are obtained from each face image; this approach is referred to as the Free-Parts approach. The extra observations from both these techniques are used to perform holistic feature distribution modelling and Free-Parts feature distribution modelling respectively. It is shown that the feature distribution modelling of these features leads to an improved 3D face verification system and an effective 2D face verification system. Using these two feature distribution techniques classifier score fusion is then examined. This thesis also examines methods for performing classifier fusion score fusion. Classifier score fusion attempts to combine complementary information from multiple classifiers. This complementary information can be obtained in two ways: by using different algorithms (multi-algorithm fusion) to represent the same face data for instance the 2D face data or by capturing the face data with different sensors (multimodal fusion) for instance capturing 2D and 3D face data. Multi-algorithm fusion is approached as combining verification systems that use holistic features and local features (Free-Parts) and multi-modal fusion examines the combination of 2D and 3D face data using all of the investigated techniques. The results of the fusion experiments show that multi-modal fusion leads to a consistent improvement in performance. This is attributed to the fact that the data being fused is collected by two different sensors, a camera and a laser scanner. In deriving the multi-algorithm and multi-modal algorithms a consistent framework for fusion was developed. The consistent fusion framework, developed from the multi-algorithm and multimodal experiments, is used to combine multiple algorithms across multiple modalities. This fusion method, referred to as hybrid fusion, is shown to provide improved performance over either fusion system on its own. The experiments show that the final hybrid face verification system reduces the False Rejection Rate from 8:59% for the best 2D verification system and 4:48% for the best 3D verification system to 0:59% for the hybrid verification system; at a False Acceptance Rate of 0:1%.

Naga is an animated short about a lively dragon that roams about the lands embracing it’s surroundings dearly. It discovered a barren land while out exploring and was saddened by the sight. After pondering for a while, it then realized it could revive the land with it’s ability to summon rain using it's dragon ball. The short blends traditional animation and computer animation, where the look is similar to 2D animation but the character and a few environment elements are done in 3D. Software utilized to complete the short were Autodesk Maya, Adobe Photoshop, Clip Studio Paint, Adobe After Effects, and Adobe Premiere Pro. The short showcases a stylized Chinese traditional ink painting style, key frame animation, and particle effects.

Visual tracking of unknown objects in unconstrained video-sequences is extremely challenging due to a number of unsolved issues. This thesis explores several of these and examines possible approaches to tackle them. The unconstrained nature of real-world input sequences creates huge variation in the appearance of the target object due to changes in pose and lighting. Additionally, the object can be occluded by either parts of itself, other elements of the scene, or the frame boundaries. Observations may also be corrupted due to low resolution, motion blur, large frame-to-frame displacement, or incorrect exposure or focus of the camera. Finally, some objects are inherently difficult to track due to their (low) texture, specular/transparent nature, non-rigid deformations, etc. Conventional trackers depend heavily on the texture of the target. This causes issues with transparent or untextured objects. Edge points can be used in cases where standard feature points are scarce; these however suffer from the aperture problem. To address this, the first contribution of this thesis explores the idea of virtual corners, using pairs of non-adjacent line correspondences, tangent to edges in the image. Furthermore, the chapter investigates the possibility of long-term tracking, introducing a re-detection scheme to handle occlusions while limiting drift of the object model. The outcome of this research is an edge-based tracker, able to track in scenarios including untextured objects, full occlusions and significant length. The tracker, besides reporting excellent results in standard benchmarks, is demonstrated to successfully track the longest sequence published to date. Some of the issues in visual tracking are caused by suboptimal utilisation of the image information. The object of interest can easily occupy as few as ten or even one percent of the video frame area. This causes difficulties in challenging scenarios such as sudden camera shakes or full occlusions. To improve tracking in such cases, the next major contribution of this thesis explores relationships within the context of visual tracking, with a focus on causality. These include causal links between the tracked object and other elements of the scene such as the camera motion or other objects. Properties of such relationships are identified in a framework based on information theory. The resulting technique can be employed as a causality-based motion model to improve the results of virtually any tracker. Significant effort has previously been devoted to rapid learning of object properties on the fly. However, state-of-the-art approaches still often fail in cases such as rapid out-of-plane rotations, when the appearance changes suddenly. One of the major contributions of this thesis is a radical rethinking of the traditional wisdom of modelling 3D motion as appearance change. Instead, 3D motion is modelled as 3D motion. This intuitive but previously unexplored approach provides new possibilities in visual tracking research. Firstly, 3D tracking is more general, as large out-of-plane motion is often fatal for 2D trackers, but helps 3D trackers to build better models. Secondly, the tracker’s internal model of the object can be used in many different applications and it could even become the main motivation, with tracking supporting reconstruction rather than vice versa. This effectively bridges the gap between visual tracking and Structure from Motion. The proposed method is capable of successfully tracking sequences with extreme out-of-plane rotation, which poses a considerable challenge to 2D trackers. This is done by creating realistic 3D models of the targets, which then aid in tracking. In the majority of the thesis, the assumption is made that the target’s 3D shape is rigid. This is, however, a relatively strong limitation. In the final chapter, tracking and dense modelling of non-rigid targets is explored, demonstrating results in even more generic (and therefore challenging) scenarios. This final advancement truly generalises the tracking problem with support for long-term tracking of low texture and non-rigid objects in sequences with camera shake, shot cuts and significant rotation. Taken together, these contributions address some of the major sources of failure in visual tracking. The presented research advances the field of visual tracking, facilitating tracking in scenarios which were previously infeasible. Excellent results are demonstrated in these challenging scenarios. Finally, this thesis demonstrates that 3D reconstruction and visual tracking can be used together to tackle difficult tasks.

This thesis describes two novel catheter-based 2D-3D cardiac image registration algorithms for overlaying preoperative 3D MR or CT data onto intraoperative fluoroscopy, and fusing electroanatomical data onto clinical images. The work is intended for use in cardiac catheterisation procedures. To fulfil this objective, the algorithms must be accurate, robust and minimally disruptive to the clinical workflow. The first algorithm relies on the catheterisation of vessels of the heart and registers by minimising a vessel-radius-weighted distance between the catheters and corresponding vessel centrelines. A novelty here is a global-fit search strategy that considers all vessel branches during registration, adding robustness and avoiding manual branch selection. Another contribution to knowledge is an analysis of catheter configurations for registration. Results show that accuracy is highly dependent on the catheter configuration, and that using a coronary vessel (CV) with the aorta (Ao) was most accurate, yielding mean 3D target registration errors (TRE) between 0.55 and 7.0 mm with phantom data. Using two large-diameter vessels was least accurate, with TRE between 10 and 43 mm, and should be avoided. When applied to clinical data, registrations with the CV/Ao configuration resulted an estimated mean 2D-TRE of 5.9 mm, on average. The second 2D-3D registration algorithm extends the novelty of exploring catheter configurations by registering using catheters looped inside chambers of the heart. In phantom experiments, two-view registration yielded an average accuracy of 4.0 mm 3D-TRE (7.8-mm capture range). Using a single view, average reprojection distance was 2.7 mm (6.0-mm capture range). Application of the algorithm to a clinical dataset resulted in an estimated average 2D-TRE of 10 mm. Single view registrations are ideal when biplane X-ray acquisition is undesirable and for correcting bulk patient motion. In current practice, registration is performed manually. The algorithms in this thesis can register with comparable accuracy to manual registration, but are automated and can therefore fit better with the clinical workflow.

Este trabalho apresenta novos métodos para análise de formas tridimensionais dentro do contexto de visão computacional, destacando-se o uso das transformadas wavelets 1D, 2D e 3D, as quais proporcionam uma análise multi-escala das formas estudadas. As formas analisadas se dividem em três tipos diferentes, dependendo da sua representação matemática: f(t)=(x(t),y(t),z(t)), f(x,y)=z e f(x,y,z)=w. Cada tipo de forma é analisado por um método melhor adaptado. Primeiramente, tais formas passam por uma rotina de pré-processamento e, em seguida, pela caracterização por meio da aplicação das transformadas wavelet 1D, 2D e 3D para as respectivas formas. Esta aplicação nos permite extrair características que sejam invariantes à rotação e translação, levando em consideração alguns conceitos matemáticos da geometria diferencial. Destaca-se também neste trabalho a não obrigatoriedade de parametrização das formas. Os resultados obtidos a partir de formas extraídas de imagens médicas e dados biológicos, que justificam este trabalho, são apresentados.
This work presents new methods for three-dimensional shape analysis in the context of computational vision, being emphasized the use of 1D, 2D and 3D wavelet transforms, which provide a multiscale analysis of the studied shapes. The analyzed shapes are divided in three different types depending on their representation: f(t)=(x(t),y(t),z(t)), f(x,y)=z and f(x,y,z)=w. Each type of shape is analyzed by a more suitable method. Firstly, such shapes undergo a pre-processing procedure followed by the characterization using the 1D, 2D or 3D wavelet transform, depending on its representation. This application allows to extract features that are rotation- and translation-invariant, based on some mathematical concepts of differential geometry. In this work, we emphasize that it is not necessary to use the parameterized version of the 2D and 3D shapes. The experimental results obtained from shapes extracted from medical and biological images, that corroborate the introduced methods, are presented.

L’angioplastie coronarienne est une intervention guidée par rayons X et réalisée par voie endovasculaire qui restaure le diamètre des vaisseaux coronaires réduit par accumulation de corps gras au sein de leurs parois. Lors de cette procédure, un guide est introduit à partir d’une artère qui peut être située au poignet ou au niveau de l’aine. La technique consiste alors à amener dans le vaisseau pathologique un ballonnet gonflable dans la zone rétrécie, localisée grâce à l’injection préalable d’un produit de contraste. Le gonflement du ballonnet élargit l’artère et s’accompagne en général de la pose d’un stent, structure métallique capable de renforcer la paroi. Le geste clinique peut être facilité en intégrant aux images le détail de la nature de la paroi, information disponible avec un scanner 3D. Le but de cette thèse est de proposer une méthode de recalage déformable pour superposer cette information 3D à des images angiographiques 2D per-opératoire en déformant le modèle 3D afin qu’il suive la dynamique cardiaque capturée dans les images angiographiques. Nous introduisons un algorithme de segmentation capable de segmenter automatiquement les vaisseaux principaux dans ces images angiographiques. Ensuite, nous présentons une approche de suivi du vaisseau 3D pathologique dans une séquence 2D+t combinant appariements et déformation d’une courbe spline. Enfin, nous décrivons l’extension au suivi d'un arbre vasculaire 3D, représenté par un arbre dont les arcs sont des courbes splines, dans une séquence 2D+t. Nous avons privilégié les approches applicables avec une seule projection angiographique, bien adapté au déroulement usuel des procédures cliniques. La performance des algorithmes a fait l’objet d’évaluations quantitatives sur des données réelles incluant 30 images pour la segmentation et 23 séquences pour le recalage
Coronary angioplasty is an X-ray guided intervention, which aims at recovering the diameter of coronary vessels when the accumulation of fat in the vessel wall reduced it. During this procedure, a guide-wire is inserted in the blood vessel located at the wrist or groin. This guide-wire brings into the pathologic vessel a balloon at the level of the fat accumulation, thanks to a previous contrast injection which highlights the lesion. The balloon is inflated and very frequently a thin mesh tube of metallic wires (stent), which is wrapped around the balloon, is then expanded during the balloon inflation. The procedure could benefit from additional information on the nature of the inner wall, available on 3D CT scan. The aim of the thesis is to propose a dynamic registration to superimpose this 3D information onto the intraoperative 2D angiographic sequence, by deforming the 3D model so that it can follow the cardiac motion captured thanks to the angiographic images. We introduce a segmentation algorithm able to automatically segment the main vessels of the angiographic images. Then, we present a tracking approach of the 3D pathologic vessel in a 2D+t sequence combining pairings and the deformation of a spline curve. Finally, we describe the extension to the 3D vascular tree tracking represented by a tree, whose edges are spline curves, in a 2D+t sequence. We favored approaches that are applicable to a single angiographic projection, which is well adapted to the usual process of clinical procedures. All the proposed methods have been tested on real data, consisting of 30 angiographic images for the segmentation algorithm and 23 angiographic sequences for the registration algorithms

In the field of industrial automation large savings can be realized if position andorientation of an object is known. Knowledge about an objects position and orien-tation can be used by advanced robotic systems to be able to work with complexitems. Specifically 2D-objects are a big enough sub domain to motivate specialattention. Traditionally this problem has been solved with large mechanical sys-tems that forces the objects into specific configurations. Besides being expensive,taking up a lot of space and having great difficulty handling fragile items, thesemechanical systems have to be constructed for each particular type of object. Thisthesis explores the possibility of using registration algorithms from computer vi-sion based on 3D-data to find flat objects. While systems for locating 3D objectsalready exists they have issues with locating essentially flat objects since theirpositioning is mostly a function of their contour. The thesis consists of a briefexamination of 2D-algorithms and their extension to 3D as well as results fromthe most suitable algorithm.
Inom fältet industriautomation kan stora besparingar realiseras om man kännertill position och orientering av föremål i deras leverade skick. Kunskap om dettatillåter avancerade robotsystem att arbeta med komplicerade föremål. Specifikt2D föremål är en tillräckligt stor underdomän för att motivera speciallösningar.Traditionellt har det här problemet lösts med stora mekaniska system, som, utö-ver att de är dyra, tar upp mycket yta och har svårt att hantera ömtåliga objektäven måste konstrueras som en specifik lösning för varje objektstyp. Denna upp-sats undersöker möjligheten att använda sig av registreringsalgoritmer baseradepå datorseende i 3D-data för att hitta platta föremål. Det finns system som han-terar lokalisering av 3D objekt men de har problem med att hantera essentieltplatta föremål då deras positionering främst är en funktion av deras kontur. Upp-satsen består av en undersökning av några 2D-algoritmer och deras uttökning till3D samt resultat från en implementation som fungerar väl.

Image segmentation of organs in medical images using model-based approaches requires a priori information which is often given by manually tagging landmarks on a training set of shapes. This is a tedious, time-consuming, and error prone task. To overcome some of these drawbacks, several automatic methods were devised. Identification of the same homologous set of points in a training set of object shapes is the most crucial step in Active Shape Modelling, which has encountered several challenges. The most crucial among these are: (C1) defining and characterizing landmarks; (C2) obtaining landmarks at the desired level of detail; (C3) ensuring homology; (C4) generalizing to n>2 dimensions; (C5) achieving practical computations. This thesis proposes several novel modelling techniques attempting to meet C1-C5. In this process, this thesis makes the following key contributions: the concept of local scale for shapes; the idea of allowing level of detail for selecting landmarks; the concept of equalization of shape variance for selecting landmarks; the idea of recursively subdividing shapes and letting the sub-shapes guide landmark selection, which is a very general n-dimensional strategy; the idea of virtual landmarks, which may be situated anywhere relative to, not necessarily on, the shape boundary; a new compactness measure that considers both the number of landmarks and the number of modes selected as independent variables. The first of three methods uses the c-scale shape descriptor, based on the new concept of curvature-scale, to automatically locate mathematical landmarks on the mean of the training shapes. The landmarks are propagated to the training shapes to establish correspondence among shapes. Since all shapes of the same family do not necessarily present exactly the same shape features, another novel method was devised that takes into account the real shape variability existing in the training set and that is guided by the strategy of equalization of the variance observed in the training set for selecting landmarks. By incorporating the above basic concepts into modelling, a third family of methods with numerous possibilities was developed, taking into account shape features, and the variability among shapes, while being easily generalized to the 3D space. Its output is multi-resolutional allowing landmark selection at any lower resolution trivially as a subset of those found at a higher resolution. The best strategy to use within the family will have to be determined according to the clinical application at hand. All methods were evaluated in terms of compactness on two data sets - 40 CT images of the liver and 40 MR images of the talus bone of the foot. Further, numerous artificial shapes with known salient points were also used for testing the accuracy of the proposed methods. The results show that, for the same number of landmarks, the proposed methods are more compact than manual and equally spaced annotations. Besides, the accuracy (in terms of false positives and negatives and the location of landmarks) of the proposed shape descriptor on artificial shapes is considerably superior to a state-of-the-art scale space approach to finding salient points on shapes.

This thesis presents a study on rotation in 2 dimensional and 3 dimensional discrete spaces. In computer science, using floating numbers is problematic due to computation errors. Thus we chose during this thesis to work only in discrete space. In the field of computer vision, the rotation is a transformation required for many applications. Using discretized Euclidean rotation gives bad results. Then, it is necessary to develop new rotation methods adapted to the discrete spaces. We mainly studied the hinge angles that represent the discontinuity of the rotation in the discrete space. Indeed, it is possible to perform two rotations of the same digital image with two angles that are slightly different and obtain the same result. This is captured by hinge angles. Using these angles allow us to describe a discrete rotation that gives the same results than the discretized Euclidean rotation without using floating numbers. They also allow describing an incremental rotation that performs all possible rotations of a given digital image. Using hinge angles can also be extended to the rotations in 3 dimensional discrete spaces. The extension requires the multi-grids that are rotation planes containing three sets of parallel lines. These parallel lines represent the discontinuities of the rotation in 3D discrete space. Thus they are useful to describe the hinge angles in rotation planes. Multi-grids allow obtaining the same results in 3D discrete rotations than the results obtained in 2D discrete rotations. This thesis presents a study on rotation in 2 dimensional and 3 dimensional discrete spaces. In computer science, using floating numbers is problematic due to computation errors. Thus we chose during this thesis to work only in discrete space. In the field of computer vision, the rotation is a transformation required for many applications. Using discretized Euclidean rotation gives bad results. Then, it is necessary to develop new rotation methods adapted to the discrete spaces. We mainly studied the hinge angles that represent the discontinuity of the rotation in the discrete space. Indeed, it is possible to perform two rotations of the same digital image with two angles that are slightly different and obtain the same result. This is captured by hinge angles. Using these angles allow us to describe a discrete rotation that gives the same results than the discretized Euclidean rotation without using floating numbers. They also allow describing an incremental rotation that performs all possible rotations of a given digital image. Using hinge angles can also be extended to the rotations in 3 dimensional discrete spaces. The extension requires the multi-grids that are rotation planes containing three sets of parallel lines. These parallel lines represent the discontinuities of the rotation in 3D discrete space. Thus they are useful to describe the hinge angles in rotation planes. Multi-grids allow obtaining the same results in 3D discrete rotations than the results obtained in 2D discrete rotations

Tiivistelmä. Siltojen tulopenkereiden stabiliteettilaskenta tehdään pääosin 2D-menetelmällä, jossa laskentapoikkileikkauksen oletetaan jatkuvan äärettömän pitkänä. 2D-laskenta soveltuu kuitenkin penkereen pituussuuntaisen stabiliteetin tarkasteluun huonosti, koska kuormituksen ja penkereen paikallisuutta ei pystytä huomioimaan 2D-mallissa. Diplomityön tavoitteena oli pyrkiä hahmottamaan, kuinka paljon 2D-laskentamenetelmillä saadut tulokset poikkeavat 3D-laskentamenetelmiin verrattuna ja mitkä tekijät vaikuttavat voimakkaimmin eri laskentamenetelmillä saatujen tulosten välisiin eroihin. Lisäksi tavoitteena oli pyrkiä antamaan laskentatulosten perusteella suositus päätyvastuksen huomioimisesta 2D LEM menetelmässä. Työssä vertailtiin 2D LEM ja 3D LEM -menetelmillä saatujen kokonaisvarmuuskertoimien eroja penkereen pituussuunnassa vaihtelevan muotoisilla liukupinnoilla. Tarkastelu tehtiin vesistö- ja alikulkukohteeseen kolmessa pohjamaatyypissä ja kolmella pengerleveydellä käyttäen Bishopin yksinkertaistettua menetelmää. Tarkasteltavat penkereet olivat yksi- ja kaksiraiteinen ratapenger sekä leveä moottoritiepenger. Vakavuus laskettiin sekä pelkälle penkereelle että penkereelle, jonka päällä oli pintakuorma. Kahdessa pohjamaatyypissä käytettiin suljetun tilan parametreja (löyhä ja sitkeä savi) ja yhdessä avoimen tilan parametreja (siltti). 2D LEM -menetelmän varmuuskertoimet laskettiin ilman päätyvastusta ja päätyvastuksen kanssa. Valituista tapauksista tehtiin vertailulaskelmat 3D FEM SRM -menetelmällä. Paalulaatan pituuden mitoitusta tutkittiin yhdessä tapauksessa 2D LEM ja 3D LEM -laskentamenetelmien välillä. Vertailulaskelma eurokoodin DA3 MRT -osavarmuuskertoimia soveltaen laadittiin yhdelle laskentatapaukselle. Kapeiden penkereiden tapauksessa 2D- ja 3D-laskentamenetelmillä saatujen varmuuskertoimien välille muodostui suuria eroja kummassakin kohteessa, kun käytettiin suljetun tilan parametrien pohjamaatyyppejä. 2D LEM menetelmällä saatiin näissä tapauksissa merkittävästi 3D-menetelmiä pienempiä varmuuskertoimia. Suurimmillaan 3D LEM -menetelmällä saatu varmuuskerroin oli yli kaksinkertainen verrattuna 2D LEM menetelmän varmuuskertoimeen ilman päätyvastusta. Ilman penkereen päälle tulevaa pintakuormaa, 2D LEM -menetelmällä päätyvastus huomioiden saatiin valituilla parametreilla likimain yhtä suuria varmuuskertoimia kuin 3D LEM -menetelmällä vapaan muotoisilla liukupinnoilla. Kun penkereen päällä oli lisäksi pintakuorma, 2D LEM menetelmällä päätyvastuksen kanssa saatiin pienempiä varmuuskertoimia kuin 3D LEM -menetelmällä vapaan muotoisilla liukupinnoilla. Leveimmällä penkereellä erot eivät olleet enää pääsääntöisesti merkittäviä. Penkereen leveyden vaikutus varmuuskertoimien eroihin oli selkeä: kapeimmalla, yksiraiteisella ratapenkereellä 2D ja 3D -laskentamenetelmien tulosten välinen ero oli suurin ja se pieneni pengerleveyden kasvaessa. Myös pintakuorma vaikutti varmuuskertoimien erojen suuruuteen kapeimmilla penkereillä: varmuuskertoimien ero oli suurempi penkereellä pintakuorman kanssa kuin penkereellä ilman pintakuormaa. Pohjamaan suuremmalla suljetun leikkauslujuuden arvolla ei ollut selvää vaikutusta laskentamenetelmien välisten varmuuskertoimien erojen pienenemiseen. Siltti-pohjamaatyypillä varmuuskertoimien erot olivat selvästi pienemmät kuin suljetun tilan parametrien pohjamaatyypeillä. Elementtimenetelmällä saatujen vertailulaskelmien varmuuskertoimet noudattelivat kaikissa lasketuissa tapauksissa 3D LEM -menetelmän vapaan muotoisten liukupintojen varmuuskertoimia. Paalulaatan pituustarkastelussa paalulaatan pituudelle tuli merkittäviä eroja, kun pituuden määrittämisen perusteena käytettiin 1,8 kokonaisvarmuuskerroinvaatimusta. Vaadittava paalulaatan pituus vaihteli 10 m ja yli 55 m välillä. Lyhyin paalulaatta olisi vaadittu 3D LEM -menetelmän perusteella ja pisin 2D LEM -menetelmän perusteella ilman päätyvastusta. Varmuuskertoimien ero vaihteli 2D LEM ja 3D LEM -menetelmien välillä kuormitustapauksesta riippuen myös siinä tapauksessa, kun 2D LEM -menetelmässä huomioitiin päätyvastus. Päätyvastuksen laskentaan ei siis ole olemassa yhdenmukaisia vakioparametreja, joilla saataisiin 3D LEM -menetelmää vastaavia varmuuskertoimia kaikissa tapauksissa. Paalulaatan pituustarkastelussa 2D LEM -menetelmällä päätyvastuksen kanssa, varmuuskerroin kasvoi jyrkästi paalulaatan pituuden funktiona nousten nopeasti suuremmaksi kuin 3D LEM -menetelmän varmuuskerroin vapaan muotoisilla liukupinnoilla. Lisätutkimustarpeena havaittiin muun muassa kerroksellisen maaperän ja maaperän lujuuden vaikutuksen sekä pengerleveyden ja kuormituksen yhteyden tarkempaa selvittämistä päätyvastusparametrien valintaan. Lisäksi 3D LEM -menetelmän soveltuvuus stabiliteetin laskentaan epäsymmetrisissä tapauksissa vaikeissa pohjaolosuhteissa olisi syytä selvittää.Comparison of 2D and 3D slope stability analysis methods in the case of bridge approach embankments. Abstract. Slope stability analysis of bridge approach embankments is done mainly by using 2D methods, which suppose calculation cross section to be infinitely long. However, 2D methods are not ideal for studying the embankment’s stability in longitudinal direction, because it is impossible to model the locality of an embankment and loading in a 2D model. The aim of this master’s thesis work was to study how much the results of 2D methods diverge compared to more realistic 3D situation, and which factors have the greatest influence on the difference. Another goal was to give recommendations for the definition of end effect in 2D LEM method. In this work, differences in factor of safety were compared in the longitudinal direction of embankment obtained by 2D LEM and 3D LEM methods. Varying slip surface shapes were utilized in a watercourse and an underpass case, using Bishop simplified method with three different soil types and three different embankments widths. The investigated embankments were a single-track railway, a double-track railway, and a wide motorway embankment. Factor of safety was calculated with and without a surface load. Undrained shear strength parameters were used with two soil types (soft and firm clay) and in one case friction and cohesion was used for subsoil (silt). Factor of safety was calculated in 2D LEM method with and without an end effect. For selected cases, reference calculations were performed with 3D FEM SRM method. Dimensioning of pile slab length was studied in one case using 2D LEM and 3D LEM methods. Reference calculation using Eurocode DA3 MRT partial factors was conducted in one case. There were big differences in the factor of safety between the 2D and 3D methods in both watercourse and underpass cases with the two narrowest embankments, when undrained shear strength soil types were used. In these cases, factor of safety values calculated with a 2D LEM method were significantly lower than with 3D methods. At the maximum, factor of safety calculated with a 3D LEM method was more than twice greater compared to 2D LEM method without an end effect. When there was no surface load on top of the embankment, the factors of safety calculated using a 2D LEM method with an end effect were approximately the same compared to the values obtained with 3D LEM method using non-spherical slip surfaces. When there was a surface load on the top embankment, factor of safety calculated using a 2D LEM method with an end effect was lower than the factor of safety calculated with 3D LEM method using non-spherical slip surfaces. In the wide motorway embankment, the difference was not significant. The impact of embankment width to the differences in factors of safety was obvious: the difference was the greatest in the narrowest, single-track embankment, and it became smaller when the width of the embankment increased. Also, the surface load had an impact on the differences in factors of safety in railway embankments: the difference was greater in an embankment with the surface load than with an embankment without the surface load. An increase of undrained shear strength of soil did not have an explicit impact on the reduction of safety factor difference between calculation methods. In the silt subsoil, the differences in factors of safety were clearly smaller than when utilizing undrained shear strength. Factors of safety calculated using a finite element method were close to the 3D LEM values with non-spherical slip surfaces in all calculated cases. The length of pile supported embankment slab varied a lot when a requirement of 1.8 was used for stability safety factor. The required length of pile slab varied from 10 m to over 55 m. The shortest pile slab length was based on 3D LEM method and the longest based on 2D LEM method without end effect. The difference in safety factor between the 2D LEM and 3D LEM methods varied depending on loading case, also when end effect was considered in the 2D LEM method. Thus, there are no constant parameters in 2D LEM end effect calculation that would produce factor of safety corresponding to 3D LEM method. In a pile slab length study, greater factor of safety values were obtained using 2D LEM method with end effect than with 3D LEM method using non-spherical slip surfaces. More research is needed to determine the impact of stratified soil base, impact of soil strength, as well as impact of loading intensity on determination of end effect calculation parameters. Also, the suitability of 3D LEM method in asymmetric cases in loose soil base should be clarified.

Cardiovascular disease is one of the leading causes of the morbidity and mortality in the western world today. Many different imaging modalities are in place today to diagnose and investigate cardiovascular diseases. Each of these, however, has strengths and weaknesses. There are different forms of noise and artifacts in each image modality that combine to make the field of medical image analysis both important and challenging. The aim of this thesis is develop a reliable method for segmentation of vessel structures in medical imaging, combining the expert knowledge of the user in such a way as to maintain efficiency whilst overcoming the inherent noise and artifacts present in the images. We present results from 2D segmentation techniques using different methodologies, before developing 3D techniques for segmenting vessel shape from a series of images. The main drive of the work involves the investigation of medical images obtained using catheter based techniques, namely Intra Vascular Ultrasound (IVUS) and Optical Coherence Tomography (OCT). We will present a robust segmentation paradigm, combining both edge and region information to segment the media-adventitia, and lumenal borders in those modalities respectively. By using a semi-interactive method that utilizes "soft" constraints, allowing imprecise user input which provides a balance between using the user's expert knowledge and efficiency. In the later part of the work, we develop automatic methods for segmenting the walls of lymph vessels. These methods are employed on sequential images in order to obtain data to reconstruct the vessel walls in the region of the lymph valves. We investigated methods to segment the vessel walls both individually and simultaneously, and compared the results both quantitatively and qualitatively in order obtain the most appropriate for the 3D reconstruction of the vessel wall. Lastly, we adapt the semi-interactive method used on vessels earlier into 3D to help segment out the lymph valve. This involved the user interactive method to provide guidance to help segment the boundary of the lymph vessel, then we apply a minimal surface segmentation methodology to provide segmentation of the valve.

The field of biomaterials has seen huge development over the past decade with enormous efforts invested in discovering materials with improved biocompatibility, application and versatility. Polymers can display many properties that make them ideal biomaterials, such as their potential flexibility, low weight, low cost and biodegradability. Moreover, they can be prepared in a wide variety of compositions and forms and be readily fabricated into various shapes and structures. Polymer microarrays represent an efficient high-throughput platform for the screening and discovery of new materials compared to conventional assays with advantages such as high-density screening, internal consistency of assays and the requirement for only small quantities of material. The first part of this thesis describes work in the area of diabetes research with a focus on how dysfunctional β-cells could be replaced by the transplantation of β-cells obtained from pluripotent stem cells. To achieve this aim, high numbers of β-cells are required. A polymer microarray screening approach was used to identify a number of polymers that promoted the attachment of pancreatic progenitor cells and enhanced cell proliferation. Multiple scale-up fabrication techniques were assessed to establish the most suitable approach and surface for long term cell culture leading to the obtainment of reproducible in situ polymerised polymer layers with enhanced binding properties toward pancreatic progenitor cells. These surfaces have the potential to support cell adhesion and proliferation and could find potential use in the industrial sector to increase the production of pancreatic progenitor cells in vitro. In the second part, efforts were made to gain a better understanding of the maturation of β-cells and their behaviour, with the development of 3D hydrogels based on the previously identified polymers. In this scenario, parameters such as stiffness and porosity were evaluated to identify the best environmental conditions to support 3D cell culturing of pancreatic progenitor cells. Several approaches were tested to generate scaffolds with suitable stiffness and porosity leading to the obtainment of scaffolds based on the previously identified polymer composition and with controlled porosity and stiffness. These scaffolds could represent a suitable environment to allow a better understanding of cell organisation and regulation. In a third avenue of work, arrays of 3D biocompatible materials, which were tailored for varying elasticity, hardness, and porosity (to provide the necessary physical cues to control cellular functions) were fabricated. In this chapter, details of the development of an array of eighty 3D double-network hydrogel features are reported. The array features can be produced as single or double networks and modulated in terms of stiffness, viscoelasticity and porosity to assess cell response to materials with a wide range of properties. The final part of the thesis describes the development and screening of polymeric materials to allow a better understanding of cell–surface interactions with various cell types. To investigate the correlation between cell attachment and the nature of the polymer, a series of random and block copolymers were synthesised and examined for their abilities to attach and support the growth of human cervical cancer cells (HeLa) and human embryonic kidney cells (HEK293T), with attachment modelled on monomer ratios, arrangement, and polymer chain length. The results of this screening showed differences between block copolymers and random copolymers in cell adhesion and provide interesting insight into the improvement of polymer coatings for cell culture.

Attributed Relational Graph (ARG) is a powerful representation for model based object recognition due to its inherent robustness in handling noisy and incomplete data. In the past few years, the availability of efficient ARG matching algorithms and their theoretical underpinnings have greatly contributed to many successful applications of ARG representation in tackling high level vision problems. During my past three year investigation into object recognition using ARG representation, we have developed a number of novel theories and techniques in the subject area. Some are image processing techniques which help to segment and generate primitive features for building ARG representation (Chapter 2 and 4). Some are about projective invariance in ARG representations (Chapter 3 and 5). Some are about new ARG matching algorithms (Chapter 6). This thesis serves as a summary document of these theories and techniques. The most important contributions of our work to the domain of computer vision, in my opinion, are in two areas: Firstly, in the area of projective invariant ARG representation for object recognition. Here, we demonstrated for the first time, a way to systematically derive ARG representation for objects under complex projective transform by exploiting the knowledge of invariance. The methodology developed by us is a sound strategy that generates ARG representations with a number of desirable and provable properties, amongst which, the most important one is the ability to capture global transformation constraint using binary relations only. The approach significantly reduces the heuristic nature of designing relational measurements and paves the way for wider application of ARG representation in 2D and 3D object recognition. Secondly, in the area of ARG matching. A new mathematical framework for deterministic relaxation algorithms was developed to overcome a number of problems appeared in the existing theories and practises of efficient ARG labelling. A novel labelling algorithm was proposed based on the new theoretical framework. The algorithm has a number of desirable properties compared to existing algorithms. In particular, the resulting algorithm delivers more consistent, faithful-to-observation results in the presence of ambiguities and multiple interpretations compared to other algorithms.

Dans cette thèse, nous abordons les défis de la rareté des annotations et la fusion de données hétérogènes tels que les nuages de points 3D et images 2D. D’abord, nous adoptons une stratégie de conduite de bout en bout où un réseau de neurones est entraîné pour directement traduire l'entrée capteur (image caméra) en contrôles-commandes, ce qui rend cette approche indépendante des annotations dans le domaine visuel. Nous utilisons l’apprentissage par renforcement profond où l'algorithme apprend de la récompense, obtenue par interaction avec un simulateur réaliste. Nous proposons de nouvelles stratégies d'entraînement et fonctions de récompense pour une meilleure conduite et une convergence plus rapide. Cependant, le temps d’apprentissage reste élevé. C'est pourquoi nous nous concentrons sur la perception dans le reste de cette thèse pour étudier la fusion de nuage de points et d'images. Nous proposons deux méthodes différentes pour la fusion 2D-3D. Premièrement, nous projetons des nuages de points LiDAR 3D dans l’espace image 2D, résultant en des cartes de profondeur éparses. Nous proposons une nouvelle architecture encodeur-décodeur qui fusionne les informations de l’image et la profondeur pour la tâche de complétion de carte de profondeur, améliorant ainsi la résolution du nuage de points projeté dans l'espace image. Deuxièmement, nous fusionnons directement dans l'espace 3D pour éviter la perte d'informations dû à la projection. Pour cela, nous calculons les caractéristiques d’image issues de plusieurs vues avec un CNN 2D, puis nous les projetons dans un nuage de points 3D global pour les fusionner avec l’information 3D. Par la suite, ce nuage de point enrichi sert d'entrée à un réseau "point-based" dont la tâche est l'inférence de la sémantique 3D par point. Sur la base de ce travail, nous introduisons la nouvelle tâche d'adaptation de domaine non supervisée inter-modalités où on a accès à des données multi-capteurs dans une base de données source annotée et une base cible non annotée. Nous proposons une méthode d’apprentissage inter-modalités 2D-3D via une imitation mutuelle entre les réseaux d'images et de nuages de points pour résoudre l’écart de domaine source-cible. Nous montrons en outre que notre méthode est complémentaire à la technique unimodale existante dite de pseudo-labeling
In this thesis, we address the challenges of label scarcity and fusion of heterogeneous 3D point clouds and 2D images. We adopt the strategy of end-to-end race driving where a neural network is trained to directly map sensor input (camera image) to control output, which makes this strategy independent from annotations in the visual domain. We employ deep reinforcement learning where the algorithm learns from reward by interaction with a realistic simulator. We propose new training strategies and reward functions for better driving and faster convergence. However, training time is still very long which is why we focus on perception to study point cloud and image fusion in the remainder of this thesis. We propose two different methods for 2D-3D fusion. First, we project 3D LiDAR point clouds into 2D image space, resulting in sparse depth maps. We propose a novel encoder-decoder architecture to fuse dense RGB and sparse depth for the task of depth completion that enhances point cloud resolution to image level. Second, we fuse directly in 3D space to prevent information loss through projection. Therefore, we compute image features with a 2D CNN of multiple views and then lift them all to a global 3D point cloud for fusion, followed by a point-based network to predict 3D semantic labels. Building on this work, we introduce the more difficult novel task of cross-modal unsupervised domain adaptation, where one is provided with multi-modal data in a labeled source and an unlabeled target dataset. We propose to perform 2D-3D cross-modal learning via mutual mimicking between image and point cloud networks to address the source-target domain shift. We further showcase that our method is complementary to the existing uni-modal technique of pseudo-labeling

L’imagerie biologique a réalisé des progrès significatifs durant les dernières décennies. Les récentes innovations portent sur la manipulation et la visualisation de cellules uniques avec une résolution spatiale de l’ordre du nanomètre. Une technologie d’imagerie récente, l’imagerie «sans lentille », est particulièrement prometteuse car elle combine une bonne résolution spatiale, un champ de vision étendu, une simplicité d’utilisation, un coût abordable et la possibilité de travailler sur des échantillons exempts de marqueurs spécifiques. En imagerie sans lentille, le système optique classiquement utilisé pour constituer l’image de l’échantillon est remplacé par des algorithmes informatiques qui s’appuient sur les propriétés de cohérence spatiale de la lumière. Dans cette thèse, deux approches différentes de microscopie sans lentille sont considérées : l’holographie numérique en ligne et l’holographie par transformée de Fourier.Deux prototypes d’imagerie, construits selon ces principes, sont présentés. Ils offrent une résolution de l’ordre du micron, ainsi que la possibilité de retrouver les informations relatives à l’amplitude spatiale et à la phase du champ optique. Cela permet la réalisation de reconstructions pseudo-3D d’objets volumétriques à partir d’un unique hologramme. Les deux dispositifs ont d’abord été caractérisés avec des échantillons de référence. Par la suite, des expériences d’applications ont été testées pour estimer la capacité des dispositifs à répondre à des problématiques concrètes dans le domaine de la biologie, grâce à la haute résolution, l’imagerie en temps réel et la reconstruction 3D.L’objectif de cette thèse est également de développer une nouvelle plateforme qui intègre, dans une puce microfluidique, d’une part un système permettant la manipulation de cellules par diélectrophorèse, et d’autre part un masque optique pour la visualisation des cellules par imagerie sans lentille. Le principe de fonctionnement est basé sur le déplacement des cellules en milieu liquide et la séparation des cellules dans le champ de vision du microscope en utilisant un champ électrique induit par des électrodes spécifiques. Le masque optique permet de définir le champ de vision du microscope et de créer les faisceaux de référence nécessaires pour l’imagerie par holographie par transformée de Fourier. Le principal avantage de ce système électro-optique pour l’imagerie cellulaire réside dans sa capacité à fournir une plateforme d’imagerie compacte qui regroupe précision et sensibilité. Les champs d’applications de cette plateforme sont variés. Une application concrète qui découle immédiatement des premières expériences présentées dans ce manuscrit serait l’analyse du comportement des cellules et de leurs modifications morphologiques lors d’un processus électrochimique de diélectrophorèse.L’un des challenges majeurs dans le domaine de la microscopie est de réduire les coûts de fabrication. Les deux types de microscopes sans lentille présentés dans cette thèse visent à introduire dans le monde scientifique des outils d’imagerie permettant d’obtenir une haute résolution à un faible coût et sans marquage. Par ailleurs, la puce microfluidique est une première démonstration de plateforme intégrée pour l’analyse des cellules en temps réel dans un dispositif de type « Lab-on-a-chip »
Biological imaging has made tremendous progresses these last decades. The latest developments concern manipulating and imaging single cells with nanometer spatial resolutions. A recent category of imaging techniques, called lensless microscopy, are very promising because they combine very good spatial resolutions in a large field-of-view, simplicity of use and low cost, while operating on label free samples. In this thesis two different lensless approaches are considered: digital in-line holography (DILH) and Fourier transform holography (FTH). In lensless imaging, the usual optical system used to form the sample’s image are remove and replace by numerical algorithms using the light spatial coherence properties.Two imaging prototypes, built on these principles, are presented. They offer (sub ) micrometer scale resolutions, and offer the possibility to retrieve both spatial amplitude and phase information of the optical field. This allows to achieve pseudo-3D reconstruction of volumetric objects from a single 2D hologram. Both devices were first characterized with reference samples. In a second step, real applications, relevant to selected biological problems, were performed to assess the devices’ performances towards high resolution, real time imaging and 3D.This thesis objective is also to develop a new platform directly integrating in a single chip a microfluidics system for biological cell handling by dielectrophoresis and an optical mask for cell visualization by lensless microscopy. Its working principle is based on cell transport in a liquid media by microfluidics, cell separation in the microscope field of view by the electric field induced by specific electrodes, and simultaneous cell imaging by Fourier Transform Holography. The main advantage of such coupled electro-optical system for cell imaging and analysis are the improved control, the precision and sensitivity regarding cell morphology all together merged in a compact imaging platform. The capability of the platform can be extended to analysis of cells’ behavior and morphologic deviation during the electrochemical processes of DEP.A major challenge in microscopy field is to reduce the production costs. The two types of lensless microscopy presented in this thesis aims to introduce new imaging tools that allows scientists to obtain low-cost high-resolution images in label-free conditions. Additionally, the microfluidics chip is a first demonstration of a new integrated platform for cell live analysis into a single Lab-on-a-chip device

Abstract: Oil pollution and spills cause serious damage to marine ecosystems and coastal environments. Currently, oily waters recuperated form a spill must be shipped onshore for treatment. This limits the volume of water that can be treated during a spill. There is a need to develop technologies to treat oily waters below 15 ppm (parts per million) at the site of the spill. Synthetic membrane technologies are widely used in water treatment and purification. They can offer an on-site solution to contaminated oily water treatment in oil production and spills. The suitability of a membrane for use in this application is determined by the type of material used in its fabrication. Compared to polymeric membranes, inorganic membranes are inert to microbiological degradation, offer high chemical and thermal resistance, and can easily be backflushed and cleaned once fouled. However, inorganic membranes consisting of metal oxides are heavier and more expensive than polymeric membranes, due to their bulky and brittle ceramic support layers. This limits their application when the overall weight of a process unit is of concern. A newly developed 2D/3D material, named twinned alumina nanosheets (TAN), has recently been used to make dynamic membranes. The nanoplatelets forming TAN have a length of 4 µm, a width of 1 µm, and a thickness of 100 nm. They have a very high permeability, a 0.2 µm-pore size and a porosity up to 88% due to their low nanosheet volume. These unique characteristics make TAN a very promising material to form membrane selective layers. However, they must be supported on a very open layer in order to take advantage of their high porosity. In this work, a composite membrane was produced with a selective layer of 2D/3D alumina nanoplatelets deposited onto stainless steel meshes and ceramic supports. The structure of the TAN in the selective layer was reinforced with binders. The main objective of this work was to verify the adhesion of the TANs onto the support. The crystallization of TAN was optimized to obtain an open 2D/3D structure. This structure was then deposited on a stainless-steel mesh. The mesh was pretreated by electrochemical etching to achieve a re-entrant surface. The mesh was immersed in an etching solution and placed parallel to a conductive graphite plate under a constant electric potential of 5V for 4 min. Aqueous solutions of silica sol and colloidal silver were tested as binding agents. They were deposited on the mesh with TAN and sintered for 4 hrs. Experiments were performed on testing stainless steel meshes with different opening sizes and comparing different calcination temperatures. The best sintering temperature was 800°C for a mesh with an opening size of 35µm. The synthesized membrane was challenged with a suspension of 10 ppm bentonite clay at a constant pressure of 100 mbar. The integral structure of a TAN membrane produced with a 2.5wt% silica binder was maintained after backflushing. The 2.5wt% silica membrane had a high flux and the particle filtration process for this membrane was modelled as pore constriction and intermediate blocking, indicating that backflushing provided the deep cleaning of pores. According to the SEM images, the 2.5wt% silica membrane preserved the integral structure of the TAN, while the pores tended to fill with silica at higher silica concentrations. The effective pore size of the 2.5wt% silica membrane was estimated to be the smallest, which is approximately 0.53 μm. The 7.5wt% silica membrane had half the permeate flux of the other membranes, because of the high concentration of binder filling the pores of the TAN selective layer. The SiO 2 binder had a positive effect in reinforcing the TAN particles. The flux of the membrane did not increase after backflushing indicating that the selective layer of the membrane was securely bound to the stainless steel mesh. The membrane exhibited flux decline between backflushings indicating that particles were retained on its surface. SEM images taken after the filtration showed that this membrane completely released bentonite particles form its pores. Tests were also performed with a membrane having two TAN coatings on the wire mesh. This reduced the flux but did not improve the retention of fine particles. Colloidal silver was found to be a poor binding agent as particles were released particles from its selective layer. Silica was a highly successful binding agent while colloidal silver was not. TAN was also successfully deposited onto ceramic supports. It was also retained on top of the membrane after backflushing. The results of this work demonstrate that TANs reinforced and bound with silica are a promising type of material to form membrane selective layers. These layers have an open pore structure with a three-dimensional channel connectivity on both stainless steel and ceramic supports. The selective layer was successfully bound to the stainless steel supports. If the pore size of this membrane were to be reduced, it would meet the requirements for use at the site of an oil spill to treat contaminated waters as it does not need the heavier supports found in traditional ceramic membranes. Résumé: La pollution et les déversements d'hydrocarbures causent de graves dommages aux écosystèmes marins et aux environnements côtiers. À l'heure actuelle, les eaux huileuses récupérées d'un déversement doivent être expédiées à terre pour leur décontamination. Ceci limite le volume d’eau contaminé qui peut être traité. Il est nécessaire de développer des technologies permettant de traiter les eaux huileuses en dessous de 15 ppm (parties par million) sur le site du déversement. Les technologies membranaires sont largement utilisées dans le traitement et la purification de l'eau. La possibilité de se servir d’une membrane dans cette application est déterminée par les matériaux utilisés dans sa fabrication. Comparées aux membranes polymères, les membranes inorganiques sont inertes vis-à-vis de la dégradation microbiologique, offrent une résistance chimique et thermique élevée et peuvent facilement être rincées et nettoyées une fois encrassées. Cependant, les membranes inorganiques constituées d'oxydes métalliques sont plus lourdes et plus coûteuses que les membranes polymères, en raison de leurs couches de support en céramique volumineuses et cassantes. Cela limite leur application lorsque le poids total d'une unité de traitement est préoccupant. Un matériau 2D/3D récemment développé, appelé TAN (Twinned Alumina Nanosheets), a récemment été utilisé dans la formation de membranes dynamiques. Les nano-plaquettes formant les TAN ont une longueur de 4 µm, une largeur de 1 µm et une épaisseur de 100 nm. Ils ont une très haute perméabilité, une taille de pores de 0,2 µm et une porosité allant jusqu'à 88% en raison du faible volume des nanofeuilles. Ces caractéristiques uniques font du TAN un matériau très prometteur pour la formation de couches sélectives de membranes. Cependant, ils doivent être déposes sur une couche très ouverte afin de tirer parti de leur grande porosité. Au cours de ce travail, une membrane composite a été réalisée avec une couche sélective de nanoplaques d’alumine 2D / 3D (TAN) déposées sur deux types de supports; des mailles en acier inoxydable et des supports en céramique. La structure du TAN dans la couche sélective a été renforcée avec des liants. L'objectif principal de ce travail était de vérifier l'adhérence des TAN sur le support. La cristallisation des TAN a été optimisée pour obtenir une structure 2D/3D ouverte. Cette structure a ensuite été déposée sur un treillis en acier inoxydable. Les mailles ont été prétraitées pour obtenir une surface réentrante. Le maillage a été immergé dans une solution de gravure et placé parallèlement à une plaque de graphite conductrice sous un potentiel électrique constant de 5 V pendant 4 min. Des solutions aqueuses de sol de silice et d’argent colloïdal ont été testées en tant que liants. Ils ont été déposés sur la maille et frittés pendant 4 heures. Des expériences ont été effectuées sur des mailles en acier inoxydable avec différentes tailles d’ouverture et températures de calcination. La meilleure température de frittage était de 800 ° C pour un treillis ayant une taille d'ouverture de 35 µm. La membrane synthétisée a été mise à l’essai avec une suspension de 10 ppm d'argile bentonite à une pression constante de 100 mbar. La structure intégrale de la membrane couche de TAN produite avec un liant à 2,5wt% de silice a été maintenue après les tests de perméabilité. La structure 3D poreuse a tendance à se remplir de silice à des concentrations de silice supérieures à 2,5wt%. La taille effective des pores de la membrane produite avec 2,5wt% de liant de silice a été estimée à 0,53 µm. Le flux de la membrane n'a pas augmenté après le rinçage, indiquant que la couche sélective de la membrane était liée de manière sûre au maillage en acier inoxydable. La membrane présentait un déclin de flux entre les rinçages indiquant que des particules étaient retenues à sa surface. Les images au microscope à balayage prises après la filtration ont montré que cette membrane libère complètement les particules de bentonite de ses pores. Des essais ont également été réalisés avec une membrane comportant deux revêtements TAN sur le treillis métallique. Cela réduit le flux mais n'améliore pas la rétention des particules fines. L'argent colloïdal s'est avéré être un agent de liaison médiocre car des particules sont libérées de sa couche sélective. La silice était un liant très efficace, contrairement à l'argent colloïdal. Le TAN a également été déposé avec succès sur des supports en céramique. Il est également resté sur la membrane après le rinçage à contre-courant. Les résultats de ce travail démontrent que les TAN renforcés avec un liant de silice sont un type de matériau prometteur pour former des couches sélectives, avec des structures à pores ouverts possédant une connectivité de canal tridimensionnelle, sur des supports en acier inoxydable et en céramique. La couche sélective a été liée avec succès au support en acier inoxydable. Si la taille des pores de cette membrane devait être réduite, elle pourrait être utilisée sur le site d'un déversement d'hydrocarbures pour traiter les eaux contaminées car elle ne nécessite pas les supports plus lourds que l'on trouve dans les membranes de céramique traditionnelles.

This thesis investigates robust and fast methods for single and cooperative 2D/3D image mosaicing to enhance field of view of images by joining them together. Image mosaicing is underlined by the process of image registration and a significant portion of the contributions of this work are dedicated to it. Image features are identified as a solution to the problem of image registration that uses feature-to-feature matching between images to solve for inter-image transformations. We have developed a novel two signature distribution based feature descriptor that combines grey level gradients and a colour histogram. This descriptor is robust to illumination changes and shows better matching accuracy compared to state of the art. Furthermore, we introduce a feature clustering technique that uses colour codes assigned to each feature to group them together. This allows fast and accurate feature matching as the search space is reduced. Taking into account feature location uncertainty we have introduced a novel information fusion technique to reduce this error by covariance intersection. This reduced error location is consequently fed to an H∞ filter taking into account system uncertainty for parameter estimation. We show that this technique outperforms costly nonlinear optimisation techniques. We have also developed a novel coupled filtering scheme based on H∞ filtering that estimates inter-image geometric and photometric transformations simultaneously. This is shown to perform better than standard least square techniques. Furthermore, we have introduced time varying parameter estimation using recursive techniques that facilitate in tracking changing parameters of inter-image transformations, suitable for image mosaicing between moving platforms. A method for rapid 3D scene reconstruction is developed that uses homographic lines between images for semi-dense pixel matching. Triangular meshes are then used for a complete visualisation of the scene and to fill in the gaps. To tackle cooperative mosaicing scenarios, additional methods are presented that include descriptor compression using principal components and 3D scene merging using the trifocal tensor. Capabilities of the proposed techniques are illustrated with real world images.

Kunderna i en webbshop kan ibland uppleva den sortens inköp som diffusa och osäkra. Osäkerheten beror på att man inte kan testa, känna eller se produkten i verkligheten. Produkterna i webbshopar presenteras ofta i 2D-format i form av ett foto och visas då i en vinkel. Genom att tillföra webbshopen nya egenskaper som att kunna undersöka varorna och titta på dem från olika håll, kan kundernas osäkerhet minskas.

Valet av produkt föll i det här arbetet på möbler på grund av att de utgör en varugrupp som anses som svåra att handla via nätet.

Frågeställningen har varit om 3D-format, i motsats till 2D-format, kan leda till att kunderna lättare kan avgöra ett inköpsbeslut i en webbshop genom att känslan för hur möblerna ser ut ökas? Uppfattningen om möblerna har i det här fallet ökats genom en interaktiv 3D-prototyp. Med hjälp av prototypen kan möblerna undersökas genom att vrida, förstora, öppna, stänga och ändra utföranden på dem. På detta sätt får kunderna ut mer information om möblernas utseende, storlek och funktion än från en traditionell webbshop i 2D. Prototypen, som är tänkt att användas i samband med en webbutik, skapades för att ett användartest skulle kunna utföras. Testet gick ut på att undersöka frågeställningen.

Resultatet av undersökningen visar att drygt hälften av testdeltagarna ansåg att de fått så pass mycket information från prototypen, att de kunde avgöra om de ville köpa produkten eller inte. Resultatet visade även att det hade betydelse om man inte handlat via nätet förut. Det har lett till att man, trots upplevelsen av positiva egenskaper hos verktyget (i vissa fall), inte kände sig bekväm med den sortens inköp. Dessa personer kunde även urskilja färre egenskaper hos båda presentationerna än vad de vana Internethandlarna kunde. Största orsaken till de osäkra testdeltagarnas tvekan var att de ville se möblerna i verkligheten.

Slutsatsen blev ändå att produktpresentation genom interaktiv 3D upplevs som mer intressant och underhållande än en 2D-presentation. Det konstaterades också att majoriteten var positiv till att möbler presenterades i 3D även om det inte var det mest ultimata sättet.

Die technologische Entwicklung im Bereich der Echokardiographie hat in der letzten Dekade neue Methoden zur objektiven Erfassung der regionalen linksventrikulären Wandbewegung ermöglicht. Speckle Tracking erfasst die myokardiale Deformation durch die Positionsänderung einzelner Bildpunkte von einem Bild des analysierten Datensatzes zum nächsten. Diese Methode ist dem Gewebedoppler überlegen, insbesondere wegen ihrer Unabhängigkeit vom Anlotungswinkel. Zwei-dimensionale (2D) Speckle Tracking Analysen wurden für die klinische Praxis validiert. Die drei-dimensionale (3D) Echokardiographie erlaubt inzwischen Speckle Tracking Analysen von 3D Datensätzen, welche jedoch für die klinische Praxis noch nicht ausreichend validiert sind. Bei Patienten mit normaler regionaler linksventrikulärer Wandbewegung (N=37), sowie bei Patienten mit ischämie-bedingten Wandbewegungsstörungen (N=18) wurden 3D und 2D Speckle Tracking Analysen durchgeführt. Die Vergleichbarkeit der beiden Methoden hinschtlich der Quantifizierung von normalen und pathologischen Wandbewegungsmustern wurde anhand dieser Messungen geprüft. Des weiteren wurde der Einfluss der Bildrate und Bildqualität drei-dimensionaler Datensätze auf die Vergleichbarkeit beider Methoden analysiert. Es zeigte sich eine gute Vergleichbarkeit des 2D und 3D Speckle Tracking in der Diagnostik eingeschränkter linksventrikulärer systolischer Funktion, sowie in der Lokalisationsdiagnostik umschriebener Wandbewegungsstörungen. 2D und 3D Speckle Tracking sind jedoch noch nicht als gleichwertige Methoden anzusehen. Die Bildqualität, generell bei beiden Modalitäten - jedoch speziell bei 3D Datensätzen, sowie die Bildrate der 3D Datensätze zeigen signifikante Einflüsse auf die 3D Strain Analysen. Eine korrekte Standardisierung der analysierten Aufnahmen und eine optimale Bildqualität sind wichtige Faktoren, die die Zuverlässigkeit des 2D und 3D Speckle Trackings bestimmen.

This research examines the effects of using 3D virtual shopping malls and shows the user interaction and experience in this type of virtual environments. Technology developments and the use of the Internet made users to expect more services, which help to improve the user's life. Using a 3D virtual environment and adding a shopping idea to it is considered as very interesting, especially if it reflects the real world in a virtual reality that make its users attached to it. Furthermore, the possibility of having an avatar to represent themselves in computer-mediated virtual environment help users to explore the virtual environment. Another advantage is that users have the possibility to invite friends and to navigate inside a 3D Virtual-shopping mall not alone but together, which gives the user the capability to socialise inside the mall. Besides, changing of the customer's profile and chatting with friends is supported, too. Moreover, the transformation process from a two-dimensional environment to a three-dimensional environment is considered to be convenient for both, customers and mall owners.   I have used phenomenology to investigate this new phenomenon. In addition the data was gathered by interviewing several participants from different educational levels as well as business owners of 3D virtual malls. The research will show the potential of using a 3D virtual shopping mall from the user's perspective as well as from the business' owners and having a 3D virtual shopping mall is not only considered entertaining, but also convenient.   As a future work I recommend to interview the merchants in this research to grasp this idea completely.

Notre travail concerne l'étude, le calcul et la simplification des squelettes d'objets 2D et 3D. Le squelette d'un objet est une figure mince, centrée dans la forme et qui en résume l'aspect. Il est utile pour la description et la reconnaissance de formes, la quantification, la mise en correspondance, etc. Dans un premier temps, nous recensons les différentes techniques de calcul du squelette. La très grande majorité d'entre elles travaille sur des images binaires avec des outils de la géométrie discrète. Or, dernièrement, une nouvelle famille de méthodes, appelées méthodes continues a vu le jour. Le squelette est approché à l'aide du graphe de Voronoï d'un échantillonnage de la frontière, et se calcule par des moyens propres à la géométrie algorithmique. Notre intérêt s'est porté sur cette nouvelle approche et les problèmes qui s'y rattache. Pour commencer, nous proposons une formulation des méthodes continues à l'aide du squelette d'une union finie de sphères. En effet, nous montrons que le squelette d'une union finie de sphères se construit de façon exacte à l'aide d'éléments très simples comme des segments de droite en 2D et des polygones en 3D. La construction du squelette nécessite de pouvoir interpoler par des facettes triangulaires un ensemble de points localisés sur la frontière d'un objet. Nous proposons une méthode, fondée sur le calcul du graphe de Delaunay et dont nous montrons la convergence en 2D. Enfin, des méthodes de simplification du squelette sont présentées. Elles permettent de sélectionner les branches correspondant à des renflements significatifs de la forme et conduisent en 3D soit à des squelettes surfaciques, soit à des squelettes filiformes selon les besoins de l'utilisateur. Pour finir, nous décrivons une application qui valide notre approche, et l'illustre sur des données biologiques

This thesis describes the theoretical bases, development and testing of an integrated moving object detection framework in 2D and 3D scenes. The detection problem is analyzed in stationary and non-stationary camera sequences and different algorithms are developed for each case. Two methods are proposed in stationary camera sequences: background extraction followed by differencing and thresholding, and motion detection using optical flow field calculated by &ldquo
Kanade-Lucas Feature Tracker&rdquo
. For non-stationary camera sequences, different algorithms are developed based on the scene structure and camera motion characteristics. In planar scenes where the scene is flat or distant from the camera and/or when camera makes rotations only, a method is proposed that uses 2D parametric registration based on affine parameters of the dominant plane for independently moving object detection. A modified version of the 2D parametric registration approach is used when the scene is not planar but consists of a few number of planes at different depths, and camera makes translational motion. Optical flow field segmentation and sequential registration are the key points for this case. For 3D scenes, where the depth variation within the scene is high, a parallax rigidity based approach is developed for moving object detection. All these algorithms are integrated to form a unified independently moving object detector that works in stationary and non-stationary camera sequences and with different scene and camera motion structures. Optical flow field estimation and segmentation is used for this purpose.

The goal of this thesis is to gain a deep understanding of inverse problems arising from 2D image and 3D surface reconstruction, and to design effective techniques for solving them. Both computational and theoretical issues are studied and efficient numerical algorithms are proposed. The first part of this thesis is concerned with the recovery of 2D images, e.g., de-noising and de-blurring. We first consider implicit methods that involve solving linear systems at each iteration. An adaptive Huber regularization functional is used to select the most reasonable model and a global convergence result for lagged diffusivity is proved. Two mechanisms---multilevel continuation and multigrid preconditioning---are proposed to improve efficiency for large-scale problems. Next, explicit methods involving the construction of an artificial time-dependent differential equation model followed by forward Euler discretization are analyzed. A rapid, adaptive scheme is then proposed, and additional hybrid algorithms are designed to improve the quality of such processes. We also devise methods for more challenging cases, such as recapturing texture from a noisy input and de-blurring an image in the presence of significant noise. It is well-known that extending image processing methods to 3D triangular surface meshes is far from trivial or automatic. In the second part of this thesis we discuss techniques for faithfully reconstructing such surface models with different features. Some models contain a lot of small yet visually meaningful details, and typically require very fine meshes to represent them well; others consist of large flat regions, long sharp edges (creases) and distinct corners, and the meshes required for their representation can often be much coarser. All of these models may be sampled very irregularly. For models of the first class, we methodically develop a fast multiscale anisotropic Laplacian (MSAL) smoothing algorithm. To reconstruct a piecewise smooth CAD-like model in the second class, we design an efficient hybrid algorithm based on specific vertex classification, which combines K-means clustering and geometric a priori information. Hence, we have a set of algorithms that efficiently handle smoothing and regularization of meshes large and small in a variety of situations.

This thesis involves the study of four woodcuts from the twentieth century German Expressionist movement. The study of these woodcuts inspires and informs the translation of these artworks from two-dimensional works on paper to three-dimensional computer generated models. The goal of this thesis is to produce work that is derivative of the chosen woodcuts rather than create replicas of these artworks. The work undertaken utilizes Maya, a 3D software, and Mental Ray, a production quality rendering software. The final results are presented through stills, image sequences, and prints. The methods used to attain the goal of this thesis can provide a framework for artists who would like to achieve a similar woodcut look in their digital three-dimensional work.

We study contact representations of graphs in the plane and in 3D space, where vertices are represented by polygons or polyhedra and each edge is represented by a common boundary between two polygons or polyhedra. In the weighted version of the problem, we find contact representations with the additional restriction that the areas for the polygons or the volumes for the polyhedra realize some pre-specified value for the vertices. We address different variants of the problem depending on the types of polygons or polyhedra (convex or non-convex, axis-aligned or not), types of contacts (proper contacts with common boundaries of non-zero lengths in 2D or non-zero areas in 3D or improper contacts where common boundaries of zero lengths or areas are allowed), and whether holes are allowed in the representation or not. In the plane we mainly focus on the weighted version of the problem. We find optimal (in terms of polygonal complexity) contact representations for planar graphs (both for axis-aligned and non-axis-aligned polygons) and some subclasses of planar graphs. With non-axis-aligned polygons we show that non-convex polygons with 4 sides are sometimes necessary and always sufficient for proportional contact representation of a planar graph, when point contacts are allowed; otherwise for proper contacts 7-sided polygons are sometimes necessary and always sufficient. We give a linear-time algorithm in each case to compute the optimal representation. We also give quadratic-time algorithms to construct optimal proportional contact representations for (2, 0)-sparse graphs (with triangles for improper contacts and with convex quadrilaterals for proper contacts). For maximal outerplanar graphs proportional contact representation with triangles can also be computed in linear time. In case only axis-aligned polygons are used, we show that 8 sides are sometimes necessary and always sufficient for a planar graph. While we do not have a polynomial-time algorithm to construct such a representation, we give a linear-time algorithm to compute representation with 10-sided axis-aligned polygons. We also give linear-time construction algorithms for optimal proportional contact representations with 8-sided polygons for planar 3-trees and Hamiltonian maximal planar graphs, and with rectangles for maximal outerplanar graphs. For contact representation with 3D polyhedra, we consider both the weighted and the unweighted variants of the problem for both planar and non-planar graphs and have some preliminary results. We find several subclasses of planar graphs that have contact representations using cubes or proportional boxes. We also consider (improper) contact representations using tetrahedra, and show that planar graphs, complete bipartite and tripartite graphs, and complete graphs with at most 10 vertices have contact representations with tetrahedra. We also addressed variants of this problem using only unit regular tetrahedra or considering contacts only between apices of the tetrahedra or using both restrictions.

An integrated horizontally two- and fully three-dimensional numerical model system has been developed based on a combined unstructured and σ-coordinate grid to simulate the flow and water quality process in large water bodies with a focus on the three dimensional behaviours at specific areas. The model is based on the time dependent Reynolds-Averaged Navier-Stokes equations with a non-hydrostatic pressure distribution and a baroclinic force being incorporated in the three dimensional (3D) model. The two sub models interact dynamically during the solution procedure with no time-step restriction due to integration. The main idea is to use a fractional step algorithm for each model and then integrate the two models fraction by fraction. Hybrid 2D-3D finite volume cells have been introduced for the link nodes which are partly in the 2D domain and partly in the 3D domain. Thus an interpolation/averaging procedure at the interface and domain overlapping is no longer needed. The 3D model uses the projection method for pressure calculation. The advection equation is solved by the semi-Lagrangian method. Other components are solved via the finite element - finite volume (FV) method. The water surface is determined implicitly through a global matrix equation created by assembling the domain's matrices. The cell integrals are calculated analytically to eliminate a common source of numerical diffusion due to the use of approximation techniques for the FV integrals. The horizontal gradients of the density and shear stresses are calculated on true horizontal planes, in order to avoid artificial velocity and diffusion in highly stratified flows. Neumann interpolation elements with virtual nodes have been introduced at Neumann type of boundaries for more accuracy. The integrated model has been verified using analytical solutions and benchmark test cases, including the Ekman velocity distribution, wind driven circulation, lock exchange and integrated 2D-3D flows in basin. The results show the model is capable of the model for accurate simulation and implicit 2D-3D integration. Keywords: integrated modelling, hydrodynamic numerical model, non-hydrostatic, unstructured mesh, hybrid finite element finite volume method.

The aim of the present work was to build a novel 3D object recognition system capable of classifying man-made and natural objects based on single 2D views. The approach to this problem has been one motivated by recent theories on biological vision and multiresolution analysis. The project's objectives were the implementation of a system that is able to deal with simple 3D scenes and constitutes an engineering solution to the problem of 3D object recognition, allowing the proposed recognition system to operate in a practically acceptable time frame. The developed system takes further the work on automatic classification of marine phytoplanktons, carried out at the Centre for Intelligent Systems, University of Plymouth. The thesis discusses the main theoretical issues that prompted the fundamental system design options. The principles and the implementation of the coarse data channels used in the system are described. A new multiresolution representation of 2D views is presented, which provides the classifier module of the system with coarse-coded descriptions of the scale-space distribution of potentially interesting features. A multiresolution analysis-based mechanism is proposed, which directs the system's attention towards potentially salient features. Unsupervised similarity-based feature grouping is introduced, which is used in coarse data channels to yield feature signatures that are not spatially coherent and provide the classifier module with salient descriptions of object views. A simple texture descriptor is described, which is based on properties of a special wavelet transform. The system has been tested on computer-generated and natural image data sets, in conditions where the inter-object similarity was monitored and quantitatively assessed by human subjects, or the analysed objects were very similar and their discrimination constituted a difficult task even for human experts. The validity of the above described approaches has been proven. The studies conducted with various statistical and artificial neural network-based classifiers have shown that the system is able to perform well in all of the above mentioned situations. These investigations also made possible to take further and generalise a number of important conclusions drawn during previous work carried out in the field of 2D shape (plankton) recognition, regarding the behaviour of multiple coarse data channels-based pattern recognition systems and various classifier architectures. The system possesses the ability of dealing with difficult field-collected images of objects and the techniques employed by its component modules make possible its extension to the domain of complex multiple-object 3D scene recognition. The system is expected to find immediate applicability in the field of marine biota classification.

Thesis (M.S.)--University of North Carolina at Chapel Hill, 2008.
Title from electronic title page (viewed Dec. 11, 2008). "... in partial fulfillment of the requirements for the degree of Master of Science in the Department of Oral and Maxillofacial Radiology." Discipline: Oral and Maxillofacial Radiology; Department/School: Dentistry.

Humans appear to solve complex vision tasks in an almost effortless manner, as compared to their computer counterparts. One major reason for this is the intelligent cooperation between the sensory and the motor system, which is facilitated by development of motor skills that help to shape visual information that is relevant to a specific vision task. This dynamic interaction of sensory-motor components in biological systems can be a great inspiration to how artificial systems, such as robots could use their visual mechanism to interacts with their world. In this thesis, we seek to explore an approach to active vision inspired by biological evolution, which does not use a predefined framework or assumptions, but develops motor strategies for a given task through progressive adaptation of the evolutionary method. Thus, this kind of approach will give freedom to artificial systems in the discovery of eye movement strategies that may be useful to biological systems but are not known to us. The contributions of this thesis are: 1. We used this type of active vision system for more complex images taken from the camera of the iCub robot. 2. We demonstrated the effectiveness of the active vision system in a more realistic setting for 3D object categorisation using the humanoid robot (iCub) platform. 3. We extended the applicability of the system to the 3D environment for indoor and outdoor environment classification task using the iCub platform. 4. We extended the system with pre-processing using Uniform Local Binary Patterns [1] in both 2D and 3D environment categorisation tasks. 5. We further extended the system with pre-processing using Histogram of Oriented Gradients [2] for classification tasks in the 2D and 3D environments. Analysis of the results from the system shows that the model was able to complete discrimination tasks through: (i) exploiting sensory-motor coordination to experience sensory stimuli that facilitates the classification tasks; (ii) an indication of integration of perceptual information over time.

Oral keratinocyte behaviour were analysed in two and three dimensional cultures of an immortalised human H400 cellline and primary rat keratinocytes (PRKs) using a novel method of quantitative microscopy, RT-PCR data and immunohistochemistry profiles. Monolayer cultures were established in high and low calcium media at different cell densities and analysed prior to generating 3D organotypic cultures (OCs) onde-epidermalised dermis (DED), polyethylene terephthalate porous membrane (PET) and collagen gels for up to 14 days.H400 and PRKs proliferation in monolayer cultures was greater in low calcium medium compared with high calcium medium.Gene expression analysis indicated that adhesion and structural molecules including E-cadherin, plakophilin, desmocollin-3, desmogleins-3 and cytokeratins-1, -5, -6, -10, -13 were up-regulated by days 6 and 8 compared with day 4in high calcium medium. Immunohistochemical profiles and gene expression data of OCs on DED recapitulated those of normal oral epithelium. The final thickness of OCs as well as the degree of maturation/stratification was significantly greater on DED compared with other scaffolds used. Quantitative microscopy approaches enabled unbiased architectural characterisation of OCs and the ability to relate stratified organotypic epithelial structures to the normal oral mucosa. H400 and PRK OCs on DED at the air liquid interface demonstrated similar characteristics in terms of gene expression and protein distribution to the normal tissue architecture.