Active contours have been proven to be a powerful semiautomatic image segmentation approach, that seems to cope with many applications and different image modalities. However, they exhibit inherent drawbacks, including the sensibility to contour initialization due to the limited capture range of image edges and problems with concave boundary regions. The Gradient Vector Flow replaces the traditional image force and provides an enlarged capture range as well as enhanced concavity extraction capabilities, but it involves an expensive computational effort and considerably increased memory requirements at the time of computation. In this paper, we present an enhancement of the active contour model to facilitate semiautomatic contour detection in huge images. We propose a tile-based image decomposition accompanying an image force computation scheme on demand in order to minimize both computational and memory requirements. We show an efficient implementation of this approach on the basis of general purpose GPU processing providing for continuous active contour deformation without a considerable delay.

<div><div><div><p>Modeling microscopy images and extracting information from them are important problems in the fields of physics and material science. </p><p><br></p><p>Model-based methods, such as marked point processes (MPPs), and machine learning approaches, such as convolutional neural networks (CNNs), are powerful tools to perform these tasks. Nevertheless, MPPs present limitations when modeling objects with irregular boundaries. Similarly, machine learning techniques show drawbacks when differentiating clustered objects in volumetric datasets.</p><p> </p><p>In this thesis we explore the extension of the MPP framework to detect irregularly shaped objects. In addition, we develop a CNN approach to perform efficient 3D object detection. Finally, we propose a CNN approach together with geometric regularization to provide robustness in object detection across different datasets.</p><p><br></p><p>The first part of this thesis explores the addition of boundary energy to the MPP by using active contours energy and level sets energy. Our results show this extension allows the MPP framework to detect material porosity in CT microscopy images and to detect red blood cells in DIC microscopy images.</p><p><br></p><p>The second part of this thesis proposes a convolutional neural network approach to perform 3D object detection by regressing objects voxels into clusters. Comparisons with leading methods demonstrate a significant speed-up in 3D fiber and porosity detection in composite polymers while preserving detection accuracy.</p><p><br></p><p>The third part of this thesis explores an improvement in the 3D object detection approach by regressing pixels into their instance centers and using geometric regularization. This improvement demonstrates robustness when comparing 3D fiber detection in several large volumetric datasets.</p><p><br></p></div></div></div><div><div><div><p>These methods can contribute to fast and correct structural characterization of large volumetric datasets, which could potentially lead to the development of novel materials.</p></div></div></div>

This work is rooted in a need for dermatologists to explore skin characteristicsin depth. The inuence of skin disease such as acne in dermal tissues is stilla complex task to assess. Among the possibilities, high frequency ultrasoundimaging is a paradigm shift to probe and characterizes upper and deep dermis.For this purpose, a cohort of 58 high-frequency 3D images has been acquiredby the French laboratory Pierre Fabre in order to study acne vulgaris disease.This common skin disorder is a societal challenge and burden aecting late adolescentsacross the world. The medical protocol developed by Pierre Fabre wasto screen a lesion every day during 9 days for dierent patients with ultrasoundimaging. The provided data features skin epidermis and dermis structure witha fantastic resolution. The strategy we led to study these data can be explainedin three steps. First, epidermis surface is detected among artifacts and noisethanks to a robust level-set algorithm. Secondly, acne spots are located on theresulting height map and associated to each other among the data by computingand thresholding a local variance. And eventually potential inammatorydermal cavities related to each lesion are geometrically and statistically characterizedin order to assess the evolution of the disease. The results presentan automatic algorithm which permits dermatologists to screen acne vulgarislesions and to characterize them in a complete data set. It can hence be a powerfultoolbox to assess the eciency of a treatment. / Detta arbete är grundat i en dermatologs behov att undersöka hudens egenskaperpå djupet. Påverkan av hudsjukdomar så som acne på dermala vävanderär fortfarande svårt att bedöma. Bland möjligheterna är högfrekvent ultraljudsavbildningett paradigmskifte för undersökning och karakterisering av övre ochdjupa dermis. I detta syfte har en kohort av 58 högfrekventa 3D bilder förvärvatsav det Franska laboratoriet Pierre Fabre för att studera sjukdomen acne vulgaris.Denna vanliga hudsjukdom är en utmaning för samhället och en bördasom påverkar de i slutet av tonåren över hela världen. Protokollet utvecklatav Pierre Fabre innebar att undersöka en lesion varje dag över 9 dagar förolika patienter med ultraljudavbildning. Den insamlade datan visar hudens epidermisoch dermis struktur med en fantastiskt hög upplösning. Strategin vianvände för att studera denna data kan förklaras i tre steg. För det första,hittas epidermis yta bland artifakter och brus tack vare en robust level-set algoritm.För det andra, acne äckar hittas på höjdkartan och associeras tillvarandra bland mätdatan genom en tröskeljämförelse över lokala variationer.Även potentiellt inammatoriska dermala hålrum relaterade till varje lesion blirgeometriskt ochj statistiskt kännetecknade för att bedöma sjukdomens förlopp.Resultaten framför en automatisk algoritm som gör det möjligt för dermatologeratt undersöka acne vulgaris lesioner och utmärka de i ett dataset. Detta kandärmed vara en kraftfull verktygslåda för att undersöka inverkan av en behandlingtill denna sjukdom.

La segmentation de régions d'intérêt dans le cadre de l'analyse d'images médicales et biomédicales reste encore à ce jour un challenge en raison notamment de la variété des modalités d'acquisition et des caractéristiques associées (bruit par exemple).Dans ce contexte particulier, cet exposé présente une méthode de segmentation de type contour actif dont l ‘énergie associée à l'obtention de l'équation d'évolution s'appuie sur une mesure de similarité entre les densités de probabilités (en niveau de gris) des régions intérieure et extérieure au contour au cours du processus itératif de segmentation. En particulier, nous nous intéressons à la famille particulière des alpha-divergences. L'intérêt principal de cette méthode réside (i) dans la flexibilité des alpha-divergences dont la métrique intrinsèque peut être paramétrisée via la valeur du paramètre alpha et donc adaptée aux distributions statistiques des régions de l'image à segmenter ; et (ii) dans la capacité unificatrice de cette mesure statistique vis-à-vis des distances classiquement utilisées dans ce contexte (Kullback- Leibler, Hellinger...). Nous abordons l'étude de cette mesure statistique tout d'abord d'un point de vue supervisé pour lequel le processus itératif de segmentation se déduit de la minimisation de l'alpha-divergence (au sens variationnel) entre la densité de probabilité courante et une référence définie a priori. Puis nous nous intéressons au point de vue non supervisé qui permet de s'affranchir de l'étape de définition des références par le biais d'une maximisation de distance entre les densités de probabilités intérieure et extérieure au contour. Par ailleurs, nous proposons une démarche d'optimisation de l'évolution du paramètre alpha conjointe au processus de minimisation ou de maximisation de la divergence permettant d'adapter itérativement la divergence à la statistique des données considérées. Au niveau expérimental, nous proposons une étude comparée des différentes approches de segmentation : en premier lieu, sur des images synthétiques bruitées et texturées, puis, sur des images naturelles. Enfin, nous focalisons notre étude sur différentes applications issues des domaines biomédicaux (microscopie confocale cellulaire) et médicaux (radiographie X, IRM) dans le contexte de l'aide au diagnotic. Dans chacun des cas, une discussion sur l'apport des alpha-divergences est proposée. / In the particular field of Computer-Aided-Diagnosis, the segmentation of particular regions of interest corresponding usually to organs is still a challenging issue mainly because of the various existing for which the charateristics of acquisition are very different (corrupting noise for instance). In this context, this PhD work introduces an original histogram-based active contour segmentation using alpha-divergence family as similarity measure. The method keypoint are twofold: (i) the flexibility of alpha-divergences whose metric could be parametrized using alpha value can be adaptedto the statistical distribution of the different regions of the image and (ii) the ability of alpha-divergence ability to enbed standard distances like the Kullback-Leibler's divergence or the Hellinger's one makes these divergences an interesting unifying tool.In this document, first, we propose a supervised version of proposed approach:. In this particular case, the iterative process of segmentation comes from alpha-divergenceminimization between the current probability density function and a reference one which can be manually defined for instance. In a second part, we focus on the non-supervised version of the method inorder to be able.In that particular case, the alpha-divergence maximization between probabilitydensity functions of inner and outer regions defined by the active contour is maximized. In addition, we propose an optimization scheme of the alpha parameter jointly with the optimization of the divergence in order to adapt iteratively the divergence to the inner statistics of processed data. Furthermore, a comparative study is proposed between the different segmentation schemes : first, on synthetic images then, on natural images. Finally, we focus on different kinds of biomedical images (cellular confocal microscopy) and medical ones (X-ray) for computer-aided diagnosis.

La segmentation d'images est un domaine important dans le traitement d'images et un grand nombre d'approches différentes ent été développées pendant ces dernières décennies. L'approche des contours actifs est un des plus populaires. Dans ce cadre, cette thèse vise à développer des algorithmes robustes, qui peuvent segmenter des images avec des inhomogénéités d'intensité. Nous nous concentrons sur l'étude des énergies externes basées région dans le cadre des ensembles de niveaux. Précisément, nous abordons la difficulté de choisir l'échelle de la fenêtre spatiale qui définit la localité. Notre contribution principale est d'avoir proposé une échelle adaptative pour les méthodes de segmentation basées sur les statistiques locales. Nous utilisons l'approche d'Intersection des Intervalles de Confiance pour définir une échelle position-dépendante pour l'estimation des statistiques image. L'échelle est optimale dans le sens où elle donne le meilleur compromis entre le biais et la variance de l'approximation polynomiale locale de l'image observée conditionnellement à la segmentation actuelle. De plus, pour le model de segmentation basé sur une interprétation Bahésienne avec deux noyaux locaux, nous suggérons de considérer leurs valeurs séparément. Notre proposition donne une segmentation plus lisse avec moins de délocalisations que la méthode originale. Des expériences comparatives de notre proposition à d'autres méthodes de segmentation basées sur des statistiques locales sont effectuées. Les résultats quantitatifs réalisés sur des images ultrasonores de simulation, montrent que la méthode proposée est plus robuste au phénomène d'atténuation. Des expériences sur des images réelles montrent également l'utilité de notre approche. / Image segmentation is an important research area in image processing and a large number of different approaches have been developed over the last few decades. The active contour approach is one of the most popular among them. Within this framework, this thesis aims at developing robust algorithms, which can segment images with intensity inhomogeneities. We focus on the study of region-based external energies within the level set framework. We study the use of local image statistics for the design of external energies. Precisely, we address the difficulty of choosing the scale of the spatial window that defines locality. Our main contribution is to propose an adaptive scale for local region-based segmen¬tation methods. We use the Intersection of Confidence Intervals approach to define this pixel-dependent scale for the estimation of local image statistics. The scale is optimal in the sense that it gives the best trade-off between the bias and the variance of a Local Polynomial Approximation of the observed image conditional on the current segmenta¬tion. Additionally, for the segmentation model based on a Bayesian interpretation with two local kernels, we suggest to consider their values separately. Our proposition gives a smoother segmentation with less mis-localisations Chan the original method.Comparative experiments of our method to other local region-based segmentation me¬thods are carried out. The quantitative results, on simulated ultrasound B-mode images, show that the proposed scale selection strategy gives a robust solution to the intensity inhomogeneity artifact of this imaging modality. More general experiments on real images also demonstrate the usefulness of our approach.

In this thesis, we propose and evaluate two novel scale-based decomposable representations of shape for the segmentation and morphometric analysis of anatomical structures in medical imaging. We propose two representations that are adapted to a particular class of anatomical structures and allow for a richer shape description and a more fine-grained control over the deformation of models based on these representations, when compared to previous techniques. In the first part of this thesis, we introduce the concept of a scale-space shape filter for implicit shape representations that measures the deviation from a tubular shape in a local neighborhood of points, given a particular scale of analysis. We use these filters for the segmentation of blood vessels, and introduce the notion of segmentation with a soft shape prior, where the segmented model is not globally constrained to a predefined shape space, but is penalized locally if it deviates strongly from a tubular structure. Using this filter, we derive a region-based active contour segmentation algorithm for tubular structures that penalizes leakages. We present results on synthetic and real 2D and 3D datasets. In the second part of this thesis, we present a novel multi-scale parametric shape representation using spherical wavelets. Our proposed shape representation encodes shape variations in a population at various scales to be used as prior in a probabilistic segmentation framework. We derive a probabilistic active surface segmentation algorithm using the multi-scale prior coefficients as parameters for our optimization procedure. One nice benefit of this algorithm is that the optimization method can be applied in a coarse-to-fine manner. We present results on 3D sub-cortical brain structures. We also present a novel method of statistical surface-based morphometry based on the use of non-parametric permutation tests and the spherical wavelet shape representation. As an application, we analyze two sub-cortical brain structures, the caudate nucleus and hippocampus.

The purpose of this study is to develop an automatic facial feature extraction system, which is able to identify the detailed shape of eyes, eyebrows and mouth from facial images. The developed system not only extracts the location information of the features, but also estimates the parameters pertaining the contours and parts of the features using parametric deformable templates approach. In order to extract facial features, deformable models for each of eye, eyebrow, and mouth are developed. The development steps of the geometry, imaging model and matching algorithms, and energy functions for each of these templates are presented in detail, along with the important implementation issues. In addition, an eigenfaces based multi-scale face detection algorithm which incorporates standard facial proportions is implemented, so that when a face is detected the rough search regions for the facial features are readily available. The developed system is tested on JAFFE (Japanese Females Facial Expression Database), Yale Faces, and ORL (Olivetti Research Laboratory) face image databases. The performance of each deformable templates, and the face detection algorithm are discussed separately.

This thesis explores two possible ways of detecting soft plaque present in the coronary arteries, using CTA imagery. The coronary arteries are vessels that supply oxidized blood to the cardiac muscle and are thus important for the proper functioning of heart. Cholesterol or reactive oxygen species from cigarette smoke and other toxins may get adhered to the walls of coronary arteries and trigger chronic inflammation that leads to formation of the soft plaque. When the soft plaque grows bigger in volume, it occludes the blood flow to the cardiac muscle and finally results in ischemic heart attack. Moreover, smaller plaque can easily rupture due to the blood flow in arteries and can result in complications such as stroke. Hence there is a need to detect the soft plaque using non-invasive or minimally invasive techniques. In CTA imagery, the cardiac muscle appears as a dark gray color, while the blood appears as dull white color and the the calcified plaque appears as bright white. The soft plaque has an intensity which falls between the intensity level of the blood and cardiac muscle, making it difficult to directly segment the soft plaque using standard segmentation methods. Soft plaque in its advanced stages forms a concavity in the blood lumen. A watershed based segmentation method was used to detect the presence of this concavity which in turn identifies the location of the soft plaque. For segmenting the soft plaque at its earlier stages, a novel segmentation technique was used. In this technique the surface is evolved based on a region-based energy calculated in the local neighborhood around each point on the evolving surface. This method seems to be superior to the watershed based segmentation method in detecting smaller plaque deposits.

Les contours actifs sont parmi les méthodes de segmentation d'images les plus utilisées et de nombreuses implémentations ont vu le jour durant ces 25 dernières années. Parmi elles, l'approche greedy est considérée comme l'une des plus rapides et des plus stables. Toutefois, quelle que soit l'implémentation choisie, les résultats de segmentation souffrent grandement en présence d'occlusions, de concavités ou de déformation anormales de la forme. Si l'on dispose d'informations a priori sur la forme recherchée, alors son incorporation à un modèle existant peut permettre d'améliorer très nettement les résultats de segmentation. Dans cette thèse, l'inclusion de ce type de contraintes de formes dans un modèle de contour actif explicite est proposée. Afin de garantir une invariance à la rotation, à la translation et au changement d'échelle, les descripteurs de Fourier sont utilisés. Contrairement à la plupart des méthodes existantes, qui comparent la forme de référence et le contour actif en cours d'évolution dans le domaine d'origine par le biais d'une transformation inverse, la méthode proposée ici réalise cette comparaison dans l'espace des descripteurs. Cela assure à notre approche un faible temps de calcul et lui permet d'être indépendante du nombre de points de contrôle choisis pour le contour actif. En revanche, cela induit un biais dans la phase des coefficients de Fourier, handicapant l'invariance à la rotation. Ce problème est résolu par un algorithme original. Les expérimentations indiquent clairement que l'utilisation de ce type de contrainte de forme améliore significativement les résultats de segmentation du modèle de contour actif utilisé. / Active contours are widely used for image segmentation. There are many implementations of active contours. The greedy algorithm is being regarded as one of the fastest and stable implementations. No matter which implementation is being employed, the segmentation results suffer greatly in the presence of occlusion, context noise, concavities or abnormal deformation of shape. If some prior knowledge about the shape of the object is available, then its addition to an existing model can greatly improve the segmentation results. In this thesis inclusion of such shape constraints for explicit active contours is being implemented. These shape priors are introduced through the use of robust Fourier based descriptors which makes them invariant to the translation, scaling and rotation factors and enables the deformable model to converge towards the prior shape even in the presence of occlusion and contextual noise. Unlike most existing methods which compare the reference shape and evolving contour in the spatial domain by applying the inverse transforms, our proposed method realizes such comparisons entirely in the descriptor space. This not only decreases the computational time but also allows our method to be independent of the number of control points chosen for the description of the active contour. This formulation however, may introduce certain anomalies in the phase of the descriptors which affects the rotation invariance. This problem has been solved by an original algorithm. Experimental results clearly indicate that the inclusion of these shape priors significantly improved the segmentation results of the active contour model being used.

Le coeur de cette thèse porte sur la modélisation géométrique et introduit une méthode robuste d'extraction de formes tubulaires à partir de nuages de points. Nous avons choisi de tester nos méthodes dans le contexte applicatif de la foresterie pour mettre en valeur la robustesse de nos algorithmes.Nos méthodes intègrent les normales aux points, il est donc nécessaire de les pré-calculer. Notre premier développement a alors consisté à présenter une méthode rapide d'estimation de normales. Pour ce faire nous avons approximé localement la géométrie du nuage de points en utilisant des "patchs" lisses dont la taille s'adapte à la complexité locale des nuages de points.Nos travaux se sont ensuite concentrés sur l’extraction robuste de formes tubulaires dans des nuages de points occlus, bruités et de densité inhomogène. Nous avons développé une variante de la transformée de Hough que nous avons couplé à une proposition de contours actifs indépendants de leur paramétrisation. Notre méthode a été validée en environnement forestier pour reconstruire des troncs d'arbre afin d'en relever les qualités par comparaison à des méthodes existantes.La reconstruction de troncs d'arbre ouvre d'autres questions dont la segmentation des arbres d'une placette forestière. Nous proposons également une méthode de segmentation pour isoler les différents objets d'un jeu de données.Durant nos travaux nous avons utilisé des approches de modélisation pour répondre à des questions géométriques, et nous les avons appliqué à des problématiques forestières. Il en résulte un pipeline de traitements cohérent qui, bien qu'illustré sur des données forestières, est applicable dans des contextes variés. / The core of this thesis concerns geometric modelling and introduces a fast and robust method for the extraction of tubular shapes from point clouds. We chose to test our method in the difficult applicative context of forestry in order to highlight the robustness of our algorithms.Our methods integrate normal vectors, thus they have to be pre-computed. Our first development consisted in the development of a fast normal estimation method on point cloud. To do so, we locally approximated the point cloud geometry using smooth "patches" of points which size adapts to the local complexity of the point cloud geometry.We then focused our work on the robust extraction of tubular shapes from dense, occluded, noisy point clouds suffering from non-homogeneous sampling density. We developed a variant of the Hough transform and combined this research with a new definition of parametrisation-invariant active contours. We validated our method in complex forest environments with the reconstruction of tree stems to emphasize its advantages and compare it to existing methods.Tree stem reconstruction also opens new perspectives halfway in between forestry and geometry such as the segmentation of trees from a forest plot. Therefore we propose a segmentation approach capable of isolating objects inside a point cloud.During our work we used modelling approaches to answer geometric questions and we applied our methods to forestry problems. Therefore, our studies result in a processing pipeline adapted to forest point cloud analyses, but the general geometric algorithms we propose can also be applied in various contexts.