INTRODUCTION: Despite the advances in radiotherapy treatment delivery, target volume delineation remains one of the greatest sources of error in the radiotherapy delivery process, which can lead to poor tumour control probability and impact clinical outcome. Contouring assessments are performed to ensure high quality of target volume definition in clinical trials but this can be subjective and labour-intensive. This project addresses the hypothesis that computational segmentation techniques, with a given prior, can be used to develop an image-based tumour delineation process for contour assessments. This thesis focuses on the exploration of the segmentation techniques to develop an automated method for generating reference delineations in the setting of advanced lung cancer. The novelty of this project is in the use of the initial clinician outline as a prior for image segmentation. METHODS: Automated segmentation processes were developed for stage II and III non-small cell lung cancer using the IDEAL-CRT clinical trial dataset. Marker-controlled watershed segmentation, two active contour approaches (edge- and region-based) and graph-cut applied on superpixels were explored. k-nearest neighbour (k-NN) classification of tumour from normal tissues based on texture features was also investigated. RESULTS: 63 cases were used for development and training. Segmentation and classification performance were evaluated on an independent test set of 16 cases. Edge-based active contour segmentation achieved highest Dice similarity coefficient of 0.80 ± 0.06, followed by graphcut at 0.76 ± 0.06, watershed at 0.72 ± 0.08 and region-based active contour at 0.71 ± 0.07, with mean computational times of 192 ± 102 sec, 834 ± 438 sec, 21 ± 5 sec and 45 ± 18 sec per case respectively. Errors in accuracy of irregularly shaped lesions and segmentation leakages at the mediastinum were observed. In the distinction of tumour and non-tumour regions, misclassification errors of 14.5% and 15.5% were achieved using 16- and 8-pixel regions of interest (ROIs) respectively. Higher misclassification errors of 24.7% and 26.9% for 16- and 8-pixel ROIs were obtained in the analysis of the tumour boundary. CONCLUSIONS: Conventional image-based segmentation techniques with the application of priors are useful in automatic segmentation of tumours, although further developments are required to improve their performance. Texture classification can be useful in distinguishing tumour from non-tumour tissue, but the segmentation task at the tumour boundary is more difficult. Future work with deep-learning segmentation approaches need to be explored.

A displasia cortical focal (DCF) é uma das causas mais frequentes de epilepsia refratária. Na clínica, diferentes informações são usadas para localizar o foco epileptogênico, mas nenhum método é autossuficiente para evidenciar o local original das crises, associado com a presença da DCF. Embora haja relatos na literatura indicando alterações no padrão de distribuição de tons de cinza e morfologia dos voxels decorrentes da DCF, algumas limitações dos métodos desenvolvidos ainda impedem a utilização clínica. Nossa proposta foi investigar a capacidade de identificar DCF através de análises de espessura cortical e padrões de textura em imagens estruturais de Ressonância Magnética (RM), validando os métodos desenvolvidos a partir uma base de imagens retrospectiva, cujo tecido epileptogênico já havia sido ressecado e a DCF confirmada em análise histológica. A caracterização das DCF foi feita a partir da segmentação automática de tecido cortical saudável em conjunto com a segmentação manual da DCF feita por um especialista, e consiste na geração de mapas de característica e extração de valores de distribuições para comparação em análise estatística. Investigamos também a eficácia da detecção de DCF através do uso de algoritmos de aprendizado de máquina para classificação automática. Obtivemos precisão 0,81 e sensitividade 0,87, colocando o método desenvolvido em par com outros métodos presentes na literatura. Entretanto, foi identificada uma grande dependência do desempenho de métodos de pré-processamento, como corregistro e segmentação automática. / Focal Cortical Dysplasia (FCD) is one of the most frequent causes of refractory epilepsy. In clinical procedures, the information gathered from different techniques is used in order to locate the epileptogenic focus, associated with the presence of FCD. However, there is no self sufficient method to evidence the presence and location of such lesions and especially its extension. Although there are reports indicating change in gray scale intensity patterns and voxel morphology in the presence of DCF, limitations in developed methods still prevent their clinical use. Our proposal was to investigate the capability of identifying FCD through cortical thickness and texture patter analysis in structural MRI images, validating developed methods by utilizing a retrospective base of images from patients that were subjected to surgery, with the FCD being confirmed in histological analysis. Characterization of FCD was achieved from automatic segmentation of healthy cortex and manual segmentation of FCD tissue made by an specialist, and consists in the generation of texture or structural feature maps and comparison of distribution values in healthy or FCD tissue with statistical analysis. We also investigate the efficiency of FCD detection with Machine Learning automatic classification, obtaining precision of 0,81 and sensitivity of 0,87, placing our method on par with other methods in the literature. However, there is a major performance dependency of proposed method with pre-processing steps, like registration and automatic segmentation.

O avanço tecnológico recente tem atraído tanto a comunidade acadêmica quanto o mercado para a investigação de novos métodos, técnicas e linguagens formais para a área de Projeto de Engenharia. A principal motivação é o atendimento à demanda para desenvolver produtos e sistemas cada vez mais completos e que satisfaçam as necessidades do usuário final. Necessidades estas que podem estar ligadas, por exemplo, à análise e reconhecimento de objetos que compõe uma imagem pela sua textura, um processo essencial na automação de uma enorme gama de aplicações como: visão robótica, monitoração industrial, sensoriamento remoto, segurança e diagnóstico médico assistido. Em vista da relevância das inúmeras aplicações envolvidas e pelo fato do domínio de aplicação ser muito próximo do contexto do desenvolvedor, é apresentada uma proposta de um processo de design baseado no Projeto Axiomático como sendo o mais indicado para esta situação. Especificamente, se espera que no estudo de caso da análise de textura haja uma convergência mais rápida para a solução - se esta existir. No estudo de caso, se desenvolve uma nova concepção de arquitetura de rede neural artificial (RNA), auto-organizável, com a estrutura espacial bidimensional da imagem de entrada preservada, tendo a extração e reconhecimento/classificação de textura em uma única fase de aprendizado. Um novo conceito para o paradigma da competição entre os neurônios também é estabelecida. O processo é original por permitir que o desenvolvedor assuma concomitantemente o papel do cliente no projeto, e especificamente por estabelecer o processo de sistematização e estruturação do raciocínio lógico do projetista para a solução do problema a ser desenvolvido e implementado em RNA. / The recent technological advance has attracted the industry and the academic community to research and propose methods, seek for new techniques, and formal languages for engineering design in order to respond to the growing demand for sophisticated product and systems that fully satisfy customers needs. It can be associated, for instance, with an application of object recognition using texture features, essential to a variety of applications domains, such as robotic vision, industrial inspection, remote sensing, security and medical image diagnosis. Considering the importance of the large number of applications mentioned before, and due to their characteristic where both application and developer domain are very close to each other, this work aims to present a design process based on ideas extracted from axiomatic design to accelerate the development for the classical approach to texture analysis. Thus, a case study is accomplished where a new conception of neural network architecture is specially designed for the following proposal: preserving the two-dimensional spatial structure of the input image, and performing texture feature extraction and classification within the same architecture. As a result, a new mechanism for neuronal competition is also developed as specific knowledge for the domain. In fact, the process proposed has some originality because it does take into account that the developer assumes also the customers role on the project, and establishes the systematization process and structure of logical reasoning of the developer in order to develop and implement the solution in neural network domain.

Visual recognition is the problem of learning visual categories from a limited set of samples and identifying new instances of those categories, the problem is often separated into two types: the specific case and the generic category case. In the specific case the objective is to identify instances of a particular object, place or person. Whereas in the generic category case we seek to recognize different instances that belong to the same conceptual class: cars, pedestrians, road signs and mugs. Specific object recognition works by matching and geometric verification. In contrast, generic object categorization often includes a statistical model of their appearance and/or shape.This thesis proposes a computer vision system for detecting and tracking multiple targets in videos. A preliminary work of this thesis consists on the adaptation of color according to lighting variations and relevance of the color. Then, literature shows a wide variety of tracking methods, which have both advantages and limitations, depending on the object to track and the context. Here, a deterministic method is developed to automatically adapt the tracking method to the context through the cooperation of two complementary techniques. A first proposition combines covariance matching for modeling characteristics texture-color information with optical flow (KLT) of a set of points uniformly distributed on the object . A second technique associates covariance and Mean-Shift. In both cases, the cooperation allows a good robustness of the tracking whatever the nature of the target, while reducing the global execution times .The second contribution is the definition of descriptors both discriminative and compact to be included in the target representation. To improve the ability of visual recognition of descriptors two approaches are proposed. The first is an adaptation operators (LBP to Local Binary Patterns ) for inclusion in the covariance matrices . This method is called ELBCM for Enhanced Local Binary Covariance Matrices . The second approach is based on the analysis of different spaces and color invariants to obtain a descriptor which is discriminating and robust to illumination changes.The third contribution addresses the problem of multi-target tracking, the difficulties of which are the matching ambiguities, the occlusions, the merging and division of trajectories.Finally to speed algorithms and provide a usable quick solution in embedded applications this thesis proposes a series of optimizations to accelerate the matching using covariance matrices. Data layout transformations, vectorizing the calculations (using SIMD instructions) and some loop transformations had made possible the real-time execution of the algorithm not only on Intel classic but also on embedded platforms (ARM Cortex A9 and Intel U9300).

Agricultural production is becoming increasingly important as the world demand increases. On the other hand, there are several factors threatening that production such as the climate change. Therefore, monitoring and management of different parameters affecting the production are important. The current study is dedicated to two key parameters, namely agricultural land cover and soil-moisture mapping using X- and L-Band Synthetic Aperture Radar (SAR) data. Land-cover mapping plays an essential role in various applications like irrigation management, yield estimation and subsidy control. A model of multi-direction/multi-distance texture analysis on SAR data and its use for agricultural land cover classification was developed. The model is built and implemented in ESRI ArcGIS software and integrated with “R Environment”. Sets of texture measures can be calculated on a plot basis and stored in an attribute table for further classification. The classification module provides various classification approaches such as support vector machine and artificial neural network, in addition to different feature-selection methods. The model has been tested for a typical Mid-European agricultural and horticultural land use pattern south to the town of Pirna (Saxony/Germany), where the high-resolution SAR data, TerraSAR-X and ALOS/PALSAR (HH/HV) imagery, were used for land-cover mapping. The results indicate that an integrated classification using textural information of SAR data has a high potential for land-cover mapping. Moreover, the multi-dimensional SAR data approach improved the overall accuracy. Soil moisture (SM) is important for various applications such as crop-water management and hydrological modelling. The above-mentioned TerraSAR-X data were utilised for soil-moisture mapping verified by synchronous field measurements. Different speckle-reduction techniques were applied and the most representative filtered image was determined. Then the soil moisture was calculated for the mapped area using the obtained linear regression equations for each corresponding land-cover type. The results proved the efficiency of SAR data in soil-moisture mapping for bare soils and at the early growing stage of fieldcrops. / Landwirtschaftliche Produktion erlangt mit weltweit steigender Nahrungsmittelnachfrage zunehmende Bedeutung. Zahlreiche Faktoren bedrohen die landwirtschaftliche Produktion wie beispielsweise die globale Klimaveränderung einschließlich ihrer indirekten Nebenwirkungen. Somit ist das Monitoring der Produktion selbst und der wesentlichen Produktionsparameter eine zweifelsfrei wichtige Aufgabe. Die vorliegende Studie widmet sich in diesem Kontext zwei Schlüsselinformationen, der Aufnahme landwirtschaftlicher Kulturen und den Bodenfeuchteverhältnissen, jeweils unter Nutzung von Satellitenbilddaten von Radarsensoren mit Synthetischer Apertur, die im X- und L-Band operieren. Landnutzungskartierung spielt eine essentielle Rolle für zahlreiche agrarische Anwendungen; genannt seien hier nur Bewässerungsmaßnahmen, Ernteschätzung und Fördermittelkontrolle. In der vorliegenden Arbeit wurde ein Modell entwickelt, welches auf Grundlage einer Texturanalyse der genannten SAR-Daten für variable Richtungen und Distanzen eine Klassifikation landwirtschaftlicher Nutzungsformen ermöglicht. Das Modell wurde als zusätzliche Funktionalität für die ArcGIS-Software implementiert. Es bindet dabei Klassifikationsverfahren ein, die aus dem Funktionsschatz der Sprache „R“ entnommen sind. Zum Konzept: Ein Bündel von Texturparametern wird durch das vorliegende Programm auf Schlagbasis berechnet und in einer Polygonattributtabelle der landwirtschaftlichen Schläge abgelegt. Auf diese Attributtabelle greift das nachfolgend einzusetzende Klassifikationsmodul zu. Die Software erlaubt nun die Suche nach „aussagekräftigen“ Teilmengen innerhalb des umfangreichen Texturmerkmalsraumes. Im Klassifikationsprozess kann aus verschiedenen Ansätzen gewählt werden. Genannt seien „Support Vector Machine“ und künstliche neuronale Netze. Das Modell wurde für einen typischen mitteleuropäischen Untersuchungsraum mit landwirtschaftlicher und gartenbaulicher Nutzung getestet. Er liegt südlich von Pirna im Freistaat Sachsen. Zum Test lagen für den Untersuchungsraum Daten von TerraSAR-X und ALOS/PALSAR (HH/HV) aus identischen Aufnahmetagen vor. Die Untersuchungen beweisen ein hohes Potenzial der Texturinformation aus hoch aufgelösten SAR-Daten für die landwirtschaftliche Nutzungserkennung. Auch die erhöhte Dimensionalität durch die Kombination von zwei Sensoren erbrachte eine Verbesserung der Klassifikationsgüte. Kenntnisse der Bodenfeuchteverteilung sind u.a. bedeutsam für Bewässerungsanwendungen und hydrologische Modellierung. Die oben genannten SAR-Datensätze wurden auch zur Bodenfeuchteermittlung genutzt. Eine Verifikation wurde durch synchrone Feldmessungen ermöglicht. Initial musste der Radar-typische „Speckle“ in den Bildern durch Filterung verringert werden. Verschiedene Filtertechniken wurden getestet und das beste Resultat genutzt. Die Bodenfeuchtebestimmung erfolgte in Abhängigkeit vom Nutzungstyp über Regressionsanalyse. Auch die Resultate für die Bodenfeuchtebestimmung bewiesen das Nutzpotenzial der genutzten SAR-Daten für offene Ackerböden und Stadien, in denen die Kulturpflanzen noch einen geringen Bedeckungsgrad aufweisen.

La visió és probablement el nostre sentit més dominant a partir del qual derivem la majoria d'informació del món que ens envolta. A través de la visió podem percebre com són les coses, on són i com es mouen. En les imatges que percebem amb el nostre sistema de visió podem extreure'n característiques com el color, la textura i la forma, i gràcies a aquesta informació som capaços de reconèixer objectes fins i tot quan s'observen sota unes condicions totalment diferents. Per exemple, som capaços de distingir un mateix objecte si l'observem des de diferents punts de vista, distància, condicions d'il·luminació, etc.La Visió per Computador intenta emular el sistema de visió humà mitjançant un sistema de captura d'imatges, un ordinador, i un conjunt de programes. L'objectiu desitjat no és altre que desenvolupar un sistema que pugui entendre una imatge d'una manera similar com ho realitzaria una persona. Aquesta tesi es centra en l'anàlisi de la textura per tal de realitzar el reconeixement de superfícies. La motivació principal és resoldre el problema de la classificació de superfícies texturades quan han estat capturades sota diferents condicions, com ara distància de la càmera o direcció de la il·luminació. D'aquesta forma s'aconsegueix reduir els errors de classificació provocats per aquests canvis en les condicions de captura.En aquest treball es presenta detalladament un sistema de reconeixement de textures que ens permet classificar imatges de diferents superfícies capturades en diferents condicions. El sistema proposat es basa en un model 3D de la superfície (que inclou informació de color i forma) obtingut mitjançant la tècnica coneguda com a 4-Source Colour Photometric Stereo (CPS). Aquesta informació és utilitzada posteriorment per un mètode de predicció de textures amb l'objectiu de generar noves imatges 2D de les textures sota unes noves condicions. Aquestes imatges virtuals que es generen seran la base del nostre sistema de reconeixement, ja que seran utilitzades com a models de referència per al nostre classificador de textures.El sistema de reconeixement proposat combina les Matrius de Co-ocurrència per a l'extracció de característiques de textura, amb la utilització del Classificador del veí més proper. Aquest classificador ens permet al mateix temps aproximar la direcció d'il·luminació present en les imatges que s'utilitzen per testejar el sistema de reconeixement. És a dir, serem capaços de predir l'angle d'il·luminació sota el qual han estat capturades les imatges de test. Els resultats obtinguts en els diferents experiments que s'han realitzat demostren la viabilitat del sistema de predicció de textures, així com del sistema de reconeixement. / This thesis is concerned with the application of texture analysis to discriminate between textured surfaces. The main motivation is the problem of classifying textured surfaces imaged under varying geometries, i.e. distance from the sensor and illumination direction, as well as the necessity of finding reliable methods of reducing classification errors caused by changes in the geometry's properties. In texture analysis one must distinguish between image texture and surface texture. Image texture is what appears in the 2D image of a physical object, while surface texture refers to the variation of the physical and geometric properties of the imaged surface which give rise to the image texture. Changes in the imaging geometry can significantly alter the appearance of the surface, implying significant variations in the image texture. And one still has to perform the task of recognition from the image texture. In this thesis, after analysing different strategies, we integrate the surface texture information derived by colour photometric stereo (CPS) into a complete model-based texture classification system. Photometric stereo is the technique which allows us to obtain surface texture information from a few images of the same surface imaged under various illumination directions. Basically, the main idea of our strategy consists of creating, by means of the surface texture information, a virtual' database of image textures against which we compare unknown test images in order to classify them. Note that we do not use the surface texture information directly to perform classification, but we use it to create new images which are the references for our training and classification process. Furthermore, the classification system allows us to guess the approximate direction of the illumination used to capture the test images.The proposed prediction methods, as well as the model-based texture classification system, are tested and evaluated. A set of real surface textures containing a wide variety of relatively smooth and very rough surfaces are used in this thesis as our image database.

Submitted by MARIA JACQUELINE ATOCHE BRAVO (jacqui_mab@hotmail.com) on 2016-05-13T13:28:28Z No. of bitstreams: 1 disertacao__Jacqui.pdf: 8482416 bytes, checksum: 2325158a94282088f873ac31bbd97305 (MD5) / Approved for entry into archive by Felipe Augusto Arakaki (arakaki@reitoria.unesp.br) on 2016-05-16T12:30:13Z (GMT) No. of bitstreams: 1 bravo_mja_me_sjrp.pdf: 8482416 bytes, checksum: 2325158a94282088f873ac31bbd97305 (MD5) / Made available in DSpace on 2016-05-16T12:30:13Z (GMT). No. of bitstreams: 1 bravo_mja_me_sjrp.pdf: 8482416 bytes, checksum: 2325158a94282088f873ac31bbd97305 (MD5) Previous issue date: 2016-03-31 / Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) / No presente trabalho, apresenta-se uma revisão sobre os principais abordagens para análise e classificação de texturas, entre eles o descritor LBP (Local Binary Pattern), o descritor LFP (Local Fuzzy Patterm) e o descritor MSLMP (Multi-scale Local Mapped Pattern), o qual é uma extensão multiescalar do descritor LMP (Local Mapped Pattern). Resultados anteriores presentes na literatura, indicaram que o MSLMP conseguiu resultados superiores aos mencionados anteriormente. Neste trabalho propõe-se uma análise mais abrangente sobre sua viabilidade para concluir que o MSLMP é mais eficaz que os anteriores. Essa análise é feita alterando-se a Matriz de Pesos para os pixels limiarizados. Para avaliar seu desempenho, foi utilizada a base de texturas do Album de Brodatz. Após processá-la pelo descritor MSLMP, com cada uma das matrizes de Pesos propostas neste trabalho, foram comparadas as taxas de acertos alcançadas usando a distância Chi-quadrado. Resultados experimentais mostram um valor de sensibilidade melhor para o descritor MSLMP em comparação aos outros descritores presentes na literatura. / This work, presents a review about the main techniques for analysis and classification of textures, including the LBP descriptor (Local Binary Pattern), the descriptor LFP (Local Fuzzy Pattern) and the descriptor MSLMP (Multi-Scale Local Mapped Pattern), which is a multi-scale extension of the LMP method (Local Mapped Pattern). Previous results present in the literature, indicated that the MSLMP achieved better results than those mentioned above. This work proposes a more comprehensive analysis of its feasibility to conclude that this descriptor is more effective than the others. This analysis is done by changing the weight matrix for the thresholding pixels. To evaluate its performance, it was used the texture base of the Brodatz album. After processing it by the descriptor MSLMP with each of the weights matrices proposed in this work, the achieved hit rates were compared by using the distance Chi-square. Experimental results show a better sensitivity value for MSLMP descriptor in comparison of other descriptors present in the literature. / CNPq: 131632/2014-0

La présente étude concerne l’évaluation de l’impact du séchage par air chaud (HAD), lyophilisation (FD) et « swell drying » (SD), procédé couplant le séchage par air chaud avec le procédé de Détente Instantanée Contrôlée (DIC), sur les fraises (Fragaria var. Camarosa). Il s’agit de comparer et de contraster les performances des procédés et la qualité du produit fini séché en termes des cinétiques de séchage et de réhydratation, de contenus en molécules bioactives et activité antioxydante, et des paramètres caractéristiques de texture comme croquant et croustillant. Les résultats obtenus ont montré que le procédé de SD comparé aux procédés classiques de séchage et de lyophilisation, réduit d’une façon importante le temps de séchage ainsi que les coûts d’opération. D’autre part, SD conserve la qualité nutritionnelle des fraises en gardant leur contenu en composants bioactifs et en augmentant leur disponibilité. De plus, une corrélation importante entre la capacité antioxydante et le contenu total d’anthocyanes a été établie. D’autre part, les fraises séchées par SD ont montré une très intéressante macro et micro-structure. Les produits ont présenté une haute expansion et une croustillance significative due au phénomène de micro-alvéolation par décompression instantanée par DIC. D’ailleurs, il a été possible de mesurer les caractéristiques instrumentales de croustillance/croquance des échantillons finaux séchés. Grâce à la possibilité de modifier, contrôler et optimiser les paramètres opératoires du procédé DIC, il a été possible d’obtenir un produit du type « snack » croustillant avec une très haute valeur nutritionnelle. / The aim of this study was to evaluate the effect of hot air drying (HAD), freeze-drying (FD) and swell drying (SD), which is a coupling of hot air drying to instant controlled pressure drop, (DIC) on the strawberry (Fragaria var. Camarosa) to compare and to contrast its quality in terms of drying and rehydration kinetics, bioactive compounds and its antioxidant activity, and texture parameters as crunchy and crispy features. The obtained results shown that SD method helped to reduce the drying time leading to a low-cost processing compared with classical hot air drying and freezedrying. SD globally preserved the strawberry’s nutritional value and bioactive compounds, increasing their availability. Moreover, a strong correlation between antioxidant activity and total anthocyanin content was established in SD strawberries. On the other hand, the swell-dried strawberries showed an interesting macro and micro-structure. They presented a high expansion ratios and significant crispness provoked by the micro-alveolation phenomenon induced as consequence of the instant decompression process in the DIC treatment. Moreover, it was possible to instrumentaly measure the crispy/crunchy features of the final dried samples. By assessing such crispy and healthy contents of fruit “snacking”, it was possible to modify, control, and optimize DIC operating parameters. And, it can be designed according to the industrial or consumer needs.

Cette thèse porte sur des canevas méthodologiques paramétriques pour la modélisation de champs de tenseurs de structure locaux (TSL) calculés sur des images texturées. Estimé en chaque pixel, le tenseur de structure permet la caractérisation de la géométrie d’une image texturée à travers des mesures d’orientation et d’anisotropie locales. Matrices symétriques semi-définies positives, les tenseurs de structure ne peuvent pas être manipulés avec les outils classiques de la géométrie euclidienne. Deux canevas statistiques riemanniens, reposant respectivement sur les espaces métriques a ne invariant (AI) et log-euclidien (LE), sont étudiés pour leur représentation. Dans chaque cas, un modèle de distribution gaussienne et de mélange associé sont considérés pour une analyse statistique. Des algorithmes d’estimation de leurs paramètres sont proposés ainsi qu’une mesure de dissimilarité. Les modèles statistiques proposés sont tout d’abord considérés pour décrire des champs de TSL calculés sur des images texturées. Les modèles AI et LE sont utilisés pour décrire des distributions marginales de TSL tandis que les modèles LE sont étendus afin de décrire des distributions jointes de TSL et de caractériser des dépendances spatiales et multi-échelles. L’ajustement des modèles théoriques aux distributions empiriques de TSL est évalué de manière expérimentale sur un ensemble de textures composées d’un spectre assez large de motifs structuraux. Les capacités descriptives des modèles statistiques proposés sont ensuite éprouvées à travers deux applications. Une première application concerne la reconnaissance de texture sur des images de télédétection très haute résolution et sur des images de matériaux carbonés issues de la microscopie électronique à transmission haute résolution. Dans la plupart des cas, les performances des approches proposées sont supérieures à celles obtenues par les méthodes de l’état de l’art. Sur l’espace LE, les modèles joints pour la caractérisation des dépendances spatiales au sein d’un champ de TSL améliorent légèrement les résultats des modèles opérant uniquement sur les distributions marginales. La capacité intrinsèque des méthodes basées sur le tenseur de structure à prendre en considération l’invariance à la rotation, requise dans beaucoup d’applications portant sur des textures anisotropes, est également démontrée de manière expérimentale. Une deuxième application concerne la synthèse de champs de TSL. A cet e et, des approches mono-échelle ainsi que des approches pyramidales multi-échelles respectant une hypothèse markovienne sont proposées. Les expériences sont effectuées à la fois sur des champs de TSL simulés et sur des champs de TSL calculés sur des textures réelles. Efficientes dans quelques configurations et démontrant d’un potentiel réel de description des modèles proposés, les expériences menées montrent également une grande sensibilité aux choix des paramètres qui peut s’expliquer par des instabilités d’estimation sur des espaces de grande dimension. / This thesis proposes and evaluates parametric frameworks for modelling local structure tensor (LST) fields computed on textured images. A texture’s underlying geometry is described in terms of orientation and anisotropy, estimated in each pixel by the LST. Defined as symmetric non-negative definite matrices, LSTs cannot be handled using the classical tools of Euclidean geometry. In this work, two complete Riemannian statistical frameworks are investigated to address the representation of symmetric positive definite matrices. They rely on the a ne-invariant (AI) and log-Euclidean (LE) metric spaces. For each framework, a Gaussian distribution and its corresponding mixture models are considered for statistical modelling. Solutions for parameter estimation are provided and parametric dissimilarity measures between statistical models are proposed as well. The proposed statistical frameworks are first considered for characterising LST fields computed on textured images. Both AI and LE models are first employed to handle marginal LST distributions. Then, LE models are extended to describe joint LST distributions with the purpose of characterising both spatial and multiscale dependencies. The theoretical models’ fit to empirical LST distributions is experimentally assessed for a texture set composed of a large diversity of patterns. The descriptive potential of the proposed statistical models are then assessed in two applications. A first application consists of texture recognition. It deals with very high resolution remote sensing images and carbonaceous material images issued from high resolution transmission electron microscopy technology. The LST statistical modelling based approaches for texture characterisation outperform, in most cases, the state of the art methods. Competitive texture classification performances are obtained when modelling marginal LST distributions on both AI and LE metric spaces. When modelling joint LST distributions, a slight gain in performance is obtained with respect to the case when marginal distributions are modelled. In addition, the LST based methods’ intrinsic ability to address the rotation invariance prerequisite that arises in many classification tasks dealing with anisotropic textures is experimentally validated as well. In contrast, state of the art methods achieve a rather pseudo rotation invariance. A second application concerns LST field synthesis. To this purpose, monoscale and multiscale pyramidal approaches relying on a Markovian hypothesis are developed. Experiments are carried out on toy LST field examples and on real texture LST fields. The successful synthesis results obtained when optimal parameter configurations are employed, are a proof of the real descriptive potential of the proposed statistical models. However, the experiments have also shown a high sensitivity to the parameters’ choice, that may be due to statistical inference limitations in high dimensional spaces.

L’anesthésie loco-régionale présente une alternative intéressante à l’anesthésie générale dans de nombreuses interventions chirurgicales. L’atout majeur de cette technique est qu’elle réduit grandement les scores de douleurs et améliore par la même la mobilité post-opératoire. L’anesthésie locorégionale écho-guidée (UGRA) devient aujourd’hui, la méthode de référence dans le domaine de l’anesthésie, offrant de nombreux avantages par rapport aux autres méthodes comme la neurostimulation. Cependant, cette technique nécessite en contrepartie un apprentissage spécifique afin d’éviter des complications sévères liées à une erreur de localisation visuelle du nerf dans les images échographiques. L’objectif de cette thèse est de faciliter et de sécuriser la pratique de l’anesthésie loco-régionale écho-guidée. Dans un premier temps, nous avons proposé une méthode de détection du nerf mettant en oeuvre un algorithme qui suite à un prétraitement à partir de filtres fréquentielles, réalise une analyse de texture par apprentissage. Dans ce cadre, deux nouvelles approches ont été explorées : l’une concerne la caractérisation du nerf qui s’appuie sur la prise en compte du bruit présent dans une image ultrasonore, bruit ayant été au préalable atténué partiellement. L’autre propose une technique de sélection des caractéristiques mettant en avant celles qui sont les moins redondantes et les plus pertinentes. Dans un second temps, après étude fine du comportement variable de la morphologie du nerf tout au long d’une séquence d’images ultrasonores, nous avons développé un modèle dynamique ayant comme paramètres des informations en lien avec la cohérence temporelle de la position, de la forme et la confiance de classification des ROI potentielles afin de générer une segmentation robuste. Il est proposé également dans cette partie, un nouveau modèle de forme prenant en compte un ensemble d’intervalles de points de repères du contour, permettant ainsi de s’adapter aux variations de la forme du nerf dans le temps. / Regional anesthesia presents an interesting alternative or complementary act to general anesthesia in many surgical procedures. It reduces pain scores, improves postoperative mobility and facilitates earlier hospital discharge. Ultrasound-Guided Regional Anesthesia (UGRA) has been gaining importance in the last few years, offering numerous advantages over alternative methods of nerve localization (neurostimulation or paraesthesia). However, nerve detection is one of the most difficult tasks that anesthetists can encounter in the UGRA procedure. The context of the present work is to provide practitioners with a method to facilitate and secure the practice of UGRA. However, automatic detection and segmentation in ultrasound images is still a challenging problem in many medical applications. This work addresses two main issues. The first one, we propose an algorithm for nerve detection and segmentation in ultrasound images, this method is composed of a pre-processing, texture analysis and machine learning steps. In this part of work, we explore two new approaches ; one to characterize the nerve and the second for selecting the minimum redundant and maximum relevant features. The second one, we studied the nerve detection in consecutive ultrasound frames. We have demonstrated that the development of an algorithm based on the temporal coherence of the position, the shape and the confidence measure of the classification, allows to generate a robust segmentation. In this work, we also propose a new model of shape based on a set of intervals landmarks able to adapt to the nerve shape under a morphological variations.