Image segmentation is an important research area in digital image processing with several applications in vision-guided autonomous robotics, product quality inspection, medical diagnosis, the analysis of remotely sensed images, etc. The aim of image segmentation can be defined as partitioning an image into homogeneous regions in terms of the features of pixels extracted from the image. Image segmentation methods can be classified into four main categories: 1) clustering methods, 2) region-based methods, 3) hybrid methods, and 4) bayesian methods. In this thesis, major image segmentation methods belonging to first three categories are examined and tested on typical images. Moreover, improvements are also proposed to well-known Recursive Shortest-Spanning Tree (RSST) algorithm. The improvements aim to better model each region during merging stage. Namely, grayscale histogram, joint histogram and homogeneous texture are used for better region modeling.

No description available.

Dans cette thèse, nous étudions le problème de l'inférence bayésienne dans les graphes factoriels, en particulier les codes LDPC, quasiment résolus par les algorithmes de message-passing. Nous réalisons en particulier une étude approfondie du Belief Propagation (BP) dont la question de la sous-optimalité est soulevée dans le cas où le graphe factoriel présente des boucles. A partir de l'équivalence entre le BP et l'approximation de Bethe en physique statistique qui se généralise à l'approximation basée régions, nous détaillons le Generalized Belief Propagation (GBP), un algorithme de message-passing entre des clusters du graphe factoriel. Nous montrons par des expériences que le GBP surpasse le BP dans les cas où le clustering est réalisé selon les structures topologiques néfastes qui empêchent le BP de bien décoder, à savoir les trapping sets. Au-delà de l'étude des performances en termes de taux d'erreur, nous confrontons les deux algorithmes par rapport à leurs dynamiques face à des événements d'erreur non triviaux, en particulier lorsqu'ils présentent des comportements chaotiques. Par des estimateurs classiques et originaux, nous montrons que l'algorithme du GBP peut dominer l'algorithme du BP. / This thesis addresses the problem of inference in factor graphs, especially the LDPC codes, almost solved by message-passing algorithms. In particular, the Belief Propagation algorithm (BP) is investigated as a particular message-passing algorithm whose suboptimality is discussed in the case where the factor graph has a loop-like topology. From the equivalence between the BP and the Bethe approximation in statistical physics that is generalized to the region-based approximation, is detailed the Generalized Belief Propagation algorithm (GBP), a message-passing algorithm between clusters of the factor graph. It is experimentally shown to surpass the BP in the cases where the clustering deals with the harmful topological structures that prevents the BP from rightly decoding any LDPC code, namely the trapping sets. We do not only confront the BP and the GBP algorithms according to their performance from the point of view of the channel coding with the error-rate, but also according to their dynamical behaviors for non-trivial error-events for which both algorithms can exhibit chaotic beahviors. By means of classical and original dynamical quantifiers, it is shown that the GBP algorithm can overcome the BP algorithm.

The popularization of social media, combined with the widespread use of smartphones and the use of advanced equipment in hospitals and medical centers has generated single and sequences of complex data, including images of high quality and in large quantity. Providing appropriate tools to extract meaningful knowledge from such data is a big challenge, and taking advantage of existing approaches to find patterns from images can be meaningful. While many potential techniques have been proposed to analyze images, most of the processing performed by image mining techniques consider the entire image. Thus, regions that are not of interest are considered in the analysis step, without proper distinction and consequently damaging most tasks. This doctorate PhD research has the following thesis: The analysis of image regions, combined to additional information, leads to more accurate mining results regarding the entire image, and also helps the processing of sequences of images, speeding-up costly pipelines and making it possible to infer knowledge from objects movement. We evaluate this thesis in three application scenarios. In the first scenario, we analyzed regions of images from emergency situations, gathered from social media and which depict smoke regions. We were able to segment smoke regions and improve the classification of smoke images by up to 23%, compared to global approaches. In the second scenario, we worked with images from the medical context, containing Interstitial Lung Diseases (ILD). We classified the images considering the uncertainty of each lung region to contain different abnormalities, representing the obtained results with a heat map visualization. Our approach was able to outperform its competitors in the classification of lung regions by up to four of five classes of abnormalities. In the third scenario, we dealt with sequences of microscopic images depicting embryos being developed over time. Using region-based information of images, we were able to track and predict cells over time and build their motion vector. Our approaches showed an improvement of up to 57% in quality, and a speed-up of the tracking pipeline by up to 81:9%. Therefore, this PhD research contributed to the state-of-the-art by introducing methods of region-based image analysis for the three aforementioned application scenarios. / A popularização de redes sociais e o uso generalizado de smartphones e equipamentos avançados em hospitais têm gerado dados complexos e sequências de dados, tais como imagens de alta qualidade, em grande quantidade. Fornecer ferramentas apropriadas para extrair conhecimento útil de tais dados é um grande desafio, e tirar vantagem de abordagens existentes para encontrar padrões em imagens pode ser significativo. Enquanto diversas técnicas em potencial têm sido propostas para analisar imagens, grande parte dessas técnicas consideram a imagem inteira na análise. Assim, regiões que não são de interesse são consideradas na etapa de análise, sem distinção apropriada e consequentemente prejudicando diversas tarefas. Esta pesquisa de Doutorado baseou-se na seguinte tese: A análise de regiões de imagens, combinada com informações adicionais, leva a resultados de mineração mais precisos em relação à imagem inteira, ajudando também no processamento de sequências de imagens, acelerando pipelines custosos e tornando possível inferir conhecimento do movimento de objetos. Essa tese foi avaliada em três cenários de aplicação. No primeiro cenário, foram analisadas regiões de imagens de situações de emergência, obtidas por meio de redes sociais e que apresentavam regiões de fumaça. Os métodos propostos são capazes de segmentar regiões de fumaça e melhorar a classificação global de imagens em até 23% em comparação ao estado da arte. No segundo cenário, foram abordadas imagens do contexto médico, contendo doenças pulmonares intersticiais. As imagens foram classificadas considerando a incerteza de cada região do pulmão em conter diferentes anormalidades, representando os resultados obtidos por meio de uma visualização baseada em mapas de calor. A abordagem proposta foi melhor que os competidores na tarefa de classificação de regiões pulmonares, apresentando melhores resultados em até quatro de cinco anormalidades. No terceiro cenário, foram tratadas de sequências de imagens microscópicas, exibindo embriões se desenvolvendo ao longo do tempo. Com o uso de informações das imagens baseadas em regiões, foi possível rastrear e predizer trajetórias de células ao longo do tempo, e também construir o vetor de movimento das mesmas. As abordagens propostas mostraram uma melhora de até 57% em qualidade, e uma melhora de tempo no pipeline de rastreamento de até 81:9%. Esta tese de Doutorado contribuiu para o estado da arte introduzindo métodos de análise de imagem baseados em região para os três cenários de aplicação mencionados anteriormente.

Data Distribution Management (DDM) deals with two basic problems: how to distribute data generated at the application layer among underlying nodes in a distributed system and how to retrieve data back whenever it is necessary. This thesis explores DDM in two different network environments: peer-to-peer (P2P) overlay networks and cluster-based network environments. DDM in P2P overlay networks is considered a more complete concept of building and maintaining a P2P overlay architecture than a simple data fetching scheme, and is closely related to the more commonly known associative searching or queries. DDM in the cluster-based network environment is one of the important services provided by the simulation middle-ware to support real-time distributed interactive simulations. The only common feature shared by DDM in both environments is that they are all built to provide data indexing service. Because of these fundamental differences, we have designed and developed a novel distributed data structure, Hierarchically Distributed Tree (HD Tree), to support range queries in P2P overlay networks. All the relevant problems of a distributed data structure, including the scalability, self-organizing, fault-tolerance, and load balancing have been studied. Both theoretical analysis and experimental results show that the HD Tree is able to give a complete view of system states when processing multi-dimensional range queries at different levels of selectivity and in various error-prone routing environments. On the other hand, a novel DDM scheme, Adaptive Grid-based DDM scheme, is proposed to improve the DDM performance in the cluster-based network environment. This new DDM scheme evaluates the input size of a simulation based on probability models. The optimum DDM performance is best approached by adapting the simulation running in a mode that is most appropriate to the size of the simulation.

Data Distribution Management (DDM) deals with two basic problems: how to distribute data generated at the application layer among underlying nodes in a distributed system and how to retrieve data back whenever it is necessary. This thesis explores DDM in two different network environments: peer-to-peer (P2P) overlay networks and cluster-based network environments. DDM in P2P overlay networks is considered a more complete concept of building and maintaining a P2P overlay architecture than a simple data fetching scheme, and is closely related to the more commonly known associative searching or queries. DDM in the cluster-based network environment is one of the important services provided by the simulation middle-ware to support real-time distributed interactive simulations. The only common feature shared by DDM in both environments is that they are all built to provide data indexing service. Because of these fundamental differences, we have designed and developed a novel distributed data structure, Hierarchically Distributed Tree (HD Tree), to support range queries in P2P overlay networks. All the relevant problems of a distributed data structure, including the scalability, self-organizing, fault-tolerance, and load balancing have been studied. Both theoretical analysis and experimental results show that the HD Tree is able to give a complete view of system states when processing multi-dimensional range queries at different levels of selectivity and in various error-prone routing environments. On the other hand, a novel DDM scheme, Adaptive Grid-based DDM scheme, is proposed to improve the DDM performance in the cluster-based network environment. This new DDM scheme evaluates the input size of a simulation based on probability models. The optimum DDM performance is best approached by adapting the simulation running in a mode that is most appropriate to the size of the simulation.

Data Distribution Management (DDM) deals with two basic problems: how to distribute data generated at the application layer among underlying nodes in a distributed system and how to retrieve data back whenever it is necessary. This thesis explores DDM in two different network environments: peer-to-peer (P2P) overlay networks and cluster-based network environments. DDM in P2P overlay networks is considered a more complete concept of building and maintaining a P2P overlay architecture than a simple data fetching scheme, and is closely related to the more commonly known associative searching or queries. DDM in the cluster-based network environment is one of the important services provided by the simulation middle-ware to support real-time distributed interactive simulations. The only common feature shared by DDM in both environments is that they are all built to provide data indexing service. Because of these fundamental differences, we have designed and developed a novel distributed data structure, Hierarchically Distributed Tree (HD Tree), to support range queries in P2P overlay networks. All the relevant problems of a distributed data structure, including the scalability, self-organizing, fault-tolerance, and load balancing have been studied. Both theoretical analysis and experimental results show that the HD Tree is able to give a complete view of system states when processing multi-dimensional range queries at different levels of selectivity and in various error-prone routing environments. On the other hand, a novel DDM scheme, Adaptive Grid-based DDM scheme, is proposed to improve the DDM performance in the cluster-based network environment. This new DDM scheme evaluates the input size of a simulation based on probability models. The optimum DDM performance is best approached by adapting the simulation running in a mode that is most appropriate to the size of the simulation.

Data Distribution Management (DDM) deals with two basic problems: how to distribute data generated at the application layer among underlying nodes in a distributed system and how to retrieve data back whenever it is necessary. This thesis explores DDM in two different network environments: peer-to-peer (P2P) overlay networks and cluster-based network environments. DDM in P2P overlay networks is considered a more complete concept of building and maintaining a P2P overlay architecture than a simple data fetching scheme, and is closely related to the more commonly known associative searching or queries. DDM in the cluster-based network environment is one of the important services provided by the simulation middle-ware to support real-time distributed interactive simulations. The only common feature shared by DDM in both environments is that they are all built to provide data indexing service. Because of these fundamental differences, we have designed and developed a novel distributed data structure, Hierarchically Distributed Tree (HD Tree), to support range queries in P2P overlay networks. All the relevant problems of a distributed data structure, including the scalability, self-organizing, fault-tolerance, and load balancing have been studied. Both theoretical analysis and experimental results show that the HD Tree is able to give a complete view of system states when processing multi-dimensional range queries at different levels of selectivity and in various error-prone routing environments. On the other hand, a novel DDM scheme, Adaptive Grid-based DDM scheme, is proposed to improve the DDM performance in the cluster-based network environment. This new DDM scheme evaluates the input size of a simulation based on probability models. The optimum DDM performance is best approached by adapting the simulation running in a mode that is most appropriate to the size of the simulation.

Given satellite images with accompanying pixel classifications and elevation data, we propose different solutions to object detection. The first method uses hierarchical clustering for segmentation and then employs different methods of classification. One of these classification methods used domain knowledge to classify objects while the other used Support Vector Machines. Additionally, a combination of three Support Vector Machines were used in a hierarchical structure which out-performed the regular Support Vector Machine method in most of the evaluation metrics. The second approach is more conventional with different types of Convolutional Neural Networks. A segmentation network was used as well as a few detection networks and different fusions between these. The Convolutional Neural Network approach proved to be the better of the two in terms of precision and recall but the clustering approach was not far behind. This work was done using a relatively small amount of data which potentially could have impacted the results of the Machine Learning models in a negative way.

Abstract<p>Since 1999, Very High spatial Resolution satellite data (Ikonos-2, QuickBird and OrbView-3) represent the surface of the Earth with more detail. However, information extraction by multispectral pixel-based classification proves to have become more complex owing to the internal variability increase in the land-cover units and to the weakness of spectral resolution. <p>Therefore, one possibility is to consider the internal spectral variability of land-cover classes as a valuable source of spatial information that can be used as an additional clue in characterizing and identifying land cover. Moreover, the spatial resolution gap that existed between satellite images and aerial photographs has strongly decreased, and the features used in visual interpretation transposed to digital analysis (texture, morphology and context) can be used as additional information on top of spectral features for the land cover classification.<p>The difficulty of this approach is often to transpose the visual features to digital analysis.<p>To overcome this problem region-based classification could be used. Segmentation, before classification, produces regions that are more homogeneous in themselves than with nearby regions and represent discrete objects or areas in the image. Each region becomes then a unit analysis, which makes it possible to avoid much of the structural clutter and allows to measure and use a number of features on top of spectral features. These features can be the surface, the perimeter, the compactness, the degree and kind of texture. Segmentation is one of the only methods which ensures to measure the morphological features (surface, perimeter.) and the textural features on non-arbitrary neighbourhood. In the pixel-based methods, texture is calculated with mobile windows that smooth the boundaries between discrete land cover regions and create between-class texture. This between-class texture could cause an edge-effect in the classification.<p><p>In this context, our research focuses on the potential of land cover region-based classification of VHR satellite data through the study of the object extraction capacity of segmentation processes, and through the study of the relevance of region features for classifying the land-cover classes in different kinds of Belgian landscapes; always keeping in mind the parallel with the visual interpretation which remains the reference.<p><p>Firstly, the results of the assessment of four segmentation algorithms belonging to the two main segmentation categories (contour- and region-based segmentation methods) show that the contour detection methods are sensitive to local variability, which is precisely the problem that we want to overcome. Then, a pre-processing like a filter may be used, at the risk of losing a part of the information. The “region-growing” segmentation that uses the local variability in the segmentation process appears to be the best compromise for the segmentation of different kinds of landscape.<p>Secondly, the features calculated thanks to segmentation seem to be relevant to identify some land-cover classes in urban/sub-urban and rural areas. These relevant features are of the same type as the features selected visually, which shows that the region-based classification gets close to the visual interpretation. <p>The research shows the real usefulness of region-based classification in order to classify the land cover with VHR satellite data. Even in some cases where the features calculated thanks to the segmentation prove to be useless, the region-based classification has other advantages. Working with regions instead of pixels allows to avoid the salt-and-pepper effect and makes the GIS integration easier.<p>The research also highlights some problems that are independent from the region-based classification and are recursive in VHR satellite data, like shadows and the spatial resolution weakness for identifying some land-cover classes.<p><p>Résumé<p>Depuis 1999, les données satellitaires à très haute résolution spatiale (IKONOS-2, QuickBird and OrbView-3) représentent la surface de la terre avec plus de détail. Cependant, l’extraction d’information par une classification multispectrale par pixel devient plus complexe en raison de l’augmentation de la variabilité spectrale dans les unités d’occupation du sol et du manque de résolution spectrale de ces données. Cependant, une possibilité est de considérer cette variabilité spectrale comme une information spatiale utile pouvant être utilisée comme une information complémentaire dans la caractérisation de l’occupation du sol. De plus, de part la diminution de la différence de résolution spatiale qui existait entre les photographies aériennes et les images satellitaires, les caractéristiques (attributs) utilisées en interprétation visuelle transposées à l’analyse digitale (texture, morphologie and contexte) peuvent être utilisées comme information complémentaire en plus de l’information spectrale pour la classification de l’occupation du sol.<p><p>La difficulté de cette approche est la transposition des caractéristiques visuelles à l’analyse digitale. Pour résoudre ce problème la classification par région pourrait être utilisée. La segmentation, avant la classification, produit des régions qui sont plus homogène en elles-mêmes qu’avec les régions voisines et qui représentent des objets ou des aires dans l’image. Chaque région devient alors une unité d’analyse qui permet l’élimination de l’effet « poivre et sel » et permet de mesurer et d’utiliser de nombreuses caractéristiques en plus des caractéristiques spectrales. Ces caractéristiques peuvent être la surface, le périmètre, la compacité, la texture. La segmentation est une des seules méthodes qui permet le calcul des caractéristiques morphologiques (surface, périmètre, …) et des caractéristiques texturales sur un voisinage non-arbitraire. Avec les méthodes de classification par pixel, la texture est calculée avec des fenêtres mobiles qui lissent les limites entre les régions d’occupation du sol et créent une texture interclasse. Cette texture interclasse peut alors causer un effet de bord dans le résultat de la classification.<p><p>Dans ce contexte, la recherche s’est focalisée sur l’étude du potentiel de la classification par région de l’occupation du sol avec des images satellitaires à très haute résolution spatiale. Ce potentiel a été étudié par l’intermédiaire de l’étude des capacités d’extraction d’objet de la segmentation et par l’intermédiaire de l’étude de la pertinence des caractéristiques des régions pour la classification de l’occupation du sol dans différents paysages belges tant urbains que ruraux. / Doctorat en sciences agronomiques et ingénierie biologique / info:eu-repo/semantics/nonPublished