CoordenaÃÃo de AperfeiÃoamento de Pessoal de NÃvel Superior / Sistemas de recuperaÃÃo de imagens baseada em conteÃdo (do termo em inglÃs, Content-Based Image Retrieval - CBIR) que operam em bases com grande volume de dados constituem um problema relevante e desafiador em diferentes Ãreas do conhecimento, a saber, medicina, biologia, computaÃÃo, catalogaÃÃo em geral, etc. A indexaÃÃo das imagens nestas bases pode ser realizada atravÃs de conteÃdo visual como cor, textura e forma, sendo esta Ãltima caracterÃstica a traduÃÃo visual dos objetos em uma cena. Tarefas automatizadas em inspeÃÃo industrial, registro de marca, biometria e descriÃÃo de imagens utilizam atributos da forma, como os cantos, na geraÃÃo de descritores para representaÃÃo, anÃlise e reconhecimento da mesma, possibilitando ainda que estes descritores se adequem ao uso em sistemas de recuperaÃÃo. Esta tese aborda o problema da extraÃÃo de caracterÃsticas de formas planares binÃrias a partir de cantos, na proposta de um detector multiescala de cantos e sua aplicaÃÃo em um sistema CBIR. O mÃtodo de detecÃÃo de cantos proposto combina uma funÃÃo de angulaÃÃo do contorno da forma, a sua decomposiÃÃo nÃo decimada por transformada wavelet ChapÃu Mexicano e a correlaÃÃo espacial entre as escalas do sinal de angulaÃÃo decomposto. A partir dos resultados de detecÃÃo de cantos, foi realizado um experimento com o sistema CBIR proposto, em que informaÃÃes locais e globais extraÃdas dos cantos detectados da forma foram combinadas à tÃcnica DeformaÃÃo Espacial DinÃmica (do termo em inglÃs, Dynamic Space Warping), para fins de anÃlise de similaridade formas com tamanhos distintos. Ainda com este experimento foi traÃada uma estratÃgia de busca e ajuste dos parÃmetros multiescala de detectores de cantos, segundo a maximizaÃÃo de uma funÃÃo de custo. Na avaliaÃÃo de desempenho da metodologia proposta, e outras tÃcnicas de detecÃÃo de cantos, foram empregadas as medidas PrecisÃo e RevocaÃÃo. Estas medidas atestaram o bom desempenho da metodologia proposta na detecÃÃo de cantos verdadeiros das formas, em uma base pÃblica de imagens cujas verdades terrestres estÃo disponÃveis. Para a avaliaÃÃo do experimento de recuperaÃÃo de imagens, utilizamos a taxa Bullâs eye em trÃs bases pÃblicas. Os valores alcanÃados desta taxa mostraram que o experimento proposto foi bem sucedido na descriÃÃo e recuperaÃÃo das formas, dentre os demais mÃtodos avaliados. / Content-based image retrieval (CBIR) applied to large scale datasets is a relevant and challenging problem present in medicine, biology, computer science, general cataloging etc. Image indexing can be done using visual information such as colors, textures and shapes (the visual translation of objects in a scene). Automated tasks in industrial inspection, trademark registration, biostatistics and image description use shape attributes, e.g. corners, to generate descriptors for representation, analysis and recognition; allowing those descriptors to be used in image retrieval systems. This thesis explores the problem of extracting information from binary planar shapes from corners, by proposing a multiscale corner detector and its use in a CBIR system. The proposed corner detection method combines an angulation function of the shape contour, its non-decimated decomposition using the Mexican hat wavelet and the spatial correlation among scales of the decomposed angulation signal. Using the information provided by our corner detection algorithm, we made experiments with the proposed CBIR. Local and global information extracted from the corners detected on shapes was used in a Dynamic Space Warping technique in order to analyze the similarity among shapes of different sizes. We also devised a strategy for searching and refining the multiscale parameters of the corner detector by maximizing an objective function. For performance evaluation of the proposed methodology and other techniques, we employed the Precision and Recall measures. These measures proved the good performance of our method in detecting true corners on shapes from a public image dataset with ground truth information. To assess the image retrieval experiments, we used the Bullâs eye score in three public databases. Our experiments showed our method performed well when compared to the existing approaches in the literature.

The analysis of microstructural shapes is an underutilized tool in the field of materials science. Typical observations of morphology are qualitative, rather than quantitative, which prevents the identification of relationships between shape and the mechanical properties of a material. Recent advances in the fields of computer vision and high-dimensional analysis have made computer-based shape characterization feasible on a variety of materials. In this work, the relationship between microstructural shapes, and the properties and function of the material as a whole, is explored using moment invariants as global shape descriptors. A diifferent relationship is examined in each of three material systems: how the three-dimensional shapes of cells in the cotyledons of the plant Arabidopsis Thaliana can be used to identify cell function; the two-dimensional shapes of additive manufacturing feedstock powder and the ability to distinguish between images of powders from different samples; and the two-dimensional shapes of ' precipitates and their influence on the creep resistance of single crystal nickel-base superalloys. In the case of Arabidopsis Thaliana cotyledon cells, three-dimensional Zernike and Cartesian moment invariants were used to quantify morphology, and combined with size and orientation information. These feature sets were then analyzed using unsupervised and supervised machine learning methods. Moderate success was found using unsupervised methods, indicating that natural delineations in the data correlate to cell roles to some degree. Using supervised methods, a success rate of 90% was possible, indicating that these features can be used to identify cell function. The ability of two-dimensional Cartesian moment invariants to distinguish meaningful features in particles of additive manufacturing feedstock was tested by using these features to classify images of feedstock. Ultimately, simple histogram matching methods were unsuccessful, likely because they rely on the most common particles to draw conclusions. A bag-of-words method was used, which uses high-dimensional visualization and clustering techniques to classify individual particles by common features. Histograms of particle clusters are then used to represent each image. This method was far more successful, and a correct classification rate of up to 90% was found, and comparable rates were discovered using invariants which describe the shapes only broadly. This indicates that moment invariants are an effective measure of the morphologies of these types of particles, and can be used to classify powder shapes, which control many properties which are relevant to the additive manufacturing process. In the case of the superalloys, it has been shown that the shape distribution of ' precipitates can be tracked using second order moment invariants. In addition, several loworder moment invariants are shown to correlate to creep resistance in four alloys examined, which supports the idea that the shape of precipitates plays role in determining creep resistance in these alloys.

Many applications for 3d object analysis and visualization need to deal with object features. The most common features on triangulated surfaces are ridges and valleys which are interesting for matching, segmentation, smoothing, non-photo-realistic rendering and many other tasks. In some cases these features are interrupted or corrupted otherwise. In this paper we present an approach to close such gaps in feature regions. For this purpose a vectorfield on the surface, based on the detected features, is computed. The analysis of this vector field yields to possible junction points. These junction points are connected (guided by the vector field) to close the interrupted feature regions.

L'analyse de la morphologie axonale est un problème important en neuroscience. Diverses études ont montré que les caractéristiques morphologiques de ces structures donnent des informations sur son fonctionnement et permettent la caractérisation d'états pathologiques. En conséquence, il est important de développer des méthodes pour étudier leurs formes et quantifier leurs différences structurelles.Dans cette thèse on propose une méthode pour la comparaison des arbres axonaux qui inclue des informations topologiques et géométriques. La méthode est fondée sur la théorie des formes élastiques. Avec cette approche, nous pouvons exhiber le chemin géodésique entre deux formes et la forme moyenne d'un ensemble d'échantillons. En outre, nous proposons un schéma de classification à partir de cette métrique que nous comparons à l'état de l'art. Finalement, nous proposons un modèle stochastique pour la simulation de la croissance axonale défini par une chaîne de Markov. Il considère 2 processus principaux qui modélisent l'élongation et forme de l'axone et la génération des branches. Le processus de croissance dépend de différentes variables, dont un champ externe d'attraction généré par certaines molécules dans l'environnement. Les deux techniques proposées ont été validées sur une base d'images de microscopie confocale de neurones chez la Drosophile. Des neurones normaux et modifiés génétiquement ont été considérés. Les résultats montrent que la méthode de comparaison proposée fournit de meilleurs résultats que les méthodes décrites dans la littérature. De plus, les paramètres du modèle donnent des informations sur le processus de croissance de chaque population d'axones. / The morphological analysis of axonal trees is an important problem in neuroscience. It has been shown that the morphological characteristics of thesestructures provide information on their functioning and allows the characterization of pathological states. Therefore, it is of great importance to develop methods to analyze their shape and to quantify differences between structures. In this thesis we propose a method for the comparison of axonal trees that takes into account both topological and geometrical information. Using this method, which is based on the Elastic Shape Analysis Framework, we can compute the geodesic path between two axons and the mean shape of a population of trees. In addition, we derive a classfication scheme based on this metric and compare it with state of the art approaches. Finally, we propose a 2D discrete stochastic model for the simulation of axonal biogenesis. The model is defined by a third order Markov Chain and considers two main processes: the growth process that models the elongation and shape of the neurites and the bifurcation process that models the generation of branches. The growth process depends, among other variables, on an external attraction field. Both techniques were validated on a database of real fluorescent confocal microscopy images of neurons within Drosophila fly brains. Both normal neurons and neurons in which certain genes were inactivated have been considered. Results show that the proposed comparison method obtains better results that other methods found in the literature, and that the model parameter values provide information about the growth properties of the populations.

Supporting user designs of 3D contents remains a challenge in geometric modeling. Various modeling tools have been developed in recent years to facilitate architectural designs and artistic creations. However, these tools require both modeling skills and raw creativity. Instead of creating models from scratch, one of the most popular choices is to extract intrinsic patterns from exemplar inputs (e.g., shape collections and sketches), to produce creative models while preserving the patterns. The mod- eling process contains two main stages, analysis and synthesis. The analysis of input models is usually performed at component level, especially for man-made objects that can be decomposed into several semantic parts, for example, the seat and handles of a bicycle. The synthesis stage recombines parts of shapes to generate new models that usually have topological or geometric variations. In this thesis, we propose three design tools aimed at easing the modeling process. We focus on man-made objects and scenes such as buildings and furniture, as the functionality of such shapes can be analyzed at the component level. A relation graph, which is commonly used in shape analysis, can then be built to represent the input shapes. In our work, the graph nodes denote the elements of a model (i.e., rooms, shape parts, and strokes respectively), while the edges capture the intrinsic relations between connected elements. With the use of graph representations, we extract and present controllable components to users for supporting their designs. The emphasis of our work is on three aspects. Firstly, we propose a framework for supporting interior layout design, which allows users to manipulate the produced floor plans, i.e., changing the scales of rooms and their positions as well. When the user modifies the topology of a layout, the corresponding layout graph is updated and the room geometries are optimized under certain constraints, e.g., user specified scales, the adjacency of rooms, and fabrication considerations (i.e., economic construction cost). Secondly, we introduce replaceable substructures as arrangements of shape components that can be interchanged while ensuring boundary consistency. Based on the shape graphs that encode the structures of input models, we propose new automatic operations to discover replaceable substructures across models or within a model. We enforce a pair of subgraphs matching along their boundaries so that switching two subgraphs results in topological variations. Thirdly, we develop an interactive system that supports a freeform design by interpreting user sketches. 3D contents can be extracted from input strokes with or without user annotations. Our system accepts user strokes, analyzes their contacts and vanishing directions with respect to an anchored image, and projects 2D strokes to 3D space via a multi- stage optimization on spatial canvas selection. We demonstrate the computational approaches on a range of example models and design studies.

No description available.

Computing correspondences or maps between shapes is one of the oldest problems in Computer Graphics and Geometry Processing with a wide range of applications from deformation transfer, statistical shape analysis, to co-segmentation and exploration among a myriad others. A good map is supposed to be continuous, as-bijective-as-possible, accurate if there are ground-truth corresponding landmarks given, and lowdistortionw.r.t. different measures, for example as-conformal-as-possible to preserve the angles. This thesis contributes to the area of non-rigid shape matching and map space exploration in Geometry Processing. Specifically, we consider the discrete setting, where the shapes are discretized as amesh structure consisting of vertices, edges, and polygonal faces. In the simplest case, we only consider the graph structure with vertices and edges only. In this thesis, we design algorithms to compute soft correspondences between discrete shapes. Specifically, (1)we propose different regularizers, including orientation-preserving operator and the Resolvent Laplacian Commutativity operator, to promote the shape correspondences in the functional map framework. (2) We propose two refinement methods, namely BCICP and ZoomOut, to improve the accuracy, continuity, bijectivity and the coverage of given point-wisemaps. (3)We propose a tree structure and an enumeration algorithm to explore the map space between a pair of shapes that can update multiple high-quality dense correspondences.

Neural fields, also known as neural implicit representations, are powerful for modeling 3D shapes. They encode shapes as continuous functions mapping 3D coordinates to scalar values like the signed distance function (SDF) or occupancy probability. Neural fields represent complex shapes using an MLP. The MLP takes spatial coordinates, undergoes nonlinear transformations, and approximates the continuous function of the neural field. During training, the MLP's weights are learned through backpropagation. This PhD thesis presents novel methods for shape representation learning and generation with neural fields. The first part introduces an interpretable and high-quality reconstruction method for neural fields. A neural network predicts labeled points, improving surface visualization and interpretability. The method achieves accurate reconstruction even with rendered image input. A binary classifier, based on predicted labeled points, represents the shape's surface with precision. The second part focuses on shape generation, a challenge in generative modeling. Complex data structures like oct-trees or BSP-trees are challenging to generate with neural networks. To address this, a two-step framework is proposed: an autoencoder compresses the neural field into a fixed-size latent space, followed by training generative models within that space. Incorporating sparsity into the shape autoencoding network reduces dimensionality while maintaining high-quality shape reconstruction. Autoregressive transformer models enable the generation of complex shapes with intricate details. This research explores the potential of denoising diffusion models for 3D shape generation. The latent space efficiency is improved by further compression, leading to more efficient and effective generation of high-quality shapes. Remarkable shape reconstruction results are achieved, even without sparse structures. The approach combines the latest generative model advancements with novel techniques, advancing the field. It has the potential to revolutionize shape generation in gaming, manufacturing, and beyond. In summary, this PhD thesis proposes novel methods for shape representation learning, generation, and reconstruction. It contributes to the field of shape analysis and generation by enhancing interpretability, improving reconstruction quality, and pushing the boundaries of efficient and effective 3D shape generation.

The majority of car crash deaths occur in the front seats because the majority of occupants sit in the front seats. Traditionally, the rear seats were safer than the front seats because a front seated occupant would be closer to rigid structures such as the steering wheel, and they would be closer to the location of the impact. Therefore, government crash test regulations as well as academic and industry testing up to this point have principally focused on the front seats. Since the beginning of efforts to make cars safer, innovations were applied to the front seats first. Only some of these safety innovations have transitioned into the rear seats. Over the years, the front seats have gotten much safer due to advanced seatbelts with pretentioners and load limiters, airbags surrounding the driver, and structural changes to the vehicle frame to prevent intrusion into the occupant compartment. At the same time, occupant safety in the rear seats has also improved, however at only a fraction of the improvement of the front seats. With modern vehicles, the front seats have actually become safer than the rear seats for certain occupants and specific crash types (e.g., adult occupants in frontal crash). The lagging performance of the rear seats represents a problem because thousands of rear-seated occupants are injured or killed each year. With the rise in autonomous driving systems, the amount of occupants sitting in the rear seats, and therefore sustaining injury, could increase dramatically. In this dissertation, rear seats of a range of current vehicles were reconstructed to examine injury risk with the finite element models of two anthropomorphic test devices. These models showed a wide range of injury risks in the reconstructed seats. They were also able to show results similar to sled impact tests with the same vehicles. Knowledge gained from these reconstructions was then used to perform parametric studies on key variables that influence injury risk in the rear seats. From the parametric studies, it was found that the seat back angle, the width of the seatbelt anchors, and the presence of a seatbelt pretensioner had the largest influences on the injury risk. One of the injury mechanisms prevalent in the rear seats is submarining. Submarining likelihood and injury probability is difficult to predict with anthropomorphic test devices; however, human body models can help to improve injury prediction in these cases. To improve the injury prediction capability of human body models, several additions to the models are necessary. This dissertation outlines the investigation of spleen and kidney shapes through statistical shape analysis. This type of analysis allows more customizable human body models which could better capture the injury probability to these organs for a wider range of the population. Finally, subject-specific models of ribs were created to investigate factors affecting the predictive capability of finite element models. The findings and methodology from this body of work have the ability to add critical contributions to the understanding of injury risk and injury mechanisms in the rear seats. / Doctor of Philosophy / The majority of car crash deaths occur in the front seats because the majority of occupants sit in the front seats. Traditionally, the rear seats were safer than the front seats because a front seated occupant would be closer to hard objects such as the steering wheel, and they would be closer to the location of the impact. Therefore, government crash test regulations as well as academic and industry testing up to this point have principally focused on the front seats. Since the beginning of efforts to make cars safer, technology such as seatbelts and airbags were applied to the front seats first. Only some of this technology has been added into the rear seats. Over the years, the front seats have gotten much safer due to all the work focused on the front seats. At the same time, the rear seats have also improved, however at only a fraction of the improvement of the front seats. With modern vehicles, the front seats have actually become safer than the rear seats in some cases. The lagging performance of the rear seats represents a problem because thousands of rear-seated occupants are injured or killed each year. With the rise in self driving cars, the amount of occupants sitting in the rear seats, and therefore sustaining injury, could increase dramatically. In this dissertation, rear seats of a range of current vehicles were reconstructed to examine injury risk with the models of two crash test dummies. These models showed a wide range of injury risks in the reconstructed seats. They were also able to show results similar to physical tests with the same vehicles. Knowledge gained from this work was then used to help look at key variables that influence injury risk in the rear seats. It was found that the angle of the seat back, the width of the seatbelt anchors, and the presence of advanced seatbelts had the largest influences on the injury risk. One of the injury mechanisms prevalent in the rear seats is submarining, where the seatbelt slides up off the hips. Submarining likelihood and injury probability is difficult to predict with crash test dummies; however, human body models can help to improve injury prediction in these cases. To improve the injury prediction capability of human body models, several additions to the models are necessary. This dissertation outlines the investigation of spleen and kidney shapes to allow more customizable human body models which could better capture the injury probability to these organs for a wider range of the population. Finally, subject-specific models of ribs were created to investigate factors affecting the predictive capability of rib models. The findings and methodology from this body of work have the ability to add critical contributions to the understanding of injury risk and injury mechanisms in the rear seats.

The number of women diagnosed with the breast cancer continues to rise year on year. Breast cancer is now the most common type of cancer in the UK, with over 55000 cases reported last year. In most cases, mammography is the first step towards diagnosing breast cancer. However, it continues to have many practical limitations as compared to more sophisticated modalities such as MRI. The relatively low cost of mammography, together with the ever increasing risk of women contracting the disease, has led to many developed countries having a breast screening program. These routine breast screens are taken at different points in time and are called temporal mammograms. Currently, a radiologist tends to qualitatively assess temporal mammograms and look for any abnormalities or suspicious regions that might be of a concern. In this thesis, we develop an automatic shape analysis model that can detect and quantify such changes inside the breast. This will not only help in early diagnosis of the disease, which is key to survival, but will potentially aid prognosis and post treatment care. The core to this thesis is the use of Circular Integral Invariants. We explore its multi-scale properties and use it for image smoothing to reduce image noise and enhance features for segmentation. We implement, modify and enhance a segmentation method which previously has been successfully used to acquire breast regions of interest. We applied such Integral Invariants for shape description, to be used for shape matching as well as for subdividing shapes into sub-regions and quantifying the differences between two such shapes. We combine boundary information with the information from inside a shape, thus eccentrically transforming shapes before describing their structure. We develop a novel false positives reduction method based on Integral Invariants scale space. A second aspect of the thesis is the evaluation of and emphasis on the use of breast density maps against the commonly used intensity maps or x-rays. We find density maps sufficient to use in clinical practice. The methods developed in this thesis aim to help clinicians in making diagnostic decision at the point of case. Our shape analysis model is easy to compute, fast and very general in nature that could be deployed in a wide range of applications, beyond mammography.