[pt] Diversas aplicações, em matemática, computação gráfica, medicina, geofísica e outras áreas, têm explorado a representação de sólidos por superfícies de contorno, em particular malhas poligonais. As malhas podem aproximar com muita precisão as propriedades geométricas da superfície de contorno de um sólido e ainda guardar importantes propriedades topológicas das superfícies como gênero, bordo e conexidade. Devido à grande complexidade dessas malhas, elas são geralmente processadas em meios computacionais usando alguma estrutura de dados. Essas estruturas guardam, além da geometria da malha, informações de incidências e adjacências entre os elementos da malha e exigem uma capacidade de armazenamento e processamento em função da complexidade da malha. Apesar da evolução dos recursos computacionais disponíveis para a manipulação destas estruturas, malhas extremamente complexas com milhões de elementos inviabilizam o armazenamento, processamento e transmissão de sua estrutura de dados nos meios computacionais. Muitas pesquisas recentes estão voltadas para a obtenção de processos de simplificação de malhas que permitam representar a mesma superfície com menos elementos na estrutura de dados e processos de compressão que codifiquem os modelos em formatos menores para efeitos de transmissão e armazenamento em mídia. Neste trabalho, desenvolvemos operadores, em uma estrutura de dados compacta, para a simplificação de malhas através da decimação de células da superfície. Objetivamos, com esses operadores, obter uma malha menos complexa que preserve as propriedades topológicas da superfície original e ainda, controlar as propriedades geométricas como volume, área e aspecto visual da mesma. Apresentamos ainda algumas aplicações para os processos de simplificação desenvolvidos com esses operadores. / [en] Many applications, in mathematics, computer graphics, medical imaging, geophysics and others, have used the representation of solids by their boundary surface, usually polygonal meshes. Those meshes can represent, with high precision, the geometric properties of the boundary surface of solid and also store important topological surface properties as genus, boundary and connected components. Because of the high complexity of such meshes, they are usually processed by the computers using specific data structures. These structures store, beyond the mesh geometry, information about incidence and adjacency relations among the mesh elements. They require computational resources for storage and processing according to the mesh complexity. Even with the development of the computational resources available for handling such structures, very large meshes with millions of elements are hard to store, to process and to exchange through the web. Many recent researches are looking for mesh simplification process that allows to represent the same surface with fewer elements and compression process to encode it in compact ways for transmition and storage. In this work, we develop topological operators, in a concise data structure, for simplifying meshes by the decimation of its cells. One of our goals, with these operators, is to obtain a mesh with a low complexity that preserves the topological properties from the original surface without loosing the control of the geometric proprieties as volume, area and visual aspect.

Dans cette thèse, nous présentons des méthodes pour construire une surface géologique optimal à partir d’une quantité énorme de points 3D extraits de cubes sismiques. Appliquer le processus à l’ensemble des points induit un risque important de contraction de la surface de sorte que l’extraction de la frontière initiale est une étape importante permettant une simplification à l’intérieur de la surface. La forme globale de la surface sera alors mieux respectée pour la reconstruction de la surface triangulaire finale. Nos propositions sont basées sur la régularité des données qui permet, même si des données sont manquantes, d’obtenir facilement les informations de voisinage. Tout d’abord, nous présentons une nouvelle méthode pour extraire et simplifier la frontière d’une surface d’élévation définie par un ensemble de voxels dans un grand volume 3D où des données sont manquantes. Deuxièmement, une méthode pour simplifier la surface à l’intérieur de sa frontière est présentée. Elle comprend une étape de simplification grossière optionnelle suivie par une étape plus fine basée sur l’étude des courbures. Nous tenons également compte du fait que la densité de données doit changer graduellement afin de recevoir à la dernière étape d’une surface triangulée avec de meilleurs triangles. Troisièmement, nous avons proposé une nouvelle méthode rapide pour trianguler la surface après simplification. / Reconstructing surfaces with data coming from an automatic acquisition technique always entails the problem of mass of data. This implies that the usual processes cannot be applied directly. Therefore, it leads to a mandatory data reduction process. An effective algorithm for a rapid processing while keeping the original model is a valuable tool for constructing an optimal surface and managing the complex data.In this dissertation, we present methods for building an optimal geological surface from a huge amount of 3D points extracted from seismic cubes. Applying the process to the whole set of points induces an important risk of surface shrinking so that the initial boundary extraction is an important step permitting a simplification inside the surface. The global surface shape will then be better kept for the reconstruction of the final triangular surface. Our proposals are based on the regularity of data which permits, even if data are missing, to easily obtain the neighboring information. Firstly, we present a new method to extract and simplify the boundary of an elevation surface given as voxels in a large 3D volume having the characteristics to be sparse. Secondly, a method for simplifying the surface inside its boundary is presented with a rough optional simplification step followed by a finer one based on curvatures. We also keep into consideration that the density of data must gradually change in order to receive in the last step a triangulated surface with better triangles. Thirdly, we have proposed a new and fast method for triangulating the surface after simplification.

The high level of realism and interaction in many computer graphic applications requires techniques for processing complex geometric models. First, we present a method that provides an accurate low-resolution approximation from a multi-chart textured model that guarantees geometric fidelity and correct preservation of the appearance attributes. Then, we introduce a mesh structure called Compact Model that approximates dense triangular meshes while preserving sharp features, allowing adaptive reconstructions and supporting textured models. Next, we design a new space deformation technique called *Cages based on a multi-level system of cages that preserves the smoothness of the mesh between neighbouring cages and is extremely versatile, allowing the use of heterogeneous sets of coordinates and different levels of deformation. Finally, we propose a hybrid method that allows to apply any deformation technique on large models obtaining high quality results with a reduced memory footprint and a high performance. / L’elevat nivell de realisme i d’interacció requerit en múltiples aplicacions gràfiques fa que siguin necessàries tècniques pel processament de models geomètrics complexes. En primer lloc, presentem un mètode de simplificació que proporciona una aproximació precisa de baixa resolució d'un model texturat que garanteix fidelitat geomètrica i una correcta preservació de l’aparença. A continuació, introduïm el Compact Model, una nova estructura de dades que permet aproximar malles triangulars denses preservant els trets més distintius del model, permetent reconstruccions adaptatives i suportant models texturats. Seguidament, hem dissenyat *Cages, un esquema de deformació basat en un sistema de caixes multi-nivell que conserva la suavitat de la malla entre caixes veïnes i és extremadament versàtil, permetent l'ús de conjunts heterogenis de coordenades i diferents nivells de deformació. Finalment, proposem un mètode híbrid que permet aplicar qualsevol tècnica de deformació sobre models complexes obtenint resultats d’alta qualitat amb una memòria reduïda i un alt rendiment.