Nowadays, virtual objects have become omnipresent. We can find them in various domains such as entertainment (movies, video games, etc.), computer-aided design or virtual reality. Our main focus in this document is the modeling of 3D objects in the domain of artistic creation, where rich images creation requires highly detailed and complex models. Subdivision surfaces, the most used surface representation in this domain, quickly become very dense as the user add details, and manual handling of the connectivity becomes too cumbersome. A standard approach to handle the complexity of such models is to separate the overall shape of the surface and the details. Although, these detail maps are often stored in bitmap images that does not provide the advantages of vectorial representation, which complicate some tasks, like animation. In this document, we present two new vectorial representations: the first one for the base surface, the second one for the detail maps. For the later, we use a vectorial representation called diffusion images that allow to create smooth or sharp variations from a small set of constraints. This enables us to represent geometry as well as color or any other parameter required for rendering, while keeping high-level controls. Our first contribution is a surface representation, called LS3, based on the combination of subdivision surfaces and point set surfaces. This approach reduces notably artifacts that subdivision surfaces produce around so called extraordinary vertices. We also present a numerical analysis of the mathematical properties of these surfaces, that show that they behave at least as well as classical subdivision schemes. Our second contribution is a solver for diffusion images that has the particularity to produce as output a denser vectorial representation which is light and fast to evaluate. We show the advantages of this approach on several examples that would be hard or impossible to produce with former methods. To conclude, we show how these two contributions can be used together to obtain a fully vectorial surface representation able to produce detailed surfaces without needing to deal with complex connectivity.
