Skip to Main content Skip to Navigation

Modélisation par bruit procédural et rendu de détails volumiques de surfaces dans les scènes virtuelles

Abstract : The growing power of graphics processing units (GPU) in mainstream computers creates a need for a higher quality and complexity of virtual scenes. Managing this complexity for natural objects such as trees or grass fields or even animals is painstaking, due to the large amount of small objects decorating their surface. The diversity of such details, mandatory for realistic rendering of natural objects, translates in a longer authoring time, a higher memory requirement and a more complex evaluation. We review in this thesis the related works on data representations and on-the-fly generation methods used for the creation and real-time rendering of details over large surfaces. We focus our study on the particular case of grass fields and fur : the fuzzy visual appearance of those surfaces is obtained by the distribution of many self-similar blades or strands, creating a pattern closely related to a noise with structural features. We first present a procedural noise that aims at spatial modeling of self-similar elements and their distribution. The elliptical Gaussian function used as a modeling primitive and the controlled non-uniform distribution of elements allows for various type of patterns to be modeled, from stochastic to near-regular one, while including structural features. The by-example analysis process based on an ellipse fitting method allows a fast configuration of the noise for patterns reproduction. We further introduce an extension of this noise model for the authoring of procedural shell textures of strand-based or more complex volumetric details. For interactive authoring of such volumetric pattern, an image-order and an object-order rendering methods are proposed, both methods being optimized for an implementation on the GPU. Our rendering methods allow for interactive visualization of a visually-convincing result.
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Monday, January 23, 2017 - 11:16:16 AM
Last modification on : Wednesday, December 22, 2021 - 11:58:02 AM
Long-term archiving on: : Monday, April 24, 2017 - 12:59:54 PM


Version validated by the jury (STAR)


  • HAL Id : tel-01443555, version 1



Nicolas Pavie. Modélisation par bruit procédural et rendu de détails volumiques de surfaces dans les scènes virtuelles. Synthèse d'image et réalité virtuelle [cs.GR]. Université de Limoges, 2016. Français. ⟨NNT : 2016LIMO0080⟩. ⟨tel-01443555⟩



Record views


Files downloads