Appearance preserving octree-textures

Abstract

Because of their geometric complexity, high resolution 3D models, either designed in high-end modeling packages or acquired with range scanning devices, cannot be directly used in applications that require rendering at interactive framerates. One clever method to overcome this limitation is to perform an appearance preserving geometry simplification, by replacing the original model with a low resolution mesh equipped with high resolution normal maps. This process visually preserves small scale features from the initial geometry, while only requiring a reduced set of polygons. However, this conversion usually relies on some kind of global or piecewise parameterization, combined with the generation of a texture atlas, a process that is computationally expensive and requires precise user supervision. In this paper, we propose an alternative method in which the normal field of a high resolution model is adaptively sampled and encoded in an octree-based data structure, that we call appearance preserving octree-texture (APO). Our main contributions are: a normal-driven octree generation, a compact encoding and an efficient look-up algorithm. Our method is efficient, totally automatic, and avoids the expensive creation of a parameterization with its corresponding texture atlas.