Recovering geometric detail by octree normal maps

Abstract

This paper presents a new approach for constructing normal maps that capture high-frequency geometric detail from dense models of arbitrary topology and are applied to the simplified version of the same models generated by any simplification method to mimic the same level of detail. A variant of loose octree scheme is used to optimally calculate the mesh normals. A B-spline surface fitting based method is employed to solve the issue of thin plate. A memory saving Breadth-First Search (BFS) order construction is designed. Furthermore, a speedup scheme that exploits access coherence is used to accelerate filtering operation. The proposed method can synthesize good quality images of models with extremely high number of polygons while using much less memory and render at much higher frame rate. © 2012 Springer-Verlag Berlin Heidelberg.