Simplification of point-sampled geometry with feature preservation

Abstract

A curvature-adaptive simplification method for point-sampled geometry (PSG) that efficiently preserves the surface features is presented. The main idea of the method consists of focusing on the edge intensities of sample points and the similarity of geometry features of sample points. Using the eigen analysis of normal voting tensor, with every point of PSG, an edge intensity is associated by which the PSG is decomposed into two components, one for the strong edge intensity and another for the non-strong edge intensity. Based on the adaptive mean-shift clustering, the second component is clustered into some clusters according to the geometric features' similarity. The first component and all these clusters are down-sampled, respectively, in combination with the mean-curvature threshold and sampling-density control to generate the simplified point set. In addition, the quality of the simplified PSG is evaluated using the error measurement method based on the moving least-squares surfaces. Experimental results show that the algorithm can achieve high-quality simplification result while efficiently preserving their features.