Salience Feature Detection for Polygonal Meshes Based on Maximum Normal Variation

Abstract

Feature detection or extraction for polygon mesh is one of the most fundamental and extensively used techniques. Applications such as mesh simplification, smoothing, parameterization, segmentation, morphing, and shape matching, etc. requires feature detection. A common technique is to identify the edge features through the dihedral angles of the two neighbouring faces adjacent to the edge. In this paper, we propose a variation of such technique that identifies feature vertices by considering the maximum normal variation of the neighbouring faces in the 1-ring neighbourhood of a vertex. By clustering and labelling the feature vertices, we may successfully identify the features of a given object as disjoint cluster of feature vertices. By means of Shortest Path or Spanning Tree with proper weighting scheme, we may further process each cluster as a feature curve that potentially enables a cut to the mesh.