Detection of Surface Creases in Range Data

We propose a fully automatic and view-independent computational procedure for detecting salient curvature extrema in range data. Our method consists of two major steps: (1) smoothing given range data by applying a nonlinear diffusion of normals with automatic thresholding; (2) using a Canny-like non-maximum suppression and hysteresis thresholding operations for detecting crease pixels. A delicate analysis of curvature extrema properties allows us to make those Canny-like image processing operations orientation-independent. The detected patterns of creases can be considered as “shape fingerprints”. The proposed method can be potentially used for shape recognition, quality evaluation, and matching purposes.