Reproducing human perception of shapes from dot patterns is a classical problem in low-level computer vision, pattern recognition, and cluster analysis. We implemented (an extended version of) the paper "Computational morphology of curves" by Luiz Henrique de Figueredo & Jonas Gomes, where the problem of curve morphology reconstruction ("given a set of 2D points, find a subset which we can sort in an order compatible with the natural trace of the curve, as perceived by humans") is approached with graph theoretic and differential topology tools.

Noisy point cloud	Delaunay triangulation
Euclidean Minimum Spanning Tree	Connected components (after bridge deletion)
Noise filtered	Closed components (after loop closure)

Presentation Slides: [PDF file - 99K]

Source Code (pretty rough version...): [gzipped tar archive - 5.7M] (use it by your own risk... :D)