Collapsible Spikes and the Correspondence Problem in Point Clouds
- Day - Time: 30 October 2017, h.15:00
- Place: Area della Ricerca CNR di Pisa - Room: C-40
- Edgar Chavez (CICESE, Mexico)
A useful representation of many digital objects is a set of points in two or three dimensions. Examples of this are keypoints in images, e.g. SIFT/SURF, superpixels, high energy points in the spectrum of audio files and minutiae in fingerprints, to name a few. Point clouds can be subject to affine, elastic or projective transformations plus noise, insertions and deletions. When a large set of digital objects is stored, and a new query object arrives, a subproblem to be solved is to find which points in the collection match the query. Since the points are indistinguishable, we can use the geometry of a neighborhood of them to detect matching points.
In this talk we describe a new distance to compute a matching score of individual points using collapsible spikes and give efficient algorithms to solve it.