Graph reconstruction from noisy scans based on persistence

27^th September 2016, 13:00 Ashton Lecture Theater
Dr. Vitaliy Kurlin
Department of Computer Science and Materials Innovation Factory, University of Liverpool

Abstract

The work is based on the on-going collaboration with Microsoft Research Cambridge. For a cloud of points in any metric space, a Homologically Persistent Skeleton captures the 1-dimensional topology of the cloud across all scales (Computer Graphics Forum, 2015). We define a new subgraph of this skeleton giving a topologically correct reconstruction that closely approximates an unknown graph given by a noisy sample of points under weak
restrictions. More retails including papers and software are available at http://kurlin.org.