Edge-cycles: A qualitative sketch representation to support recognition
Matthew D. McLure, Scott Friedman, Andrew Lovett, Kenneth D. Forbus · 2011
Qualitative representations can play an important role in sketch understanding, by providing stable relational descriptions that support learning for recognition. A common approach is to represent a sketch of an object as a set of edges and relations between edges. However, hand-drawn sketches are generally noisy, with unintended gaps, jitter, and other glitches that cause artifacts in edge representations. We describe a new higher-level representation, edge-cycles, that is more stable than edgebased representations, and demonstrate that it provides significantly better results in learning to classify hand-drawn sketches of everyday objects.