Robust Sketched Symbol Fragmentation using Templates

Analysis of sketched digital ink is often aided by the division of stroke points into perceptually-salient fragments based on geometric features. Fragmentation has many applications in intelligent interfaces for digital ink capture and manipulation, as well as higher-level symbolic and structural analyses. It is our intuitive belief that the most robust fragmentations closely match a user²s natural perception of the ink, thus leading to more effective recognition and useful user feedback. We present two optimal fragmentation algorithms that fragment common geometries into a basis set of line segments and elliptical arcs. The first algorithm uses an explicit template in which the order and types of bases are specified. The other only requires the number of fragments of each basis type. For the set of symbols under test, both algorithms achieved 100% fragmentation accuracy rate for symbols with line bases, >99% accuracy for symbols with elliptical bases, and >90% accuracy for symbols with mixed line and elliptical bases.