Segmentation and recognition of offline sketch scenes using dynamic programming

Abstract

Sketch recognition aims to segment and identify objects in a collection of hand-drawn strokes. In general, segmentation is a computationally demanding process since it requires searching through a large number of possible recognition hypotheses. It has been shown that, if the drawing order of the strokes is known, as in the case of online drawing, a class of efficient recognition algorithms becomes applicable. In this article, we introduce a method that achieves efficient segmentation and recognition in offline drawings by combining dynamic programming with a novel stroke ordering method. Through rigorous evaluation, we demonstrate that the combined system is efficient as promised, and either beats or matches the state of the art in well-established databases and benchmarks.