Sketch-Based Recognition System for General Articulated Skeletal Figures

Abstract

We present a new recognition system for detecting general articulated skeletal figures in sketch-based applications. We abstract drawing style from recognition by defining figures using two data models: templates and figure targets. Our system recognizes general skeletal figures consisting of lines and ellipses which may be drawn partially off the canvas with the system automatically completing the figure. Figures are modeled as graphs and are allowed to contain cycles. Subgraph matching on a graph built from the stroke input is used to perform recognition. This paper outlines our system design, key details to its proper implementation, and proposes its application to various domains of sketch recognition.