We describe a trainable, hand-drawn symbol recognizer based on a multi-layer recognition scheme. Symbols are internally represented as binary templates. An ensemble of four template classifiers ranks each definition according to similarity with an unknown symbol. Scores from the individual classifiers are then aggregated to determine the best definition for the unknown. Ordinarily, template-matching is sensitive to rotation, and existing solutions for rotation invariance are too expensive for interactive use. We have developed an efficient technique for achieving rotation invariance based on polar coordinates. This techniques also filters out the bulk of unlikely definitions, thereby simplifying the task of the multi-classifier recognition step.
Levent Burak Kara, Thomas F. Stahovich. (2004). An Image-Based Trainable Symbol Recognizer for Sketch-Based Interfaces. AAAI Fall Symposium Series 2004: Making Pen-Based Interaction Intelligent and Natural.