Hidden Markov Model Symbol Recognition for Sketch-Based Interfaces.

Abstract

A central challenge for sketch-based interfaces is robust symbol recognition. Artifacts such as sketching style, pixelized symbol representation and affine transformations are just a few of the problems. Temporal pattern recognition through Hidden Markov Models (HMM) can be used to recognize and distinguish symbols as pixel-driven gestures. The key challenges of such a system are the type and amount of necessary pre-processing, feature extraction, HMM parameter selection and optional post processing. In this paper, we describe a recognition strategy based on HMMs and include recognition results on twelve sketched symbols. In addition, we have successfully applied this methodology to a PDA sketch-based interface to control a team of robots. The symbol recognition component is used to identify sketched formations and issue commands that drive the behavior of the robot team.