A user definable two dimensional gesture recognition system has been developed. The system is intented to provide user-definable gestures such as proof reading marks and operating commands to be used in pen computers. While most of previous gesture recognizers deal with only fixed set of gesture symbols, we propose a system by which users can define own set of symbols for a particular application. The user can define gesture shapes and their hot points on which the operations occur. Additionally the user can specify operational features such as possibility of rotation invariance, mirror reflection and/or reverse direction.
A new gesture symbol is defined in the training phase. A training set of gesture drawings is used to construct models which are represented by composition of three primitive types; line, arc, circle. In recognition phase, the best matching model is found for an known drawing and its label is used as that of the unknown.
The system achieved 94% recognition accuracy in an experiment with three sets of gesture symbols drown by seven people. The system was integrated as a part of a gesture-based Hangul editor on a Pen Computing enviroment.