An input panel and recognition engine for on-line handwritten text recognition

A portable input panel and a real-time recognition engine for isolated, handwritten characters are presented. The input panel mimics the one found in standard pen-centered interfaces and the recognition engine is based on approximate Dynamic Time Warping comparisons with prototypes selected by two different fast classification procedures. The error rate obtained on the standard Pendigits task, 0.83%, is significantly lower than the error rate of other recently published techniques.