Online stroke modeling for handwriting recognition

The process of recognizing individual handwritten characters is one of classifying curves. Typically, handwriting recognition systems---even "online" systems---require entire characters be completed before recognition is attempted. This paper presents another approach for real-time recognition: certain characteristics of a curve can be computed as the curve is being written, and these characteristics are used to classify the character in constant time when the pen is lifted. We adapt an earlier approach of representing curves in a functional basis and reduce real-time stroke modelling to the Hausdorff moment problem.