We explore a characterization of a handwritten symbol, that of approximating a normalized x/y coordinate sequence from a handwriting sample as a linear combination of basis functions fit to the normalized coordinate sequence. We can describe the model for a symbol in this way, and match new input against model vectors that are close to the vector of the input.

We explored this idea using the UWO ORCCA database of handwriting samples. It suggests that similar handwritten characters do have vectors that are close together, and that polynomials of modest degree (6 or 8) do a reasonably good job. We discuss the preliminary results and what they suggest for future work, and demonstrate some of the tools (some using Maple) we used or built to conduct the exploration.

This is joint work with Stephen Watt, with contributions by Elena Smirnova and Jerliazko Polihronov of the ORCCA group of the University of Western Ontario.