For the final project in  we (I and Akihiro Yamashita) implemented stroke-based classification of handwritten digits.

The classification was performed by fitting a bezier curve to elements of the digits. The features were selected by calculating the derivatives of this Bezier curve. Although the method is fairly simple, and fast, it gave surprisingly good results. A set of slides discussing the project is given here.

The source code (MATLAB) is digitclassifier.zip (It is around 50 MB). This includes the original collected datasets, the program to collect the data and the processed datasets.

Here is a video demonstrating the project: