|
Abstract : |
We present a new method for classifying handwritten characters, in particular digits. In our approach each character in the alphabet is represented by a prototype, in particular a piecewise-linear curve. Prototypes are tted to a point-set representation of the image by least squares, allowing a free ane transformation in R 2. Every prototype is tted to a given image, and then classication is based on the matching statistics. The procedure is demonstrated on a large database of handwritten zipcodes, where we show it competes favorably with existing methods. 1, |
