A feature set for the encoding of optical images of handwritten charac
ters is presented. It allows good recognition rates when coupled with
a neural network classifier and can be easily implemented in analogue
CMOS VLSI technology. A circuit for the extraction of such features is
described.