Vector quantization for recognition of hand written numerals
K. Reiser · 2002
An extremely simple, highly parallel vector quantization method far recognizing hand written numerals is described. A perceptron learning rule was used in training model numerals composed of simple, local features. 120 images per character were used to train the system; a different set of 100 images per character was used in testing. In the 32 experiments performed to examine the effects of various feature-related parameters, a maximum correct classification rate of 96.7% was observed.>