A neural network structure for feature extraction and recognition of handwritten digits
Liming Zhang, Donghui Qu · 2002
The authors suggest some criteria of feature extraction for distinguishing numeric handwritten character. According to the criteria, 27-dimension feature vectors are chosen from a training set. Thus some of them can be obtained by neural networks. A back-propagation network is used to classify a test set. The recognition rate performance on 2000 characters written by 200 people is 94%.>