In this paper, we describe a neural-network-based handwritten alphabetic character recognition system. Feature vectors are generated using a two-pass algorithm to generate direction magnitude images followed by summing over overlapping zones. Crisp neural networks and fuzzy neural networks are trained using back-propagation. In the crisp case, the desired output is high for the correct class and low for all others. In the fuzzy case, the desired outputs are set using a fuzzy k-nearest neighbor algorithm.
|出版狀態||Published - 1992 十二月 1|
|事件||Proceedings of the 1992 Artificial Neural Networks in Engineering, ANNIE'92 - St.Louis, MO, USA|
持續時間: 1992 十一月 15 → 1992 十一月 18
|Other||Proceedings of the 1992 Artificial Neural Networks in Engineering, ANNIE'92|
|城市||St.Louis, MO, USA|
|期間||92-11-15 → 92-11-18|
All Science Journal Classification (ASJC) codes