著者

"大倉 充/塩野 充/橋本 禮治" "オオクラ ミツル/シオノ ミツル/ハシモト レイジ" "Ohkura Mitsuru/Shiono Mitsuru/Hashimoto Reiji"

雑誌

岡山理科大学紀要. A, 自然科学

巻号頁・発行日

vol.30, pp.319-327, 1994

---

"Handwritten HIRAGANA characters are classified according to whether they have DAKUTEN (or HAN-DAKUTEN) or not by the 3-layer neural networks. The input data to the networks is 25-dimensional local mesh-feature extracted from an original character pattern. The numbers of units of three layers are 25 (input-layer), 25 (hidden-layer) and 2 (output-layer). The numbers of training and unknown samples used in a classification experiment are 6900 and 5000,respectively. The average classification rate of 94(%) for the unknown samples is obtained."