CHARACTER RECOGNIZING DEVICE

PURPOSE:To enable character recognition with the high rate of a correct answer by executing learning so that the decided result of a neural network for low- order group selection can be outputted as the final character recognized result while including a character pattern candidate for which origina...

Full description

Saved in:
Bibliographic Details
Main Authors HARUKI KAZUHITO, HATANO TOSHIAKI, ONO MASAYA
Format Patent
LanguageEnglish
Published 16.07.1991
Subjects
Online AccessGet full text

Cover

Loading…
More Information
Summary:PURPOSE:To enable character recognition with the high rate of a correct answer by executing learning so that the decided result of a neural network for low- order group selection can be outputted as the final character recognized result while including a character pattern candidate for which original similarity is made maximum from each similarity. CONSTITUTION:In respective low-order group neural networks 121-125...161-165, learning and the decision of the character recognized result is executed between the character pattern candidates respectively classified for each similarity pattern, and in respective neural networks 12-16 of for low-order group selection, out of the low-order group neural networks 121-125...161-165 belonging to the respective neural networks, the neural network with the character pattern of the highest probability is learnt and decided. Further, in a neural network 11 for high-order group selection, which neural network out of the neural networks 12-16 for low-order group selection includes the character pattern candidate of the highest probability is learnt and decided. Thus, even when there are a lot of character kinds such as handwritten Japanese syllabary or a lot of the similar character patterns are included, the character recognition can be executed with the high rate of the correct answer.
Bibliography:Application Number: JP19890305021