학회논문집의 DB구축을 위 한 자동문서 인식기 설계

A Design oi the Automatic Document Recognition System for DB Construction of Institute Journal

In this paper, Character recognition system, which recognized Hangul—Document to image type using scanner or camera, designed using neural network method. After learning neural network is possible supervised—learning using Back Propagation algorithm, this system recognized the character with same processing. The neural network consists of total 17 hierarchical neural network to recognize Hangul with grapheme unit. In Preprocessing step, 1. omary image, 2. rotation- of the nltea image, 3. separate text parts and graphic parts, 4. separate each character process about input image, 5. Hangul separated 6 types with combination of vowel and consonant. In character recognition step, 1. normalized character area with 60x60 size, 2. separated fixed grapheme area for each types, 3. extracted grapheme s feature information using Mesh vector at separated grapheme area, 4. constracted graphemes recognition neural network recognize the character. we get the research result that is 98% recognition rate about Hangul document (9 sheet of paper in A4 size) was not participate in neural network learning.