SEGMENTATION OF TOUCHING CHARACTERS IN PRINTED DOCUMENT RECOGNITION

被引：29

作者：

LIANG, S ^{[1
]}

SHRIDHAR, M ^{[1
]}

AHMADI, M ^{[1
]}

机构：

[1] UNIV MICHIGAN,DEPT ELECT & COMP ENGN,DEARBORN,MI 48128

来源：

PATTERN RECOGNITION | 1994年 / 27卷 / 06期

关键词：

TOUCHING CHARACTER SEGMENTATION; DYNAMIC RECURSIVE SEGMENTATION; CHARACTER CONTEXTUAL CLASSES; FEATURE EXTRACTION; SPELLING CORRECTION;

D O I：

10.1016/0031-3203(94)90167-8

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

A new discrimination function is presented for segmenting touching characters based on both pixel and profile projections. A dynamic recursive segmentation algorithm is developed for effectively segmenting touching characters. Contextual information and spell checking are used to correct errors caused by incorrect recognition and segmentation. Based on 12 real documents, a maximum 99.85% and a minimum 99.4% recognition accuracy is achieved.

引用

页码：825 / 840

页数：16

共 17 条

[1] AUTOMATED ENTRY SYSTEM FOR PRINTED DOCUMENTS [J].