摘要： 针对进一步提高英文OCR分类器的鲁棒性进行了研究,结合传统的单模板MQDF分类器和多模板欧氏距离分类器各自的优点,提出了一种新的基于MQDF的多模板分类器设计方法。与传统分类器的对比测试证明,该文提出的这种新的分类器能够有效地提高多体英文OCR字符的单字首选正确率,并对低质量文本中的模糊和断裂字符也能保持很高的识别率。

关键词: 光学字符识别, 多模板, LDA变换, k均值聚类, MQDF分类器

Abstract: This paper focuses on robust classification research. By combining the advantages of traditional single template modified quadraticdiscriminant function classification and multi-template euclidean classification, it presents a new multi-template classification design based onMQDF. Through comparative test with the traditional classification, it proves that the new design can distinguishingly improve the preferredrecognition rate of characters,and keep high accurate recognition of darkened or splitter characters in low quality document as well.

中图分类号: 

• TP391.1