A stepwise recognition method of library cataloging cards on the basis of various kinds of knowledge

摘要

An experimental method is proposed for extracting individual data items from library cataloging cards and classify them into the cataloging item classes automatically. The basic idea for the recognition strategy is to utilize various kinds of knowledge stepwise on the basis the generation/verification process of object hypotheses. In this approach, such knowledge as layout information of card structures, relationship information among cataloging items, and property information of data items is effectual. The knowledge is not represented as same-level information, but is mutually specified with the hierarchical relationship. Lower-level knowledge is successful for recognizing the processing objects interpreted by higher-level knowledge. Though the authors' primary concern is about the understanding of library cataloging cards, the proposed framework is applicable to other documents with particular layout structures such as pamphlets, letters, office mail, office documents, articles, papers and so on.