摘要 |
PROBLEM TO BE SOLVED: To solve the problem that it is difficult to read a document name written in decorative characters such as various non-character marks and outline characters when similar documents different only in document logo marks and the like are discriminated. SOLUTION: Document logo marks and the like can be discriminated by using document feature point vector collation to absorb displacement when document images are displaced from each other. Further, in a learning process to create a document discrimination dictionary, whether an attribute of a feature point is independency or dependency is discriminated and absence of a logo mark and the like can be treated as a feature. In addition, a position of the document logo marks and the like automatically estimated among the similar documents is highlighted to improve efficiency in creating the document dictionary. COPYRIGHT: (C)2009,JPO&INPIT
|