Image-document retrieving apparatus, method of retrieving image document, program, and recording medium

An image-document retrieving apparatus that retrieves an image document containing mainly text and a graphic, the image-document retrieving apparatus comprising:

• a feature vector creating section that creates a feature vector of a page image document;
• and
• an image document discrimination section that compares the feature vector, created by the feature vector creating section, of the page image document that is an object of discrimination, with the feature vector, created by the feature vector creating section, of the page image document that is accumulated as an object of retrieval, and extracts, from page image documents that are objects of retrieval, a page image document that corresponds to the page image document that is the object of discrimination,
• the feature vector creating section including:
• a rectangular image extracting section that combines text areas of the page image document to extract a plurality of rectangular images;
• a block-division section that analyzes, on the basis of information on the plurality of rectangular images extracted by the rectangular image extracting section, a geometric structure of a page to divide the page image document into plural blocks;
• a feature block selecting section that selects, among the plural blocks, plural feature blocks that describe a feature of the page image document;
• and
• a feature calculating section that determines, as the feature of the page image document, information on the plural feature blocks selected by the feature block selecting section,
• the feature vector creating section incorporating, as an element into the feature vector of the page image document, a result of determination carried out by the feature calculating section,
• the feature block selecting section selecting n (n is an integer that is not less than 3) pieces of feature blocks;
• the feature calculating section determines, as the feature of the page image document, information on a polygon having n sides (n is an integer that is not less than 3) and having apexes at respective centers of the feature blocks, in addition to information on the n pieces of feature blocks;
• and
• the image document discrimination section carries out comparison of the information on the n pieces of feature blocks only with respect to a page image document having a feature vector that is similar in, among the features included in the feature vector, the polygon having n sides.