| 摘要 |
<P>PROBLEM TO BE SOLVED: To efficiently retrieve a similar image along with a purpose and content of each image from a document group including the image. <P>SOLUTION: An image retrieval system is provided with functions of: (1) detecting a feature amount from each image included in a document, quantizing it and determining it as a pseudo-word; (2) extracting a term from a legend of the image in the document; (3) extracting the image and its legend from a document group where both each image and its legend are similar, and giving a feature amount common in a group of the extracted images heavier weight than other feature amounts, or determining a sum total of relatively weighted feature amount vectors in the group of the extracted images to be a profile of each of the images; and (4) determining each of the images to be a document consisting of the pseudo-word and the term, and using both a query given in the image and/or in a text and the weight or the profile to retrieve an image similar to the query. <P>COPYRIGHT: (C)2013,JPO&INPIT |
