摘要 |
PROBLEM TO BE SOLVED: To generate an image dictionary with high accuracy, for use at the time of automatic addition of a label indicating a meaning of image to the image. SOLUTION: An image group related to a certain meaning is collected in advance as learning data. An area division section 102 divides each image in the learning data into areas according to the feature amount of the image. A representation area cluster extraction section 103 performs clustering to the area group obtained by the division, and extracts a representation area cluster corresponding to a plurality of objects capable of sufficiently representing the meaning among the above clusters. A representation area cluster model learning section 104 builds a representation area cluster model as the learning data, using a sample included in each representation area cluster. A weight coefficient calculation section 105 calculates a weight coefficient indicative of the degree of importance relative to the meaning of the representation area cluster model. Then, the above representation area cluster model and the information of the weight coefficient are stored into an image dictionary storage 106, as image dictionary. COPYRIGHT: (C)2009,JPO&INPIT
|