| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide technology for attribute estimation in a wide range with high accuracy even under actual environment. <P>SOLUTION: An image feature quantity is extracted from an area of a subject for estimation reflected in an input image. For each of a plurality of attributes, a projection vector obtained by projecting a vector showing the image feature quantity in a feature space is calculated by using a base of the feature space. The subject for estimation is classified into any of the plurality of attributes by using the calculated projection vector. A model which is preliminarily calculated for the classified attribute is selected as a selection model, and sub attributes to which the subject for estimation belongs is estimated from the projection vector calculated for the classified attribute by using the selection model. <P>COPYRIGHT: (C)2013,JPO&INPIT |
