| 摘要 |
<P>PROBLEM TO BE SOLVED: To achieve highly accurate similarity retrieval. <P>SOLUTION: A pivot is determined from data for registration by a pivot determination part, raw data are acquired, a feature amount is extracted from the raw data, a score is calculated as a distance or similarity between the feature amounts, a vector for an index is generated using the score for the pivot, and a Δ score is calculated as the distance or similarity between the vectors for the index. A parameter for each non-pivot including a regression coefficient is learned using data for learning, the selection order of the non-pivot is determined in the descending order of a posterior probability by logistic regression using the Δ score and the regression coefficient of data for retrieval and the non-pivot, and a retrieval result is outputted on the basis of the score of the data for the retrieval and the data for the registration. <P>COPYRIGHT: (C)2012,JPO&INPIT |
