FEATURE POINT SELECTION SYSTEM, FEATURE POINT SELECTION METHOD AND FEATURE POINT SELECTION PROGRAM

摘要

A recognition task execution means (11) for providing a feature point selection system capable of selecting suitable feature points that fit a recognition algorithm for a recognition task, for the plurality of evaluation-use images, executes the recognition task while using the significance of each of a plurality of feature point candidates upon a three-dimensional shape model. A recognition error margin evaluation means (12) evaluates a recognition error margin related to all evaluation-use images based on a difference between recognition results of the recognition task and correct response data of the recognition task for each of the evaluation-use images. A feature point significance determination means (13) defines, for recognition error margins related to all evaluation-use images, a cost function represented by a function to which a constraint condition has been added for causing the significance of insignificant feature point candidates to approach zero easily, and calculates a significance for each feature point candidate that is likely to minimize the value of the cost function. A feature point selection means (14) selects feature points to be used for the recognition task from feature point candidates upon the three-dimensional shape model on the basis of the significance of each of the feature point candidates.