Learning device, learning method, identification device, identification method, and program

摘要

Provided is a learning device including: an acquisition section that acquires a plurality of image pairs in which the same subjects appear and a plurality of image pairs in which different subjects appear; a setting section that sets feature points on one image and the other image of each image pair; a selection section that selects a plurality of prescribed feature points, which are set at the same positions of the one image and the other image, so as to thereby select a feature extraction filter for each prescribed feature point; an extraction section that extracts the features of the prescribed feature points of each of the one image and the other image by using the plurality of feature extraction filters; a calculation section that calculates a correlation between the features; and a learning section that learns a same-subject classifier on the basis of the correlation and label information.

主权项

1. A learning device, comprising:
circuitry configured to:
acquire a first plurality of image pairs in which same subjects appear and a second plurality of image pairs in which different subjects appear;set feature points on a first image and a second image of each image pair of the first plurality of image pairs and of the second plurality of image pairs;select a plurality of prescribed feature points set at corresponding same positions on the first image and the second image of each image pair of the first plurality of image pairs and of each image pair of the second plurality of image pairs, so as to thereby select feature extraction filters, which are used to extract a feature for each prescribed feature of the prescribed feature points;extract the feature for each prescribed feature of the prescribed feature points on each of the first image and the second image of each image pair of the first plurality of image pairs and of each image pair of the second plurality of image pairs by using the selected feature extraction filters;calculate a correlation between the extracted features for each image pair of the first plurality of image pairs, and for each image pair of the second plurality of image pairs; andlearn same-subject classifiers for identifying whether or not the same subjects appear, on the basis of the calculated correlation and label information representing in which pair of the first plurality of image pairs and of the second plurality of image pairs the same subject appears,wherein the same-subject classifiers comprise at least one strong classifier and at least one weak classifier, andwherein the circuitry is further configured to:
learn a same-subject classifier of the same-subject classifiers that is a strong classifier formed of a plurality of weak classifiers, through boosting, andrandomly select the prescribed feature points and the feature extraction filters whenever a weak classifier of the plurality of weak classifiers is learned.