| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide a learning device and so on capable of improving discrimination accuracy of a created discriminator, while suppressing costs of labeling for a non-labeled image created from an image with a correct label. <P>SOLUTION: A learning device 100 includes: a feature vector calculation part 110; a discriminator creation part 120 for performing creation processing of a discriminator, on the basis of a feature vector; an image creation part 130 for performing image creation processing of a non-labeled image, on the basis of a first image with a correct label; an image selection part 140 for performing image selection processing of the non-labeled image to be presented to a user; an output part 150 for outputting the selected non-labeled image to a display part 300; and a correct label setting part 160 for associating a correct label acquired from an input part 200 with the selected non-labeled image and making the same a second image with a correct label. The discriminator creation part 120 performs regeneration processing of the discriminator, on the basis of a feature vector of the first image with a correct label and a feature vector of the second image with a correct label. <P>COPYRIGHT: (C)2013,JPO&INPIT |
