APPARATUS AND METHOD FOR VIDEO RECOGNITION, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To highly precisely learn relationship between an image and language information even when only a few images associated with language information can be used. SOLUTION: A degree-of-interest image extraction part 1 extracts a degree-of-interest image that shows the degree of interest of a human being in each position in an input image that is a frame of an input video. A region-of-interest extraction part 2 extracts a region-of-interest image that shows a region that is likely to draw interest in the input image from the degree-of-interest image and the input image. A region-of-interest image feature extraction part 3 extracts the region-of-interest image features as vectors that represnt the characteristics of the image in the region-of-interest from the input image and the region-of-interest image. A region-of-interest additional information presentation part 4 gives the region-of-interest image features to the image/additional information relation model, and extracts and presents the region-of-interest additional information as additional information that describes the region-of-interest by adding the region-of-interest image features to the image/additional information relation model. COPYRIGHT: (C)2011,JPO&INPIT