摘要 |
PROBLEM TO BE SOLVED: To enable the extraction of a spatio-temporal feature amount which properly represents visual perception and movement information of a subject from a video.SOLUTION: A movement feature extraction part 54 extracts a relative movement feature amount between clusters of a combination about each combination of clusters of one image of the pair of associated frame images and clusters of the other image of the pair of the frame images to each pair of frame images on the basis of a local feature point associated with the clusters of the combination, a relevance weighting part 56 performs weighting according to relevance between the clusters of the combination, of the relative movement feature amount between the clusters of the combination on the basis of a code book, and a spatio-temporal feature amount generation part 58 extracts a spatio-temporal feature amount which properly represents visual perception and movement information of the subject by generating a spatio-temporal feature amount of an image on the basis of a matrix of N*N of the relative movement feature amount.SELECTED DRAWING: Figure 1 |