摘要 |
PROBLEM TO BE SOLVED: To calculate similarity between shapes without depending on the precision of preprocessing such as contour extraction, binarization, and without requiring processing for preliminarily obtaining ordering of feature points. SOLUTION: An image input unit 1 inputs an image. A feature point extraction unit 2 extracts feature points from the image. A feature point distribution parameter estimating and category determining unit 3-3 estimates a probability distribution parameter of extracted feature points with respect to each feature point distribution models in a categorized shape model 4 and determines a category of a shape on the basis of the estimated probability distribution parameter of feature points. A reference direction feature generating and similarity calculating unit 3-4-1 generates reference- direction features and calculates, as similarity of the shape, a distance between reference-direction features 4-2 in the categorized shape model 4 and the generated reference-direction features with respect to the determined category of the shape. COPYRIGHT: (C)2013,JPO&INPIT |