| 摘要 |
<P>PROBLEM TO BE SOLVED: To highly precisely collate images of various objects which are photographed under conditions different in attitude and illuminationof by quickly estimating a prescribed color range of light and dark parts or the like having unstable features. <P>SOLUTION: A color distribution determination means 20 determines a color distribution of an input picture, and a weight coefficient determination means 30 uses the color distribution to determine a weight coefficient of a color of the input picture, and a weighted 3-dimensional collation means 40 generates a base picture group obtained by coordinate transformation in an illumination variance space generation means 75, on the basis of attitude candidates obtained for each model by an attitude candidate determination means 60 on the basis of respective 3-dimensional object models obtained from a reference three-dimensional object model storage part 70, and a weighted comparison picture generation means 45 uses the weight coefficient obtained by the weight coefficient determination means 30 to generate comparison pictures approximating the input picture by synthesis of the base picture group, and a weighted collation/attitude 50 uses the weight coefficient to calculate degrees of similarity between comparison pictures and the input picture, and an attitude giving the highest degree of similarity is selected. Finally, a model having the highest degree of similarity is defined as a collation result. <P>COPYRIGHT: (C)2004,JPO |
