| 摘要 |
<P>PROBLEM TO BE SOLVED: To accurately extract a model to be detected from an object image including a plurality of objects. <P>SOLUTION: In the image recognition device 1, characteristic point extracting parts 10a, 10b extract characteristic points from a model image and an object image, and characteristic quantity holding parts 11a, 11b extract characteristic quantities of the individual characteristic points, and hold the quantities together with position information of the characteristic points. A characteristic quantity comparing part 12 calculates the similarity or difference by comparing the characteristic quantities and creates candidates for corresponding characteristic point pairs having a high correspondence possibility. A model pose estimating part 13 repeats an operation of projecting, in a parameter space, an affine conversion parameter determined from three pairs randomly selected from the candidates for corresponding characteristic point pairs, defines individual members of a cluster having a maximum number of members formed in the parameter space as a true characteristic point pair, obtains the affine conversion parameter by minimum square estimation using the true characteristic point pair, and outputs the model pose determined by this affine conversion parameter. <P>COPYRIGHT: (C)2005,JPO&NCIPI |
