| 摘要 |
PROBLEM TO BE SOLVED: To provide a robust, highly accurate and high-speed method for estimating corresponding points between images. SOLUTION: A corresponding point, on an image after deformation, which corresponds to a remarkable point on the image before deformation is estimated by template matching. The image before deformation is divided into two or more small areas first. Two or more templates including one or more small areas are produced. Two or more matching points are acquired by performing template matching using the produced templates for the image after deformation. With the two or more matching points being observation data, the displacement of the two or more small areas, with respect to the image after the deformation, which include the remarkable point are taken as unknown parameters, a matrix with each row including elements being respective elements of a weighting vector including elements being values proportional to the superficial dimensions of the small areas is taken as an observation matrix, and the estimated values of displacements of the two or more small areas are acquired by minimum variance estimation for a general linear regression model. The corresponding points are estimated based on the acquired estimated values. COPYRIGHT: (C)2005,JPO&NCIPI
|
