| 摘要 |
<P>PROBLEM TO BE SOLVED: To quickly detect the position of a featured point from an image. <P>SOLUTION: The information of correlation about the difference between a correct answer model, in which nodes are arranged at the correct positions of featured points and an error model in which any node is arranged at a wrong position and node featured values acquired, based on the error model, is preliminarily acquired by using a leaning image. In the case of detecting featured points from an input image, a three-dimensional model, in which the three-dimensional positions of a plurality of nodes are determined is prepared, and the respective nodes are projected on the input image, and the node featured values are acquired from projection points; error inferred quantity, showing deviation between the current position of each node and the position of the corresponding featured point, is acquired based on the node featured values and the learnt information of the correlation; and the three-dimensional positions of the respective featured points in the input image are inferred, based on the error inferred quantity and the current positions of the respective nodes. <P>COPYRIGHT: (C)2007,JPO&INPIT |
