| 摘要 |
Plural nodes are arranged at predetermined initial positions, and feature values at plural sampling points around each node are obtained as a node feature value of each corresponding node. An error estimator indicating displacement between the current position of each node and the position of corresponding feature point is obtained based on correlation information on a difference between the node feature value obtained in a state in which the plural nodes are arranged at correct positions of the corresponding feature points and the node feature value obtained in a state in which the plural nodes are arranged at wrong positions of the corresponding feature points in a learning image, correlation information on a difference between the correct position and the wrong position, and a node feature value of each node. The position of each feature point is estimated in an input image based on the error estimator and the current position of each node. |
