摘要 |
Nodes are arranged at predetermined initial point. The feature values at sampling points around each node are acquired as the node feature values of the each corresponding node. From a learning image, error estimates representing the difference between the current position of each node and the position of the corresponding feature point are acquired from the information on the correlation between the difference between the node feature value acquired in a state that the nodes are arranged at the corresponding correct positions and the node feature value acquired in a state that the nodes are arranged at wrong positions and the differences between the correct positions and wrong positions, and the node feature values of the nodes. On the basis of the error estimates and the current positions of the nodes, the positions of the feature points on the input image are estimated. |