| 摘要 |
PROBLEM TO BE SOLVED: To provide a method for recognizing an object in an image that may be composed of any number of parts movable in relation with other object. SOLUTION: In the offline stage, the method automatically learns the relative movement of the parts of a single object from a sequence of exemplary images and builds a hierarchical model including the description of the parts of the single object, the relation among the parts, and an efficient search strategy. This is executed by analyzing a change in the pose of the parts of the single image in the exemplary images. The pose can be obtained by any similarity measure about object recognition. In the online stage, the hierarchical model is used to efficiently find the whole object in a search image. During the online stage, the parts of an object are searched not in the whole image but in a limited part of a parameter space defined by the relation among the object parts in the hierarchical model that facilitates an efficient search and eliminates the need for a subsequent validity inspection step. COPYRIGHT: (C)2004,JPO
|
