摘要 |
PROBLEM TO BE SOLVED: To prevent an increase in calculation time necessary for discrimination of objects, and to quickly conduct the discrimination even if the number of the object extremely increases. SOLUTION: On a learning device 1 which registers a plurality of objects, the object is classified as a tree structure from a similarity between feature vectors of each object by an object classification means 12, and the feature vector belonging to each node of the classified tree structure is analyzed for principal component by a principal component analysis means 13, then it is accumulated in an accumulation means 14. Besides, on an identifying device 2 which discriminates the objects, the feature vector is compressed using the principal component in each accumulated node of the tree structure for the feature of the target object input from an input means 21 by a compression means 22, then a determination means 24 determines which node of a lower layer the feature belongs to, and it is out put by an output means 25. COPYRIGHT: (C)2005,JPO&NCIPI
|