| 摘要 |
A method for formulating a knowledge database of object images acquired by an imaging device furnished with a sensor, including defining N classes each including a set of objects represented by a label and a set of characteristics, learning decision rules associated with these classes, definition of K contextual information cues regarding image background fk, with k varying from 1 to K and K>1, association of one of these information cues regarding image background fk with each object, allotting of the objects into M new classes, with N<M≦̸K×N, each including a set of objects represented by a new label including an old label and one of said information cues regarding image background fk and the set of characteristics of these objects, definition for each background fk of a subset Qk of these M classes associated with this image background, learning of decision rules for each of these subsets Qk. |
