摘要 |
A clustering method for applying to a plurality of classes each including one or more images as objects, includes: performing an iterative merge on the classes according to inter-class Rank-Order distances; obtaining an intra-class convergence degree of each class obtained by the iterative merge, based on inter-object distances in the class; and grouping, for each class obtained by the iterative merge, objects in the class that have a distance smaller than the intra-class convergence degree of the class into a new class, thereby to perform an update on the classes. |
主权项 |
1. A clustering method for applying to a plurality of classes each including one or more images as objects, comprising:
performing an iterative merge on the classes according to inter-class Rank-Order distances; obtaining an intra-class convergence degree of each class obtained by the iterative merge, based on inter-object distances in the class; grouping, for each class obtained by the iterative merge, objects in the class that have a distance smaller than the intra-class convergence degree of the class into a new class, thereby to perform an update on the classes; if a number of the classes after the update is smaller than a number of the classes before the update, repeating the performing of the iterative merge, the obtaining, and the grouping, until the numbers of the classes before and after an update are the same, to obtain a clustering result, the clustering result including one or more classes each containing a plurality of objects and one or more classes each containing a single object. |