| US 7,492,971 B2 | ||
| Selection of elements strongly related to a predetermined reference element | ||
| Jun Sakuma, Tokyo-to (Japan); and Michael Edward Houle, Kawasaki (Japan) | ||
| Assigned to International Business Machines Corporation, Armonk, N.Y. (US) | ||
| Filed on May 10, 2005, as Appl. No. 11/125,494. | ||
| Claims priority of application No. 2004-141515 (JP), filed on May 11, 2004. | ||
| Prior Publication US 2006/0039612 A1, Feb. 23, 2006 | ||
| Int. Cl. G06K 9/54 (2006.01) | ||
| U.S. Cl. 382—305 | 8 Claims |

| 1. A selection device for selecting a nearest neighbor set among a plurality of elements, the nearest neighbor set being a
set of elements each having a relation to a predetermined reference element, a degree of the relation meeting a predetermined
condition, the selection device comprising:
a nearest neighbor set candidate selection unit for selecting, among the plurality of elements, a set of elements determined
that the set of elements have stronger relations to the reference element, as each of a plurality of nearest neighbor set
candidates, which are a plurality of different candidates for the nearest neighbor set;
a nearest neighbor set selection unit for selecting, as the nearest neighbor set, a set of elements having relations, degrees
of which meet the predetermined condition, to the reference element, among a union of the plurality of nearest neighbor set
candidates; and
a hierarchical structure data generation unit which sets the plurality of elements as nodes, generates a plurality of hierarchical
structure data having a data structure in which the plurality of nodes are classified into any of a plurality of hierarchies,
and generates an edge between a node classified into each of the plurality of hierarchies and another node adjacent to the
hierarchy, for each of the plurality of hierarchical structure data, based on a degree of relation between the node and the
other node,
wherein, for each of the plurality of hierarchical structure data, the nearest neighbor set candidate selection unit selects
one nearest neighbor set candidate based on the hierarchical structure data.
|