Abstract
A system for real-time object recognition and learning from known and unknown classes is presented. A three-level algorithm is described. Level I does not classify patterns. Using deterministic rules, it acts as a prefilter to eliminate obvious classes from further consideration. Level II algorithm is a modified version of the (k,k prime ) nearest neighbor algorithm designed to deal with objects that do not belong to any of the known classes. Level III algorithm is clustering-oriented. When a new pattern arrives, a clustering algorithm is activated to decide whether to assign this pattern to one of the known classes or to classify it as a new type. Experimental results with the system are reported.
Original language | English |
---|---|
Pages | 66-71 |
Number of pages | 6 |
State | Published - 1978 |
Event | Proc IEEE Comput Soc Conf Pattern Recognition Image Process - Chicago, IL, USA Duration: 31 May 1978 → 2 Jun 1978 |
Conference
Conference | Proc IEEE Comput Soc Conf Pattern Recognition Image Process |
---|---|
City | Chicago, IL, USA |
Period | 31/05/78 → 2/06/78 |