An Robust RPCL Algorithm and Its Application in Clustering of Visual Features

摘要

Clustering in the neural-network literature is generally based on the competitive learning paradigm. This paper presents a new clustering algorithm which is against initialization while meantime can find the natural prototypes in the input data, especially it could partly handle problems that Rival Penalized Competitive Learning (RPCL) algorithm have. Simulation results on synthesized data sets show that proposed method is effective and robust. Application of the proposed robust RPCL algorithm in indexing of visual features is discussed.