In this paper a clustering algorithm for sparsely sampled high-dimensional
feature spaces is proposed. The algorithm performs clustering by employing
a distance measure that compensates for differently sized clusters. A seque
ntial version of the algorithm is constructed in the form of a frequency-se
nsitive competitive learning scheme. Experiments are conducted on an artifi
cial Gaussian data set and on wavelet-based texture feature sets, where cla
ssification performance is used as a clustering significance measure. It is
shown that the proposed technique improves classification performance dram
atically for high-dimensional problems. (C) 1999 Pattern Recognition Societ
y. Published by Elsevier Science Ltd. All rights reserved.