We derive real-time global optimization methods for several clustering opti
mization problems commonly used in unsupervised texture segmentation. Speed
is achieved by exploiting the image neighborhood relation of features to d
esign a multiscale optimization technique, while accuracy and global optimi
zation properties are gained using annealing techniques. Coarse grained cos
t functions are derived for central and histogram-based clustering as well
as several sparse proximity-based clustering methods. For optimization dete
rministic annealing algorithms are applied. Annealing schedule, coarse-to-f
ine optimization and the estimated number of segments are tightly coupled b
y a statistical convergence criterion derived from computational learning t
heory. The notion of optimization scale parametrized by a computational tem
perature is thus unified with the scales defined by the image resolution an
d the model or segment complexity. The algorithms are benchmarked on Brodat
z-like microtexture mixtures. Results are presented for an autonomous robot
ics application. Extensions are discussed in the context of prestructuring
large image databases valuable for fast and reliable image retrieval. (C) 1
999 Academic Press.