SEGMENTATION OF MINERAL TEXTURES

R.J. WILSON
Department of Mathematics 
The University of Queensland 
and 
Co-operative Research Centre for 
Mining Technology and Equipment 



I. RYCHLIK
Department of Mathematical Statistics 
Lund University



Abstract
Images of mineral textures provide information on the phases within
ores.  However, in segmenting such images, the spectral values for
the different phases usually overlap and the polishing of the ore
surface often results in corruption of the data.  We present a method
using joint seeded region growing where the seeds are obtained by 
classifying so-called ``rainfall cycle'' minimum-maximum pairs,
which can be obtained using Grimaud(1992)'s dynamic.  Due to the
nature of the variability within phases, this method seems to work
better than other methods based on neighbourhood filters.