Abstract A method is introduced which enables reliable, automated mineral classification in the macroscopic scale. Polished rock samples are scanned by a color image scanner. Scanner output is split into red, green and blue component images to be evaluated by multispectral image processing methods. Different unsupervised and supervised image classification algorithms have been tested with medium-to-coarse grained crystalline rocks. Statistical evaluation of the results showed that the supervised maximum likelihood algorithm was the most robust approach: provided the minerals of interest show different hues, average mineral-phase recognition levels are approximately 90%. The result of the method is a classified raster image of mineral distribution which, besides giving rock modes, can be postprocessed by shape and fabric analysis programs or passed to 3D serial section reconstruction algorithms.
[1]
M. Pareschi,et al.
Automated evaluation of volumetric grain-size distribution from thin-section images
,
1990
.
[2]
N. J. Fortey.
Image analysis in mineralogy and petrology
,
1995
.
[4]
Andrew R. H. Swan,et al.
Image analysis of petrographic textures and fabrics using semivariance
,
1995,
Mineralogical Magazine.
[5]
C. Sotin,et al.
Determination of mineral phase percentages in granular rocks by image analysis on a microcomputer
,
1988
.
[6]
Alex B. McBratney,et al.
STRUCTURA: a C program for estimating attributes of two-phase, heterogeneous structures digitized from planar specimens
,
1991
.
[7]
R. Marschallinger,et al.
A method for three-dimensional reconstruction of macroscopic features in geological materials
,
1998
.