Extracting ore texture information using image analysis

Knowledge of ore texture and particle composition information is of vital importance to mineral processing engineers, especially to quantify mineral liberation. A great deal of information on the distribution of mineral in particles can be obtained by scanning electron microscopy (SEM) based techniques, such as a mineral liberation analyser (MLA) and quantitative evaluation of minerals by scanning electron microscopy (QEMScan). However, there is no reliable method at present that interprets two-dimensional data to obtain three-dimensional texture and thus predict mineral liberation quantitatively. Researchers have proposed models in one-, two- and three-dimension to evaluate the extent of liberation in comminuted particles. They have quantified ore texture information in terms of descriptors, such as covariance function, proximity function and linear intercept length distribution. However, measurement of such quantities is difficult in practice and hence analytical functions have been used for such descriptors based on assumed ore textures, such as Poisson polyhedra and Boolean textures. Such models are useful at best for low grade ores but not high grade ores. In this work, a reliable method of measuring the textural descriptors using image analysis techniques with SEM images has been presented. The conventional texture descriptors, namely, linear intercept distributions i(L), covariance function C(L), proximity function P(L) and phase specific line segment function ω⁽ⁱ⁾(L), have been obtained from two-dimensional images and compared with models proposed by King and Barbery. The method described in this work adopts a numerical method to determine the P(L) function based on the measured intercept length distributions, without any assumption of ore texture or particle shape. It may be applied to extract more extended ore texture data than the information that MLA and QEMScan techniques provide.