Multi-spectral image analysis is a powerful method to characterise quantitatively the mineralogy and microfabric of soils, sediments, and other particulate materials. Backscattered scanning electron microscope (SEM) images of polished, resin-impregnated samples are grouped with the corresponding X-ray elemental maps using classification methods commonly used in remote sensing. However, the resulting mineral-segmented images require processing to render them suitable for quantification. In the past, this has been done subjectively and interactively, but the new objective methods described in this paper largely eliminate this subjectivity. An intensity gradient magnitude image of the original backscattered electron image is used as the basis of an interactive erosion and dilation sequence to generate skeleton outlines defining the edges of the mineral grains. The areas defined within the skeleton areas are then classified as a particular mineral according to the predominant feature in the corresponding mineral-segmented image. Subsequent processing tackles the problems of 'holes' defined by the skeleton outlines, and the over-segmentation associated with certain classes of mineral grain. Further methods to deal with particles made up of more than one mineral are considered.
The matrix and porosity information is recombined to generate an image suitable for analysis using feature size statistics or general orientation analysis. The techniques described can be combined to permit batch processing of images. Applications of the techniques are illustrated on a soil from the East Anglian Breckland.
Tovey, N. K.; Dent, D. L.; Corbett, W. M.; and Krinsley, D. H.
"Processing Multi-Spectral Scanning Electron Microscopy Images for Quantitative Microfabric Analysis,"
Scanning Microscopy: Vol. 1992:
6, Article 25.
Available at: https://digitalcommons.usu.edu/microscopy/vol1992/iss6/25