An image analysis method to determine crystal size distributions of olivine in kimberlite |
| |
Authors: | E J Holden S Moss J K Russell M C Dentith |
| |
Institution: | (1) Centre for Exploration Targeting, School of Earth and Geographical Sciences, The University of Western Australia, 35 Stirling Highway, Crawley, WA, 6009, Australia;(2) Volcanology and Petrology Laboratory, Department of Earth and Ocean Sciences, University of British Columbia, Vancouver, BC, Canada |
| |
Abstract: | Crystal size distributions (CSDs) are a standard method of describing populations of crystals within magmatic rocks. Olivine
is the dominant phase in kimberlite (∼40–50% by volume) and features a diverse range of sizes, shapes and origins. CSDs of
olivine provide a logical means of semi-quantitatively characterising kimberlite. The CSDs can then be used to distinguish
or correlate between kimberlite bodies or to investigate processes related to ascent, emplacement and eruption. In this paper,
we present an automatic image analysis technique that provides efficient quantification of olivine CSDs within digital images
of polished slabs of kimberlite. This technique relies on a combination of algorithms for detecting regions of interest (ROI)
and for segmentation of ROIs in order to identify individual olivine crystals that are used for size distribution datasets.
The detection process identifies regions expected to be olivine using a model-based colour detection technique using Mahalanobis
distance combined with texture analysis based on local standard deviation and greyscale foreground enhancement techniques.
The segmentation process separates adjacent domains to identify individual crystals using an iterative marker-based watershed
algorithm to separate adjoined structures of varying sizes. We demonstrate the utility of automatic image analysis by comparing
CSDs for olivine derived from this method versus results from manual digitisation of olivine grains. The automatic detection
system correctly identified ∼86% of the manually detected olivine domains; ∼88% of the automatically detected regions correctly
correlate to manually defined olivine grains. Discrepancies between the two methods are mostly the result of oversimplification
of crystal margins (i.e. rounding) by manual tracing whereas automatic boundary recognition shows clear advantages in identifying
irregularities in crystal edges. Closer examination of the results shows that both methods suffer from under-representation
of smaller crystals due to: (1) human subjectivity and error in manual tracing and (2) noise removal processes in automatic
detection. Automatic detection of olivine grains is much more efficient than conventional manual tracing; manual detection
requires ∼6 h per sample versus ∼1 min for automatic analysis of the same sample. |
| |
Keywords: | Image analysis Feature extraction Mineral grain detection Crystal size distribution Olivine size distribution |
本文献已被 SpringerLink 等数据库收录! |
|