Utilisation of Enhanced Thresholding for Non-Opaque Mineral Segmentation in Optical Image Analysis

Abstract

To understand and optimise downstream processing of ores, reliable information about mineral abundance, association, liberation and textural characteristics is needed. Such information can be obtained by using Optical Image Analysis (OIA) in reflected light, which can achieve good discrimination for the majority of minerals. However, reliable automated segmentation of non-opaque minerals, such as quartz, which have reflectivity close to that of the epoxy they are embedded in, has always been problematic. Application of standard thresholding techniques for that purpose typically results in significant misidentifications. This paper presents a sophisticated segmentation mechanism, based on enhanced thresholding of non-opaque minerals developed for Commonwealth Scientific and Industrial Research Organisation’s (CSIRO) Mineral5/Recognition5 OIA software, which significantly improves segmentation in many applications. The method utilises an enhanced image view using an adjusted reflectivity scale for more precise initial thresholding, and comprehensive clean-up procedures for further segmentation improvement. For more complex cases, the method also employs specific particle border thresholding with subsequent selective erosion-based “reduction to borders”, while “particle restoration” prevents the detachment of non-opaque grains from larger particles. This method can be combined with “relief-based discrimination of non-opaque minerals” to achieve improved overall segmentation of non-opaque minerals.