Automated relief-based discrimination of non-opaque minerals in optical image analysis

• Segmentation of some non-opaque minerals such as quartz by reflectivity is an issue in automated OIA.
• Segmentation of such minerals based on optical relief is suggested instead.
• Relief-based segmentation gave promising results for well-prepared samples.
• Issues have been identified and resolved when applying the method to real-life samples.
• The algorithm has been incorporated into CSIRO Mineral4/Recognition4 OIA software.

Ore characterisation is important in order to understand the quality of ores and their behaviour during downstream processing. Many significant ore characteristics can only be determined through the use of various imaging techniques. Optical Image Analysis (OIA) is one such technique and is particularly attractive for many applications due to its low cost and high resolution. However OIA also has some limitations, one of which is the difficulty with discriminating non-opaque minerals. Some non-opaque minerals, such as quartz, are typical gangue minerals in certain types of iron ores. Even though in many cases quartz particles can be easily seen and attributed by mineralogists in polished sections, their automated discrimination has always been an issue, the reasons for which are discussed in this article. The ability to automatically discriminate quartz and other non-opaque minerals would significantly increase the value of OIA for the mineral industry.This paper describes a novel method of discriminating non-opaque minerals in the sample by their optical relief, which results in visible borders between the mineral and the epoxy resin mounting medium. An algorithm for such discrimination that has been developed for the CSIRO Mineral4/Recognition4 OIA software package is described. The algorithm is based on dynamic thresholding of the image with subsequent cleanup and enhancement to reliably determine borders between non-opaque particles and epoxy and on subsequent attribution of image areas created by these borders to either the non-opaque mineral or the epoxy resin. Further, this article discusses difficulties that may arise when applying this algorithm due to sample peculiarities and describes algorithm enhancements incorporated in Mineral4 in order to overcome these issues. The resulting software is capable of reliably discriminating non-opaque minerals in a variety of samples, including iron and manganese ores.