Imaging and image analysis are fundamental to mineral liberation analysis. A low noise, high resolution image is a prerequisite for mineral identification and quantification. The very stable BSE signals from a modern SEM can generate quality high resolution (0.1 - 0.2 microns) images of particle sections. These images allow the MLA, through its advanced image analysis techniques, to accurately discriminate the mineral phases within a particle. The principal image analysis functions used by the MLA are known as particle de-agglomeration and phase segmentation.

Particle De-agglomeration

A liberation analysis using the MLA involves the setting of particles into a mould (typically 30 mm diameter) with epoxy resin to form a hardened block. Typical particle sizes range from 10 µm to 1 mm and are preferably of a defined narrow size fraction. The block is then ground to expose a representative cross section of particles which is subsequently polished and coated with conductive carbon before being presented to the SEM. Despite precautions to prevent it, inevitably some particles in the prepared sample block will touch each other. If not recognised by the system and treated appropriately the agglomeration of particles can lead to biased liberation results. The MLA system has an automated de-agglomeration function that detects agglomerates and separates them according to a set of predetermined parameters.

The de-agglomeration function can be used both during the on-line measurement or performed off-line. Particle shape parameters determine if particles are agglomerated. The de-agglomeration procedure has three methods or criteria at its disposal to find the best separation option: 1) shadow or boundary identification, 2) linear feature recognition and 3) an erosion/dilation procedure. The operator can control the weighting applied to each of the separating criteria through a set of parameters.

Phase Segmentation

Once individual particles have been identified, the next step of the liberation analysis identifies all distinct mineral phases (or grains) and defines their boundaries accurately. This process is called phase segmentation and is performed on each individual particle. The MLA phase segmentation function outlines the regions of homogeneous grey levels within a particle BSE image. The average BSE grey value of every defined region corresponds with a mineral of unique average atomic number (AAN). The AAN determines the number of back scatter electrons emitted by the mineral and hence is directly proportional to the grey level registered in the BSE image. An example of a grey scale histogram with its peaks corresponding to the minerals in a lead-zinc ore is shown below.

Phase segmentation also involves the recognition and elimination of features of a BSE particle image that do not represent an independent phase. These artefacts can be cracks, shading, tiny voids or the dark perimeter or halo that appears around many particles.

As particle based segmentation uses the grey scale histograms of each individual particle, the influence of any grey level effects due to changes in measurement conditions, such as beam drift, are eliminated.