Variogram-Based Descriptors for Comparison and Classification of Rock Texture Images

Gonzalo Diaz, Julián Ortiz, Jorge F Silva, Rodrigo Lobos, Alvaro F. Egaña

Abstract:

Rock characterization is typically performed by geologists in mining companies and involves the analysis of several meters of drill-hole samples to describe
distinctive geological properties. In this procedure, rock texture is not typically taken
into account despite its importance given its close relation with metallurgical responses
and, therefore, all mineral processes. To support the work of geology experts, this
research seeks to obtain rock texture information, discriminating it from digital
images through image processing and machine learning techniques. For this purpose,
a geologist-labeled digital photograph database was used with different rock texture
classes (including geological textures and structures) from drill-hole samples. To characterize rock texture, three texture descriptors based on variographic information are
proposed, which summarize data contained in the image pixels, focusing on local
structural patterns that numerically describe its texture properties. Then, based on a
methodology of image texture comparison, which could be extended to classify different types of rock texture classes, a quantification of the system’s performance was
obtained. The results showed a high discrimination among common texture classes
using compact variogram-based features that outperformed previous methods applied
on the same rock texture database.