Prediction of sandstone parameters from texture

This report describes an extension of the experiment "Texture analysis of sandstone surfaces" described in NR Note BILD/08/94. The objective of the study was to investigate possible textural features that can be applied to characterize different sandstone facies. A larger data set, consisting of 11 samples with known sandstone parameters and five samples with unknown parameters, was available in this experiment. The specified parameters of interest are permeability, grain size, porosity, and contents of quartz, clay, and feldspar. The textural features that showed strongest correlation were the Local GLCM features Inverse and Contrast, and the Global GLCM features Inverse Difference Moment and Inertia. The strongest correlation was found between grain size and Inertia (95%). Some geometrical features were also extracted from segmented sandstone images, however, the correlation was not very high for any of these features. GLCM spectra were calculated by varying the distance parameter in the textural features. The spectra indicated that values in the interval 6-10 should give good discrimination between the sandstone images. Linear regression models based on the best Global and Local GLCM textural measures were calculated for the sand­stone parameters, and PLS regression was calculated on a subset of the GLCM spectra by SINTEF. A comparison between the parameters for the five samples with unknown parameters after the release of these values, shows that both models predict several of the sandstone parameters fairly well.