Point counting

Point counting is a method for estimating the composition of rocks, based on identifying the mineral or grain present at a large number (usually 300 to 500) of points in a petrographic slide.

Error in point counting
Sources of error:
 * Misidentification of minerals or grains, or counting bioclasts as 'carbonate', say
 * Not sampling volumetrically small components
 * Choosing a small step size, compared to the grain size — old rule of thumb is 1.5 times the grain size, but this is an oversimplification. The strategy should be to uniformly sample the entire sample we have collected. Be sure not to collect along a transect, for example, especially if the sample is stratified.
 * Not sampling the rock enough, failing to capture heterogeneity
 * Not accounting for prior likelihood
 * Interpretation: point count data gives a volume percent composition. It is often interpreted alongside other data like XRD analysis, which gives weight percent. You must use the density of each mineral to correct one of the datasets
 * Using different samples for correlated interpretation (e.g. doing CL or SEM or XRD or XRF or whatever on completely different samples)

Literature
Treatments by Nielsen & Brockman (1977), Demirmen (1971) , Demirmen (1972).

Bayesian approach. Should we take into account the prior likelihood of encountering the mineral? What about the likelihood of the operator spotting and identifying it correctly? Perhaps it has a cryptic habit, or a small grain size. E.g. this study of point-counting birds: Simons et al. (2009).

Basic assumption: the slide is representative of the rock. In other words: the rock is scale-free. Must be sure to sample at a spatial frequency that will capture the volumetric heterogeneity in the rock.

Computing error

 * ImageJ porosity workflow
 * Some Python goodness here.

Point count tools and software
Why no random sampling?
 * Pelcon
 * Endeeper
 * Petrog
 * Pointscan
 * Not sure if Rock.AR is available

Automatic detection

 * Livingood & Cordell (2008)