Preliminary investigation of two methods for the automatic handling of multivariate maps in microanalysis

Multivariable analytical maps are very useful for evaluating the joint distribution of several elements in a specimen. However, few tools have been developed for the handling of such multivariate data sets and for their quantitation. Scatter diagrams are limited to two or three maps, and their exploitation is often made interactively. Two ways are preliminary explored in this paper, in order to go from interactive correlation partitioning (ICP) in a two-dimensional space towards automatic correlation partitioning (ACP) in a N-dimensional space (N ⩾ 2). Of these two ways, the first is a clustering method (the K-means) working in the abstract parameter space while the second is a region growing method working in the image space. These two methods are described and tested on two simulations and a real X-ray map series. Some indications are also given for further studies which have to be conducted before such methods can be used routinely.