Elemental classification in multi-detector stem images using image analysis clustering techniques.

Abstract

The multi-detector STEM images form a multi-dimensional measurement space where picture elements of morphologically distinct regions cluster and can be separated for classification using image processing clustering techniques. These images are highly correlated and improve the classification only at the high cost of adding dimensionality. Data reduction techniques which take the class separation into account can be used to compress the useful information carried by these images into a few components to which clustering techniques can be successfully applied. At high magnification, a slight displacement is sometimes observed, which can slightly impair the classification result. The redundancy of the information in the quadrant images suggests their use for phase retrieval - which, added to an energy loss data channel, may greatly improve the classification.