Quantitative interpretation of HRTEM images using multivariate statistics: the case of the (γ, γ′)-interface in a Ni base superalloy

We present a two-stage procedure for the analysis of high resolution transmission electron microscopy (HRTEM) images. First the original image, or more precisely the underlying unit cells, is decomposed according to correspondence analysis (CA), a branch of multivariate statistics (MS), into a certain number of “eigencells” with respective coordinates. CA orders these eigencells in a hierarchical way according to the information content via the associated eigenvalues. The initial image can then be partially reconstructed by considering only those eigencells carrying the relevant information present in the original image and discarding by this way “useless” information (“noise”). Second a reduced set of coordinates equivalent to this partially reconstructed image is then submitted to several classification algorithms known in remote sensing (RS). The specificities of RS and its relationship to other methods will be highlighted. The combination of these two techniques will be applied in order to localize the (γ, γ′)-interface in a Ni base superalloy.