Developments in processing image sequences for elemental mapping.

Abstract

Elemental mapping consists in searching the distribution of a given chemical species over an extended specimen area, with relation to topographical or structural features. It can be done with EELS core signals from a combination of several energy filtered images. One major problem encountered in the processing of such sequences of images lies in the extrapolation errors due to a difficult estimate of the background below the characteristic signal. The chosen method must be sufficiently reliable to avoid the risk of both "false positive" and "false negative" values: the first category may stem from spurious signals or from a non-satisfactory fit of the background. The second category is mainly due to a limited sensitivity. The EELS signal is often much weaker than the background intensity; an extrapolation error can therefore transform a negative value into a positive one, or vice versa. The purpose of the present contribution is to check the validity of the processing at different levels: i) different mathematical models of background; ii) different types of fitting procedures (one-parameter and two-parameters fits); iii) different fitting methods and several associated manipulations, such as a quasi local estimation of the involved fitting parameters. The statistical validity of those techniques is discussed through several tests on real images obtained from different specimens (Co/CeO2 catalysts, ferritin molecules, U and Tb staining clusters). Progress is made on the way of quantitative elemental mapping at a given confidence level, and towards the identification of single atoms.