The effects of pre-processing of image data on self-modeling image analysis

The use of chemical imaging of secondary ion mass spectrometry (SIMS) data for self-modeling image analysis (SIA) has special challenges because of the following reasons: (a) At higher counting rates, the data are non-linear. (b) The heteroscedastic nature of the noise causes structure in the data which gives rise to extra components. (c) There is a high amount of noise in SIMS data and outliers often cause problems. This paper will discuss an adaptation of a pre-processing method to correct for heteroscedastic noise and a method to minimize the effect of outlying pixels. Examples will be given of the following: (a) Different mixtures of palmitic and stearic acid on aluminum foil. (b) A film coating of polyvinyl acetate (PVA) and polystyrene (PS). (c) A sample of copper and nickel and a fused layer. Copyright (C) 2008 John Wiley & Sons, Ltd.