Segmentation and interpretation of multivariate chemical image data

Both techniques micro-XRF and TOF-SIMS produce enormous amounts of data when used in imaging mode. It was the aim of this dissertation to develop and/or investigate a means to perform the semi-automatic segmentation of multivariate image data sets into regions of chemical relevant phases. In chapter 1 an introduction of principal component analysis (PCA) was given. PCA proves to be a relevant technique to reduce the dimensions of the data set. It reveals the relationships between the chemical images. Chapter 2 discussed the classical data transformation (pre-processing techniques and the wavelet transformation technique). As it turned out, data sets need to be pre-processed in order to remove and reduce experimental artifacts. Chapter 3 discussed the clustering of large sized data sets. In chapter 4 multivariate data analysis was performed on 2 micro-XRF data sets. In view of image processing, in particular the automated segmentation of micro-XRF images into regions of phases, the usefulness of principal component analysis, wavelet compression and clustering techniques on the image sets was demonstrated. It was concluded that pre-processing techniques influence the importance of chemical element in the data set. The square root pre-processing technique offers relative advantage for low intensity signals (lower concentration) and is reasonable compromise. The segmentation results of the wavelet and PCA compressed data set still reveals the chemical phases present in the micro-XRF data set. Chapter 5 discussed the multivariate data analysis of TOF-SIMS image data sets. Four image data sets were studied. Wavelets were used to reduce the images themselves and PCA was used to reduce the number of images. Data analysis of the image data sets revealed that the application of data reduction techniques retains chemical information. Wavelet compression removes the pixel-noise and PCA removes the variable noise. Segmentations show to be less dependent on the selected cluster technique.