Investigating preprocessing of multivariate images in combination with principal component analysis

Principal component analysis (PCA) is quite widely used multivariate technique for finding interpretations of the variance-covariance structure, and to reduce the dimensionality, of the investigated (image) data set. However, PCA is not always used in a straightforward manner, it is quite often combined with preprocessing of th data. An overview of different possibilities used. mainly in the remote sensing area, and investigations on the effects for a couple of cases, are presented. In an application example using a Landsat TM scene, the scene is subject to preprocessing combined with PCA, and the result is investigated. It is concluded that objective measures, possibly in terms of signal-to-noise ratios, are needed in order to handle the situation of obtaining several sets of PC images from one original image data set.