On the roles of PCA and ICA in data fusion

The authors consider the roles of PCA (principal component analysis) and ICA (independent component analysis) in the fusion of several data sources in remote sensing. With high dimensional data from each source they may be tempted to use PCA or ICA to obtain a small subset of most significant components. Smaller feature sets not only can reduce the computation requirements but also can mitigate the Hughes phenomenon. However how they use the resulting feature sets from several data sources can make significant difference in the overall classification performance. Also fusion can lead to image enhancement. Several experiments have been performed to illustrate the conditions that PCA and ICA can be most useful in data fusion.