Two novel extensions to traditional
correlation analysis applied to
change detection in earth observation image data
Two extensions to traditional canonical correlation analysis are described. One consists of an iterative scheme in which increasing weight is put on observations that show a high degree of similarity between the two sets of variables involved. The other extension is based on replacing correlation as the measure of association between the two sets by the information theory concept of mutual information which in turn is entropy based. Both extensions lead to more computer intensive algorithms than the traditional method. Also, both extensions are applied to detect change over time in bi-temporal, multispectral earth observation image data.