An example of principal component analysis applied to correlated images

The use of principal component analysis (PCA), also known as singular value decomposition (SVD), is a powerful tool that is frequently applied to the classification of hyperspectral images in remote sensing. Unfortunately, the utility of the resulting PCA may depend on the resolution of the original image, i.e., too coarse-grained of an image may result in inaccurate major principal components. This work presents an example of how the major principal component obtained from the PCA of a low-resolution image may be refined to obtain a more accurate estimate of the major principal component. The more accurate estimate is obtained by recursively performing a PCA on only those pixels that contribute strongly to the major principal component.