Jeff Lewis at UCLA told me he teaches principal components with an image reconstruction example. This got me inspired to try it myself. A snapshot appears below, showing how the image quality improves quickly with a relatively small number of principal components. A full, Sweaved write up is here, making use of the biOps package


Zero Inflated Models and Generalized Linear Mixed Models with R.
Zuur, Saveliev, Ieno (2012).