Typical use on PCA/SVD on image, what you do is treat each image as a vector, create a matrix out of a collection of images, and then do PCA/SVD on the matrix to analyze the distribution of the images, normalize, get the eigen-images (principal components), etc.
