
152 Filters, Kernels and Fields
variance into the first three principal components. Chapter 9 will illustrate
the use of kernel PCA for change detection with (simulated) nonlinear data.
4.5 Gibbs–Markov random fields
Random fields are freq ue ntly invoked to describe prior expectations in a B ayes-
ian approach to image analysis (Winkler, 1995; Li, 2001 ). The following brief
introduction will serve to make their use more plausible in the unsupervised
land cover classification context that we will meet in Chapter 8. The devel-
opment adheres close ly to Li (2001), but in a notation specific to that used
later in the treatment of image classification.
Image ...