Skip to main content
. 2011 Oct 20;2011:846312. doi: 10.1155/2011/846312

Figure 4.

Figure 4

The training phase of our CRF model. We follow the same procedure as in our PP model up until the posteriors are estimated. We then calculate pixel likelihoods and use these as node features in our CRF model. We infer CRF parameters using gradient descent.