A new Bayesian method is developed for segmenting objects in 2D images. An efficient prior and likelihood are initially derived for a deformable template, which is optimised by a new stochastic Markov Chain Monte Carlo method. The representation of the prior and the likelihood, and the resulting posterior allow inference to be carried out either with respect to the model or the image. This permits a combined modle-guided and data guided approach,which may be advantageous in many situations.
展开▼
