Markov chain Monte Carlo methods for clustering of image features

The identification of centres of clustering is of interest in many areas of applications, for instance edge detector output has to be grouped into meaningful curves. The authors argue that stochastic geometry models are helpful both in providing models for clustering and as a prior distribution to combat overestimation of the number of clusters and to improve robustness. The general idea in connection with object recognition was proposed by Baddeley and van Lieshout (1993) and van Lieshout (1993). Independently, in an epidemiological context, a different Gibbs sampler technique for detection of cluster centres in a Cox process was developed by Lawson (1993).<
>