Fast automatic unsupervised image segmentation and curve detection in spatial point patterns

There is a growing need for image analysis methods which can process large image databases quickly and with limited human input. I propose a method for segmenting greyscale images which automatically estimates all necessary parameters, including choosing the number of segments. This method is both fast and general, and it does not require any training data. The EM and ICM algorithms are used to fit an image model and compute a pseudolikelihood; this pseudolikelihood is used in a modified form of the Bayesian Information Criterion (BIC) to automatically select the number of segments. A consistency result for this approach is proven and several example applications are shown. A method for automatically detecting curves in spatial point patterns is also presented. Principal curves are used to model curvilinear features; BIC is used to automatically select the amount of smoothing. Applications to simulated minefields and seismological data are shown.