In this report, an approach to spatial clustering based on Markov Random Fields will be proposed. The report has two main foci, multi-layer clustering, performed on geofloral data, as well as improved initialisation, performed on image data.

A superior technique for clustering of spatial data in multiple layers is designed, using logistic regression, with the capabability to handle both categorical and numerical data.

A group of initialisation techniques for image clustering is developed using histograms and selection heuristics, with a focus on the initial colour model prior to segmentation having as different models for each segment as possible.