AAU Student Projects - visit Aalborg University's student projects portal
A master's thesis from Aalborg University
Book cover


Initialisation and Multi-Layer Clustering of Spatial Data: A Better Method

Authors

;

Term

4. term

Education

Publication year

2011

Submitted on

Pages

104

Abstract

I denne afhandling præsenterer vi en metode til klyngedannelse af rumlige data baseret på Markov Random Fields (MRF), et statistisk rammeværk der modellerer, hvordan nabopunkter påvirker hinanden. Arbejdet har to dele. Først undersøger vi flerlags klyngedannelse og anvender det på geoflorale data (geografisk refererede plantedata). Vi designer en MRF-baseret tilgang, der bruger logistisk regression og kan håndtere både kategoriske og numeriske variabler, så forskellige typer information kan kombineres på tværs af lag. Dernæst ser vi på, hvordan man bedst initialiserer klyngedannelse i billeder. Vi udvikler en række initialiseringsteknikker baseret på histogrammer og enkle udvælgelsesheuristikker. Målet er at vælge en initial farvemodel, hvor segmenterne starter med så tydeligt forskellige farveprofiler som muligt, for at understøtte den efterfølgende segmentering.

This thesis presents a method for clustering spatial data based on Markov Random Fields (MRFs), a statistical framework that models how nearby locations influence each other. The work has two parts. First, we study multi-layer clustering and apply it to geofloral data (geographically referenced plant data). We design an MRF-based approach that uses logistic regression and can handle both categorical and numerical variables, enabling different types of information to be combined across layers. Second, we address how to initialize image clustering. We develop a set of initialization techniques based on histograms and simple selection heuristics. The goal is to choose an initial color model in which segments start with as distinct color profiles as possible, to support subsequent segmentation.

[This abstract was generated with the help of AI]