Unsupervised image classification using isodata and fuzzy C-Means

Kek, Soo Ling (2014) Unsupervised image classification using isodata and fuzzy C-Means. Project Report. UTeM. (Submitted)

Text UNSUPERVISED IMAGE CLASSIFCATION USING.pdf Download (280kB)

Abstract

The research is focus on unsupervised of satellite image classification to classify an image into thematic image by using two techniques: Iterative Self-Organizing Data Technique Algorithm (ISODATA) and Fuzzy C-Mean (FCM). ISODATA is an extension of K-Means algorithm but ISODATA determines the number of clusters dynamically. ISODATA tries to find the best cluster centroids through the iterative approach until it meet some convergence criteria. Besides, ISODATA involves splitting and merging of the resulting clusters based on the user pre-specified thresholds. For splitting situation, when a cluster standard deviation above a pre-specified threshold, a cluster will be split into two while for merging situation, when the distance between the centroids is below another pre-specified threshold, two clusters are merged. FCM is the most popular fuzzy clustering technique which allows one data point to belong to two or more groups or clusters with different membership degrees between 0 and 1. The aim of FCM is to find the cluster centers that minimize a dissimilarity function. Thus the research is necessary to know the how the classification work and which of the techniques can produce the better output after classify the satellite image. Moreover, the research will compare the results generated between the two different techniques. The research is carried out using Matlab R2010a and at the end of this research, the result will show which techniques will produce the better output after doing the comparison of the final outputs.

Actions (login required)

View Item