Abstract
A clustering Method with context information based on Genetic Algorithms(GA)is proposed. It is found that the proposed method allows to remove isolated clusters from clustered results which are usually occurred in the satellite image clustering. This is because of utilizing context information which derived from the image in concern, other than spectral information. On the other hand, context information may affect to clustering performance when the ground cover targets have a line likeness feature such as roads and rivers so that a specific treatment have to be done for such targets before applying the proposed contextual GA clustering. The clustering performance is evaluated through a comparison with a clustered result based on visual perception with Maximum Likelihood method for both simulated data and Landsat TM imagery data. It is found that the effectiveness of the proposed method depends on separeability or divergence between clusters.
