Image Integration Based Ant Colony System for Multiband Satellite Image Classification

The motivation we address in this paper is to find out a generic method used to classify conceptual satellite image taken in multiband imagery. Predefined training image with different imagery bands is considered to test the proposed classification method. The Korhunen-Loeve (KL) transform is first employed to create newly integrated image with dense information and best contrast due to the information of all used bands are concentrated in one integral image. Then, the integrated image is partitioned into variable sized blocks using hybrid horizontal-vertical (HHV) partitioning method. The size of blocks is determined automatically according to the spectral uniformity measurements. Later, ant colony optimization (ACO) is used to find out the optimal number of classes that may exist in the image, and then classify the image in terms of the discovered classes. It was found that the obtained classification results by ACO are in a good agreement with the actual training data, which ensure the success of the proposed method and the effective performance of the classification.