Multi-Band Image Classification Using Klt and Fractal Classifier

Abstract

In this paper, a multi-spectral satellite image of known training areas taken in six bands is used. The adopted classification method suggests transforming the six bands into new six other bands using KL transform. The information of the image is redistributed to be concentrated in the first transformed band, the information is decreases gradually in the remaining bands. Thus, the first three transformed bands that carried most of image features can only be used to achieve more accurate classification. The first chosen band was partitioned into blocks by quadtree method. Then the most popular fractal feature namely lacunarity was estimated for each block to classify them. The results showed six classes, which were greatly identifying the training areas existing in the image. The classification score was found about 91% when the contribution weight of the first band is 90%. This score increases lightly by increasing the weight of contributing the first band.