SATELLITE IMAGES CLASSIFICATION BASED FRACTAL FEATURES

In this paper, a TM-multi-spectral satellite images is adopted in a purpose of supervised classification. The traditional method of the segmentation namely Quad tree is applied as pre processing step. For each segmented block, the fractal features (fractal dimension and lacunarity)s are determined to be used as a maximum likelihood classifier. The results showed that the fractal dimension has not certainly able to classify the segmented blocks while the lacunarity gave good classification results. In general, the fractal geometry was found an efficient parameter for describing the image. The results show that the over all classification accuracy is 85.5%.