Classification of remote sensing images via fractal discriptores

In image analysis and recognition of natural scenes, fractal geometry takes an important place among other techniques by analyzing, coding and extracting of important features to represent the digital image. In this paper, we discussed the problem of classification of natural urban images based on some statistical approaches, such as co-occurrence matrix and the principle of variogram function, which provides a more descriptive way to analyze different textures. Moreover, a new fractal descriptor is proposed as one of the statistical approaches to serve as a feature extractor in classification of different land cover image types. The performance of this descriptor is investigated through hierarchal clustering that show encouraging results to some classical textural analysis method in terms of computing and accuracy.