On comparing multifractal and classical features in minimum distance classification of AVHRR imagery

The ability to distinguish between different types of surfaces is the strength of texture descriptors in the analysis of satellite imagery. Although the most common analytical means are based on co-occurrence analysis, considerable progress has been made in understanding the role of fractal and multifractal analysis in remote sensing. After indicating the limitations of using fractal dimensions as the only texture descriptor and introducing the concept of multifractal geometry, we consider the effectiveness of using multifractal and second-order fractal features in image classification. In particular, we present the results of comparing two supervised classifications of an Advanced Very High Resolution Radiometer (AVHRR) image of Scotland using classical texture features and multifractal second-order fractal ones. In terms of percentage correct and Khat statistics, this study provides evidence, with a confidence limit of 95%, that classifications using multifractal and second-order fractal features are more accurate than those using classical features. The classification algorithm used for this study is a typical minimum distance classifier.