Comparison of efficiency of extraction of built-up areas in aerial images using fractal analysis and morphological granulometry

The paper presents a comparison of results of the automatic extraction of built-up areas, based on fractal analysis and granulometric maps, in the aerial images. Built-up areas as a land-use class can be clearly seen in an aerial or satellite image, due to its high granularity, but for the same reason they are very difficult to extract using a “traditional” non-contextual, pixel-based classification. Both approaches presented in the paper, using fractal analysis and morphological granulometry, base generally on a pixel-based classification, but performed on images reviously processed using these two types of processes. Fractal analysis consists in an empirical computing of fractal dimension of parts of an image, using a box-counting method. Such an approach generates an image where pixel values are equal to a fractal dimension values of their neighbourhood. Since we can interpret a fractal dimension as a level of granularity, a simple reclassification of such an image can improve a performance of an automatic extraction of built-up area effectively. The approach based on a morphological granulometry creates a number of granulometric maps – images where pixel values mean an amount of objects of certain size in a set neighbouring fragment of an image. This way a number of these images can be processed using a pixel-based classification, to perform an effective extraction of built-up areas in an image. The results of the presented approaches have been compared to the reference mask obtained basing on a visual interpretation of the image.