Fractal analysis of a classified landsat scene

Remotely sensed images tend to be spatially very complicated, revealing regions of homogeneously classified pixels with quite convoluted perimeters. Fractal analysis, the study of complicated phenomena manifesting self-similarity at many scales, is suited to the description of the form and sizes of these regions. I apply fractal analysis to the patterns created by eight land-cover classes from a Landsat TM image of northwest Vermont. The results suggest that forests manifest high fractal dimension and large regions, agricultural activities have large regions with fractal dimension inversely related to the intensity of cultivation, and urban land cover yields small regions with relatively high fractal dimension. Analysis of the individual urban regions provides a data structure in the form of a raster-based GIS which can be used to investigate the location and description of individual regions and to diagnose the reliability of the classification and labeling processes.