Fractal approaches in texture analysis and classification of remotely sensed data: comparisons with spatial autocorrelation techniques and simple descriptive statistics

There has been growing interest in the application of fractal geometry to observe spatial complexity of natural features at different scales. This study utilized three different fractal approaches—isarithm, triangular prism, and variogram—to characterize texture features of urban land-cover classes in highresolution image data. For comparison purpose and to better evaluate the efficiency of fractal approaches in image classification, spatial autocorrelation techniques (Moran’s I and Geary’s C ), simple standard deviation, and mean of the selected features were also examined in this study. The discriminant analysis was carried out to discriminate between classes of urban land cover on the basis of texture measures (variables). This study demonstrated that the spatial autocorrelation approach was superior to the fractal approaches. In some cases, simple standard deviation and mean value of the samples gave better accuracy than all or some of the fractal approaches. The results obtained from this analysis suggest that fractal-based textural discrimination methods are applicable but these methods alone may be ineffective in extracting texture features or identifying different land-use and land-cover classes in remotely sensed images.