| Fractional Brownian motion (fBm) holds intuitive appeal for modeling natural visual textures. In addition to synthesizing artificial textures based upon fBm, fractal characteristics may be extracted from actual images, to quantify aspects of texture within them. Recent research has demonstrated that fBm may be used to model and subsequently characterize visual textures. However, the vast majority of existing investigations have relied upon the fractal dimension, variants of the fractal dimension and lacunarity, all of which assume that the observed image fits the fractal model. The error between the actual image and the fractal model should be considered. Here, the fractal error is introduced as a texture feature.; Although fractal models apply to a wide class of textures, particular emphasis is made here on aerial imagery, which few studies have included in their analyses. A localized version of the fractal error measurement was developed for the segmentation of aerial imagery. This operator was incorporated into a general purpose aerial image segmentation software package and evaluated. These successful results were compared against the previous choice of texture features for representing natural textures in aerial images, the "busyness" operator. In addition to segmentation, another important aspect of the scene evaluation process is the identification of stable features within an aerial image. Here, the fractal error also performed favorably, locating the stable, man-made objects in certain types of aerial images. |
