Quadratic self-correlation: An improved method for computing local fractal dimension in remote sensing imagery

We present a new method for computing the local fractal dimension in remote sensing imagery. It is based on a novel way of estimating the quadratic self correlation (or 2D Hurst coefficient) of the pixel values. The method is thoroughly tested with a set of synthetic images an also with remote sensing imagery to assess the usefulness of the techniques for unsupervised image segmentation. We make a comparison with other estimators of the local fractal dimension. Quadratic self-correlation methods provide more accurate results with synthetic images, and also produce more robust and fit segmentations in remote sensing imagery. Even with very small computation windows, the methods prove to be able to detect borders and details precisely.