Texture Segmentation via Haar Fractal Feature Estimation

Abstract We examine an approach to texture segmentation that uses the fractal dimensions along the 1-D cross sections of 2-D texture data as image features, where an effective Haar transform fractal estimation algorithm is utilized. The major advantage of the Haar fractal estimator is its computational efficiency along with robustness. The method is fast due to the pyramid structure of the Haar transform and nearly optimal in the maximum likelihood sense for fractional Brownian motion (fBm) data. We compare the low complexity of this new algorithm with the complexity of existing fractal feature extraction techniques, and test our new method on fBm data, real Brodatz textures, and natural scenes.