Histogram and average neighborhood structure of local lacunarity and fractal features for noise-robust and rotation-invariant texture description

This paper presents a rotation-invariant and noise-robust texture descriptor based on fractal analysis of digital images. In the proposed method, the fractal dimension and lacunarity of local regions in the given texture image is first computed, and considered as local fractal features. The histogram of the computed features is then calculated. To extract global features, a global map called average neighborhood structure is computed, which represents the average similarity between the computed local fractal features of any given image pixel and its surrounding neighbors within a specific window. A number of rotation-invariant features are then extracted from the obtained map. The extracted global features along with the calculated histogram is finally used for texture description and classification. Extensive experiments were conducted on two well-known texture datasets (Outex TC10 and CUReT) in the presence of different levels of Gaussion noise. The obtained results indicate that the proposed method achieves high texture classification accuracy, especially in the presence of high levels of Gaussian noise.