Fractal dimension of maximum response filters applied to texture analysis

Texture provides fundamental features for several applications in computer vision and it is known to be an important cue for human vision. To improve the description of texture in different viewpoints (e.g., scale and orientation), we propose to extract fractal descriptors from invariant filter responses space. Experimental results on four well-known texture datasets show the effectiveness of the proposed method. In the experiments, our method has provided better results than relevant texture methods, including the standard fractal descriptors. In addition, experimental results have shown the robustness of our method in a dataset with high noise.