Hexagonal wavelet decomposition for texture characterization

This paper demonstrates new possibilities in texture characterization. offered by hexagonal spectral decomposition schemes. We have described, theoretically analyzed and compared image decompositions based on separable and hexagonal sampling. For both decompositions, feature extraction is performed with the fillers derived from the same 1D wavelet family, and textures are characterized by a set at channel variances calculated at the output at the corresponding filter bank. Experiments with a large number of textures have shown that in standard working conditions both decompositions perform equally, but that hexagonal transform is optimal for line structural analysis and performs well in the presence of noise. Finally, hexagonal decomposition was incorporated in a simple region growing technique, showing its use in a texture segmentation.