A wavelet transform approach to texture analysis

Traditional texture analysis algorithms focus too much on the local coupling between image pixels. Time/frequency analytical tools such as the Gabor and wavelet transforms can efficiently characterize the coupling of different scales in textures, and help to overcome this difficulty. However, the conventional wavelet transform, which has a finer resolution in the lower frequency channels, does not work properly for textured images, since textures are quasi-periodic signals whose dominant frequencies are located in the middle frequency regions. In the present research, a tree-structured wavelet transform is proposed. An adaptive procedure is developed to zoom into any frequency channels with significant information so that the decomposition can be further performed. The application of the new transform to texture classification is demonstrated