Rotation invariant texture classification by ridgelet transform and frequency–orientation space decomposition

A new rotation invariant feature extraction method in the ridgelet transform domain for texture classification is proposed. Ridgelet transform can be divided into two stages: the Radon transform stage and the 1-D wavelet transform stage. According to the Projection-Slice theorem, the Radon transform actually provides information about the image data on a polar-grid in the frequency domain. This is ideal for rotation invariant feature extraction. Furthermore, by using wavelets that have compact support in the frequency domain, we can actually use ridgelet transform to achieve frequency–orientation decompositions for the given image data, which is similar to the multi-channel filtering technique. This makes the proposed method very effective in capturing texture properties for classification. Experimental results show a good performance achieved by the proposed method.