A study on methods for rotation invariant image recognition based on texture characteristic

Three kinds of rotation invariant image classification recognition algorithms based on texture characteristic are proposed. All the methods proposed are based on rotation-to-shift. First, the texture image is transformed by log-polar transform or Radon transform to convert the rotation to shift, then filter the transformed image using dual-tree complex wavelet transform(DT-CWT) or discrete stationary wavelet transform(SWT) which is shift invariant to eliminate the shift. The rotation invariant feature vector is composed of the energies of the filter subbands and the SVM algorithm is used to classify at last. The paper experiments three feature extraction methods: log-polar transform combined DT-CWT, Radon transform combined DT-CWT and Radon transform combined SWT. Analyze the experiment results and compare the best one with other rotation invariant texture classification algorithm, the experiment results show that it can improve the classification rate effectively.