Translation, rotation and scale invariant texture characterization method for retrieval

Retrieval and classification of distorted texture images is a challenging and important problem in real world image analysis and understanding. This paper proposes a new texture characterization method applying for image retrieval and classification which is robust to geometric distortions. The geometric distortions include translation, rotation and scale changes of textures. A log-polar transform of autocorrelation image which expresses regularity and continuity is introduced to eliminate the entire geometric distortions. Next, the texture feature is extracted using statistics of wavelet packet In the simulations, the robustness to geometric distortions is verified. The effectiveness of our characterization method is demonstrated in texture image retrieval experiments.