Rotation Invariant Texture Image Retrieval Based on Log-Polar and NSCT

Abstract

In order to solve the problem of rotation invariant texture image retrieval, an image retrieval algorithm based on Log-Polar and nonsubsampled contourlet transform (NSCT) is proposed. Log-Polar transform was first applied to texture image to convert the rotation to translation. Then, translation invariant NSCT was employed to decompose the transformed images. Standard deviations, energies and entropies of the different directional subbands features were calculated to make up a 45-dimension feature vector. Finally, Canberra distance was used to compute the comparability of this feature vector to retrieval images. The experiment results showed that the proposed algorithm was able to obtain a higher recall rate.

Extracted Key Phrases

7 Figures and Tables

Cite this paper

@article{Xu2013RotationIT, title={Rotation Invariant Texture Image Retrieval Based on Log-Polar and NSCT}, author={Bing Xu and Zhende Li and Shiyi Xie}, journal={JDIM}, year={2013}, volume={11}, pages={470-475} }