Abstract:

A new kind of texture image retrieval algorithm based on Radon and wavelet transform is presented. Firstly, the Radon transform is used to deal with retrieval image. Centering the Radon projection of image by its center of mass can eliminate the shift effect. The statistical value of projection can be used to normalize the projection matrix. Then the translation- , rotation- and scale-invariant wavelet coefficients of the projection can be attained by using a TSI wavelet transform along and t directions. Secondly, the energy of every sub frequency bands is extracted to be the texture feature of image. The Gaussian model is used to normalize the different sub-characters distance to the texture feature of image. Thirdly, the texture similarities of images are computed by the Canberra distance. The image retrieval experiments indicate that this method is robustness to the noises in image’s similarity retrieval and has a higher retrieval-rate than other methods’.

Key words: content-based image retrieval, wavelet transform, Radon transform, texture feature