Abstract: A new texture feature is proposed using the Radon and wavelet transform according to the geometric property of Radon transform and a new image retrieval algorithm based on the texture feature is presented. First, the orientation-normalized image is achieved by rotating its principal axis to the horizontal direction. It is turned to obtain the rotation-invariant project after the orientation-normalized image is done by the Radon transform. Second, the adoptive wavelet decomposition, which has the property of translation-and scale-invariance to the project of Radon transform, is constructed. Finally, the energy of every sub frequency band is extracted to be the texture feature of the image. The image retrieval experiments indicate that this method has the property of shift, scale and rotation invariance and has a higher retrieval-rate of 3.8％ than the RIM method’s.

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