Abstract: This paper proposes a novel rotation-invariant texture image retrieval algorithm based on nonsubsampled contourlet transform (NSCT) and matrix F-norm against the common rotation problems for image retrieval. By calculating the F-norm of image subbands decomposed by NSCT, the texture feature elements are extracted. For each scale, the feature elements are re-ordered ascending by the sum of mean and standard deviation of each image subband, the similarity of two images is calculated by the feature vector similarity weighting of different scales. Experiments are conducted on the image set produced by Brodatz database, and the result demonstrates superiority of the proposed method.

Key words: texture image retrieval, nonsubsampled contourlet transform, F-norm