Rotation-invariant Texture Retrieval by Combining Log-polar and Nonsubsampled Contourlet Features

This paper analyzes the spectrum influence between Radon transform and Log-polar transform when rotation effect is eliminated. The average retrieval performance of wavelet and NSCT with different retrieval parameters is also studied. Based on which, the authors design a multi-scale and multi-orientation texture transform spectrum, as well as rotation-invariant feature vector and its measurement criteria. Then a new two-level rotation-invariant texture retrieval algorithm based on no-parameter statistic features is proposed. Experiments on Brodatz image database show that the algorithm proposed in this paper is appropriate for main orientation capturing and detail information description. Besides, the combination of this two-level progressive retrieval strategy and multi-scale analysis method can effectively improve retrieval efficiency compared with traditional algorithms and ensure a high precision as well.