Shift-invariant texture retrieval using P- contourlet

• Authors: Jiuwen Zhang;Jincai Mi;Tongfeng Zhang
• Venue: Proceedings of the International Conference on Advances in Computing, Communications and Informatics
• Year: 2012

This paper presents a new kind of shiftable complex Contourlet transform, P-Contourlet, which has the characteristics of multiresolution, higher direction selectivity and low redundant. The P-Contourlet transform first project a real signal into an analytic signal and then Contourlet transform is applied on it. As an analytic signal itself has the property of shift-invariance, the P- Contourlet transform has higher shift-invariance than Contourlet transform which is applied on real signals. The multiresolution and higher direction selectivity of the P-Contourlet transform inherits directly from the Contourlet transform. The P- Contourlet transform has 8/3 redundancy for the analytic signal of a real signal is complex, two times to the Contourlet which is up to 4/3 redundancy, much less than the NSCT and less than the PDTDFB. Unlike the NSCT and PDTDFB, the P-Contourlet transform has a simple structure to implement and has higher computation efficiency. The projection is implemented by convolution the real signal and a projecting filter, and the projecting filter is obtained from shifting 90° in phase to a half-band low-pass orthogonal filter. The projection process maps the frequency spectrum [(-π,-π) ~ (π,π)] of an image to [(-π,π) ~ (π,π)] (or[π,-π) ~ (π,π)]) to suppress negative frequency along single axis. This is a kind of frequency band limited in two-dimension and thus reducing the aliasing in succeeding directional subbands. We apply the P-Contourlet on texture retrieval and the experimental results indicate that it outperforms other approaches.