Invariant pattern recognition using ring-projection and dual-tree complex wavelets

A novel descriptor is proposed for invariant pattern recognition by using ring-projection and dual-tree complex wavelets. The ring-projection takes the summation of all pixels that lie on the circle with radius r and centre at the centroid of the pattern. This transforms the pattern from a 2-D image to a 1-D signal, so less memory is needed and it is faster than existing 2-D descriptors in the recognition process. Dual-tree complex wavelet transform is applied to the ring-projection since it has the approximate shift-invariant property, which is very important in pattern recognition. Experimental results show that the proposed descriptor in this paper is a feasible choice for invariant pattern recognition.