Image recognition using a projection-based wavelet moment

In this paper, a new wavelet invariant moment based on image projections is proposed. Firstly, Radon transform is used to project the image to 1-D space. Then, a translation and scaling invariant adaptive 1-D wavelet transform is introduced, and the projection data is transformed to wavelet domain to get a feature matrix with translation and scaling invariance. Since the feature matrix is influenced to be cyclically shifted of rows when image rotates, Singular Value Decomposition (SVD) is employed and singular value vector is invariant under image translation, scaling and rotation. Simulation results are given to obtain insight into the efficiencies offered by this method.