Natural-Ordered Complex Hadamard Transform Based Shape Description and Retrieval

Natural-ordered complex Hadamard Transform (NCHT) is a novel discrete orthogonal transform and NCHT power spectrum is invariant to the cyclic shift of transformed data. In this paper, NCHT is used to derive new shape descriptor, namely NCHT based descriptor (NCHD) for image retrieval. In NCHD each shape contour is sampled first and expressed by a complex signature which captures both local contour variations and global information of shape. And then NCHT is applied to these signatures. NCHD is formed by the resulting transform coefficients to describe shape features. The image retrieval performance of the proposed descriptor is evaluated on the standard MPEG-7 shape database using standard measurement, and the experimental results show that the proposed NCHD outperforms Fourier descriptor and the Wavelet Fourier descriptor.