Computation of moment invariants and Hadamard transform using neural net
Yiquan Wu, Zhaoda Zhu · 2002
In this paper, computation of moment invariants and Hadamard transform using the Tank and Hopfield linear programming neural net is proposed. First, the relationship between the one-dimensional moments and the one-dimensional Hadamard transform (1D HT) is derived. One can compute the moments of grey level image through 2N 1D HT's except for a negligible amount of addition, shift and multiplication operations. Then, the neural net to compute the 1D HT is shown. Because the HT matrix H satisfies H=H/sup T/ and H/sup 2/=NI, a closed-form solution of the time evolution of the neural net can be found. A proof is given that the neural net will find a result arbitrarily close to the correct HT of the input data in hundreds of nanoseconds. The proposed HT implementation is speedy, simple and robust. The proposed approach will be expected to find wide practical applications that require computing moments and HT.>