Application of SVD to 2-D spectral estimation

In this paper, a new method by which some modern techniques of one dimension (1-D) spectral estimation can be extended to two dimension (2-D) cases is presented. The main point of the new method is to find the optimum separable approximation of a given autocorrelation matrix in the least square sense, so that 2-D spectral estimation can be reduced to 1-D problem. It is proved in this paper that finding the optimum separable approximation of a matrix in the least square sense is equivalent to finding separable representation of the matrix by singular value decomposition (SVD). Finally, some results of experiments are shown to illustrate the performance of the new method and to compare with other 2-D spectral estimation methods.