ABSTRACT

In Part I, we first present the tensor approach and its improvement for divi­ ding the calculation of the 2-D Fourier transform into the minimal number of short transforms. The main advantages of the tensor approach are the following. The approach is universal because it can be implemented for computing other discrete unitary transforms, such as the Hadamard, cosine, and Hartley transforms. The tensor approach requires fewer operations of multiplication than the existent al­ gorithms. The computation of the 2-D transforms can be efficiently performed in parallel, and it can be used with any new, fast 1-D unitary transform.