Karhunen-Loeve transform using neural networks

The optimality of Karhunen-Loeve transform (KLT) over other transforms has been well known, together with the difficulty in implementing practical KLT systems. The wide applications of the transform deserve a new investigation on realizing such systems by using artificial neural networks. In this paper the KLT is known to be equivalent to a constrained optimization problem by maximizing covariance of output signals with the constraint of orthonormality. A neural network is then developed which can converge to the basis vectors of the transform