Novel approaches to reconstruction of two dimensional signals observed under additive noise are investigated. The techniques make use of third-order statistics and their noise insensitivity and phase preserving properties. Conditions and algorithms for reconstruction of two-dimensional signals from the fourier transform of the third-order statistics (bispectra) are investigated. Advantages and disadvantages of the proposed method with other existing methods are established through theoretical analysis and simulation. The techniques are applied to image registration and reconstruction of moving objects.
