Neural Networks for Image Restoration from the Magnitude of Its Fourier Transform
IWANN '01 Proceedings of the 6th International Work-Conference on Artificial and Natural Neural Networks: Bio-inspired Applications of Connectionism-Part II
The self-duality of discrete short-time Fourier transform and its applications
IEEE Transactions on Signal Processing
| Hi-index | 0.02 |
The Gerchberg-Saxton (GS) algorithm and its generalizations have been the main tools for phase retrieval. Unfortunately, it has been observed that the reconstruction using these algorithms does not always converge to the correct result even if the desired solution satisfies the uniqueness condition. In this paper, we propose a new deautocorrelation algorithm and a few auxiliary techniques. We recommend that a combination of the iterative Fourier transform (IFT) algorithm with our new algorithm and techniques can improve the probability of success of phase retrieval. A pragmatic procedure is illustrated. Different reconstruction examples that are difficult to reconstructed using the single IFT algorithm are used to show the robustness and effectiveness of the new combination of algorithms. If the given Fourier modulus data contain no noise, it is sometimes possible to get a perfect reconstruction. Even when the signal-to-noise ratio (SNR) of the Fourier modulus data is only 10 dB, a meaningful result remains reachable for our examples. A concept concerning the intrinsic ambiguity of phase retrieval is suggested. We emphasize the necessity of verification of the solution, since the available phase retrieval algorithms are incompetent for distinguishing between an intrinsically ambiguous solution and the true solution