Forward/backward modeling Modeling of the free space wave field propagation by the developed discrete diffraction transform in the Fourier (F-DDT) and spatial domain (M-DDT) is presented. The reconstruction of a whole object wave field is performed from a noisy complex-valued observation given at the measurement planes. It is shown that the presented discrete-to-discrete model is aliasing free and accurate for pixel-wise invariant wave field distributions at the object and measurement planes. The object reconstruction is compared with the conventional convolution and angular spectrum decomposition (ASD) methods with a clear advantage of DDT.
Multi-plane phase retrieval The iterative phase-retrieval algorithm is developed in terms of the variational maximum likelihood using the augmented Lagrangian technique. The maximum likelihood based formulation enables the optimal solution of the phase-retrieval problem for noisy observations. The developed augmented Lagrangian based (AL) algorithm demonstrates high efficiency and significantly better reconstruction quality with respect to conventional Gerchberg-Saxton-Fienup methods. An advanced performance is obtained if a prior on the object wave field to be reconstructed is given.
Sparse reconstruction It is recognized, that the effects, arising due to the forward free space propagation, can not be properly compensated by only backward operators. Because of that the reconstructed wave fields are noisy and typically corrupted by different artifacts: e.g. blurring, "waves" on borders, etc. The disturbances of the optical track are hard to be specified but could be suppressed by filtering. Sparse modeling of the wave field is one of the effcient numerical techniques for imaging. It is found that the used spatially adaptive regularization using BM3D-frames improves the wave field reconstruction quality and gives significant enhancement of imaging. It is shown that the parallel structure of the algorithm allows increasing the computational performance by even a commodity GPU what is crucial for processing of large images.