Phase retrieval with complexity guidance

Abstract

Iterative phase retrieval methods based on the Gerchberg-Saxton (GS) or Fienup algorithm require a large number of iterations to converge to a meaningful solution. For complex-valued or phase objects, these approaches also suffer from stagnation problems where the solution does not change much from iteration to iteration but the resultant solution shows artifacts such as presence of a twin. We introduce a complexity parameter ζ that can be computed directly from the Fourier magnitude data and provides a measure of fluctuations in the desired phase retrieval solution. It is observed that when initiated with a uniformly random phase map, the complexity of the Fienup solution containing stagnation artifacts stabilizes at a numerical value that is much higher than ζ. We propose a modified Fienup algorithm that uses a controlled sparsity enhancing step such that in every iteration the complexity of the resulting solution is explicitly made close to ζ. This approach which we refer to as complexity guided phase retrieval (CGPR) is seen to significantly reduce the number of phase retrieval iterations required for convergence to a meaningful solution and automatically addresses the stagnation problems. The CGPR methodology can enable new applications of iterative phase retrieval that are considered practically difficult due to large number of iterations required for a reliable phase recovery.

Robust phase retrieval with complexity-guidance for coherent X-ray imaging

“…For the sake of completeness, we start by describing the complexity parameter ζ 0 which was introduced in [14]. As stated above, ζ 0 is a measure of fluctuations present in pixel values of an object ρ(x, y), where the co-ordinate r is expanded as (x, y) .…”

“…As observed in [14,15], the complexity parameter can be used to guide the application of constraints in the object domain and provides artifact free phase retrieval solution which degrades benignly with increasing noise. In [14,15], the CGPR methodology was mainly developed with simulated noisy data in combination with the Fienup's hybrid input-output (HIO) algorithm. The complexity guidance idea is put to test for the first time in this work for experimental data obtained from the CXIDB database.…”

Robust Phase Retrieval with Complexity-Guidance for Coherent X-Ray Imaging

“…In recent works [14,15], we have proposed a novel approach that we call "complexity-guided phase retrieval" (CGPR) that is meant to address the typical stagnation problems with phase retrieval algorithms. This methodology uses a complexity parameter which is computed directly from the Fourier intensity data and provides a measure of fluctuations in the desired phase retrieval solution.…”

“…This methodology uses a complexity parameter which is computed directly from the Fourier intensity data and provides a measure of fluctuations in the desired phase retrieval solution. As observed in [14,15], the complexity parameter can be used to guide the application of constraints in the object domain and provides artifact free phase retrieval solution which degrades benignly with increasing noise. In [14,15], the CGPR methodology was mainly developed with simulated noisy data in combination with the Fienup's hybrid input-output (HIO) algorithm.…”

“…As observed in [14,15], the complexity parameter can be used to guide the application of constraints in the object domain and provides artifact free phase retrieval solution which degrades benignly with increasing noise. In [14,15], the CGPR methodology was mainly developed with simulated noisy data in combination with the Fienup's hybrid input-output (HIO) algorithm. The complexity-guidance idea is put to test for the first time in this work for experimental data available from the CXIDB database.…”