Using machine learning for pixel apodization to suppress image duplicates in Fourier holography

Liquid-crystal-on-silicon (LCoS) spatial light modulators (SLMs) are common devices in computer-generated holography. However, because of their periodic pixel structure, reconstructing Fourier holograms using LCoS SLMs leads to the presence of unwanted image replicas. Previous research showed that the appearance of such image duplicates can be countered by applying apodization, either in amplitude or phase, to each pixel of the modulator. Here, we employ machine learning tools for generation of complex apodizing patterns, optimised for desired areas of high intensity in the reconstructed light field. A set of varying target areas is considered, and both phase and amplitude masks are created for each such case. We compare the impact of amplitude, phase, and combined masks on the resulting intensity distribution at the imaging plane. We show the successful removal of image duplicates in hologram reconstructions.