Converting amplitude holograms into complex and phase-only holograms using deep neural network-based converters

Abstract

Amplitude holograms are computationally efficient but generate unwanted conjugate and direct lights. This paper presents a deep neural network-based converter for amplitude to complex and phase-only hologram conversion. The proposed method uses real-to-real diffraction calculations, which are faster than conventional complex diffraction calculations, to generate 3D layer holograms. The deep neural network then predicts the imaginary hologram from the amplitude hologram. Compared to a conventional method based on the angular spectrum method, the proposed method accelerates the computation of a 3D layer hologram by approximately 1.4 times. The proposed method can accurately predict complex holographic images.