Reconstruction of off-axis electron holograms using a neural net

In off-axis electron holography, the information about the specimen contained in amplitude and phase of the image wave is recorded by superimposing a reference wave to the image wave and recording the resulting interference pattern, called a hologram, The image wave is usually retrieved from this hologram by a filtering process in Fourier space. An alternative approach is to calculate the image wave pixel by pixel from the intensities in the hologram. In this paper, we describe a neural network that performs this step, As neural networks can represent nonlinear relations between the input and output data, this reconstruction process can use the nonlinear part of the information contained in the hologram and is thus less sensitive to noise.