Light field phase restoration based on the stochastic gradient descent algorithm for expanding the image reconstruction depth range of a computer-generated hologram

In this paper, we propose a novel, to our knowledge, light field phase restoration method. The proposed method consists of obtaining the light field from multiple depth planes constituting a three-dimensional object and applying a stochastic gradient descent algorithm to the depth transformation process between them. The light field can be converted into computer-generated holograms by Fourier transforming each orthographic view image and rearranging them in the angular spectrum domain. In this process, computer-generated holograms converted from a light field multiplied with random phase information can reconstruct an aliasing-free image only in a narrow depth range near the light field acquisition plane. In contrast, the phase information restored by the proposed method overcomes this limitation and reconstructs high-quality images in an extended depth range. The light field with restored phase information reconstructs an image with a peak signal-to-noise ratio value improved by approximately 27.27% at a depth plane 10 cm away from the light field focal plane, compared to the case with random phase information. This paper presents an efficient approach to restore light field phase information using the proposed method and the image reconstruction results obtained through simulation and experiment.