Abstract
Since Lohman et al. invented computer-generated holography, several methods have been tried to improve its image quality. We describe a method based on a feedback neural network that is designed to find the optimum (minimum-error) hologram by minimizing its error function. In this network, neurons are allowed to take only two output states, which directly represent the binary states of the corresponding pixels on the hologram. Thus, the number of neurons in the network equals the number of pixels in the hologram. The threshold value for each neuron and the connection strength between the neurons are determined by the Fourier transform of the desired wave front and the window function, respectively. The window function specifies the area of interest on the reconstruction plane, inside of which the image quality is to be evaluated. We have calculated several holograms by this network; starting from a random initial state and a 512 × 512 pixel hologram, the neural network typically converged after 2 hours of calculation (on a Sun SPARC station). The resultant holograms were recorded on an optical disk. The optically reconstructed images have excellent signal-to-noise ratios inside the specified reconstruction area.
© 1990 Optical Society of America
PDF ArticleMore Like This
O. K. Ersoy and C. E. Bubb
MFF5 OSA Annual Meeting (FIO) 1990
Zhelun Shen, Guanglin Yang, and Haiyan Xie
W5A.30 Digital Holography and Three-Dimensional Imaging (DH) 2022
M. Hossein Eybposh, Changjia Cai, Diptodip Deb, Miguel A. B. Schott, Longtian Ye, Gert-Jan Both, Srinivas C. Turaga, Jose Rodriguez-Romaguera, and Nicolas C. Pégard
DW5A.4 3D Image Acquisition and Display: Technology, Perception and Applications (3D) 2023