The combination of Computer-Generated Holography (CGH) and deep learning has opened the possibility to generate both real-time and high-quality holograms. However, the widely-used data-driven deep learning method faces the problem of the large number of labeled training datasets generated by traditional algorithms, such as Gerchberg–Saxton (GS) iterative algorithm. It always takes a long time and limits the training performance of the network. In this work, we propose a model-driven neural network for high-fidelity Phase-Only Hologram (POH) generation. The Fresnel diffraction process is introduced as the physical model, which makes the network can automatically learn the latent encodings of POHs in an unsupervised way. Furthermore, the sub-pixel convolution upsampling method effectively improves the reconstruction quality. Once the training is completed, the POH of any two-dimensional image can be quickly generated. The calculation time is one to two orders of magnitude faster.
Deep learning networks are widely used for optical problems solving in recent years. With the explosive developments of deep learning, learning-based computer-generated holography (CGH) has become an effective way to achieve real-time high-quality holographic display. Various learning-based methods have been proposed to accelerate the computation and improve the reconstruction quality. Deep neural networks (DNNs) have a great influence on the research of holography for their high quality and high computing speed. We focus on the rapid progress on DNN-based CGH in recent years and give our introduction to the principles of CGH as well as the structure of the deep neural networks frequently used in CGH, including U-Net, ResNet and GAN, etc. We introduce the developments of the learning-based CGH and express the expectation of the prospect.
The Gerchberg–Saxton (GS) algorithm is a widely employed algorithm for phase-only hologram (POH) generation. However, the POHs which can strictly satisfy the amplitude constraints on the object and the holographic plane may not exist or be obtained, resulting in speckle noise and reduction of the reconstruction quality. Relaxing the amplitude constraint during the iterations is an effective method to solve the above problem. In this work, a GS-double amplitude freedom (GS-DAF) algorithm is proposed. The amplitude constraint relaxation is realized by both the combined amplitude constraint and the support constraint. The spherical initial phase and oversampling method are applied to further improve the optical reconstruction quality of the GS-DAF algorithm. An enhanced reconstruction quality with less speckle noise has been achieved.
KEYWORDS: 3D displays, Holography, Computer generated holography, Cameras, 3D modeling, 3D acquisition, 3D image processing, Image processing, Sensors, Optical simulations
One of the challenges that has been faced by holographic 3D display is the lack of 3D contents. To deal with the shortage of 3D contents, a 2D-to-3D algorithm is presented for the holographic 3D display. The 2D images are firstly classified into 3 categories by the features. The depth maps for different categories are obtained by different calculation models accordingly. The computer-generated holograms (CGHs) are calculated by the layer-based angular-spectrum approach. These CGHs can be reconstructed with obvious 3D depth cues in simulations and experiments.
An autoencoder neural network is proposed for real-time phase-only CGH generation. As an unsupervised learning method, the input and output of the autoencoder are both the original images, which dispenses with calculating corresponding holograms. It could automatically learn the encoding of phase-only holograms during the training period. Once the training is completed, the phase-only hologram of any two-dimensional image can be quickly generated. The calculation time is 1-2 orders of magnitude faster than the traditional iterative algorithms and the reconstructed image quality is improved.
Compared with other phase-only hologram generation approach, the bidirectional error diffusion algorithm eliminates the need for random phase and iteration, which significantly shortens the calculation time and improves the reconstructed image quality by reducing the influence of speckle noise. The basic concept of error diffusion method is compensating for the effect of removing amplitude information by diffusing the errors to their neighborhood unprocessed pixels. Based on this method, an optimization algorithm for phase-only hologram generation is proposed. The amplitudes of the errors used for diffusion are optimized by introducing a changeable coefficient for each different hologram. The simulation results show that the reconstructed image quality with the proposed optimization algorithm can be increased by up to 4 dB. Smoother amplitude information on the hologram plane can be obtained, thus less information is lost in the phase-only hologram generation process.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.