Clustering fast optimization strategy for holographic video displays

Computer-generated holography (CGH), as a cutting-edge technology in the field of three-dimensional displays, significantly enhances visual realism and viewing comfort due to its unique parallax and depth perception capabilities. This advantage makes it a revolutionary solution for applications such as virtual reality (VR/AR) headsets and automotive head-up displays (HUD). Spatial light modulators also play a crucial role in this technology, enabling precise and flexible control through functions such as real-time display, dynamic updates, wavefront modulation, and three-dimensional information reconstruction, thereby meeting diverse application requirements.

Article Information:

Computer-generated holography (CGH) is an advanced three-dimensional (3D) display technology. Although the stochastic gradient descent (SGD) method is effective for holographic optimization, its application to holographic video displays is computationally expensive, as each frame requires separate optimization. To address this issue, this paper proposes a novel clustering optimization strategy to accelerate the SGD process for holographic video displays. The method leverages the inherent similarities between video frames by first jointly optimizing shared features, followed by optimizing unique features specific to individual frames, thereby effectively reducing redundant computations. The process involves clustering video frames based on common features and using the optimized hologram of the cluster center image, combined with global scale factors, as initial conditions for each frame within the cluster. Numerical simulations and optical experiments demonstrate that this method improves computational efficiency by approximately twofold, significantly enhancing the feasibility of holographic video display technology for broader applications.

The following outlines part of the experimental process and results:

This paper validates the computational efficiency of the proposed clustering optimization strategy for hologram generation through numerical simulations and optical experiments. All computer-generated holography (CGH) algorithms were implemented using Python 3.7.1 and PyTorch 1.10.2, running on a system equipped with a 13th-generation Intel Core i7-13700KF processor and an NVIDIA GeForce RTX 4070 graphics card (supporting CUDA 11.6). The optical experimental setup, as shown in Figure 2, utilized an FSLM-2K70-P02 spatial light modulator (SLM) with a resolution of 1920×1080 pixels and a pixel pitch of 8 µm to load the holograms. A 532 nm solid-state laser was used as the light source. The distance between the SLM plane and the target plane was set to 250 mm. The final images were captured using a Nikon D810 DSLR camera equipped with a 4f system to eliminate higher-order diffraction and zero-order noise. For comparative analysis, a 700-frame video clip of Nobita Rabbit was resized to a resolution of 1600×880 pixels as the target. The phase variables for all algorithms were set to 10 cluster center images (n), with a unified learning rate of 0.02. Reconstruction quality was evaluated using peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM).

Figure 1: Experimental Setup

Figure 2: Schematic of the Clustering Optimization Strategy. This method groups video frames by extracting their common features, then uses the optimized hologram of each cluster's center image and the optimized s-factor as the initial conditions for the frames within that cluster.

Figure 3: Simulation results for the first frame after 0, 15, 50, and 100 iterations. The first row shows the optimization process without the clustering optimization strategy, starting from a randomized phase with the factor s set to 1.0. The third row shows the optimization process using the clustering optimization strategy, which begins with the optimized hologram and the factor s of the cluster center image.

Figure 4: Simulation and optical results for frames 140, 280, 420, 560, and 700, each iterated 100 times.

Figure 5: Total number of iterations across 700 frames.

The parameter specifications of the phase-type spatial light modulator used in this experiment are as follows:

Note: Different models vary in the range of phase modulation and light utilization efficiency. For specific requirements, please contact the sales manager in your region for details.

Additionally, our company offers a version with the same model featuring high reflectivity and high optical utilization efficiency. The specific parameters are as follows:

This is a paragraph

Written at the End

The spatial light modulator, based on computer control, can precisely achieve light modulation effects by simply inputting image data. In dynamic three-dimensional display applications, it can quickly adjust holograms to present continuously changing 3D images based on different scene requirements, without the need for complex physical adjustments or replacements as with traditional optical components. Additionally, it is easy to operate, highly reproducible, and capable of maintaining stable light modulation performance over extended periods.

Article Information:
https://doi.org/10.1364/OL.542604