角江 崇

Single-pixel imaging allows for high-speed imaging, miniaturization of optical systems, and imaging over a broad wavelength range, which is difficult by conventional imaging sensors, such as pixel arrays. However, a challenge in single-pixel imaging is low image quality in the presence of undersampling. Deep learning is an effective method for solving this challenge; however, a large amount of memory is required for the internal parameters. In this study, we propose single-pixel imaging based on a recurrent neural network. The proposed approach succeeds in reducing the internal parameters, reconstructing images with higher quality, and showing robustness to noise.

近年，深層学習についての研究が盛んに行われている．その中でも，再帰型ニューラルネットワークはネットワーク内に再帰構造をもつ深層学習の一種で，時系列データの扱いに優れていることから音声認識や自然言語処理の分野で高い成績を残している．再帰型ニューラルネットワークに限らず深層学習では，高い性能の反面，計算コストが多い．そのため，クラウドコンピューティングやGraphics Processing Unitを用いて計算する研究が進められている．しかし近年では，応答性の観点から，この計算をクラウドコンピューティングのように別の場所に送信し計算するのではなく，エッジ側での処理が求められるようになってきている．そこで本論文では，回路を自由に書き換えることができるField Programmable Gate ArrayにRNNの推論器と学習器を実装し，回路の評価と，実際に音声識別を行った結果を評価した．

Digital holography is a promising display technology that can account for all human visual cues, with many potential applications i.a. in AR and VR. However, one of the main challenges in computer generated holography (CGH) needed for driving these displays are the high computational requirements. In this work, we propose a new CGH technique for the efficient analytical computation of lines and arc primitives. We express the solutions analytically by means of incomplete cylindrical functions, and devise an efficiently computable approximation suitable for massively parallel computing architectures. We implement the algorithm on a GPU (with CUDA), provide an error analysis and report real-time frame rates for CGH of complex 3D scenes of line-drawn objects, and validate the algorithm in an optical setup.

We developed dedicated processors for holography named 'HORN,” which accelerate hologram calculation by parallelizing many calculation units. If we reduce the calculation accuracy of the units, the circuit size of the units decreases, which in turn leads to an increase in the number of the units; however, owing to this the image quality degrades. In this study, we propose a dedicated processor with low accuracy units assisted by a deep neural network (DNN). After performing the hologram calculation in the processor, we restore a high accuracy hologram from the low accuracy one using the DNN.

Recently, a calculation method involving sparse point spread functions in the short-time Fourier transform (STFT) domain was proposed. In this paper, a dedicated processor using the STFT algorithm is described, which is implemented on a field-programmable gate array. All the operations in this algorithm are implemented using fixed-point arithmetic. Since this algorithm includes a trigonometric function and an error function, lookup tables (LUTs) are utilized to reduce the calculation costs. We have devised a dedicated circuit architecture that allows parallel operations. In addition, a central processing unit could generate holograms using the STFT-based algorithm with fixed-point arithmetic and LUTs at a higher speed than the generation using floating-point arithmetic.

Recently, a calculation method involving sparse point spread functions in the short-time Fourier transform(STFT) domain was proposed. In this paper, a dedicated processor using the STFT algorithm is described, which is implemented on a field-programmable gate array. All the operations in this algorithm are implemented using fixed-point arithmetic. Since this algorithm includes a trigonometric function and an error function, lookup tables (LUTs) are utilized to reduce the calculation costs. We have devised a dedicated circuit architecture that allows parallel operations. In addition, a central processing unit could generate holograms using the STFT-based algorithm with fixed-point arithmetic and LUTs at a higher speed than the generation using floating-point arithmetic. (C) 2020 Optical Society of America

This study aims to improve the image quality of holographic three-dimensional (3-D) images based on the wavefront recording plane (WRP) method. In this method, we place a WRP close to the 3-D objects to reduce the propagation distance of light from the objects to the WRP. The conventional WRP method has been implemented only under conditions that did not cause aliasing noise. This study proposes a WRP method with a limiting diffraction region from the WRP to the hologram such that we can perform the WRP method under any condition. As a result, we succeeded in improving the image quality of the 3-D images based on the WRP method.

This study aims to improve the image quality of holographic three-dimensional (3-D) images based on the wavefront recording plane (WRP) method. In this method, we place a WRP close to the 3-D objects to reduce the propagation distance of light from the objects to the WRP. The conventional WRP method has been implemented only under conditions that did not cause aliasing noise. This study proposes a WRP method with a limiting diffraction region from the WRP to the hologram such that we can perform the WRP method under any condition. As a result, we succeeded in improving the image quality of the 3-D images based on the WRP method. (C) 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement

Digital holography is attracting attention because it can record instantaneous three-dimensional (3D) information and record dynamic phenomena. However, when recording highspeed phenomena, the frame rate ranges from tens of thousands to hundreds of millions, and the calculation time of the reconstructed images is a problem. We have developed a special-purpose computer for high-speed 3D imaging using digital holography. The developed special-purpose computer has four calculation modules and has achieved a calculation time 68 times faster than that of a personal computer with 48 cores. With the developed computer, a total of 32 reconstructed images can be calculated in 0.69 ms from four holograms of 128 x 128 pixels with eight varying depths. (C) The Authors. Published by SPIE under a Creative Commons Attribution 4.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI.

In this study, we compare the ray-tracing method with the look-up table (LUT) method in order to optimize computer-generated hologram (CGH) calculation based on the wavefront recording plane (WRP) method. The speed of the WRP-based CGH calculation largely depends on implementation factors, such as calculation methods, hardware, and parallelization method. Therefore, we evaluated the calculation time and image quality of the reconstructed three-dimensional (3D) image by using the ray-tracing and LUT methods in the central processing unit (CPU) and graphics processing unit (GPU) implementations. Thereafter, we performed several implementations by changing the number of object points and the distance from 3D objects to the WRP. Furthermore, we confirmed different characteristics between CPU and GPU implementations.

In this study, we compare the ray-tracing method with the look-up table (LUT) method in order to optimize computer-generated hologram (CGH) calculation based on the wavefront recording plane (WRP) method. The speed of the WRP-based CGH calculation largely depends on implementation factors, such as calculation methods, hardware, and parallelization method. Therefore, we evaluated the calculation time and image quality of the reconstructed three-dimensional (3D) image by using the ray-tracing and LUT methods in the central processing unit (CPU) and graphics processing unit (GPU) implementations. Thereafter, we performed several implementations by changing the number of object points and the distance from 3D objects to the WRP. Furthermore, we confirmed different characteristics between CPU and GPU implementations. (C) 2020 Optical Society of America

© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement Here, we managed to reconstruct a three-dimensional color video of a point-cloud object using a projection-type holographic display with a holographic optical element as an optical screen. The holographic optical element has the function of an off-axis concave mirror and has been created by the wavefront printer digitally. We defined and implemented an algorithm to reconstruct a three-dimensional image at a chosen position considering the specification of the holographic optical element designed digitally. We successfully demonstrated a reconstruction of the color video in question, composed of three-dimensional images through the holographic optical element.

A complex amplitude hologram can reconstruct perfect light waves. However, as there are no spatial light modulators that are able to display complex amplitudes, we need to use amplitude, binary, or phase-only holograms. The images reconstructed from such holograms will deteriorate; to address this problem, iterative hologram optimization algorithms have been proposed. One of the iterative algorithms utilizes a blank area to help converge the optimization; however, the calculation time and memory usage involved increases. In this study, we propose to reduce the computational complexity and memory usage of the iterative optimization using scaled diffraction, which can calculate light propagation with different sampling pitches on a hologram plane and object plane. Scaled diffraction can introduce a virtual blank area without using physical memory. We further propose a combination of scaled diffraction-based optimization and conventional methods. The combination algorithm improves the quality of a reconstructed complex amplitude while accelerating optimization.

© 2020 Optical Society of America We propose a phase retrieval method using axial diffraction patterns under planar and spherical wave illuminations. The proposed method uses a ptychographic iterative engine (PIE) for the phase retrieval algorithm. The proposed approach uses multiple diffraction patterns. Thus, adjusting the alignment of each diffraction pattern is mandatory, and we propose a method to adjust the alignment. In addition, a random selection of the measured diffraction patterns is used to further accelerate the convergence of the PIE-based optimization. To confirm the effectiveness of the proposed method, we compare the conventional and proposed methods using a simulation and optical experiments.

Creating computer-generated holograms (CGHs) is computationally costly, and many high-speed calculation algorithms have been proposed to address this problem. Recently, a calculation method using sparse point spread functions (PSFs) in the short-time Fourier transform (STFT) domain has been proposed. Since the PSFs are sparse in the STFT domain, only a small fraction of STFT coefficients need to be calculated. Further, in order to obtain the STFT coefficient, fast Fourier transform is generally necessary, but this new method can obtain the STFT coefficient by analytical means. This means that the number of calculations required can be greatly reduced. This paper describes the implementation of an STFT-based CGH calculation algorithm on a field-programmable gate array. All operations in this algorithm were implemented using fixed point arithmetic. Since this algorithm includes a trigonometric function and an error function, we used look-up tables (LUTs) to reduce calculation costs. We have devised a dedicated circuit architecture that allows parallel operations. As a secondary effect, a central processing unit was able to generate holograms using the STFT-based CGH calculation algorithm with fixed point arithmetic and LUTs, faster than by using floating point arithmetic.

One hologram calculation method is the random phase-free method. When this method is used for an amplitude hologram, a reproduced image with a high-quality image can be obtained. However, when the random phase-free method is used for the kinoform, the reproduced image is degraded. In this study, we applied the kinoform encoding method proposed by Li to the random phase-free method to improve the reproduced image. The effectiveness of the proposed method was compared with that of the conventional method via simulations and optical experiments. Additionally, the parameters were optimized by the simulations, and the effectiveness was verified by numerical experiments.

<p>ホログラフィは，光の波面を記録・再生する技術である．コンピュータ上での数値計算によって生成されたホログラムを計算機合成ホログラムという．ホログラムの計算手法の一つにランダム位相フリー法があり，これを振幅ホログラムに用いた場合，画質の良い再生像を得ることができる．しかし，位相型ホログラムにランダム位相フリー法を用いた場合，再生像が劣化するという問題があった．本研究ではランダム位相フリー法に，Liらによる位相型ホログラムのエンコード法を適用し，再生像の改善を図る．シミュレーションおよび光学実験により提案手法と従来手法の再生像画質の比較を行い，有効性を確認した．</p>

We demonstrate real-time three-dimensional (3D) color video using a color electroholographic system with a cluster of multiple-graphics processing units (multi-GPU) and three spatial light modulators (SLMs) corresponding respectively to red, green, and blue (RGB)-colored reconstructing lights. The multi-GPU cluster has a computer-generated hologram (CGH) display node containing a GPU, for displaying calculated CGHs on SLMs, and four CGH calculation nodes using 12 GPUs. The GPUs in the CGH calculation node generate CGHs corresponding to RGB reconstructing lights in a 3D color video using pipeline processing. Real-time color electroholography was realized for a 3D color object comprising approximately 21,000 points per color.

We propose a numerical simulation model to calculate a hologram of light-in-flight recording by holography. The proposed model is based on not only ray tracing but also finite-difference time-domain method. We succeeded in numerically reconstructing light pulse propagation with total reflection from the hologram calculated by the proposed model.

We evaluated calculation times of computer-generated holograms based on wavefront recording plane method using several implementations in the combination of look-up table method and direct calculation method in order to realize real-time electro-holography system. We confirmed that there are different characteristics between CPU and GPU implementations.

To realize interactive operation of 3D image projected on HOE screen, we calculated and displayed the holograms from the data of light-ray information which was loaded depending on the position of the finger detected by the motion sensor.

In this paper, we constructed a time-division reproduction system of holographic projection using a DMD (Digital Mirror Device). We succeeded in removing the direct light in projected images and enlarging the projected images by changing a sampling pitch of the original image.

We demonstrated real-time electroholographic 3-D movie reconstruction using spatiotemporal division multiplexing technique on a multiple GPU cluster system including 13 GPUs connected through gigabit ethernet network. We succeeded to display reconstructed 3-D movie consisting of 912,462 object points.

For realizing electroholography, a compact and high-performance computer is required. In this study, we implemented highly parallel special-purpose computer for electroholography on system on a chip. As a result, we succeeded in speeding up calculation 200 times faster than a CPU and a GPU.

To reconstruct a desired three-dimensional (3-D) image in the projection-type color holographic display with the holographic optical element (HOE), we shifted coordinates of the point-cloud object theoretically. As a result, we experimentally succeeded in reconstructing the color 3-D image by the optical system including the HOE screen.

We proposed cost-effective portable holographic projector composed of a portable digital micromirror device board and a single board computer. Consequently, the proposed projector succeeded to project the reconstructed video at 60 fps.

We propose a calculation reduction method for computational holograms using angular redundancy of light field by color space conversion. The angular redundancy could be enhanced by the properties of color space. We confirmed that the computational complexity can be reduced by about 20%.