Phase hologram rapid generation method based on conjugate gradient descent

Optimizing the hologram generation process by conjugate gradient descent method, the problem of slow hologram calculation speed in the prior art is solved, and faster generation speed and higher quality image reconstruction are achieved.

CN119006624APending Publication Date: 2024-11-22SICHUAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310570487.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-05-19
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In the existing computing holographic display technology, the rapid generation of high-quality phase holograms still has the problem of slow calculation speed and cannot meet the real-time computing requirements.

Method used

Using a method based on conjugation gradient descent, the random phase is initialized, the diffraction algorithm is used to reconstruct, the loss function and conjugation gradient are calculated, the line search step size is updated, and iterative optimization is performed until the satisfactory reconstruction quality or the number of set times is achieved.

Benefits of technology

Generate better quality images within the same optimization time, keep the peak signal-to-noise ratio high, or have faster generation speeds at the quality that other methods can achieve.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006624A_ABST
    Figure CN119006624A_ABST
Patent Text Reader

Abstract

The invention provides a phase hologram rapid generation method based on conjugate gradient descent. The method comprises a hologram generation part and a hologram reconstruction part. In the generation process of the method, a random phase is initialized, reconstruction is carried out through a diffraction algorithm, loss value calculation of a reconstructed image and a target image is carried out, a first-order gradient is obtained, then a conjugate gradient is calculated, step length selection is carried out through a line search algorithm, a hologram phase value is updated, and the updated hologram is subjected to the operation again. A set quality threshold value or optimization frequency is met, and finally an optimized hologram is obtained; compared with a stochastic gradient descent method and a Wirtinger Hologram method, the hologram obtained through the method has shorter calculation time under the same reconstruction quality, and the speed-up ratio of 1.4 times is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a holographic display technology, in particular to a method for quickly generating a phase hologram based on conjugate gradient descent. Background Art

[0002] As an ideal three-dimensional display method, computer-generated holographic display can provide a realistic sense of stereo without the need for complex optical equipment, and has always received great attention. Although computer-generated holographic display technology has been proposed in the past few decades, high-quality holographic reconstruction images have always relied on carefully designed generation algorithms, and it is still a challenge to quickly obtain accurate phase holograms. Currently, various optimization methods, such as the Gerchberg-Saxton algorithm and the stochastic gradient descent (SGD) method, have been proposed one after another, and the quality of holographic image reconstruction obtained has been significantly improved. However, these methods still have the problem of long computing speed and cannot meet the requirements of real-time computing. Therefore, there is still great research potential in how to continue to improve the computing speed based on the reconstruction quality that can be achieved by the current methods. Summary of the invention

[0003] This method aims to solve the problem that the current hologram calculation speed is not ideal and needs to be further improved. A phase hologram fast generation method based on conjugate gradient descent is proposed. The proposed method regards the hologram generation process as an optimization process and uses conjugate gradient descent to generate the required hologram phase. It is a novel hologram fast generation method. Compared with the current SGD method, Wirtinger Hologram (WH) and other methods can continue to reduce the required generation time on the basis of the existing reconstruction quality.

[0004] The method for rapidly generating a phase hologram based on conjugate gradient descent is as follows: Figure 1 As shown in the code flow chart, the generation process is specifically divided into seven steps: ① Initialize a random phase as the hologram Ho, ② Reconstruct the hologram Ho using the diffraction algorithm to obtain the reproduced light field U, ③ Calculate the loss function of the reproduced light field U and the target image I, ④ According to the loss value, calculate its conjugate gradient p of the hologram phase θ k ⑤ According to the line search method, determine the appropriate step size α k , update the phase of the hologram, ⑥ reconstruct the updated hologram again, repeat ③-⑤ operations to update, ⑦ when the set optimization times are met or the reconstruction quality is satisfactory, the final hologram is obtained. The reconstruction process is divided into two steps: ① loading the hologram, ② receiving the holographic reconstructed image.

[0005] The initializing a random phase as the hologram Ho refers to generating a random phase distribution within the range of 0-2π using a random function as the initial hologram Ho=exp(jθ).

[0006] The reconstructing the hologram Ho by using a diffraction algorithm to obtain a reproduced light field U refers to reconstructing the hologram Ho by using a diffraction algorithm to obtain a reconstructed light field U, wherein the diffraction algorithm adopts an angular spectrum algorithm, which is expressed as P(Ho)=IFFT[FFT(Ho) *T], and T is a transfer function, which is expressed as exp{j2πd / λ*sqrt[1-λ 2 *(fx^2+fy^2)]}, d is the diffraction distance, λ is the wavelength, fx and fy are the frequency domain coordinates, sqrt is the square root, FFT and IFFT are the forward and inverse fast Fourier transform.

[0007] The calculation of the loss function of the reproduced light field U and the target image I refers to bringing the reconstructed complex amplitude image and the target image into the loss function to calculate and obtain the loss value L .

[0008] According to the loss value, the conjugate gradient p of the hologram phase θ is calculated. k It means to bring the calculated loss value L into the conjugate gradient calculation formula to get p k , the conjugate gradient p k The expression is: k ={-▽ θ L , k=1; -▽ θ L +β k P k-1 , k>=2}, k is the number of optimizations, ▽ θ L is the first-order gradient, β k is the conjugate coefficient, expressed as (▽ θk L ) T *(▽ θk L- ▽ θk-1 L ) / ||▽ θk-1 L || 2 , where T is the matrix transpose and ||.|| is the Euclidean distance.

[0009] The method of determining a suitable step length αk according to the line search method and updating the phase of the hologram refers to determining a suitable step length αk according to the calculated conjugate gradient using the line search method in numerical optimization. k , and perform θ k+1 =θ k+α k *p k Phase update.

[0010] The updated hologram is reconstructed again, and operations ③-⑤ are repeated. The updating means that the updated hologram is reconstructed again, and the first three steps are continued to calculate the loss value, conjugate gradient and step size, and the phase is updated to continuously optimize the hologram phase.

[0011] The obtaining of the final hologram when the set number of optimizations is met or a satisfactory reconstruction quality is achieved means that when the number of optimizations meets the set number or the reconstruction quality reaches a set threshold, the optimization process is stopped and the hologram of the previous iteration is retained as the final optimized hologram.

[0012] The loading of the hologram refers to loading the hologram onto the corresponding modulator.

[0013] The received holographic reconstructed image refers to receiving a holographic reconstructed image at a corresponding diffraction distance.

[0014] The beneficial effect of this method is that the proposed method for fast hologram generation is an iterative algorithm based on conjugate gradient descent. Compared with the current SGD method, the Wirtinger Hologram (WH) method, this method can reproduce images with better quality within the same optimization time, and the peak signal-to-noise ratio remains at a high value, or this method has a faster generation speed while maintaining the quality that other methods can achieve. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Attached Figure 1 The following is the code flow chart of this method.

[0017] Attached Figure 2 are the quality comparison results of the fixed optimization times of the present invention, (a) (b) are the average peak signal-to-noise ratio and structural similarity, and (c) (d) are the specific quality indicators of the target image reconstruction in the DIV-2K dataset.

[0018] Attached Figure 3 The time comparison results of the present invention under a fixed number of optimization times are as follows: (a) average running time, and (b) optimization time comparison of a specific image.

[0019] Attached Figure 4 These are the quality results of the present invention under different iteration times.

[0020] Attached Figure 5 This is the comparison result of the present invention at different resolutions.

[0021] Attached Figure 6 These are simulation results of the present invention and other methods.

[0022] Attached Figure 7 This is the optical result of the present invention and other methods. DETAILED DESCRIPTION

[0023] A typical embodiment of a method for fast generation of phase holograms based on conjugate gradient descent of the present invention is described in detail below, and the method is further described in detail. It is necessary to point out here that the following embodiments are only used to further illustrate the method and cannot be understood as limiting the protection scope of the method. Personnel skilled in the art in the field may make some non-essential improvements and adjustments to the method based on the above content of the method, which still fall within the protection scope of the present invention.

[0024] The present invention proposes a method for quickly generating a phase hologram based on conjugate gradient descent. The method comprises two parts: hologram generation and hologram reconstruction.

[0025] The method for quickly generating a phase hologram based on conjugate gradient descent is as follows: Figure 1 As shown in the code flow chart, the generation process is specifically divided into seven steps: ① Initialize a random phase as the hologram Ho, ② Reconstruct the hologram Ho using the diffraction algorithm to obtain the reproduced light field U, ③ Calculate the loss function of the reproduced light field U and the target image I, ④ According to the loss value, calculate its conjugate gradient p of the hologram phase θ k ⑤ According to the line search method, determine the appropriate step size α k , update the phase of the hologram, ⑥ reconstruct the updated hologram again, repeat ③-⑤ operations to update, ⑦ when the set optimization times are met or the reconstruction quality is satisfactory, the final hologram is obtained. The reconstruction process is divided into two steps: ① loading the hologram, ② receiving the holographic reconstructed image.

[0026] The initializing a random phase as the hologram Ho refers to generating a random phase distribution within the range of 0-2π using a random function as the initial hologram Ho=exp(jθ).

[0027] The reconstructing the hologram Ho by using a diffraction algorithm to obtain a reproduced light field U refers to reconstructing the hologram Ho by using a diffraction algorithm to obtain a reconstructed light field U, wherein the diffraction algorithm adopts an angular spectrum algorithm, which is expressed as P(Ho)=IFFT[FFT(Ho) *T], and T is a transfer function, which is expressed as exp{j2πd / λ*sqrt[1-λ 2 *(fx^2+fy^2)]}, d is the diffraction distance, λ is the wavelength, fx and fy are the frequency domain coordinates, sqrt is the square root, FFT and IFFT are the forward and inverse fast Fourier transform.

[0028] The calculation of the loss function of the reproduced light field U and the target image I refers to bringing the reconstructed complex amplitude image and the target image into the loss function to calculate and obtain the loss value L .

[0029] According to the loss value, the conjugate gradient p of the hologram phase θ is calculated. k It means to bring the calculated loss value L into the conjugate gradient calculation formula to get p k , the conjugate gradient p k The expression is: k ={-▽ θ L , k=1; -▽ θ L +β k P k-1 , k>=2}, k is the number of optimizations, ▽ θ L is the first-order gradient, β k is the conjugate coefficient, expressed as (▽ θk L ) T *(▽ θk L- ▽ θk-1 L ) / ||▽ θk-1 L || 2 , where T is the matrix transpose and ||.|| is the Euclidean distance.

[0030] According to the line search method, determine the appropriate step length α k , updating the phase of the hologram means determining the appropriate step size α based on the calculated conjugate gradient using the line search method in numerical optimization k , and perform θ k+1 =θ k +α k *p k Phase update.

[0031] The updated hologram is reconstructed again, and operations ③-⑤ are repeated. The updating means that the updated hologram is reconstructed again, and the first three steps are continued to calculate the loss value, conjugate gradient and step size, and the phase is updated to continuously optimize the hologram phase.

[0032] The obtaining of the final hologram when the set number of optimizations is met or a satisfactory reconstruction quality is achieved means that when the number of optimizations meets the set number or the reconstruction quality reaches a set threshold, the optimization process is stopped and the hologram of the previous iteration is retained as the final optimized hologram.

[0033] The loading of the hologram refers to loading the hologram onto the corresponding modulator.

[0034] The received holographic reconstructed image refers to receiving a holographic reconstructed image at a corresponding diffraction distance.

[0035] In the example of the present invention, the loss function uses the square error, that is, L is ∑(s*|U|-I)^2, s is the energy proportional coefficient, expressed as ∑I / |U|^2, and the corresponding first-order gradient is expressed as ▽ θ L =real[-jexp(-jθ)P -1 (▽ U L )], where real() is the real part, P -1 () is the inverse diffraction algorithm, ▽ U L =2(|U| 2 -I)*2U, the wavelength is 671nm, the hologram resolution is set to 1920x1080, and the area of ​​interest is 1600x800. Figure 1 A pseudo code flow chart of the method is given. Figure 2 The comparison results of the three methods under a fixed number of optimizations are given. From the bar graph, it can be seen that the proposed method can achieve the same reconstruction quality as SGD, which is higher than the WH method. Figure 3 Given their corresponding running time comparisons, the present invention has less computation time. Figure 4 The quality change results of the present invention under different optimization times are given. When a certain number of times is reached, the quality of the holographic image reconstructed by this method will tend to be stable.

[0036] Figure 5 The comparison results of the three methods at different resolutions are given. It can be seen that with the increase of resolution, the calculation time required by other methods will gradually increase, while the optimization quality will gradually decrease. In contrast, the peak signal-to-noise ratio of the image quality reconstructed by the method of the present invention is above 30dB. Figure 6 and Figure 7 The corresponding simulation results and optical results are given. The optimization times of WH, SGD and the present invention are 200, 500 and 30 respectively. From the perspective of peak signal-to-noise ratio and structural similarity, the present method has the highest quality. Compared with the optical results, the present method has good contrast and slightly higher quality indicators.

Claims

1. A method for fast generation of phase hologram based on conjugate gradient descent, characterized in that: The method includes two parts: hologram generation and hologram reconstruction; the hologram generation process is specifically described as follows: step 1, using a random function to generate a random phase distribution in the range of 0-2π as an initial hologram Ho=exp(jθ); step 2, reconstructing the hologram Ho using a diffraction algorithm to obtain a reconstructed light field U, wherein the diffraction algorithm uses an angular spectrum algorithm, expressed as P(Ho)=IFFT[FFT(Ho) *T], T is a transfer function, expressed as exp{j2πd / λ*sqrt[1-λ 2 *(fx^2+fy^2)]}, d is the diffraction distance, λ is the wavelength, fx and fy are frequency domain coordinates, sqrt is the square root, FFT and IFFT are the forward and inverse fast Fourier transforms; Step 3, bring the reconstructed complex amplitude image and the target image into the loss function, calculate, and get the loss value L; Step 4, bring the calculated loss value L into the conjugate gradient calculation formula to get p k , the conjugate gradient p k The expression is: k ={-▽ θ L , k=1; -▽ θ L +β k P k-1 , k>=2}, k is the number of optimizations, ▽ θ L is the first-order gradient, β k is the conjugate coefficient, expressed as (▽ θk L ) T *(▽ θk L- ▽ θk-1 L ) / || ▽ θk-1 L || 2 , where T is the matrix transpose and ||.|| is the Euclidean distance; Step 5, based on the calculated conjugate gradient, use the line search method in numerical optimization to determine the appropriate step size α k , and perform θ k+1 =θ k +α k *p k phase update; step six, reconstructing the updated hologram again, and continuing the first three steps, calculating the loss value, conjugate gradient and step size, updating the phase, and continuously optimizing the hologram phase; step seven, when the number of optimizations meets the set number of times, or the reconstruction quality reaches the set threshold, stop the optimization process, and retain the hologram of the previous iteration as the final optimized hologram; the hologram reconstruction process is specifically described as follows: step one, loading the hologram onto the corresponding modulator; step two, receiving the hologram reconstructed image at the corresponding diffraction distance.

2. The loss function and diffraction algorithm mentioned in claim 1, characterized in that: The loss function may be a square error, a mean square error, a structural similarity, or other loss functions, and the diffraction algorithm may be an angular spectrum algorithm, a Fresnel diffraction algorithm, a spherical algorithm, or other diffraction algorithms.