Complex dynamic scene total variation regularized GS phase retrieval enhanced ghost imaging method
By using two-dimensional Walsh ghost imaging reconstruction and the fully variational regularized GS phase retrieval method, the problem of phase information acquisition in complex scenes of ghost imaging technology is solved, and high-fidelity complex amplitude imaging is achieved, which is suitable for quantitative imaging in complex dynamic scenes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JILIN UNIVERSITY
- Filing Date
- 2026-05-14
- Publication Date
- 2026-06-12
AI Technical Summary
Existing ghost imaging techniques struggle to acquire complex amplitude information, especially phase information, with high fidelity in complex scenes. Traditional methods are sensitive to noise and lack robustness, while deep learning methods are deficient in terms of physical interpretability and data generalization ability.
A two-dimensional Walsh ghost imaging reconstruction combined with a fully variational regularized GS phase retrieval method is adopted. Through iterative algorithms and Fourier transforms, combined with fully variational regularized denoising, high-fidelity amplitude and phase information is obtained, and the light field distribution is reconstructed using angular spectral diffraction theory.
It significantly improves the accuracy and stability of phase recovery, and can acquire high-fidelity amplitude and phase information under low sampling rate and strong noise conditions. It is suitable for quantitative imaging scenarios such as observation of transparent biological samples and three-dimensional morphology measurement. It has high computational efficiency and is easy to deploy in engineering.
Smart Images
Figure CN122199338A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of image processing technology, and in particular relates to a fully variational regularized GS phase retrieval enhancement ghost imaging method for complex dynamic scenes. Background Technology
[0002] Ghost imaging, also known as correlation imaging or quantum imaging, is a novel imaging technique based on higher-order correlations in the light field rather than direct detection. In 1995, Shi Yanhua and others experimentally verified the quantum concept of "achieving non-local imaging using entangled photons" for the first time using entangled photon pairs generated by spontaneous parametric downconversion, marking the birth of ghost imaging. This technique achieves "object-image separation," overturning the traditional understanding of imaging.
[0003] In recent years, ghost imaging technology has been deeply integrated with computational imaging, information theory, and compressed sensing theory, giving rise to the "computational ghost imaging" paradigm. By replacing the reference optical path with a known, programmable spatial modulation mode and combining sparse sampling with optimized reconstruction algorithms, the system can recover images with high quality at extremely low sampling rates. Furthermore, the introduction of methods such as deep learning has further enhanced its capabilities in noise suppression and super-resolution reconstruction.
[0004] However, despite the progress made by ghost imaging in intensity imaging in complex scenes, the acquisition of complex amplitude information, especially high-fidelity position data, remains a core bottleneck restricting its quantitative and precise applications. Existing phase retrieval methods face severe challenges in complex scenes: traditional intensity transmission equations rely on paraxial and slowly varying conditions and are sensitive to noise; iterative phase retrieval algorithms are prone to artifacts under low signal-to-noise ratio conditions; and techniques such as ghost holography face the problems of system complexity and insufficient robustness. Deep learning and other methods still have concerns regarding physical interpretability and data generalization ability. Summary of the Invention
[0005] The purpose of this invention is to provide a fully variational regularized GS phase retrieval enhanced ghost imaging method for complex dynamic scenes, aiming to solve the problems mentioned in the background art.
[0006] The embodiments of the present invention are implemented as follows: a fully variational regularized GS phase retrieval enhanced ghost imaging method for complex dynamic scenes includes the following steps:
[0007] Step 1: Two-dimensional Walsh ghost imaging reconstruction to obtain the amplitude information of the ghost imaging reconstruction image of the target object;
[0008] Step 2: GS phase retrieval based on total variation regularization, using ghost imaging to reconstruct the amplitude information of the image, and obtaining the phase information through an iterative algorithm;
[0009] Step 3: Complex amplitude reconstruction. Using the amplitude information obtained in Step 1 and the phase information obtained in Step 2, a complex amplitude is constructed, and the light field distribution after propagation is reconstructed based on the angular spectrum diffraction theory to output the final image.
[0010] In a further technical solution, step 1 includes the following specific steps:
[0011] A preset random speckle pattern sequence is loaded, a structured illumination field is generated by a digital micromirror device, the modulated light field illuminates the target object, and the total light intensity signal is recorded by a barrel detector to obtain a barrel detection value sequence.
[0012] The ghost image reconstruction image is obtained by calculating the second-order correlation function, and the initial correlation image is obtained. Given by the following formula:
[0013]
[0014] In the formula, It represents the number of measurements in each iteration; It is the first The barrel detection value measured in this step; It is the first The second measurement is at the pixel location. The intensity value at that location; and These are pixel coordinates;
[0015] Will The initial representation is a related image, denoted as . Then normalize it:
[0016]
[0017] In the formula, The image is after normalization; For the initial associated image; For associated images The minimum value of all pixel values; For associated images The maximum value of all pixel values;
[0018] After normalization, The image is transformed to the [0,1] interval; finally, the expression for reconstructing the image is:
[0019]
[0020] In the formula, To reconstruct the image.
[0021] A further technical solution is that, in step 1, the number of measurements in each iteration... Take 880.
[0022] In a further technical solution, in step 2, the complex numerical expression of the reconstructed image is:
[0023]
[0024] In the formula, To reconstruct the complex values of the image; Image amplitude, For the initial phase, It is a natural constant. It is the imaginary unit.
[0025] In a further technical solution, step 2 of the iterative algorithm includes the following specific steps:
[0026] Iteration using the GS algorithm In each iteration, the Fourier transform is first applied to obtain the frequency domain expression:
[0027]
[0028] In the formula, This represents the image in the frequency domain. for The image after the next iteration For Fourier transform operators;
[0029] By replacing the frequency domain amplitude, we obtain the first... The mathematical expression for the frequency domain complex amplitude distribution of the next iteration is:
[0030]
[0031] In the formula, The updated frequency domain complex amplitude; for The model; Indicates taking The phase angle;
[0032] The inverse Fourier transform back to the spatial domain gives its spatial domain expression as follows:
[0033]
[0034] In the formula, This is the updated spatial domain representation of the frequency domain complex amplitude; This is the inverse operation of the Fourier transform;
[0035] Extract its spatial domain phase; the spatial domain phase expression is as follows:
[0036]
[0037] In the formula, Spatial domain representation of phase; Indicates taking The phase;
[0038] By constraining the spatial domain amplitude, we obtain the image, whose expression is:
[0039]
[0040] In the formula, The image obtained after amplitude constraint;
[0041] Finally, total variational regularization is applied for noise reduction, resulting in the image expression:
[0042]
[0043] In the formula, for The image after the next iteration For total variation regularization strength, For total variation regularization terms The gradient.
[0044] In a further technical solution, step 3 includes the following specific steps:
[0045] Will Image after the second iteration As the input image, let it be... ;
[0046] Performing a Fourier transform on the input image yields its frequency domain expression:
[0047]
[0048] In the formula, for The frequency domain expression; Input image;
[0049] The expression for the angular spectrum propagation operator is:
[0050]
[0051] In the formula, For angular spectrum propagation operators; and Frequency domain coordinates; The wavelength of light; For the distance of propagation;
[0052] The final image expression is:
[0053]
[0054] In the formula, This is the final output of the reconstructed complex value.
[0055] A further technical solution is that the method is implemented through the following experimental system: including a computer, and a laser emitter, a modulation device, a front lens, a target object, a rear lens, and a barrel detector arranged in sequence;
[0056] The computer is connected to both the modulation device and the barrel detector.
[0057] The fully variational regularized GS phase retrieval enhanced ghost imaging method for complex dynamic scenes provided in this invention has the following beneficial effects:
[0058] (1) The total variation regularization is introduced into the GS phase recovery framework. A TV denoising step is added after each spatial domain update, which effectively suppresses noise and phase artifacts. This overcomes the problems of traditional GS algorithms being prone to getting trapped in local extrema, slow convergence, and sensitivity to noise, and significantly improves the accuracy and stability of phase recovery.
[0059] (2) It extends the ghost imaging system from intensity imaging to complex amplitude imaging, enabling it to acquire high-fidelity amplitude and phase information simultaneously. Combined with angular spectrum diffraction theory, it realizes arbitrary propagation and reconstruction of the light field, which is suitable for quantitative imaging scenarios such as transparent biological sample observation, wavefront sensing, and three-dimensional morphology measurement.
[0060] (3) It has good robustness to low sampling rate and strong noise. The algorithm only involves Fourier transform and gradient operation, which has high computational efficiency, does not require training data, and is easy to deploy in complex dynamic scenarios. Attached Figure Description
[0061] Figure 1 A flowchart of the fully variational regularized GS phase retrieval enhanced ghost imaging method for complex dynamic scenes provided in an embodiment of the present invention;
[0062] Figure 2 This is a diagram of the experimental system architecture;
[0063] Figure 3 25 frames of raw ghost imaging images;
[0064] Figure 4 The image optimized for phase recovery after 25 frames;
[0065] Figure 5 Comparison of BRISQUE index before and after phase recovery;
[0066] Figure 6 Comparison of NRSS index before and after phase recovery.
[0067] In the attached diagram, 1 is a laser emitter; 2 is a modulation device; 3 is a front lens; 4 is a target object; 5 is a rear lens; 6 is a barrel detector; and 7 is a computer. Detailed Implementation
[0068] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0069] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.
[0070] like Figure 1 As shown, an embodiment of the present invention provides a method for enhancing ghost imaging in complex dynamic scenes using total variation regularized GS phase retrieval, comprising the following steps:
[0071] Step 1: Two-dimensional Walsh ghost imaging reconstruction;
[0072] Experimental system architecture as follows Figure 2 As shown, it includes a computer 7, and a laser emitter 1, a modulator 2, a front lens 3, a target object 4, a rear lens 5, and a barrel detector 6 arranged sequentially. The computer 7 loads a preset random speckle pattern sequence and generates a structured illumination field through a digital micromirror device (DMD), i.e., the modulator 2. The modulated light field is focused by the front lens 3 to illuminate the target object 4. The light wave carrying the object's transmission information is collected by the rear lens 5, and the total light intensity signal is recorded by the barrel detector 6. The computer 7 achieves precise synchronization between DMD pattern switching and barrel detector 6 sampling, acquiring a barrel detector value sequence of N=880 frames. The ghost imaging reconstructed image is obtained by calculating the second-order correlation function.
[0073] Initial related image Given by the following formula:
[0074]
[0075] In the formula, It represents the number of measurements in each iteration; It is the first The barrel detection value measured in this step; It is the first The second measurement is at the pixel location. The intensity value at that location; and These are pixel coordinates.
[0076] Will The initial representation is a correlation image, denoted as... Then normalize it:
[0077]
[0078] In the formula, The image is after normalization; For the initial associated image; For associated images The minimum value of all pixel values; For associated images The maximum value of all pixel values.
[0079] After normalization, The image is transformed to the [0,1] interval to ensure consistent contrast. Finally, the reconstructed image is expressed as:
[0080]
[0081] In the formula, To reconstruct the image.
[0082] Step 2: GS phase recovery based on total variational regularization;
[0083] The amplitude information of the image is reconstructed using ghost imaging, and the phase information is obtained through total variation regularized GS phase recovery;
[0084] The complex numerical expression for the reconstructed image is:
[0085]
[0086] In the formula, To reconstruct the complex values of the image; Image amplitude, For the initial phase, It is a natural constant. It is the imaginary unit.
[0087] Iteration using the GS algorithm In each iteration, the Fourier transform is first applied to obtain the frequency domain expression:
[0088]
[0089] In the formula, This represents the image in the frequency domain. for The image after the next iteration This is the Fourier transform operator.
[0090] By replacing the frequency domain amplitude, we obtain the first... The mathematical expression for the frequency domain complex amplitude distribution of the next iteration is:
[0091]
[0092] In the formula, The updated frequency domain complex amplitude; for The model; Indicates taking The phase angle.
[0093] The inverse Fourier transform back to the spatial domain gives its spatial domain expression as follows:
[0094]
[0095] In the formula, This is the updated spatial domain representation of the frequency domain complex amplitude; This is the inverse operation of the Fourier transform.
[0096] Extract its spatial domain phase; the spatial domain phase expression is as follows:
[0097]
[0098] In the formula, Spatial domain representation of phase; Indicates taking The phase.
[0099] By constraining the spatial domain amplitude, we obtain the image, whose expression is:
[0100]
[0101] In the formula, This is the image obtained after amplitude constraint.
[0102] Finally, total variational regularization is applied for noise reduction, resulting in the image expression:
[0103]
[0104] In the formula, for The image after the next iteration For total variation regularization strength, For total variation regularization terms The gradient.
[0105] Step 3: Complex amplitude reconstruction;
[0106] The complex amplitude is constructed using the amplitude information from step 1 and the phase information from step 2. The propagated light field distribution is reconstructed based on the angular spectrum diffraction theory, and the final image is output.
[0107] Will Image after the second iteration As the input image, let it be... .
[0108] Performing a Fourier transform on the input image yields its frequency domain expression:
[0109]
[0110] In the formula, for The frequency domain expression; The input image.
[0111] The expression for the angular spectrum propagation operator is:
[0112]
[0113] In the formula, For angular spectrum propagation operators; and Frequency domain coordinates; The wavelength of light; For the distance of propagation.
[0114] The final image expression is:
[0115]
[0116] In the formula, This is the final output of the reconstructed complex value.
[0117] To verify this method, relevant experiments were conducted, enhancing 25 consecutively changing images to simulate image quality enhancement under complex dynamic scenes and complex optical conditions.
[0118] Figure 3 This is a collection of 25 raw ghost images. Figure 4 The image optimized for phase recovery from 25 frames. Figure 5 For the comparison of BRISQUE index before and after phase recovery, Figure 6 A comparison of NRSS indices before and after phase recovery. Based on... Figure 5 Comparison results of BRISQUE metrics and Figure 6 The comparison results of the NRSS index show that the image quality is effectively improved through optimization using this method.
[0119] comprehensive Figure 3 and Figure 4 Visual effects Figure 5 and Figure 6 Based on objective metrics, this method successfully implements a fully variational regularized GS phase retrieval enhanced ghost imaging method applicable to complex dynamic scenes.
[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for enhancing ghost imaging with total variational regularized GS phase retrieval in complex dynamic scenes, characterized in that, Includes the following steps: Step 1: Two-dimensional Walsh ghost imaging reconstruction to obtain the amplitude information of the ghost imaging reconstruction image of the target object; Step 2: GS phase retrieval based on total variation regularization, using ghost imaging to reconstruct the amplitude information of the image, and obtaining the phase information through an iterative algorithm; Step 3: Complex amplitude reconstruction. Using the amplitude information obtained in Step 1 and the phase information obtained in Step 2, a complex amplitude is constructed, and the light field distribution after propagation is reconstructed based on the angular spectrum diffraction theory to output the final image.
2. The method for enhancing ghost imaging in complex dynamic scenes using total variational regularized GS phase retrieval as described in claim 1, characterized in that, Step 1 includes the following specific steps: A preset random speckle pattern sequence is loaded, a structured illumination field is generated by a digital micromirror device, the modulated light field illuminates the target object, and the total light intensity signal is recorded by a barrel detector to obtain a barrel detection value sequence. The ghost image reconstruction image is obtained by calculating the second-order correlation function, and the initial correlation image is obtained. Given by the following formula: In the formula, It represents the number of measurements in each iteration; It is the first The barrel detection value measured in this step; It is the first The second measurement is at the pixel location. The intensity value at that location; and These are pixel coordinates; Will The initial representation is a related image, denoted as . Then normalize it: In the formula, The image is after normalization; For the initial associated image; For associated images The minimum value of all pixel values; For associated images The maximum value of all pixel values; After normalization, The image is transformed to the [0,1] interval; finally, the expression for reconstructing the image is: In the formula, To reconstruct the image.
3. The method for enhancing ghost imaging in complex dynamic scenes using total variational regularized GS phase retrieval as described in claim 2, characterized in that, In step 1, the number of measurements in each iteration Take 880.
4. The method for enhancing ghost imaging in complex dynamic scenes using total variation regularized GS phase retrieval as described in claim 2, characterized in that, In step 2, the complex numerical expression of the reconstructed image is: In the formula, To reconstruct the complex values of the image; Image amplitude, For the initial phase, It is a natural constant. It is the imaginary unit.
5. The method for enhancing ghost imaging in complex dynamic scenes using total variational regularized GS phase retrieval as described in claim 4, characterized in that, In step 2, the iterative algorithm includes the following specific steps: Iteration using the GS algorithm In each iteration, the Fourier transform is first applied to obtain the frequency domain expression: In the formula, This represents the image in the frequency domain. for The image after the next iteration For Fourier transform operators; By replacing the frequency domain amplitude, we obtain the first... The mathematical expression for the frequency domain complex amplitude distribution of the next iteration is: In the formula, The updated frequency domain complex amplitude; for The model; Indicates taking The phase angle; The inverse Fourier transform back to the spatial domain gives its spatial domain expression as follows: In the formula, This is the updated spatial domain representation of the frequency domain complex amplitude; This is the inverse operation of the Fourier transform; Extract its spatial domain phase; the spatial domain phase expression is as follows: In the formula, Spatial domain representation of phase; Indicates taking The phase; By constraining the spatial domain amplitude, we obtain the image, whose expression is: In the formula, The image obtained after amplitude constraint; Finally, total variational regularization is applied for noise reduction, resulting in the image expression: In the formula, for The image after the next iteration For total variation regularization strength, For total variation regularization terms The gradient.
6. The method for enhancing ghost imaging in complex dynamic scenes using total variation regularized GS phase retrieval as described in claim 5, characterized in that, Step 3 includes the following specific steps: Will Image after the second iteration As the input image, let it be... ; Performing a Fourier transform on the input image yields its frequency domain expression: In the formula, for The frequency domain expression; Input image; The expression for the angular spectrum propagation operator is: In the formula, For angular spectrum propagation operators; and Frequency domain coordinates; The wavelength of light; For the distance of propagation; The final image expression is: In the formula, This is the final output of the reconstructed complex value.
7. The method for enhancing ghost imaging in complex dynamic scenes using total variation regularized GS phase retrieval as described in claim 1, characterized in that, The method is implemented through the following experimental system, which includes a computer and, in sequence, a laser emitter, a modulation device, a front lens, a target object, a rear lens, and a barrel detector. The computer is connected to both the modulation device and the barrel detector.