A general-purpose computer hologram generation method, system, device, medium and product based on structured propagation

Abstract

The application discloses a general-purpose computer hologram generation method and system based on structured propagation, equipment, medium and product, and relates to the field of computer holography. The method comprises the following steps: decoupling a zero-frequency reference diagonal spectrum method transfer function to construct a structured propagation model; obtaining a target scene graph and a laser wavelength; based on the target scene graph and the laser wavelength, a structured propagation model is used to obtain a structure normalized input field; the structure normalized input field is input into a hologram generation model to obtain a pure phase hologram; and the hologram generation model is obtained by training a U-net network through a sample input field and a sample pure phase hologram. The application can improve the accuracy of hologram generation.

Description

Technical Field

[0001] This application relates to the field of computational holography, and in particular to a general computational hologram generation method, system, device, medium, and product based on structured propagation. Background Technology

[0002] Computer Generated Holography (CGH) aims to reconstruct three-dimensional light fields by computing holograms using algorithms. With the introduction of deep learning, neural network-based generation methods have become mainstream due to their real-time performance.

[0003] In some cases, physical diffraction models are often used directly as a layer in a network or as a data preprocessing step. However, this approach presents three main problems: 1) High parameter sensitivity: The transfer function of the standard physical diffraction model contains a global phase term that rotates rapidly with distance and wavelength. Even small changes in this global phase term can cause severe oscillations and nonlinear distortions in the complex plane. 2) Poor generalization ability: Pre-trained models built on neural networks struggle to adapt to this drastic feature drift, often resulting in models that are tied to specific distances and wavelengths. Once the physical configuration changes, the pre-trained model fails. 3) Artifacts in reconstructed images: When processing 3D scenes, the sensitivity of the standard propagation operator to parameters can induce severe numerical ringing between layers, resulting in speckle noise filling the background of the reconstructed image. Summary of the Invention

[0004] The purpose of this application is to provide a general computational hologram generation method, system, device, medium, and product based on structured propagation, which can improve the accuracy of hologram generation.

[0005] To achieve the above objectives, this application provides the following solution.

[0006] Firstly, this application provides a hologram generation method based on structured propagation, comprising: A structured propagation model is constructed by decoupling the transfer function based on the zero-frequency reference diagonal spectrum method. Obtain the target scene image and laser wavelength; Based on the target scene diagram and the laser wavelength, the structured propagation model is used to obtain the structured normalized input field; The normalized input field of the structure is input into the hologram generation model to obtain a pure phase hologram; the hologram generation model is obtained by training the U-net network with the sample input field and the sample pure phase hologram.

[0007] Secondly, this application provides a hologram generation system based on structured propagation, comprising: A building unit is used to decouple based on the zero-frequency reference diagonal spectrum transfer function and construct a structured propagation model; the structured propagation model is obtained by decoupling using the zero-frequency reference diagonal spectrum transfer function. The acquisition unit is used to acquire the target scene image and the laser wavelength; The structured unit is used to obtain a structured normalized input field based on the target scene map and the laser wavelength, using the structured propagation model. The generation unit is used to input the structure-normalized input field into the hologram generation model to obtain a pure phase hologram; the hologram generation model is obtained by training the U-net network with the sample input field and the sample pure phase hologram.

[0008] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the hologram generation method based on structured propagation described in the first aspect.

[0009] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the hologram generation method based on structured propagation described in the first aspect.

[0010] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the hologram generation method based on structured propagation as described in the first aspect.

[0011] According to the specific embodiments provided in this application, this application has the following technical effects: This application decouples the traditional angular spectrum transfer function through a zero-frequency reference to construct a structured propagation model, stripping away the non-informational global phase oscillation term. Then, based on the target scene map, the structured propagation model is used to obtain multiple structured phase terms, so that the target scene map retains only the structured phase terms, thereby removing the phase chaos caused by physical parameters. This fundamentally solves the problem of perturbation in the input data (structured normalized input field) of the hologram generation model, thus realizing high-fidelity hologram generation under arbitrary physical configuration and improving the accuracy of hologram generation. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is an application environment diagram for a hologram generation method based on structured propagation.

[0014] Figure 2 This is a flowchart illustrating a hologram generation method based on structured propagation.

[0015] Figure 3 This is a detailed flowchart of a hologram generation method based on structured propagation.

[0016] Figure 4 The diagram shows the structured propagation mechanism; (a) is a diagram of the coupling bottleneck in standard propagation; (b) is a diagram of the structured decoupling strategy; and (c) is a diagram of structured propagation and scale invariance.

[0017] Figure 5 Here are schematic diagrams of the hologram generation model; (a) is a schematic diagram of the triangular enhancement module; (b) is a schematic diagram of the convolutional residual module; and (c) is an overall schematic diagram of the hologram generation model.

[0018] Figure 6 The diagram shows a comprehensive evaluation of generalization capability; (a) is a diagram showing the comparison of six-axis radar images; (b) is a diagram showing quantitative verification within a continuous distance range of 2 to 500 mm; (c) is a diagram showing visual reconstruction of a natural scene; and (d) is a diagram showing 2D visual reconstruction of a binary target.

[0019] Figure 7 The diagrams show high-fidelity holographic reconstruction and focusing clue verification of complex 3D scenes; (a) is a schematic diagram of the target 3D color scene and its depth map; (b) is a schematic diagram of the input field and reconstruction result under the standard propagation mode; (c) is a schematic diagram of the input field and reconstruction result under the structured propagation mode; and (d) is a schematic diagram of the digital refocusing experiment.

[0020] Figure 8 This is a schematic diagram of a hologram generation system based on structured propagation provided in this application.

[0021] Figure 9 This is a schematic diagram of the structure of a computer device provided in this application.

[0022] Reference numerals: Terminal 102, Server 104, Construction Unit-801, Acquisition Unit-802, Structured Unit-803, Generation Unit-804. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] The hologram generation method based on structured propagation provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send the target scene image and laser wavelength to server 104. After receiving the target scene image and laser wavelength, server 104 decouples the target scene image and laser wavelength based on the zero-frequency reference diagonal spectrum transfer function to construct a structured propagation model; acquires the target scene image and laser wavelength; obtains multiple structured phase terms based on the target scene image and laser wavelength using the structured propagation model; performs complex amplitude coherent superposition of the multiple structured phase terms to obtain a structured normalized input field; inputs the structured normalized input field into the hologram generation model to obtain a pure phase hologram; the hologram generation model is obtained by training the U-net network using the sample input field and sample pure phase holograms. Server 104 can feed back the obtained pure phase hologram to terminal 102. In addition, in some embodiments, the hologram generation method based on structured propagation can also be implemented by the server 104 or the terminal 102 separately. For example, the terminal 102 can directly perform structured processing on the target scene map and the laser wavelength, or the server 104 can obtain the target scene map and the laser wavelength from the data storage system and perform structured processing on the target scene map and the laser wavelength.

[0026] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0027] In one exemplary embodiment, such as Figure 2 and Figure 3 As shown, a hologram generation method based on structured propagation is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S1 to S5. Wherein: Step S1: Decouple based on the zero-frequency reference diagonal spectrum transfer function and construct a structured propagation model.

[0028] As an feasible approach, step S1 specifically includes: decoupling based on the zero-frequency reference diagonal spectrum transfer function to obtain a structured phase term; and constructing a structured propagation model based on the structured phase term.

[0029] Specifically, in the fusion research of computational holography and deep learning, the angular spectral transfer function, due to its inclusion of a global phase factor that is extremely sensitive to propagation distance, often leads to severe numerical oscillations and convergence difficulties in neural networks. The phase term of the angular spectral transfer function includes a square root term describing the axial component of the wave vector. To separate the "variable" and "invariant" aspects of the physical process, the optical field is precisely decomposed into two independent parts: the carrier wave and structural information. A zero-frequency reference is introduced into this square root term, and its mathematical form is reconstructed as a superposition of "reference (+1)" and "deviation (-1)". The physical essence of this operation is to establish a reference frame that dynamically moves with the propagation distance: the "+1" term corresponds to the zeroth-order term of the Taylor expansion, which physically represents a plane wave carrier propagating linearly along the optical axis. Although this term carries the forward transport of energy, its phase oscillates linearly with distance at a high frequency. For image reconstruction tasks that focus on the lateral intensity distribution, it not only does not contain any effective structural information, but also constitutes the main noise source leading to phase aliasing and gradient instability. Conversely, the introduction of the "-1" term performs the function of phase difference, subtracting the aforementioned high-frequency carrier from the total phase to extract higher-order diffraction terms. This remaining term describes the phase lag or lead of each spatial frequency component relative to the principal plane wave, i.e., the relative bending and deformation of the wavefront, which is the essence of the "diffraction structure" in physical optics. Through this "carrier-free" processing, the global phase oscillation, as a noise source, is discarded, retaining only the structural envelope describing the wavefront evolution. This processing not only numerically smooths the optimization surface, but more importantly, it reveals the scale invariance of the diffraction process—that is, after removing the interference of absolute phase, the evolution of the light field is controlled only by dimensionless parameters. This alignment of physical features ultimately enables neural networks to overcome the barriers of physical parameters and learn universal laws of light field propagation.

[0030] In this method, the angular spectrum transfer function is decoupled into the form of a multiplication of the global phase model and the structured propagation model, and its expression is: ,in, The transfer function of the angular spectrum method, For global phase model, This is a structured propagation model. The expression for the global phase model is: .

[0031] in, This is a global propagation model; It is an exponential function; For the distance of propagation; The wavelength is denoted as λ. The global phase term calculated by the global propagation model is identified as an oscillating factor that rotates rapidly with propagation distance and laser wavelength, and is a noise source that causes parameter sensitivity.

[0032] As an implementable approach, the structured propagation model is expressed as follows: .

[0033] in, For structured propagation models; It is an exponential function; For the light field in Spatial frequency along the axis; For the light field in Spatial frequency along the axis; For the distance of propagation; The wavelength is the laser wavelength. The structured phase term calculated by the structured propagation model carries information about the object's geometry and diffraction structure.

[0034] Specifically, Figure 4 (a) Coupling bottleneck in standard propagation: The standard transfer function tightly couples the object's structural information with a frequency-independent global phase term. This phase factor exhibits extremely high-frequency rotational characteristics with propagation distance, causing the wavefront signal to oscillate violently in the complex plane, exhibiting extreme parameter sensitivity; at the same time, it induces uncorrelated random drifts in the phase distribution, presenting disordered random characteristics. This instability significantly hinders the convergence of neural networks. Figure 4 (b) Structured decoupling strategy: Through mathematical decomposition, the global phase term is identified as a non-information oscillation source and actively stripped away, while only the structured phase term carrying the wavefront shaping function is retained to participate in the calculation. Figure 4(c) Structured Propagation and Scale Invariance: This strategy constructs a normalized physical computation environment. After eliminating interference from the global phase term, the signal returns to its intrinsic stability, and diffraction patterns under different physical configurations are aligned to scale-invariant features following a unified scaling law. This allows the network to directly grasp the general laws of optical field evolution without needing to memorize absolute physical parameters.

[0035] Step S2: Obtain the target scene image and laser wavelength.

[0036] Specifically, the target scene diagram is as follows: Figure 3 RGB-D scene diagram.

[0037] Step S3: Based on the target scene map and laser wavelength, a structured propagation model is used to obtain the structured normalized input field.

[0038] As one feasible approach, the target scene map includes: an RGB image of the target scene and a corresponding depth map. Step S3 specifically includes steps S31 to S33: Step S31: Determine the depth range based on the depth map.

[0039] Specifically, the depth range refers to the initial depth coordinates of the 3D scene relative to the holographic plane along the direction of light wave propagation (optical axis). To the end depth coordinates The spatial range.

[0040] Step S32: Divide the RGB image based on the depth range to obtain multiple parallel slices, and calculate the distance between two adjacent parallel slices.

[0041] Specifically, depending on the depth range The RGB image is discretized into multiple parallel slices, and the distance between adjacent slices is calculated. .

[0042] Step S33: Based on the distance, multiple parallel slices and laser wavelength, a structured propagation model is used to obtain the structured normalized input field.

[0043] Specifically, the slice furthest from the holographic plane The image intensity is used as the amplitude, and the initial phase is set to 0, serving as the initial complex amplitude light field. The laser slices from the farthest segment... Propagate towards the holographic plane, starting from the farthest slice The process begins with layer-by-layer replacement propagation towards the holographic plane, resulting in a structure-normalized input field.

[0044] For the Layer slice ( Values ​​from Decrease to 1) and the incident light field from the previous layer, perform wavefront replacement: check the first Pixel information of layer slices; for the first layer slice... The effective image information region of the layer slice, using the first The amplitude information of the layer slice (with phase set to 0) is forcibly replaced by the first layer slice. The complex amplitude light field of the layer slice corresponds to the complex amplitude value of the region with effective image information; for the first layer... The layer slice contains no information in the region, so the first layer is retained. The complex amplitude of the light field in the slice remains unchanged in the region without information.

[0045] Each wavefront replacement is performed using a structured propagation model. Utilizing the aforementioned structured propagation model, the complex amplitude optical field after wavefront replacement is propagated forward by an interlayer interval. and laser wavelength This yields the complex amplitude optical field of the next slice, which is transmitted to the holographic plane direction. Then, based on the holographic plane and the farthest slice... The distance between them and the laser wavelength are used to obtain the structure-normalized input field.

[0046] Step S5: Input the structure-normalized input field into the hologram generation model to obtain a pure phase hologram; the hologram generation model is obtained by training the U-net network with the sample input field and the sample pure phase hologram.

[0047] Specifically, the weight parameters of the hologram generation model are pre-trained and fixed. The hologram generation model employs a physics-driven end-to-end training strategy, eliminating the need to pre-compute ground-value holograms as labels. The specific training steps are as follows: Step T1: Construct the training dataset. Select multiple two-dimensional images as sample target images (denoted as...). The target image of the sample can be a natural image, a textured image, or an image from a specific dataset.

[0048] Step T2: Generate the structure-normalized input field (backpropagation). For each sample target image, a training wavelength and training propagation distance are randomly set or specified. Based on the training wavelength and training propagation distance, the sample target image is backpropagated using the aforementioned structured propagation model to obtain the corresponding sample structure-normalized input field (i.e., complex amplitude field).

[0049] Explanation of principle: This step simulates the process of light propagating backward from the object plane to the holographic plane and removes the global phase oscillation term, providing the network with pure structured features.

[0050] Step T3: Network forward inference. The normalized input field of the sample structure is fed into the hologram generation model to be trained (such as the U-net network). The model outputs the predicted pure phase hologram.

[0051] Step T4: Numerical Reconstruction (Forward Propagation). Construct a differentiable forward propagation model. Based on the same training wavelength and training propagation distance set in Step T2, perform forward propagation calculations on the pure phase hologram output in Step T3 to simulate the process of light waves propagating from the holographic plane back to the object plane, obtaining the numerically reconstructed complex amplitude field. Take the amplitude (or intensity) of this field to obtain the reconstructed image.

[0052] Step T5: Calculate the loss function. Calculate the difference between the reconstructed image and the sample target image to construct the loss function. In this embodiment, the L1 loss function is used to constrain the reconstruction quality. The expression of the loss function is as follows: ,in, The total number of pixels in the image. To reconstruct the image, The target image is a sample.

[0053] Step T6: Parameter Update. Based on the calculated Loss value, calculate the gradient using the backpropagation algorithm and update the weight parameters of the hologram generation model (U-net). Repeat steps T2 to T6 until the Loss converges or the preset number of training rounds is reached.

[0054] As an feasible approach, such as Figure 5 As shown in (c), the hologram generation model includes: a sequentially connected inverse pixel recombination layer, a first triangle enhancement module, a first convolutional residual module, a max pooling layer, a second triangle enhancement module, a second convolutional residual module, a transposed convolutional layer, a stitching layer, a third triangle enhancement module, and a convolutional layer. Step S5 specifically includes steps S51 to S510: Step S51: Normalize the input field of the structure (i.e. Figure 5 The complex amplitude light field is input into the inverse pixel reconstruction layer to obtain the inverse pixel reconstruction map.

[0055] Step S52: Input the inverse pixel reconstruction image into the first triangular enhancement module to obtain the first enhanced feature map.

[0056] Step S53: Input the first enhanced feature map into the first convolutional residual module to obtain the first feature map.

[0057] Step S54: Input the first feature map into the max pooling layer to obtain the first feature map after pooling.

[0058] Step S55: Input the pooled first feature map into the second triangular enhancement module to obtain the second enhanced feature map.

[0059] Step S56: Input the second enhanced feature map into the second convolutional residual module to obtain the second feature map.

[0060] Step S57: Input the second feature map into the transposed convolutional layer to obtain the magnified second feature map.

[0061] Step S58: Input the magnified second feature map into the stitching layer to obtain the stitched feature map.

[0062] Step S59: Input the spliced ​​feature map into the third triangular enhancement module to obtain the third enhanced feature map.

[0063] Step S510: Input the third enhanced feature map into the convolutional layer to obtain a pure phase hologram.

[0064] As an implementable approach, the first triangular enhancement module, the second triangular enhancement module, and the third triangular enhancement module all include: a sine layer, a cosine layer, and a splicing layer.

[0065] Specifically, such as Figure 5 As shown in (a), the first, second, and third triangular enhancement modules are all triangular enhancement modules, each including a sine layer, a cosine layer, and a concatenation layer. Taking the first triangular enhancement module as an example, the inverse pixel reconstruction image is input to the sine and cosine layers respectively to obtain a sine inverse pixel reconstruction image and a cosine inverse pixel reconstruction image. Then, the sine inverse pixel reconstruction image, the cosine inverse pixel reconstruction image, and the inverse pixel reconstruction image are concatenated through the concatenation layer to obtain the first enhanced feature map. The triangular enhancement modules are used to process periodic information.

[0066] Among them, such as Figure 5 As shown in (b), both the first and second convolutional residual modules are convolutional residual modules, and from top to bottom and left to right, they each include an input layer, a 1×1 convolutional layer, a batch normalization (BN) layer, a 3×3 convolutional layer, a BN layer, a Leaky ReLU layer, a 3×3 convolutional layer, a BN layer, a Leaky ReLU layer, and an output layer. The residual convolutional modules are used for deep feature extraction.

[0067] The hologram generation model performs nonlinear mapping operations. Since the structure-normalized input field has already undergone structure normalization, the network does not need to handle phase oscillations caused by physical parameters, but directly extracts the implicit diffraction structure features. Therefore, the hologram generation model directly generates pure phase holograms.

[0068] Based on the above methods, two embodiments are provided for comprehensive evaluation.

[0069] Example 1 like Figure 6 As shown, Figure 6The comparison of the six-axis radar charts in (a) shows that the structure-guided encoder, shown in blue, achieves state-of-the-art performance across a wide range of metrics, including distance range, resolution support, and spectral coverage, using only a single lightweight model, while the red SFO-solver and the purple HoloNet are limited to specific physical parameters. Figure 6 Quantitative verification over a continuous distance range of 2 to 500 mm in (b) shows that the method of this application overcomes the limitations of SFO being effective only within the training interval and HoloNet locking only a single focal plane, maintaining extremely stable high PSNR and SSIM indices. Figure 6 (c) and (d) 2D visual reconstruction of natural scenes and binary targets. Whether in the near field of 5mm or the far field of 200mm, the model of this application can reconstruct sharp and artifact-free clear features, which is superior to the benchmark method in terms of fidelity and background purity.

[0070] Based on the above data, it can be concluded that the method in this application achieves generalization of a single model to arbitrary physical configurations. Results: Only one neural network model is needed to adapt to arbitrary wavelengths (RGB full color) and propagation distances spanning several orders of magnitude (e.g., 2mm to 500mm+), without the need for retraining or fine-tuning for new parameters. Reason: This stems from the decoupling in step S1. By stripping away the global phase term, the intrinsic scale invariance of diffraction is revealed. Diffraction patterns under different physical configurations are reduced by the structured phase term to be subject only to a dimensionless factor. A unified form of regulation. Networks no longer need to memorize absolute physical values, but instead learn universal geometric evolutionary laws.

[0071] Example 2 like Figure 7 As shown, Figure 7 (a) 3D color scene of the target and its depth map. The scene depth covers 10 to 24 mm and is discretized into 8 parallel depth layers in the calculation. Figure 7 (b) Input field and reconstruction result under standard propagation mode: Even before inputting into the network, the backpropagation field (network input) already exhibits severe phase discontinuity and amplitude ringing due to the violent global phase oscillations introduced by the standard propagation operator. This contaminated input directly leads to obvious color distortion and stripe artifacts in the final holographic reconstruction image, with a PSNR of only 31.9 dB. Figure 7 (c) Input field and reconstruction results under structured propagation mode (using the method of this application): Structured propagation effectively removes non-physical global oscillation components, resulting in a pure structured form of the input field fed to the network, fully preserving the continuity of texture and phase. Thanks to the pure input, the model generates clear and sharp high-fidelity holograms, with PSNR improved to 38.0 dB. Figure 7(d) Digital refocusing experiment: Through numerical reconstruction at three different depths of 10mm, 16mm, and 22mm, the results demonstrate the correct depth-of-field effect—objects within the blue solid frame are clearly in focus, while objects at other depths exhibit natural defocus blur. This confirms that the generated single hologram has successfully encoded accurate and continuous three-dimensional depth information.

[0072] The above data verifies that the method of this application not only eliminates numerical computation artifacts but also improves reconstruction quality. Results: During reconstruction, the generated hologram has a clean background without speckle and eliminates interlayer crosstalk and ringing effects common in traditional tomography. Reason: This stems from steps S1 and S2. In traditional methods, high-frequency oscillations in the global phase term ( The oscillation (rad / m) can lead to serious numerical errors in discretization calculations. Structured propagation eliminates this oscillation at its mathematical root by establishing a stable phase reference frame, ensuring the purity and stability of the input network data (structure-normalized input field) on the wavefront structure.

[0073] The method in this application also significantly reduces the difficulty and complexity of network learning. Results: The network converges quickly and has high inference efficiency. Reason: This stems from step S1. The random phase drift (ChaoticPhase) introduced by the global phase term is removed, providing the network with a "drift-free" learning environment. The network no longer wastes capacity fitting meaningless phase noise, but instead focuses on learning the essential mapping between object structure and diffraction patterns.

[0074] The beneficial effects of the hologram generation method based on structured propagation proposed in this application are mainly reflected in: 1. Structured Propagation Preprocessing Stage: The traditional angular spectrum transfer function is decoupled, and the non-informational global phase term (oscillation term) is removed. Only the structured phase term is used to propagate and superimpose the various depth slices of the 3D scene in reverse. This step transforms the physical light field, which varies drastically with distance and wavelength, into a "structured normalized input field" that follows a unified scaling law.

[0075] 2. General Network Encoding Stage: The normalized input field is fed into a neural network with fixed parameters. Since the input data has removed phase chaos caused by physical parameters, the network can directly predict high-fidelity pure phase holograms without fine-tuning for specific wavelengths or distances.

[0076] 3. To break free from the constraints of physical parameters on the network, this application's method delves into the mathematical structure of the diffraction transfer function, discovering that by decoupling and stripping the global phase term of high-frequency oscillations, the structured phase term carrying the wavefront shaping function can be retained. This operation mathematically constructs a scale-invariant physical space, ensuring that the diffraction process under different physical configurations follows a unified dimensionless evolution law. Based on this, this application uses structured propagation as a front-end processing step in the hologram generation model, allowing the network to learn only general light field mapping rules, thereby achieving high-fidelity holographic generation of a single model under arbitrary physical configurations.

[0077] Based on the same inventive concept, this application also provides a hologram generation system based on structured propagation. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the hologram generation system based on structured propagation provided below can be found in the limitations of the hologram generation method based on structured propagation described above, and will not be repeated here.

[0078] In one exemplary embodiment, such as Figure 8 As shown, a hologram generation system based on structured propagation is provided, comprising: Building unit 801 is used to decouple based on the zero-frequency reference diagonal spectrum transfer function and construct a structured propagation model; the structured propagation model is obtained by decoupling using the zero-frequency reference diagonal spectrum transfer function.

[0079] The acquisition unit 802 is used to acquire the target scene image and the laser wavelength.

[0080] The structured unit 803 is used to obtain a structured normalized input field based on the target scene map and the laser wavelength using a structured propagation model.

[0081] The generation unit 804 is used to input the structure-normalized input field into the hologram generation model to obtain a pure phase hologram; the hologram generation model is obtained by training the U-net network with the sample input field and the sample pure phase hologram.

[0082] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 9As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores target scene images and laser wavelengths. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a hologram generation method based on structured propagation.

[0083] Those skilled in the art will understand that Figure 9 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0084] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0085] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0086] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0087] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0088] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0089] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0090] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A hologram generation method based on structured propagation, characterized in that, The hologram generation method based on structured propagation includes: A structured propagation model is constructed by decoupling the transfer function based on the zero-frequency reference diagonal spectrum method. Obtain the target scene image and laser wavelength; Based on the target scene diagram and the laser wavelength, the structured propagation model is used to obtain the structured normalized input field; The normalized input field of the structure is input into the hologram generation model to obtain a pure phase hologram; the hologram generation model is obtained by training the U-net network with the sample input field and the sample pure phase hologram.

2. The hologram generation method based on structured propagation according to claim 1, characterized in that, Based on the zero-frequency reference diagonal spectrum transfer function, a structured propagation model is constructed through decoupling, specifically including: The structured phase term is obtained by decoupling the transfer function based on the zero-frequency reference diagonal spectrum method. A structured propagation model is constructed based on the structured phase term.

3. The hologram generation method based on structured propagation according to claim 1, characterized in that, The expression for the structured propagation model is: ； in, For structured propagation models; It is an exponential function; For the light field in Spatial frequency along the axial direction; For the light field in Spatial frequency along the axial direction; For the distance of propagation; is the laser wavelength.

4. The hologram generation method based on structured propagation according to claim 1, characterized in that, The target scene map includes: an RGB image of the target scene and a corresponding depth image; Based on the target scene diagram and the laser wavelength, the structured propagation model is used to obtain the structured normalized input field, specifically including: Based on the depth map, determine the depth range; The RGB image is divided based on the depth range to obtain multiple parallel slices, and the distance between two adjacent parallel slices is calculated. Based on the distance, the multiple parallel slices, and the laser wavelength, a structured propagation model is used to obtain a structured normalized input field.

5. The hologram generation method based on structured propagation according to claim 1, characterized in that, The hologram generation model includes: an inverse pixel recombination layer, a first triangle enhancement module, a first convolutional residual module, a max pooling layer, a second triangle enhancement module, a second convolutional residual module, a transposed convolutional layer, a stitching layer, a third triangle enhancement module, and a convolutional layer connected in sequence. The structure-normalized input field is input into the hologram generation model to obtain a pure phase hologram, specifically including: The normalized input field of the structure is input to the inverse pixel reconstruction layer to obtain the inverse pixel reconstruction map; The inverse pixel reconstruction map is input into the first triangular enhancement module to obtain the first enhanced feature map; The first enhanced feature map is input into the first convolutional residual module to obtain the first feature map; The first feature map is input into the max pooling layer to obtain the pooled first feature map. The pooled first feature map is input into the second triangular enhancement module to obtain the second enhanced feature map; The second enhanced feature map is input into the second convolutional residual module to obtain the second feature map; The second feature map is input into the transposed convolutional layer to obtain the magnified second feature map; The magnified second feature map is input into the stitching layer to obtain the stitched feature map; The stitched feature map is input into the third triangular enhancement module to obtain the third enhanced feature map; The third enhanced feature map is input into the convolutional layer to obtain a pure phase hologram.

6. The hologram generation method based on structured propagation according to claim 1, characterized in that, The first triangular enhancement module, the second triangular enhancement module, and the third triangular enhancement module all include: a sine layer, a cosine layer, and a splicing layer.

7. A hologram generation system based on structured propagation, characterized in that, The hologram generation system based on structured propagation includes: A building unit is used to decouple based on the zero-frequency reference diagonal spectrum transfer function and construct a structured propagation model; the structured propagation model is obtained by decoupling using the zero-frequency reference diagonal spectrum transfer function. The acquisition unit is used to acquire the target scene image and the laser wavelength; The structured unit is used to obtain a structured normalized input field based on the target scene map and the laser wavelength, using the structured propagation model. The generation unit is used to input the structure-normalized input field into the hologram generation model to obtain a pure phase hologram; the hologram generation model is obtained by training the U-net network with the sample input field and the sample pure phase hologram.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the hologram generation method based on structured propagation as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the hologram generation method based on structured propagation as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the hologram generation method based on structured propagation as described in any one of claims 1-6.

Images