Light field aberration correction method and apparatus, electronic device and storage medium

Abstract

The present invention relates to the technical field of computers, and in particular to a light field aberration correction method and device, an electronic device, and a storage medium. The method comprises: obtaining current light field data; inputting the current light field data into a light field aberration correction network based on an implicit space, the light field aberration correction network then outputting the light field aberration of the current light field data and denoised current light field data; and correcting the light field aberration of the denoised current light field data. In this way, the following problems in the prior art are solved: additional optical hardware support being required for light field aberration correction, and slowed iteration speed due to high hardware requirements; inaccurate modeling due to long backend computation time; and poor real-time performance.

Description

Light field aberration correction method, device, electronic device and storage medium

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application is based on the Chinese patent application with application number 202410032943.6 and application date January 9, 2024, and claims the priority of the Chinese patent application. The entire content of the Chinese patent application is hereby introduced into this application as a reference. Technical Field

[0003] The present invention relates to the field of computer technology, and in particular to a light field aberration correction method, device, electronic equipment and storage medium. Background Art

[0004] Light-field microscopy is an important tool for three-dimensional in vivo observation. However, when observing biological samples, the uneven distribution of the sample's refractive index and the change in the refractive index of optical equipment such as the water mirror lead to the generation of optical aberrations. The quality of light-field data reconstruction is affected by optical aberrations.

[0005] Due to the inherent properties of photons and camera manufacturing, light field data inevitably contains noise; greater noise can also reduce the quality of light field data reconstruction. Currently, there are two main methods for correcting light field aberrations: sensor-based hardware correction and post-processing software correction.

[0006] In related technologies, the sensor-based hardware correction method has certain requirements for optical instruments such as sensors and deformable mirrors, and the iteration speed is often slow; the software correction method based on post-calculation uses the relationship between different angles of the light field image to estimate the aberration, and its modeling is incomplete and not real-time.

[0007] Summary of the Invention

[0008] The present invention provides a light field aberration correction method, device, electronic device and storage medium to solve the problems in the related art that light field aberration correction requires additional optical hardware support, has high hardware requirements resulting in slow iteration speed, and has long post-calculation time resulting in inaccurate modeling and poor real-time performance.

[0009] A first aspect of the present invention provides a method for correcting light field aberrations, comprising the following steps: acquiring current light field data; inputting the current light field data into a light field aberration correction network based on an implicit space, the light field aberration correction network outputting the light field aberrations of the current light field data and the denoised current light field data; and correcting the light field aberrations of the denoised current light field data.

[0010] Optionally, the light field aberration correction network includes an encoder, a decoder and a regressor, wherein the encoder encodes the current light field data into an implicit space; the decoder decodes from the implicit space to obtain denoised current light field data; and the regressor extracts the light field aberration of the current light field data from the implicit space.

[0011] Optionally, before inputting the current light field data into the light field aberration correction network based on the implicit space, the method further includes: obtaining original training data carrying noise and light field aberrations; adding target noise to the original training data to form new training data; training the encoder and the decoder according to the new training data, training the encoder and the regressor according to the original training data, and performing alternating iterative training, wherein the encoder constructs an implicit space during the training process and uses the implicit space to distinguish between noise and light field aberrations; if the training stopping condition is met, the alternating iterative training is stopped, and a light field aberration correction network based on the implicit space is obtained after the training is completed.

[0012] Optionally, the new training data The formation formula is:

[0013] Where x is the light field data with noise and aberration, D is the reversible matrix, D = αI, α is an arbitrary non-zero constant, I is the unit matrix, D T is the transpose of the reversible matrix,

[0014] , where z is a zero-mean Gaussian random vector in the noisy light field data, σ 2 =β1H(xb)+β2,

[0015] Among them, σ 2 is the background variance of the light field data, b is the fixed background caused by the sample and the camera, H is a linear low-pass filter, and β1 and β2 are constants.

[0016] A second aspect of the present invention provides a light field aberration correction device, comprising: an acquisition module for acquiring current light field data; an input module for inputting the current light field data into a light field aberration correction network based on an implicit space, the light field aberration correction network outputting the light field aberration of the current light field data and the denoised current light field data; and a correction module for correcting the light field aberration of the denoised current light field data.

[0017] Optionally, the light field aberration correction network includes an encoder, a decoder and a regressor, wherein the encoder encodes the current light field data into an implicit space; the decoder decodes from the implicit space to obtain denoised current light field data; and the regressor extracts the light field aberration of the current light field data from the implicit space.

[0018] Optionally, the input module is further used to: obtain original training data carrying noise and light field aberrations; add target noise to the original training data to form new training data; train the encoder and the decoder according to the new training data, train the encoder and the regressor according to the original training data, and perform alternating iterative training, wherein the encoder constructs an implicit space during the training process and uses the implicit space to distinguish between noise and light field aberrations; if the training stopping condition is met, the alternating iterative training is stopped, and a light field aberration correction network based on the implicit space is obtained after the training is completed.

[0019] Optionally, the new training data The formation formula is:

[0020] Where x is the light field data with noise and aberration, D is the reversible matrix, D = αI, α is an arbitrary non-zero constant, I is the unit matrix, D T is the transpose of the reversible matrix,

[0021] , where z is a zero-mean Gaussian random vector in the noisy light field data, σ 2 =β1H(xb)+β2,

[0022] Among them, σ 2 is the background variance of the light field data, b is the fixed background caused by the sample and the camera, H is a linear low-pass filter, and β1 and β2 are constants.

[0023] A third aspect of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the light field aberration correction method as described in the above embodiment.

[0024] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the light field aberration correction method as described in the above embodiment.

[0025] Therefore, the present invention has at least the following beneficial effects:

[0026] The embodiments of the present invention can quickly estimate light field aberrations only through a light field aberration correction network based on implicit space, without the need for additional optical hardware. This solves the problem of inaccurate modeling of software correction methods based on post-calculation, and corrects the light field aberrations of the denoised current light field data. This can achieve fast and accurate correction of light field aberrations, effectively improve the accuracy of light field data, and thus improve the quality of light field reconstruction.

[0027] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0029] FIG1 is a flow chart of a light field aberration correction method provided according to an embodiment of the present invention;

[0030] FIG2 is a schematic flow chart of an aberration correction method for a light field microscopy system using deep learning based on implicit space according to an embodiment of the present invention;

[0031] FIG3 is a schematic diagram of an implicit space-based light field aberration correction network for iterative training according to an embodiment of the present invention;

[0032] FIG4 is a schematic diagram of training and using an implicit space-based deep learning light field microscopy system aberration correction network according to an embodiment of the present invention;

[0033] FIG5 is a block diagram of an optical field aberration correction device according to an embodiment of the present invention;

[0034] FIG6 is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0036] The light field aberration correction method, device, electronic device and storage medium of the embodiment of the present invention are described below with reference to the accompanying drawings. In response to the problems in the related art mentioned in the above background technology that light field aberration correction requires additional optical hardware support, and the high hardware requirements lead to a slow iteration speed, and the long post-calculation time leads to inaccurate modeling and poor real-time performance, the present invention provides a light field aberration correction method, in which the light field aberration can be quickly estimated only by a light field aberration correction network based on implicit space, without the need for additional optical hardware, thereby solving the problem of inaccurate modeling of the software correction method based on post-calculation, and correcting the light field aberration of the current light field data after denoising, which can achieve fast and accurate correction of light field aberration, effectively improve the accuracy of light field data, and thus improve the quality of light field reconstruction. As a result, the problems in the related art that light field aberration correction requires additional optical hardware support, and the high hardware requirements lead to a slow iteration speed, and the long post-calculation time leads to inaccurate modeling and poor real-time performance are solved.

[0037] Specifically, FIG1 is a flow chart of a light field aberration correction method provided by an embodiment of the present invention.

[0038] As shown in FIG1 , the light field aberration correction method includes the following steps:

[0039] In step S101 , current light field data is acquired.

[0040] It can be understood that the current light field data refers to data that needs to be denoised and corrected for aberrations. The light field data can be a light field microscopic image, and usually the current light field data contains noise and / or aberrations. The embodiments of the present invention can obtain the current light field data based on a microlens array, a camera array, a coding mask, and a focus stack.

[0041] Light field data can be represented by Zernike coefficients. For example, the Zernike coefficients can be represented by a 17-dimensional vector. The first four Zernike coefficients are easily corrected by other methods. Therefore, the nth (1≤n≤17) component of the vector corresponds to the coefficient of the n+4th Zernike polynomial under the standard index.

[0042] In step S102 , the current light field data is input into a light field aberration correction network based on an implicit space, and the light field aberration correction network outputs the light field aberration of the current light field data and the denoised current light field data.

[0043] Among them, the light field aberration correction network includes an encoder, a decoder and a regressor, wherein the encoder encodes the current light field data into the implicit space; the decoder decodes the implicit space to obtain the denoised current light field data; and the regressor extracts the light field aberration of the current light field data from the implicit space.

[0044] It can be understood that the embodiments of the present invention can input the current light field data into the light field aberration correction network based on the implicit space, obtain the light field aberration of the current light field data based on the light field aberration correction network and denoise the current data, which helps to jointly optimize the two tasks of denoising reconstruction and aberration estimation. The light field aberration can be quickly estimated without the need for additional optical hardware, thereby improving work efficiency and reducing costs.

[0045] Specifically, the embodiments of the present invention alternately iteratively train a light field aberration correction network based on an implicit space. Alternating iterative training means: after jointly training the encoder and decoder, the encoder and regressor are jointly trained, and the above training process is repeated. Thus, through alternating iterative training, the embodiments of the present invention help the encoder construct a suitable implicit space, enable the decoder to better distinguish between noise and structure, and enable the regressor to better extract aberration information from the implicit space. At the same time, the decoder and regressor share an encoder, which facilitates the joint optimization of the two tasks of denoising reconstruction and aberration estimation.

[0046] In an embodiment of the present invention, before inputting current light field data into a light field aberration correction network based on an implicit space, the method further includes: obtaining original training data containing noise and light field aberrations; adding target noise to the original training data to form new training data; training an encoder and a decoder based on the new training data, training an encoder and a regressor based on the original training data, and performing alternating iterative training, wherein the encoder constructs an implicit space during the training process and uses the implicit space to distinguish between noise and light field aberrations; if a training stop condition is met, the alternating iterative training is stopped, and a light field aberration correction network based on the implicit space is obtained after the training is completed.

[0047] Among them, the target noise can be noisy aberration light field data, which can be adjusted according to actual needs; the training stop condition can be that the alternating iterative training reaches a preset number of times, and the preset number of times can be set by the user without specific limitation.

[0048] It can be understood that the embodiment of the present invention adds target noise to the original training data to form new training data, and uses the new training data to alternately iteratively train the encoder and decoder, and uses the original training data to train the encoder and regressor, which helps the encoder construct a suitable implicit space, so that the decoder can better distinguish between noise and structure, and the regressor can better extract aberration information from the implicit space, thereby improving the robustness of the light field aberration correction network based on the implicit space.

[0049] Specifically, the encoder and decoder are an end-to-end neural network structure. The encoder constructs an implicit space during training and uses the implicit space to distinguish between noise and light field aberrations. The decoder decodes the implicit space to obtain the denoised current light field data. The regressor extracts the aberration information from the implicit space and outputs it.

[0050] The light field data is input into the encoder, and the output of the regressor is a 17-dimensional vector that simulates the 5th to 21st Zernike coefficients. The phase map corresponding to this output and the phase map corresponding to the true aberration are used as the loss function for gradient descent training. The encoder receives the light field data, the decoder outputs denoised aberration light field data, and the regressor outputs predicted aberration data. After the encoder, decoder, and regressor are trained, the light field aberration correction network is used, and the decoder and regressor predictions are performed simultaneously. Because after training, the encoder learns to encode information into the implicit space, the decoder can decode from the implicit space and reconstruct the denoised aberration light field, and the regressor can estimate the aberration corresponding to the light field data. The denoised aberration light field and predicted aberration can be used for subsequent light field reconstruction.

[0051] In this embodiment of the present invention, the new training data The formation formula is:

[0052] Where x is the light field data with noise and aberration, D is the reversible matrix, D = αI, α is an arbitrary non-zero constant, I is the unit matrix, D T is the transpose of the reversible matrix,

[0053] , where z is a zero-mean Gaussian random vector in the noisy light field data, σ 2 =β1H(xb)+β2,

[0054] Among them, σ 2 is the background variance of the light field data, b is the fixed background caused by the sample and the camera, H is a linear low-pass filter, and β1 and β2 are constants.

[0055] It should be noted that theoretically, new training data In the formation formula, β1H(xb) approximates the variance caused by the Poisson noise of photons, so β1 should be 1; β2 approximates the variance caused by the Gaussian white noise of the camera, so β2 should be estimated from the background of x.

[0056] Assume that the Gaussian noise variance estimated from the background of x is In practice, when multiple sets of noise data pairs are obtained from a light field data x, for each set of noise data pairs, β1 will be randomly sampled from the interval [0.5, 1.5], and β2 will be randomly sampled from the interval This is because there are inevitable errors when obtaining approximate sample signals from noisy light field data and calibrating the background variance of light field data; therefore, randomly sampling the values ​​of β1 and β2 within a certain range can increase the robustness of the training set composed of noisy data.

[0057] In step S103 , the light field aberration of the denoised current light field data is corrected.

[0058] It can be understood that the embodiments of the present invention can denoise the current light field data and correct the light field aberrations, thereby effectively encoding the impact of noise and light field aberrations on light field reconstruction, effectively improving the accuracy of the light field data, and thus improving the quality of light field reconstruction.

[0059] The light field aberration correction method proposed in an embodiment of the present invention can quickly estimate light field aberrations using only a light field aberration correction network based on implicit space, without the need for additional optical hardware. This solves the problem of inaccurate modeling in software correction methods based on post-calculation, and corrects the light field aberrations of the denoised current light field data. This can achieve fast and accurate correction of light field aberrations, effectively improving the accuracy of light field data, and thereby improving the quality of light field reconstruction.

[0060] The light field aberration correction method will be described in detail below through a specific embodiment. As shown in FIG2 , the specific steps are as follows:

[0061] Step S1: Acquire light field data.

[0062] Light field data is obtained, namely, light field data with noisy aberrations and aberrations corresponding to the image; the aberrations are represented by Zernike coefficients.

[0063] Specifically, the aberration is represented by a 17-dimensional vector, where the nth (1≤n≤17) component of the vector corresponds to the coefficient of the n+4th Zernike polynomial under the Wyant standard index. This is because the first four Zernike coefficients can be easily corrected by other means.

[0064] Step S2: forming a noise-added data pair.

[0065] Add the specified noise to the light field data x to form a new pair of noisy data. D=αI σ 2 =β1H(xb)+β2

[0066] Here, α is an arbitrary nonzero constant, and I is the identity matrix. Multiplying them yields a reversible matrix D. This construction of D is the simplest way to form a reversible matrix and its inverse, but any other reversible matrix can theoretically replace D. b is the fixed background caused by the sample and the camera, and H is any suitable linear low-pass filter, such as a mean filter with a kernel of 5 pixels, so that the background-removed and low-pass filtered light field data H(xb) approximates the sample signal. β1 and β2 are constants related to the optical system used to acquire the light field data.

[0067] Theoretically, β1H(xb) approximates the variance caused by the Poisson noise of photons, so β1 should be 1; β2 approximates the variance caused by the Gaussian white noise of the camera, so β2 should be estimated from the background of x. Assume that the Gaussian noise variance estimated from the background of x is In practice, when multiple sets of noise data pairs are obtained from a light field data x, for each set of noise data pairs, β1 will be randomly sampled from the interval [0.5, 1.5], and β2 will be randomly sampled from the interval This is because there are inevitable errors when obtaining approximate sample signals from noisy light field data and calibrating the background variance of light field data. Therefore, randomly sampling the values ​​of β1 and β2 within a certain range can increase the robustness of the training set constructed by the noisy data.

[0068] The noise-added data pairs generated by the above formula will be used to train the encoder and decoder. After training, the encoder and decoder will be able to infer the denoised light field data from a single light field data.

[0069] S3, iterative training of the implicit space-based light field aberration correction network.

[0070] The encoder and decoder, and the encoder and regressor, are trained alternately and iteratively. The implicit space-based light field aberration correction network consists of an encoder, a decoder, and a regressor. The encoder receives light field data and encodes it into the implicit space. The decoder decodes the denoised light field data from the implicit space. The regressor extracts features from the implicit space and estimates the aberration corresponding to the light field data.

[0071] Alternating and iterative training of the encoder and decoder, and the encoder and regressor, helps the encoder construct a suitable implicit space, enabling the decoder to better distinguish noise and structure, and the regressor to better extract disparity information from the implicit space. The decoder and regressor share a common encoder, facilitating the joint optimization of denoising reconstruction and disparity estimation.

[0072] As shown in Figures 3 and 4, the specific steps of alternating iterative training are as follows:

[0073] Step S31: training the encoder and decoder.

[0074] The decoder outputs a denoised aberrated light field using supervised training using the noisy data.

[0075] Input any one of the noisy data pairs into the encoder, and the output of the decoder is exactly the same size as the input of the encoder. The output is used as the loss function with the other one of the noisy data pairs, and gradient descent training is performed.

[0076] Step S32: training the encoder and the regressor.

[0077] The noisy aberration light field and its corresponding disparity are used for supervised training, and the regressor outputs the predicted disparity.

[0078] The noisy aberration light field is input into the encoder, and the output of the regressor is a 17-dimensional vector, which simulates the 5th to 21st Zernike coefficients. The phase map corresponding to the output and the phase map corresponding to the true aberration are used as the loss function for gradient descent training.

[0079] Step S31 and step S32 are iterated alternately until a predetermined number of iterations are completed. After the encoder, decoder, and regressor are trained, a light field aberration correction network is used. The encoder receives a noisy aberration light field, the decoder outputs a denoised aberration light field, and the regressor outputs a predicted aberration. The decoder and regressor predict simultaneously because after training, the encoder learns to encode information into an implicit space. The decoder can decode and reconstruct a denoised aberration light field from the implicit space, and the regressor can estimate the aberration corresponding to the light field image. The denoised aberration light field and the predicted aberration can be used for subsequent light field reconstruction.

[0080] In summary, the embodiments of the present invention can utilize the implicit space learned by the encoder to improve the accuracy of light field aberration estimation in the light field microscopy system, process noisy light field data, and correct light field aberrations, thereby improving the quality of light field reconstruction.

[0081] Next, the light field aberration correction device according to the embodiment of the present invention will be described with reference to the accompanying drawings.

[0082] FIG5 is a block diagram of a light field aberration correction device according to an embodiment of the present invention.

[0083] As shown in FIG. 5 , the light field aberration correction device 10 includes an acquisition module 100 , an input module 200 and a correction module 300 .

[0084] Among them, the acquisition module 100 is used to obtain the current light field data; the input module 200 is used to input the current light field data into the light field aberration correction network based on the implicit space, and the light field aberration correction network outputs the light field aberration of the current light field data and the denoised current light field data; the correction module 300 is used to correct the light field aberration of the denoised current light field data.

[0085] In an embodiment of the present invention, a light field aberration correction network includes an encoder, a decoder and a regressor, wherein the encoder encodes the current light field data into an implicit space; the decoder decodes the implicit space to obtain the denoised current light field data; and the regressor extracts the light field aberration of the current light field data from the implicit space.

[0086] In an embodiment of the present invention, the input module 200 is further used to: obtain original training data carrying noise and light field aberrations; add target noise to the original training data to form new training data; train the encoder and decoder according to the new training data, train the encoder and regressor according to the original training data, and perform alternating iterative training, wherein the encoder constructs an implicit space during the training process and uses the implicit space to distinguish between noise and light field aberrations; if the training stop condition is met, the alternating iterative training is stopped, and after the training is completed, a light field aberration correction network based on the implicit space is obtained.

[0087] In this embodiment of the present invention, the new training data The formation formula is:

[0088] Where x is the light field data with noise and aberration, D is the reversible matrix, D = αI, α is an arbitrary non-zero constant, I is the unit matrix, D T is the transpose of the reversible matrix,

[0089] , where z is a zero-mean Gaussian random vector in the noisy light field data, σ 2 =β1H(xb)+β2,

[0090] Among them, σ 2 is the background variance of the light field data, b is the fixed background caused by the sample and the camera, H is a linear low-pass filter, and β1 and β2 are constants.

[0091] It should be noted that the above explanation of the embodiment of the light field aberration correction method is also applicable to the light field aberration correction device of this embodiment, and will not be repeated here.

[0092] The light field aberration correction device proposed in an embodiment of the present invention can quickly estimate light field aberrations only through a light field aberration correction network based on implicit space, without the need for additional optical hardware. This solves the problem of inaccurate modeling of software correction methods based on post-calculation, and corrects the light field aberrations of the current light field data after denoising. This can achieve fast and accurate correction of light field aberrations, effectively improve the accuracy of light field data, and thus improve the quality of light field reconstruction.

[0093] FIG6 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device may include:

[0094] A memory 601 , a processor 602 , and a computer program stored in the memory 601 and executable on the processor 602 .

[0095] When the processor 602 executes the program, the light field aberration correction method provided in the above embodiment is implemented.

[0096] Furthermore, the electronic device further includes:

[0097] The communication interface 603 is used for communication between the memory 601 and the processor 602 .

[0098] The memory 601 is used to store computer programs that can be run on the processor 602 .

[0099] The memory 601 may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.

[0100] If the memory 601, processor 602, and communication interface 603 are implemented independently, the communication interface 603, memory 601, and processor 602 can be interconnected via a bus and communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus. Buses can be divided into address buses, data buses, control buses, etc. For ease of illustration, FIG6 shows only one thick line, but this does not mean that there is only one bus or only one type of bus.

[0101] Optionally, in a specific implementation, if the memory 601, the processor 602 and the communication interface 603 are integrated on a chip, the memory 601, the processor 602 and the communication interface 603 can communicate with each other through an internal interface.

[0102] The processor 602 may be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0103] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above light field aberration correction method when executed by a processor.

[0104] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples without contradiction.

[0105] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "N" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0106] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or N executable instructions for implementing a custom logical function or step of a process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0107] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logical functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array, a field programmable gate array, etc.

[0108] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0109] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for correcting light field aberration, characterized in that Including the following steps: Obtain the current light field data; Input the current light field data into an implicit space-based light field aberration correction network, and the light field aberration correction network outputs the light field aberration of the current light field data and the denoised current light field data; Correct the light field aberration of the denoised current light field data.

2. The optical field aberration correction method according to claim 1, wherein The light field aberration correction network includes an encoder, a decoder, and a regressor, where The encoder encodes the current light field data into an implicit space; The decoder decodes from the implicit space to obtain the denoised current light field data; The regressor extracts the light field aberration of the current light field data from the implicit space.

3. The light field aberration correction method according to claim 2, wherein Before inputting the current light field data into the implicit space-based light field aberration correction network, it further includes: Obtain the original training data carrying noise and light field aberration; Add target noise to the original training data to form new training data; Train the encoder and the decoder according to the new training data, train the encoder and the regressor according to the original training data, and perform alternating iterative training, where the encoder constructs an implicit space during the training process and uses the implicit space to distinguish noise and light field aberration; If the training stop condition is satisfied, stop the alternating iterative training, and obtain an implicit space-based light field aberration correction network after the training is completed.

4. The light field aberration correction method according to claim 3, wherein The new training data is formed according to the formula: Among them, x is the light field data with noise and aberration, D is an invertible matrix, D = αI, where α is an arbitrary non-zero constant and I is the identity matrix. D T is the transpose of the invertible matrix, z ∼ N(0, σ 2 I), Where z is a zero-mean Gaussian random vector in the noisy light field data, σ 2 = β1H(x - b)+β2, where σ 2 is the background variance of the light field data, b is the fixed background caused by the sample and the camera, and H is the linear low-pass filter Wave filter, and β1 and β2 are constants.

5. An optical field aberration correction device, characterized in that, Including: An acquisition module for obtaining the current light field data; An input module for inputting the current light field data into an implicit space-based light field aberration correction network, and the light field aberration correction network outputs the light field aberration of the current light field data and the denoised current light field data; A correction module for correcting the light field aberration of the denoised current light field data.

6. The optical field aberration correction device according to claim 5, wherein The light field aberration correction network includes an encoder, a decoder, and a regressor, where The encoder encodes the current light field data into an implicit space; The decoder decodes from the implicit space to obtain the denoised current light field data; The regressor extracts the light field aberration of the current light field data from the implicit space.

7. The optical field aberration correction device according to claim 6, wherein The input module is further used for: Obtain the original training data carrying noise and light field aberration; Add target noise to the original training data to form new training data; Train the encoder and the decoder according to the new training data, train the encoder and the regressor according to the original training data, and perform alternating iterative training, where the encoder constructs an implicit space during the training process and uses the implicit space to distinguish noise and light field aberration; If the training stop condition is satisfied, stop the alternating iterative training, and obtain an implicit space-based light field aberration correction network after the training is completed.

8. The light field aberration correction device according to claim 7, wherein The new training data is formed according to the formula: Among them, x is the light field data with noise and aberration, D is an invertible matrix, D = αI, where α is an arbitrary non-zero constant and I is the identity matrix. D T is the transpose of the invertible matrix, z ∼ N(0, σ 2 I) ， Where z is a zero-mean Gaussian random vector in the noisy light field data, σ 2 = β1H(x - b)+β2 ， where σ 2 is the background variance of the light field data, b is the fixed background caused by the sample and the camera, and H is a linear low-pass filter Wave filter, and β1 and β2 are constants.

9. An electronic device, characterized in that, Including: A memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the program to implement the light field aberration correction method according to any one of claims 1-4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by a processor to implement the light field aberration correction method according to any one of claims 1-4.

Images