An image processing method based on machine learning data completion algorithm

Abstract

The present invention relates to an image processing method based on a machine learning data completion algorithm, comprising: S10, obtaining a pixel matrix of an image to be processed input into a computer; S20, constructing an initial completion matrix and Lagrange multipliers based on pixel point subscripts with known pixel values ​​in the pixel matrix; S30, based on a Lagrange function model, using a data completion algorithm to iterate the completion matrix and Lagrange multipliers for multiple times, solving the saddle point of the Lagrange function model, and obtaining a final completion matrix; wherein the data completion algorithm comprises: calculating a correction step length after obtaining an initial iteration result based on each iteration process of the primal dual algorithm, performing variable step length correction on the initial iteration result according to the correction step length, and obtaining a corrected iteration result of the iteration process, so as to improve the speed of the computer completing the image to be processed. The method of the present invention improves the computational efficiency of the existing data completion method.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an image processing method based on a machine learning data completion algorithm. Background Art

[0002] As a carrier of information, images play a very important role in both information acquisition and information transmission. However, during the process of image acquisition, processing, storage or transmission, due to problems with the acquisition equipment itself or the long transmission distance, the transmission channel is subject to greater interference, which may cause image damage and loss of some information. For images, the lost information is basically the pixel values ​​of certain pixels in the image. In order to ensure that the quality of the damaged image can meet the use requirements and retain the main features of the original clear image as much as possible, the damaged image is usually supplemented before use.

[0003] For the damaged image to be processed, one type of image processing method is to regard it as equivalent to a low-rank matrix completion problem, and use a data completion algorithm to solve the problem, so as to complete the missing pixel values ​​of the image to be processed, so as to achieve the purpose of repairing and completing the image to be processed. Among them, the primal-dual algorithm is a relatively mature data completion algorithm.

[0004] However, for practical application problems of images with high pixel counts, the computational efficiency of the image processing process using the primal-dual algorithm is low, which reduces the speed at which the computer completes the image to be processed, and the real-time performance in practical applications is poor. In particular, it performs poorly in some application scenarios with high real-time requirements such as processing video streams and real-time image data. In order to improve the computational efficiency of the computer, He et al. proposed a method of increasing the correction acceleration step in each iteration to improve the computational efficiency based on the iterative framework of the primal-dual algorithm (He BS, Yuan X M. Convergence analysis of primal-dual algorithms for a saddle-point problem: from contraction perspective. SIAM Journal on Imaging Sciences, 2012, 5: 1190-149), but the correction step size of this correction acceleration step is fixed, so the effect of improving the computational efficiency of the computer is limited. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides an image processing method based on a machine learning data completion algorithm, which improves the computational efficiency of the existing data completion method and increases the speed at which the computer completes the image to be processed.

[0007] (II) Technical solution

[0008] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0009] The embodiment of the present invention provides an image processing method based on a machine learning data completion algorithm, comprising:

[0010] S10. Obtain a pixel matrix of an image to be processed input into a computer; the image to be processed is an image in which pixel values ​​of certain pixels are missing and need to be supplemented;

[0011] S20, constructing an initial completion matrix and a Lagrange multiplier associated with the completion matrix based on pixel point subscripts with known pixel values ​​in the pixel matrix;

[0012] S30, based on a preset Lagrangian function model, using a data completion algorithm to iterate the completion matrix and Lagrangian multipliers for multiple times, solving the saddle point of the Lagrangian function model, and obtaining a final completion matrix, wherein the final completion matrix is ​​the pixel matrix of the estimated image after the completion of the image to be processed; wherein,

[0013] The data completion algorithm includes: each iteration process based on the primal-dual algorithm, after obtaining the initial iteration result of the iteration process, calculating the correction step size of the iteration process based on the corrected iteration result of the previous iteration process and the initial iteration result of the iteration process, performing variable step size correction on the initial iteration result according to the correction step size, obtaining the corrected iteration result of the iteration process, and using it for the next iteration process to improve the speed of the computer to complete the image to be processed.

[0014] Optionally, in S30, the data completion algorithm is used to perform multiple iterations on the completion matrix and the Lagrange multiplier based on a preset Lagrangian function model, including:

[0015] The Lagrangian function model is:

[0016] L(X, Y) = ||X|| * - <Y,P Ω (X)-P Ω (M)> (1);

[0017] In formula (1), M represents the pixel matrix of the image to be processed, represents the completion matrix of the estimated image, Y represents the Lagrange multiplier associated with the completion matrix X, ||X|| * represents the nuclear norm of the completed matrix X, Represents the set of pixel subscripts with known pixel values ​​in the pixel matrix M of the image to be processed, P Ω (X) represents the orthogonal projection of the completion matrix X on the set Ω, P Ω (M) represents the orthogonal projection of the pixel matrix M of the image to be processed on the set Ω, <Y,P Ω (X)-P Ω (M)> represents the Lagrange multiplier Y and the formula P Ω (X)-P Ω (M) is the inner product of

[0018] Optionally, in S30, the correction step size of the current iteration process is calculated based on the corrected iteration result of the previous iteration process and the initial iteration result of the current iteration process, including:

[0019] Calculate the correction step size of the iteration process according to formula (2);

[0020] The formula (2) is:

[0021]

[0022] In formula (2), W k Represents the iterative result of the correction of the last iterative process, Represents the initial iteration result of the current iteration process, I n and I m represents the identity matrix, k represents the number of iterations, τ∈[-1, 1) represents the combination parameter, and t>0, s>0 represent the proximal parameters of the regularization term.

[0023] Optionally, the S30 includes:

[0024] S301, iterative result based on the correction of the previous iterative process And the Lagrangian function model, according to formula (3) calculate the initial iteration result of this iteration process

[0025] The formula (3) is:

[0026]

[0027] In formula (3), k represents the number of iterations, τ∈[-1, 1) represents the combination parameter, r>0, s>0 represents the proximal parameter of the regularization term;

[0028] S302: Calculate the correction step length of this iteration process according to formula (2)

[0029] S303, based on the correction step length of this iteration process and the initial iteration results According to formula (4), the corrected iterative result of this iterative process is determined

[0030] The formula (4) is:

[0031]

[0032] In formula (4), γ∈(0,2) represents the combination parameter;

[0033] S304: Determine the corrected iteration result W of this iteration process. k+1 Whether the iteration stop condition is met,

[0034] If so, the corrected iteration result W in the current iteration process is k+1 The completion matrix X in k+1 As the final completion matrix output;

[0035] If not, jump to S301.

[0036] Optionally, in S304, the iterative result W of the correction of the current iterative process is determined. k+1 Whether the iteration stop conditions are met, including:

[0037] The iteration stop condition is:

[0038]

[0039] In formula (5), g represents the preset error accuracy.

[0040] Optionally, the value range of g is (0, 0.0001].

[0041] Optionally, before S301, the method further includes:

[0042] S300, initializing the initial completion matrix X and Lagrange multiplier Y and filling in the preset matrix Get the starting matrix for the first iteration.

[0043] Optionally, the image to be processed includes: any one of a medical image, an aerial image, and an industrial image acquired by an acquisition device, transmitted by a transmission device, or saved by a storage device.

[0044] (III) Beneficial effects

[0045] The image processing method based on the machine learning data completion algorithm proposed in the embodiment of the present invention constructs a completion matrix and associated Lagrange multipliers for the pixel matrix of the image to be processed, and uses the data completion algorithm based on the primal dual algorithm based on the preset Lagrangian function model. After obtaining the initial iteration result of the current iteration process, the correction step length of the current iteration process is calculated based on the corrected iteration result of the previous iteration process and the initial iteration result of the current iteration process, and the initial iteration result is corrected by a variable step length according to the correction step length to obtain the corrected iteration result of the current iteration process, and the saddle point of the Lagrangian function model is solved to obtain the final completion matrix, that is, the pixel matrix of the estimated image after the completion of the image to be processed. The above-mentioned iterative process based on variable step length correction can make the iteration result converge quickly, thereby improving the computing efficiency of the computer and the speed of the computer completing the image to be processed, especially having good advantages in image processing scenarios such as video streams and real-time image data. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic diagram of a flow chart of an image processing method based on a machine learning data completion algorithm provided in an embodiment;

[0047] Figure 2 A schematic diagram of the process of S30 provided in the embodiment;

[0048] Figure 3 The numerical results of the image processing methods based on PDA and MPDA in the embodiment with a sampling rate of 0.3 in terms of the number of iterations, CPU time, and algorithm accuracy are compared;

[0049] Figure 4 The numerical results of the image processing methods based on PDA and MPDA in the embodiment with a sampling rate of 0.4 in terms of the number of iterations, CPU time and algorithm accuracy are compared;

[0050] Figure 5 The numerical results of the image processing methods based on PDA and MPDA in the embodiment with a sampling rate of 0.5 in terms of the number of iterations, CPU time and algorithm accuracy are compared;

[0051] Figure 6 A comparison diagram of numerical results of the image processing methods based on PDA and MPDA in the embodiments with respect to CPU time;

[0052] Figure 7 It is a descending curve diagram of the relationship between algorithm accuracy and CPU time of the image processing method based on PDA and MPDA in the embodiment;

[0053] Figure 8It is a decline curve diagram of the relationship between algorithm accuracy and the number of iterations for the image processing method based on PDA and MPDA in the embodiment;

[0054] Fig. 9 This is a comparison diagram of the restoration effects of the image processing methods based on PDA and MPDA for an image of size 256×256 in the embodiment;

[0055] Fig.10 This is a comparison diagram of the restoration effects of the image processing methods based on PDA and MPDA for an image of size 512×512 in the embodiment;

[0056] Fig.11 The numerical results of the image processing methods based on PDA and MPDA in the embodiment for the image with a size of 256×256 in terms of the number of iterations, CPU time and algorithm accuracy are compared;

[0057] Fig.12 The numerical results of the image processing methods based on PDA and MPDA for an image of size 512×512 in the embodiment are compared in terms of the number of iterations, CPU time and algorithm accuracy. DETAILED DESCRIPTION

[0058] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0059] In order to facilitate understanding of the image processing process in the specific implementation mode, the symbols involved in the steps of the image processing method are explained below:

[0060] represents the set of all m×n real matrices;

[0061] X ij represents the (i, j) element of the matrix X;

[0062] X T Represents the transpose operation of matrix X;

[0063] r = rank(X) represents the rank of matrix X;

[0064] Tr(X) represents the trace of matrix X;

[0065] represents the nuclear norm of the matrix X; where σ k (X) is the kth largest singular value of the matrix X, that is, σ1(X)≥σ2(X≥…≥σ k (X)≥…σ r (X)>0;

[0066] represents the Frobenius norm of matrix X;

[0067] I represents the identity matrix;

[0068] <A,B> :=tr(A T B) represents the inner product operation of matrices A and B;

[0069] P Ω represents the orthogonal projection on the set Ω, satisfying

[0070] in, Represents the set of pixel subscripts with known pixel values ​​in the pixel matrix M of the image to be processed.

[0071] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0072] Embodiment 1

[0073] This embodiment provides an image processing method based on a machine learning data completion algorithm. The executor of the method of this embodiment is a computer device.

[0074] like Figure 1 As shown, the image processing method provided in this embodiment includes the following steps:

[0075] S10, obtaining a pixel matrix of an image to be processed input into a computer; the image to be processed is an image in which pixel values ​​of certain pixels are missing and need to be supplemented.

[0076] Specifically, the image to be processed includes any one of medical images, aerial images, and industrial images collected by an acquisition device, transmitted by a transmission device, or stored by a storage device. The medical image may specifically be a CT (Computed Tomography) image, an MR (Magnetic Resonanc) image, a PET (Positron Emission Computed Tomography) image, etc. The aerial image may specifically be a visible light remote sensing image, a full-color remote sensing image, a multi-spectral remote sensing image, an infrared remote sensing image, a hyperspectral remote sensing image, a lidar remote sensing image, a synthetic aperture radar remote sensing image, etc. The industrial image is an image used for industrial vision, based on optical equipment, control systems, and actuators, and is obtained for the purpose of replacing the human eye to complete the recognition, measurement, positioning, detection, and other tasks in production and manufacturing.

[0077] S20, constructing an initial completion matrix and a Lagrange multiplier associated with the completion matrix based on pixel point indices with known pixel values ​​in the pixel matrix.

[0078] S30. Based on a preset Lagrangian function model, use a data completion algorithm to iterate the completion matrix and Lagrangian multipliers for multiple times, solve the saddle point of the Lagrangian function model, and obtain a final completion matrix, which is the pixel matrix of the estimated image after the image to be processed is completed.

[0079] Among them, the data completion algorithm includes: based on each iteration process of the primal dual algorithm, after obtaining the initial iteration result of the current iteration process, the correction step size of the current iteration process is calculated based on the corrected iteration result of the previous iteration process and the initial iteration result of the current iteration process, and the initial iteration result is subjected to variable step length correction according to the correction step length to obtain the corrected iteration result of the iteration process, which is used for the next iteration process to improve the speed of the computer to complete the image to be processed.

[0080] The image processing method based on the machine learning data completion algorithm proposed in the embodiment of the present invention constructs a completion matrix and associated Lagrange multipliers for the pixel matrix of the image to be processed, and uses the data completion algorithm based on the primal dual algorithm based on the preset Lagrangian function model. After obtaining the initial iteration result of the current iteration process, the correction step length of the current iteration process is calculated based on the corrected iteration result of the previous iteration process and the initial iteration result of the current iteration process, and the initial iteration result is corrected by a variable step length according to the correction step length to obtain the corrected iteration result of the current iteration process, and the saddle point of the Lagrangian function model is solved to obtain the final completion matrix, that is, the pixel matrix of the estimated image after the completion of the image to be processed. The above-mentioned iterative process based on variable step length correction can make the iteration result converge quickly, thereby improving the computing efficiency of the computer and the speed of the computer completing the image to be processed, especially having good advantages in image processing scenarios such as video streams and real-time image data.

[0081] Embodiment 2

[0082] In order to better understand the image processing method in the first embodiment, this embodiment describes the specific steps in detail.

[0083] S10, obtaining a pixel matrix M of an image to be processed input into a computer; the image to be processed is an image in which pixel values ​​of certain pixels are missing and need to be supplemented.

[0084] S20: construct an initial completion matrix X and a Lagrange multiplier Y associated with the completion matrix X based on the pixel point indices with known pixel values ​​in the pixel matrix M.

[0085] The size of the completion matrix X is the same as that of the pixel matrix M, and for a pixel point whose pixel value in the pixel matrix M is known, the pixel value of the pixel point is directly copied to the corresponding pixel point in the completion matrix X.

[0086] S30. Based on a preset Lagrangian function model, use a data completion algorithm to iterate the completion matrix and Lagrangian multipliers for multiple times, solve the saddle point of the Lagrangian function model, and obtain a final completion matrix, which is the pixel matrix of the estimated image after the image to be processed is completed.

[0087] Specifically, the Lagrangian function model is:

[0088] L(X,Y)=||X|| * - <Y,P Ω (X)-P Ω (M)> (1);

[0089] In formula (1), M represents the pixel matrix of the image to be processed, represents the completion matrix of the estimated image, Y represents the Lagrange multiplier associated with the completion matrix X, ||X|| * represents the nuclear norm of the completed matrix X, Represents the set of pixel subscripts with known pixel values ​​in the pixel matrix M of the image to be processed, P Ω (X) represents the orthogonal projection of the completion matrix X on the set Ω, P Ω (M) represents the orthogonal projection of the pixel matrix M of the image to be processed on the set Ω, <Y,P Ω (X)-P Ω (M)> represents the Lagrange multiplier Y and the formula P Ω (X)-P Ω (M) is the inner product of

[0090] like Figure 2 As shown, based on the above Lagrangian function model, the above step S30 includes sub-steps S300 to S304, which are specifically as follows:

[0091] S300, initializing the initial completion matrix X and Lagrange multiplier Y and filling in the preset matrix Get the starting matrix for the first iteration.

[0092] S301, iterative result based on the correction of the previous iterative process And the Lagrangian function model, according to formula (3) calculate the initial iteration result of this iteration process

[0093] The formula (3) is:

[0094]

[0095] In formula (3), k represents the number of iterations, τ∈[-1,1) represents the combination parameter, and r>0, s>0 represent the proximal parameters of the regularization term.

[0096] S302: Calculate the correction step length of this iteration process according to formula (2)

[0097] The formula (2) is:

[0098]

[0099] In formula (2), W k Represents the iterative result of the correction of the last iterative process, Represents the initial iteration result of the current iteration process, I n and I mrepresents the identity matrix, k represents the number of iterations, τ∈[-1, 1) represents the combination parameter, and t>0, s>0 represent the proximal parameters of the regularization term.

[0100] S303, based on the correction step length of this iteration process and the initial iteration results According to formula (4), the corrected iterative result of this iterative process is determined

[0101] The formula (4) is:

[0102]

[0103] In formula (4), γ∈(0, 2) represents a combination parameter.

[0104] S304: Determine the corrected iteration result W of this iteration process. k+1 Whether the iteration stop condition is met,

[0105] If so, the corrected iteration result W in the current iteration process is k+1 The completion matrix X in k+1 As the final completion matrix output;

[0106] If not, jump to S301 to execute the next iteration process.

[0107] Specifically, in S304, the iterative result W of the correction of the current iterative process is determined. k+1 Whether the iteration stop conditions are met, including:

[0108] The iteration stop condition is:

[0109]

[0110] In formula (5), g represents a preset error accuracy, and the value range of g is (0, 0.0001] to ensure that the estimated image after completion has good accuracy.

[0111] In addition, in order to verify the technical effect of the image processing method provided by the embodiment of the present invention, this embodiment also carried out numerical experiments and image restoration experiments to compare the effects of the image processing method based on the data completion algorithm provided by the embodiment of the present invention (hereinafter referred to as MPDA) and the image processing method based on the existing primal-dual algorithm (hereinafter referred to as PDA).

[0112] Numerical experiments:

[0113] In this numerical experiment, an n-dimensional square matrix M is used to represent the real image. The elements in the square matrix M are filled with random numbers from a normal distribution. The value of n is set to 10000, 11000, 12000, 13000, 14000, 15000, 16000, 17000, 18000, 19000, 20000 and 30000 respectively, and the pixel matrices of 12 real images are obtained.

[0114] For the above 12 real images, sampling is performed at sampling rates p of 0.3, 0.4, and 0.5, respectively, to obtain 36 images with missing pixel values ​​to be processed, which are processed using the image processing methods based on MPDA and PDA, respectively. The subscripts of m elements of the matrix M are randomly selected to generate a set Ω, where m = p × n 2 The matrix rank is represented by r, and its values ​​are 10 and 15 respectively. Other parameters in the image processing process are set as: t = 0.005, s = 202, r = 0.2, γ = 1.5. In addition, the termination condition of the algorithm is set as:

[0115]

[0116] The precision of the algorithm is set to:

[0117]

[0118] Figures 3 to 5 The table lists the numerical results of the image processing methods based on PDA and MPDA at different sampling rates in terms of number of iterations, CPU time and algorithm accuracy. Among them, the matrix size is represented by Size(n×n), the CPU time of the image processing method is represented by time in seconds, and the number of algorithm iterations is represented by iter. In order to observe the numerical effects of the two algorithms more intuitively, Figure 6 The numerical results of the CPU time of the two algorithms are compared. It can be seen from the numerical results that compared with the image processing method based on PDA, the image processing method based on MPDA runs nearly twice as fast and is more accurate. Figures 3 to 5 In the table, the minimum time difference between the two image processing methods is 70 seconds, and the maximum time difference is 2008 seconds. As the matrix size (that is, the number of image pixels) increases, the time difference between the two image processing methods becomes more and more obvious, which also verifies the speed advantage of variable step correction acceleration in the MPDA-based image processing method for completing high-pixel images.

[0119] Figure 7 and Figure 8The decline curves of the algorithm accuracy of the two image processing methods with respect to CPU time and number of iterations are given, where n = 10000, 20000, and p = 0.3, 0.4, and 0.5. The image processing method based on MPDA outperforms the existing image processing method based on PDA in terms of both CPU time and number of iterations, and can make the completion matrix converge to the optimal solution more quickly.

[0120] Image restoration experiment:

[0121] like Fig. 9 (a) and Fig.10 As shown in (a), two high-quality images of 256 pixels × 256 pixels and 512 pixels × 512 pixels are taken as the original real images. The two real images are sampled at sampling rates p of 0.1, 0.3, and 0.5 respectively, and the following are obtained: Fig. 9 (b)(e)(h) and Fig.10 (b)(e)(h) show two groups of images to be processed. Based on the same experimental parameters as the above numerical experiments, the MPDA-based and PDA-based image processing methods are used to process the images to be processed. The completed estimated images obtained by the PDA-based image processing method are shown in Figure 2. Fig. 9 (c)(f)(i) and Fig.10 As shown in (c)(f)(i), the completed estimated image obtained by the image processing method based on MPDA is as follows Fig. 9 (d)(g)(j) and Fig.10 (d)(g)(j) are shown. At the same time, the signal-to-noise ratio (SNR) is used to evaluate the restoration effect, and the numerical comparison results of the number of iterations, CPU time and signal-to-noise ratio SNR of the two image processing methods are shown in Fig.11 and Fig.12 As shown. Among them, the expression of signal-to-noise ratio SNR is:

[0122]

[0123] Among them, M represents the pixel matrix of the image to be processed, X k+1 represents the pixel matrix of the completed image, and SNR represents the signal-to-noise ratio.

[0124] Fig. 9 and Fig.10 The experimental results show that the MPDA-based image processing method has a good restoration effect on image restoration problems of different image sizes and different sampling rates. Fig.11 and 12The experimental results in the table show that compared with the image processing method based on PDA, the image processing method based on MPDA has fewer iterations, shorter CPU time, and higher SNR value. Overall, as the sampling rate increases, the image restoration time of the two algorithms is shorter and the SNR value is higher.

[0125] Embodiment 3

[0126] Based on the image processing method provided in Embodiment 1 or 2, this embodiment further provides an image processing system based on a machine learning data completion algorithm, including:

[0127] The acquisition module is used to acquire the pixel matrix of the image to be processed input into the computer; the image to be processed is: an image in which the pixel values ​​of some pixels are missing and need to be supplemented.

[0128] The pre-construction module is used to construct an initial completion matrix and a Lagrange multiplier associated with the completion matrix based on the pixel point subscripts with known pixel values ​​in the pixel matrix.

[0129] The completion module is used to iterate the completion matrix and Lagrange multipliers multiple times based on a preset Lagrangian function model using a data completion algorithm to solve the saddle point of the Lagrangian function model to obtain a final completion matrix, which is the pixel matrix of the estimated image after the image to be processed is completed.

[0130] Among them, the data completion algorithm includes: based on each iteration process of the primal-dual algorithm, after obtaining the initial iteration result of the current iteration process, the correction step size of the current iteration process is calculated based on the corrected iteration result of the previous iteration process and the initial iteration result of the current iteration process, and the initial iteration result is subjected to variable step-size correction according to the correction step size to obtain the corrected iteration result of the iteration process, which is used for the next iteration process to improve the speed at which the computer calculates the estimated image.

[0131] Based on the image processing method provided in Embodiment 1 or 2, this embodiment also provides an image processing device based on a machine learning data completion algorithm, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in Embodiment 1 or 2.

[0132] Based on the image processing method provided in Embodiment 1 or 2, this embodiment further provides a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the method described in Embodiment 1 or 2 are implemented.

[0133] Based on the image processing method provided in Embodiment 1 or 2, this embodiment further provides a computer program product, including a computer program / instruction, which implements the steps of the method described in Embodiment 1 or 2 when executed by a processor.

[0134] Since the system / device described in the above embodiments of the present invention is a system / device used to implement the method of the above embodiments of the present invention, a person skilled in the art can understand the specific structure and deformation of the system / device based on the method described in the above embodiments of the present invention, and thus will not be described in detail here. All systems / devices used in the method of the above embodiments of the present invention belong to the scope of protection of the present invention.

[0135] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.

[0137] It should be noted that in the claims, any reference numerals placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In the claims enumerating several means, several of these means may be embodied by the same hardware. The use of the words first, second, third, etc., is for convenience of expression only and does not indicate any order. These words may be understood as part of the component name.

[0138] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" etc. 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 may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0139] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0140] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.

Claims

1. An image processing method based on a machine learning data completion algorithm, characterized in that: include: S10, obtaining a pixel matrix of an image to be processed input into a computer; The image to be processed is an image in which the pixel values ​​of some pixels are missing and need to be supplemented; S20, constructing an initial completion matrix and a Lagrange multiplier associated with the completion matrix based on pixel point subscripts with known pixel values ​​in the pixel matrix; S30, based on a preset Lagrangian function model, using a data completion algorithm to iterate the completion matrix and Lagrangian multipliers for multiple times, solving the saddle point of the Lagrangian function model, and obtaining a final completion matrix, wherein the final completion matrix is ​​the pixel matrix of the estimated image after the completion of the image to be processed; wherein, The data completion algorithm comprises: in each iteration process based on the primal-dual algorithm, after obtaining the initial iteration result of the current iteration process, calculating the correction step length of the current iteration process based on the corrected iteration result of the previous iteration process and the initial iteration result of the current iteration process, performing variable step length correction on the initial iteration result according to the correction step length, obtaining the corrected iteration result of the current iteration process, and using it for the next iteration process, so as to improve the speed of the computer to complete the image to be processed; In S30, the completion matrix and the Lagrangian multiplier are iterated multiple times using a data completion algorithm based on a preset Lagrangian function model, including: The Lagrangian function model is: L(X,Y)=||X|| * -<Y,P Ω (X)-P Ω (M)> (1); In formula (1), M represents the pixel matrix of the image to be processed, represents the completion matrix of the estimated image, Y represents the Lagrange multiplier associated with the completion matrix X, ||X|| * represents the nuclear norm of the completed matrix X, Represents the set of pixel subscripts with known pixel values ​​in the pixel matrix M of the image to be processed, P Ω (X) represents the orthogonal projection of the completion matrix X on the set Ω, P Ω (M) represents the orthogonal projection of the pixel matrix M of the image to be processed on the set Ω, <Y,P Ω (X)-P Ω (M)> represents the Lagrange multiplier Y and the formula P Ω (X)-P Ω (M) inner product operation; In the S30, the correction step length of the current iteration process is calculated based on the corrected iteration result of the previous iteration process and the initial iteration result of the current iteration process, including: Calculate the correction step size of the iteration process according to formula (2); The formula (2) is: In formula (2), W k Represents the iterative result of the correction of the last iterative process, Represents the initial iteration result of the current iteration process, I n and I m represents the identity matrix, k represents the number of iterations, τ∈[-1,1) represents the combination parameter, t>0, s>0 represents the proximal parameter of the regularization term, express and The inner product operation of ; The S30 includes: S301, iterative result based on the correction of the previous iterative process And the Lagrangian function model, according to formula (3) calculate the initial iteration result of this iteration process The formula (3) is: In formula (3), k represents the number of iterations, τ∈[-1,1) represents the combination parameter, r=0.2, s=202 represents the proximal parameter of the regularization term; S302: Calculate the correction step length of this iteration process according to formula (2) S303, based on the correction step length of this iteration process and the initial iteration results According to formula (4), the corrected iterative result of this iterative process is determined The formula (4) is: In formula (4), γ = 1.5 represents the combination parameter; S304: Determine the corrected iteration result W of this iteration process. k+1 Whether the iteration stop condition is met, If so, the corrected iteration result W in the current iteration process is k+1 The completion matrix X in k+1 As the final completion matrix output; If not, jump to S301.

2. The image processing method according to claim 1, characterized in that: In S304, the iterative result W of the correction of the current iterative process is determined. k+1 Whether the iteration stop conditions are met, including: The iteration stop condition is: In formula (5), g represents the preset error accuracy.

3. The image processing method according to claim 2, characterized in that: The value range of g is (0, 0.0001].

4. The image processing method according to claim 1, characterized in that: Prior to S301, it also included: S300, initializing the initial completion matrix X and Lagrange multiplier Y and filling in the preset matrix Get the starting matrix for the first iteration.

5. The image processing method according to claim 1, characterized in that: The image to be processed includes any one of a medical image, a space image, and an industrial image acquired by an acquisition device, transmitted by a transmission device, or stored by a storage device.

Images