Image processing method and system

Abstract

The present invention provides an image processing method and system, and relates to the field of image processing technology. The image processing method includes the following steps: constructing an image processing model based on Taylor's theorem, then acquiring and using an observed image and a real original image to train the image processing model, and obtaining a target processing model for image deblurring. Then, the image to be processed is processed using the target processing model to obtain a processed image. The present invention constructs an image processing model based on Taylor's theorem, and uses a sample image to train the image processing model, which is a target processing model for image deblurring, thereby filling the gap in anti-environmental noise deblurring technology and improving image restoration quality.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image processing method and system. Background Art

[0002] Usually, during the image acquisition process, the image will be blurred due to the optical system or movement, and the image will also generate noise from circuit and photometric factors. Based on this, the deblurring scheme is ill-posed. In most cases, the main means of deblurring schemes include two methods, such as numerical iterative algorithms and neural network algorithms. However, due to the complexity of calculation, the above numerical iterative algorithms and neural network algorithms rarely take environmental noise into consideration.

[0003] Based on this, there is an urgent need for an image processing solution to fill the gap in environmental noise-resistant deblurring technology and improve the quality of image restoration. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide an image processing solution, construct an image processing model based on Taylor's theorem, and use sample images to train the image processing model, which is a target processing model for image deblurring, thereby filling the gap in anti-environmental noise deblurring technology and improving image restoration quality.

[0005] In order to achieve the above purpose, the technical solution adopted by the embodiment of the present invention is as follows:

[0006] In a first aspect, the present invention provides an image processing method, the image processing method comprising the following steps:

[0007] Build image processing models based on Taylor's theorem;

[0008] Obtain observation images and true original images for training;

[0009] The image processing model is trained based on the observed image and the real original image to obtain the target processing model for image deblurring;

[0010] An image to be processed is acquired, and the image to be processed is processed using a target processing model to obtain a processed image.

[0011] Optionally, after the step of obtaining the observed image and the real original image for training, the following step is further included:

[0012] Preprocessing the observed image to obtain the fuzzy features of the observed image;

[0013] The fuzzy feature equation is constructed based on the fuzzy features and the real original image.

[0014] Optionally, the fuzzy feature includes a fuzzy matrix and Gaussian noise; and the step of constructing a fuzzy feature equation based on the fuzzy feature and the real original image includes:

[0015] Construct the fuzzy process vector equation based on the fuzzy matrix, Gaussian noise and the real original image;

[0016] The fuzzy process vector equation is converted into a fuzzy characteristic equation to convert the deblurring process into a minimum optimization process; wherein the solution of the fuzzy characteristic equation is used to characterize the deblurred image.

[0017] Optionally, the step of training an image processing model based on an observed image and a real original image to obtain a target processing model for image deblurring includes:

[0018] The image processing model is trained using the fuzzy feature equation until the error equation corresponding to the fuzzy feature equation converges to zero, and the target processing model is obtained.

[0019] Alternatively, the fuzzy characteristic equation is expressed as:

[0020] ;

[0021] in, is the norm equation; is the regularization parameter; ) is the image vector iterated over time; is the minimization function; is the image matrix iterated over time; is the objective function; K is the first fuzzy operator; is the first column vector.

[0022] Alternatively, the error equation is expressed as:

[0023] ;

[0024] in, is the error function; ) is the image vector iterated over time; is the first-order gradient function; is the objective function; K is the first fuzzy operator; is the conjugate transpose of the first modulus operator K; is the regularization parameter; is the identity matrix; is the first column vector.

[0025] Optionally, the target processing model is represented as:

[0026]

[0027] in, is the first-order gradient function; is the objective function; For the The image matrix corresponding to the iteration is obtained; After fast Fourier transform An image matrix; is the second fuzzy operator; is the second fuzzy operator The conjugate transpose of ; is the regularization parameter; is the first characteristic matrix; is the observation matrix; is the first adjustment parameter; is the second adjustment parameter; is the i-th image matrix after fast Fourier transform; is the second-order gradient function; is the image matrix corresponding to the fast Fourier transform; After fast Fourier transform An image matrix; After fast Fourier transform An image matrix; After fast Fourier transform An image matrix; After fast Fourier transform An image matrix; is the norm equation; After fast Fourier transform An image matrix; is the convergence parameter; vec(·) is the column vectorization function.

[0028] Optionally, the steps of constructing an image processing model based on Taylor's theorem include:

[0029] A Taylor six-point expansion design formula is constructed based on the expansion formula of Taylor's theorem to obtain an image processing model, so as to discretize the error equation corresponding to the fuzzy characteristic equation through the Taylor six-point expansion design formula.

[0030] Alternatively, the expression of the Taylor six-point expansion design formula is:

[0031] ;

[0032] in, is the image vector derivative in the iteration, is the sampling interval; For the The image vector corresponding to the iteration; For the The image vector corresponding to the iteration; For the The image vector corresponding to the iteration; For the The corresponding image vector; For the The corresponding image vector; For the The corresponding image vector; The magnitude of the higher-order error terms.

[0033] In a second aspect, the present invention further provides an image processing system, the image processing system comprising:

[0034] An image acquisition module is used to acquire observation images and real original images for training; and to acquire images to be processed;

[0035] Building blocks for constructing image processing models based on Taylor's theorem;

[0036] The processing module is used to train an image processing model based on an observed image and a real original image to obtain a target processing model for image deblurring; and use the target processing model to process the image to be processed to obtain a processed image.

[0037] The present invention provides an image processing method and system, which have the following beneficial effects:

[0038] The image processing method of the present invention comprises the following steps: constructing an image processing model based on Taylor's theorem, then acquiring and using the observed image and the real original image to train the image processing model, and obtaining a target processing model for image deblurring. Then, the image to be processed is processed using the target processing model to obtain a processed image. The present invention constructs an image processing model based on Taylor's theorem, and uses a sample image to train the image processing model, which is a target processing model for image deblurring, thereby filling the gap in the anti-environmental noise deblurring technology and improving the image restoration quality.

[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 One of the step flow charts of the image processing method provided by an embodiment of the present invention is shown;

[0042] Figure 2The second step flow chart of the image processing method provided by the embodiment of the present invention is shown;

[0043] Figure 3 A step-by-step flow chart of step 400 provided in an embodiment of the present invention is shown;

[0044] Figure 4 A schematic diagram showing the structure of a server provided by an embodiment of the present invention is shown;

[0045] Figure 5 A schematic diagram of the structure of an image processing system provided by an embodiment of the present invention is shown.

[0046] Icons: 30 - image processing system; 31 - image acquisition module; 32 - construction module; 33 - processing module. DETAILED DESCRIPTION

[0047] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0048] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0049] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0050] As described in the background technology, image deblurring technology has become the key to improving the quality of computer vision image processing. Image blur often occurs when the camera shakes and the observation of moving objects are out of focus, resulting in poor image quality, loss of texture and many important information. Existing deblurring technologies include numerical iterative algorithms and neural networks.

[0051] Taking the traditional numerical iterative algorithm as an example, the numerical iterative algorithm needs to perform block matching on complex images, which will lead to large errors. In addition, the above deblurring technologies rarely consider the influence of environmental noise.

[0052] Based on this, the present invention provides an image processing solution to fill the gap in anti-environmental noise deblurring technology and improve image restoration quality.

[0053] The above image processing scheme will be introduced in detail below.

[0054] Please refer to Figure 1 , Figure 1 A flow chart showing the steps of an image processing method in an embodiment of the present invention is shown; the image processing method comprises step 100, step 200, step 400 and step 500.

[0055] Step 100: construct an image processing model based on Taylor's theorem.

[0056] Step 200: Obtain an observed image and a real original image for training.

[0057] Step 400: Train an image processing model based on the observed image and the real original image to obtain a target processing model for image deblurring.

[0058] Step 500: Acquire an image to be processed, and process the image to be processed using a target processing model to obtain a processed image.

[0059] This embodiment constructs an image processing model based on Taylor's theorem, then obtains and uses the observed image and the real original image to train the image processing model, and obtains a target processing model for image deblurring. Then, the target processing model is used to process the image to be processed to obtain the processed image. The present invention constructs an image processing model based on Taylor's theorem, and uses sample images to train the image processing model, which is a target processing model for image deblurring, thereby filling the gap in anti-environmental noise deblurring technology and improving image restoration quality.

[0060] In this embodiment, step 100 is specifically as follows:

[0061] Based on the expansion formula of Taylor's theorem, a Taylor-6-instant design formula is constructed to obtain an image processing model, so as to discretize the error equation corresponding to the fuzzy characteristic equation through the Taylor-6-instant design formula.

[0062] Among them, the expression of Taylor's six-point expansion design formula is:

[0063] ;

[0064] in, is the image vector derivative in the iteration, is the sampling interval; For the The image vector corresponding to the iteration; For the The image vector corresponding to the iteration; For the The image vector corresponding to the iteration; For the The corresponding image vector; For the The corresponding image vector; For the The corresponding image vector; The magnitude of the higher-order error terms.

[0065] In one possible implementation, in order to achieve the above purpose, the zeroing neurodynamics ZD design evolution formula can be applied to convert the error equation into a general expression formula based on zeroing neurodynamics ZD:

[0066] ;

[0067] in, is the image vector derivative in the iteration; ) is the image vector in iteration; is the objective function; (.) -1 is the reciprocal function; is the second-order gradient function; is the regularization parameter, which is used to balance the fidelity term and the regularization term and satisfies: 5.0157*10 -5 ; g is the first column vector; is a second-order gradient, satisfying: ; K is the first fuzzy operator, that is, the fuzzy operator with cyclic block and cyclic block Toeplitz matrix on the real image matrix X(t); is the conjugate transpose of the first fuzzy operator K.

[0068] In order to avoid the influence of external disturbances and ensure that the solution meets the solution requirements, the general expression formula based on zeroing neural dynamics ZD can be converted into an expression for noise-resistant zeroing neural dynamics NTZD:

[0069] in, is the sampling period.

[0070] Based on the above Taylor six-point expansion design formula, the above expression can be discretized to obtain the image processing model in this embodiment to avoid the influence of external disturbances so that the obtained solution meets the requirements. Correspondingly, the discretized expression satisfies:

[0071] ;

[0072] in, is the image vector corresponding to the k+1th iteration; is the image vector corresponding to the i-th iteration; is the second-order gradient function; The first adjustment parameter; is the second adjustment parameter; the first adjustment parameter , the second adjustment parameter All of them are adjustable positive parameters, and the corresponding calculation formula satisfies: ; is the sampling interval; It is the third adjustment parameter and is an adjustable positive parameter.

[0073] After constructing the image processing model, it is necessary to use sample parameters, such as the observed image and the real original image used for training, to train the image processing model and obtain the target processing model for image deblurring. In order to reduce the complexity of image processing (image denoising) and improve the speed of image processing, Figure 1 Based on Figure 2 , Figure 2 Another step flow chart of the image processing method in the embodiment of the present invention is shown, that is, after the step 200 of obtaining the observed image and the real original image for training, it also includes step 300, and step 300 is: preprocessing the observed image and the real original image for training.

[0074] Please continue to refer to Figure 2 In this embodiment, step 300 includes step 301 and step 302.

[0075] Step 301: preprocess the observed image to obtain the fuzzy features of the observed image.

[0076] Step 302: construct a fuzzy feature equation based on the fuzzy feature and the real original image.

[0077] In this embodiment, the observed image is different from the real original image due to environmental noise and blur. It is assumed that the observed image is represented as: =κ When X+V, ∈ represents the observed image, ∈ is the real original image, ∈ is the additional Gaussian noise, ⊗ represents the double convolution operation, κ∈ is a fuzzy matrix. For the convenience of expression, some variables can be vectorized and represented as: the first column vector g=vec(G)∈ , the second column vector x=vec(X)∈ , the third column vector v=vec(V)∈ , where vec(·) represents column vectorization, and the first column vector g represents the observed image Column vectorization; the second column vector x represents the true original image Column vectorization; the third column vector v represents the addition of Gaussian noise Column vectorization.

[0078] Based on this, the first column vector g satisfies: g=Kx+v. Where K∈ is a cyclic block and cyclic block Toeplitz matrix blur operator on the real original image X, K can be a block Toeplitz matrix representing the linear convolution operator κ(i,j), and each submatrix K(i)∈ It is a circulant matrix. In this embodiment, the first fuzzy operator K can be expressed as:

[0079] ;

[0080] Further, in Figure 2 Based on Figure 3 , Figure 3 A step-by-step flow chart of step 400 in an embodiment of the present invention is shown; in this embodiment, the fuzzy features include a fuzzy matrix and Gaussian noise, and step 400 of constructing a fuzzy feature equation based on the fuzzy features and the real original image includes steps 401 and 402.

[0081] Step 401: construct a fuzzy process vector equation based on the fuzzy matrix, Gaussian noise and the real original image.

[0082] Step 402: Convert the fuzzy process vector equation into a fuzzy characteristic equation to convert the defuzzification process into a minimum optimization process.

[0083] The solution of the fuzzy characteristic equation is used to characterize the deblurred image.

[0084] This embodiment can simplify the defuzzification method to a minimum optimization problem. That is, the expression of the first column vector g can be converted into a first formula, which is expressed as:

[0085] ;

[0086] in, is a minimization function used to obtain the variable value that minimizes the objective function. is the norm equation.

[0087] In order to avoid the problem of large error fluctuations in the inverse problem caused by noise and other reasons, this embodiment can add constraints by using additional real original images through regularization methods. The above constraints can be Lagrange constraints, Tikhonov regularization, etc. At the same time, in order to reduce the complexity of the model, Tikhonov regularization constraints can be used, and the quadratic function of the real original image can be used as regularization, that is, the above first formula can be converted into the second formula, and the second formula is expressed as:

[0088] ;

[0089] in, is the regularization parameter, which is used to balance the fidelity term and the regularization term and satisfies: 5.0157*10 -5 .

[0090] For quadratic problems, linear equations can be directly solved by taking the gradient. To increase the computation speed, block circulant matrices can be solved in the Fourier domain.

[0091] For example, the matrix is ​​diagonalized using a 2-D discrete Fourier transform (DFT) or a fast Fourier transform (FFT).

[0092] In this embodiment, for the convenience of analysis, the second formula can be converted into a time-related form, that is, a fuzzy characteristic equation. In this embodiment, the fuzzy characteristic equation is expressed as:

[0093] ;

[0094] in, is the norm equation; ) is the image vector in iteration; is the image matrix in iteration; is the objective function; K is the fuzzy operator with cyclic block and cyclic block Toeplitz matrix on the real image matrix X(t).

[0095] Correspondingly, the error equation of the fuzzy characteristic equation is expressed as:

[0096] ;

[0097] in, is the error function; is the identity matrix.

[0098] Based on this, in this embodiment, the step 400 of training the image processing model based on the observed image and the real original image to obtain the target processing model for image deblurring is specifically as follows:

[0099] The image processing model is trained using the fuzzy feature equation until the error equation corresponding to the fuzzy feature equation converges to zero, and the target processing model is obtained. Among them, the target processing model satisfies:

[0100]

[0101] in, is the first-order gradient function; is the objective function; For the The image matrix corresponding to the iteration is obtained; After fast Fourier transform An image matrix; is the second fuzzy operator; is the second fuzzy operator The conjugate transpose of ; is the regularization parameter; is the first characteristic matrix; is the observation matrix; is the first adjustment parameter; is the second adjustment parameter; is the i-th image matrix after fast Fourier transform; is the second-order gradient function; is the image matrix corresponding to the fast Fourier transform; After fast Fourier transform An image matrix; After fast Fourier transform An image matrix; After fast Fourier transform An image matrix; After fast Fourier transform An image matrix; is the norm equation; After fast Fourier transform An image matrix; is the convergence parameter; vec(·) is the column vectorization function. In this embodiment, the convergence parameter can be 10 -13.

[0102] Step 500 is to obtain an image to be processed, and process the image to be processed using a target processing model to obtain a processed image.

[0103] This embodiment provides a solution method. It should be noted that the solution method does not limit the solution method as long as the calculation process can be realized. The corresponding calculation software includes but is not limited to Matlab and other calculation software.

[0104] In order to verify the effect of the target processing model in this embodiment, this embodiment also provides a plurality of comparative experiments. Among them, this embodiment compares the deblurring effects of 12 denoising algorithms under three different noise conditions (for example, noise-free conditions, constant noise conditions and random noise), and adopts peak signal-to-noise ratio (PSNR) and mean structural similarity (MSSIM) as evaluation indicators. The above 12 denoising algorithms include Newton iteration method (hereinafter referred to as NRI (28) for the sake of simplicity), improved Newton iteration integral method (hereinafter referred to as MNI (29) for the sake of simplicity), gradient neurodynamics (hereinafter referred to as GD (25) for the sake of simplicity), anti-noise gradient neurodynamics (hereinafter referred to as NTGD (26) for the sake of simplicity), Euler expansion zeroing neurodynamics (hereinafter referred to as EDZD (11) for the sake of simplicity), Euler expansion anti-noise zeroing neurodynamics (hereinafter referred to as EDNTZD (13) for the sake of simplicity), discrete time iteration method (hereinafter referred to as GD (25) for the sake of simplicity), and Euler expansion zeroing neurodynamics (hereinafter referred to as EDZD (11) for the sake of simplicity). The Taylor four-point expansion of noise-resistant zeroing neural dynamics (for the sake of simplicity, hereinafter referred to as DTD (12)), the Taylor five-point expansion of noise-resistant zeroing neural dynamics (for the sake of simplicity, hereinafter referred to as T4NTZD (19)), the Taylor seven-point expansion of noise-resistant zeroing neural dynamics (for the sake of simplicity, hereinafter referred to as T5NTZD), the Taylor eleven-point expansion of noise-resistant zeroing neural dynamics (for the sake of simplicity, hereinafter referred to as T7NTZD (21)), the Taylor eleven-point expansion of noise-resistant zeroing neural dynamics (for the sake of simplicity, hereinafter referred to as T11NTZD (22)) and the target processing model in this embodiment (for the sake of simplicity, hereinafter referred to as T6NTZD (23)).

[0105] Experiment 1: Normality Analysis

[0106] Please refer to Table 1, Table 2 and Table 3, which respectively represent the deblurring effect under noise-free conditions, the deblurring effect under the condition that the constant noise satisfies 100, and the deblurring effect under the condition that the random noise satisfies [50,150].

[0107] Table 1. Noise-free conditions

[0108]

[0109] Table 2. Constant noise conditions

[0110]

[0111] Table 3. Random noise conditions

[0112]

[0113] Among them, taking Table 1 as an example, it can be seen that the GD class algorithms (such as the above GD (25) and NTGD (26)) are not as good as the ZD class algorithms (such as the above EDZD (11), EDNTZD (13), DTD (12), T4NTZD (19), T5NTZD, T7NTZD (21), T11NTZD (22) and T6NTZD (23) in this embodiment) and the NI class algorithms using the second-order gradient (such as the above NRI (28) and MNI (29)). However, in terms of the MSSIM index, the NI class algorithms perform the same as the ZD class algorithms.

[0114] Taking Table 2 as an example, it can be seen that the deblurring effects of NRI (28), GD (25), EDZD (11), and DTD (12) algorithms change significantly. Taking MSSIM and PSNR as examples, NRI (28) and DTD (12) drop to about 13, and the PSNR of EDZD (11) even drops directly to 0, that is, it is completely unable to work under constant noise. GD (25) is not very sensitive to constant noise, resulting in no effect on PSNR. The restoration effects of algorithms inspired by all control principles (such as the above-mentioned MNI (29), EDNTZD (13), NTGD (26), T4NTZD (19), T7NTZD (21), T11NTZD (22), and T6NTZD (23)) are the same as when there is no noise, and are almost unaffected.

[0115] Taking Table 3 as an example, taking MSSIM and PSNR as examples, the target processing models in this embodiment are all optimal, while the NI algorithm and the GD algorithm are seriously interfered by random noise, and the other NTZD algorithms gradually become better from T7NTZD (21), T4NTZD (19), EDZD (11), T5NTZD, and T11NTZD (22).

[0116] Experiment 2: Statistical Analysis

[0117] Please refer to Tables 4 and 5. Tables 4 and 5 are data sets containing 30 test images, respectively. The test objects include the NTZD algorithms discussed above, such as EDNTZD (13), T4NTZD (19), T5NTZD, T7NTZD (21), T11NTZD (22) and T6NTZD (23). Table 4 shows the PSNR index values ​​under the noise-free condition (NONOISE) and random noise condition (RANOM NOISE) corresponding to the above algorithms; Table 5 shows the MSSIM index values ​​under the noise-free condition (NO NOISE) and random noise condition (RANOM NOISE) corresponding to the above algorithms.

[0118] It can be seen from Table 4 and Table 5 that under zero noise, the accuracy of the NTZD algorithm is relatively high, and the two indicators of PSNR and MSSIM are almost the same. Under random noise, the average PSNR values ​​of EDNTZD (13), T4NTZD (19), T5NTZD, T7NTZD (21), and T11NTZD (22) are 20.02, 19.60, 21.23, and 17.26, respectively. In this embodiment, T6NTZD (23) still maintains a high level of 23.53. Among them, the average MSSIM values ​​of EDNTZD (13), T4NTZD (19), T5NTZD, T7NTZD (21), and T11NTZD (22) are 0.1767, 0.1743, 0.2442, 0.1172, and 0.2550, respectively. In this embodiment, the average MSSIM value of T6NTZD (23) is 0.3049. Both indicators are significantly better than other algorithms.

[0119] Table 4

[0120]

[0121] Table 5

[0122]

[0123] Based on this, according to the comparative experiments of different indicators, it can be concluded that GD and NTGD algorithms use the information of the first-order gradient. Because NTGD algorithms use the first-order gradient integral term, they can resist constant noise to a certain extent. However, due to the discontinuity of random noise, the fluctuation of GD algorithms under random noise is much stronger than the information of the first-order gradient. DTD (12), NRI (28), and MNI (29) all use the second-order gradient information of the deblurring model, which is faster, but the second-order gradient is easily affected by noise, so that DTD (12) and NRI (28) can have higher performance in the absence of noise. With the help of the gradient summation term, MNI (29) can work under constant noise, but it has no resistance to random noise. ZD algorithms use the first-order derivative of the prediction of error descent. Compared with GD algorithms, NTGD algorithms are more resistant to interference in terms of the next stable point of the current error value, but they are more consistent with the integral term of the deblurring model. The step size strategy adopted by the NTZD algorithm obviously affects the fit of the integral term. According to the corresponding results of the NTZD algorithm, the step size strategy of T6NTZD (23) in this embodiment has the best performance in the face of noise.

[0124] In summary, the present invention constructs an image processing model based on Taylor's theorem, and uses sample images to train the image processing model. The target processing model for image deblurring has high noise resistance, fills the gap in environmental noise resistance deblurring technology, and improves image restoration quality.

[0125] Please refer to Figure 4 , this embodiment also provides a block diagram of a server. The server includes a memory, a processor, and a communication module. The memory, the processor, and the communication module are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0126] The memory is used to store programs or data. The memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable read-only memory (EEPROM), etc.

[0127] The processor is used to read / write data or programs stored in the memory and execute corresponding functions.

[0128] The communication module is used to establish a communication connection between the server and other communication terminals through the network, and to send and receive data through the network.

[0129] It should be understood that Figure 4 The structure shown is only a schematic diagram of the server structure, and the server may also include Figure 4 More or fewer components as shown, or with Figure 4 Different configurations are shown. Figure 4 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0130] To perform the corresponding steps in the above embodiments and various possible methods, please refer to Figure 5 , Figure 5 The structure diagram of the image processing system in the embodiment of the present invention is shown below. An implementation method of the image processing system is given below. Optionally, the image processing system can be used Figure 4 The device structure shown.

[0131] For details, please continue to see Figure 5 , the image processing system 30 comprises:

[0132] The image acquisition module 31 is used to acquire the observed image and the real original image for training; and the image to be processed;

[0133] A construction module 32 is used to construct an image processing model based on Taylor's theorem;

[0134] The processing module 33 is used to train an image processing model based on the observed image and the real original image to obtain a target processing model for image deblurring; and use the target processing model to process the image to be processed to obtain a processed image.

[0135] It should be noted that the basic principles and technical effects of the image processing system provided in this embodiment are the same as those of the above embodiments. For the sake of brief description, for parts not mentioned in this embodiment, reference may be made to the corresponding contents in the above embodiments.

[0136] Optionally, the above modules can be stored in the form of software or firmware. Figure 4 The memory shown in the figure may be fixed in the operating system (OS) of the image processing system and may be Figure 4 Meanwhile, the data and program codes required for executing the above modules can be stored in the memory.

[0137] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0138] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0139] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An image processing method, characterized in that: The image processing method comprises the following steps: Build image processing models based on Taylor's theorem; Obtain observation images and true original images for training; The image processing model is trained based on the observed image and the real original image to obtain a target processing model for image deblurring; Acquire an image to be processed, and process the image to be processed using the target processing model to obtain a processed image; The target processing model is expressed as: in, is the first-order gradient function; M(.) is the objective function; X k is the image matrix corresponding to the kth iteration; is the kth image matrix after fast Fourier transform; κ is the second fuzzy operator; κ H is the conjugate transpose of the second fuzzy operator κ; λ is the regularization parameter; is the first characteristic matrix; is the observation matrix; h1 is the first adjustment parameter; h2 is the second adjustment parameter; is the i-th image matrix after fast Fourier transform; is the second-order gradient function; is the image matrix corresponding to the fast Fourier transform; is the k-1th image matrix after fast Fourier transform; is the k-2th image matrix after fast Fourier transform; is the k-3th image matrix after fast Fourier transform; is the k-4th image matrix after fast Fourier transform; is the norm equation; is the k+1th image matrix after fast Fourier transform; a is the convergence parameter; vec(·) is the column vectorization function.

2. The image processing method according to claim 1, characterized in that: After the step of obtaining the observed image and the real original image for training, the method further includes: Preprocessing the observed image to obtain fuzzy features of the observed image; A fuzzy feature equation is constructed based on the fuzzy feature and the real original image, wherein the fuzzy feature includes a fuzzy matrix and Gaussian noise.

3. The image processing method according to claim 2, characterized in that: The step of constructing a fuzzy feature equation based on the fuzzy feature and the real original image comprises: Constructing a fuzzy process vector equation based on the fuzzy matrix, the Gaussian noise and the real original image; The fuzzy process vector equation is converted into a fuzzy characteristic equation to convert the deblurring process into a minimum optimization process; wherein the solution of the fuzzy characteristic equation is used to characterize the deblurred image.

4. The image processing method according to claim 2, characterized in that: The step of training the image processing model based on the observed image and the real original image to obtain a target processing model for image deblurring comprises: The image processing model is trained using the fuzzy feature equation until the error equation corresponding to the fuzzy feature equation converges to zero, thereby obtaining a target processing model.

5. The image processing method according to claim 2, characterized in that: The fuzzy characteristic equation is expressed as: in, is the norm equation; λ is the regularization parameter; x(t) is the image vector iterated over time; argmin(.) is the minimization function; X t is the image matrix iterated over time; M(.) is the objective function; K is the first fuzzy operator; g is the first column vector.

6. The image processing method according to claim 4, characterized in that: The error equation is expressed as: Where E(.) is the error function; x(t) is the image vector iterated over time; is the first-order gradient function; M(.) is the objective function; K is the first fuzzy operator; K H is the conjugate transpose of the first fuzzy operator K; λ is the regularization parameter; I is the unit matrix; g is the first column vector.

7. The image processing method according to claim 1, characterized in that: The step of constructing an image processing model based on Taylor's theorem comprises: A Taylor six-point expansion design formula is constructed based on the expansion formula of Taylor's theorem to obtain an image processing model, so as to discretize the error equation corresponding to the fuzzy characteristic equation through the Taylor six-point expansion design formula.

8. The image processing method according to claim 7, characterized in that: The expression of the Taylor six-point expansion design formula is: in, is the image vector derivative in iteration, τ is the sampling interval; x k+1 is the image vector corresponding to the k+1th iteration; x k is the image vector corresponding to the kth iteration; x k-1 is the image vector corresponding to the k-th iteration; x k-2 is the image vector corresponding to the k-2th generation; x k-3 is the image vector of the k-3th response; x k-4 is the image vector corresponding to the k-4th generation; O(.) is the order of magnitude of the high-order error terms.

9. An image processing system, characterized in that: The image processing system comprises: An image acquisition module is used to acquire observation images and real original images for training; and to acquire images to be processed; Building blocks for constructing image processing models based on Taylor's theorem; A processing module, used for training the image processing model based on the observed image and the real original image to obtain a target processing model for image deblurring; and using the target processing model to process the image to be processed to obtain a processed image; Wherein, the target processing model is expressed as: in, is the first-order gradient function; M(.) is the objective function; X k is the image matrix corresponding to the kth iteration; is the kth image matrix after fast Fourier transform; κ is the second fuzzy operator; κ H is the conjugate transpose of the second fuzzy operator κ; λ is the regularization parameter; is the first characteristic matrix; is the observation matrix; h1 is the first adjustment parameter; h2 is the second adjustment parameter; is the i-th image matrix after fast Fourier transform; is the second-order gradient function; is the image matrix corresponding to the fast Fourier transform; is the k-1th image matrix after fast Fourier transform; is the k-2th image matrix after fast Fourier transform; is the k-3th image matrix after fast Fourier transform; is the k-4th image matrix after fast Fourier transform; is the norm equation; is the k+1th image matrix after fast Fourier transform; a is the convergence parameter; vec(·) is the column vectorization function.

Images