CN116012263A - Image noise removing method and device, storage medium and electronic equipment - Google Patents
Image noise removing method and device, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN116012263A CN116012263A CN202310301759.2A CN202310301759A CN116012263A CN 116012263 A CN116012263 A CN 116012263A CN 202310301759 A CN202310301759 A CN 202310301759A CN 116012263 A CN116012263 A CN 116012263A
- Authority
- CN
- China
- Prior art keywords
- noise
- component
- image
- denoising model
- modeling
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 100
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 41
- 230000006870 function Effects 0.000 claims description 33
- 230000015654 memory Effects 0.000 claims description 20
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 11
- 239000000203 mixture Substances 0.000 claims description 7
- 230000003190 augmentative effect Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 8
- 230000008569 process Effects 0.000 description 19
- 238000004891 communication Methods 0.000 description 7
- 238000005259 measurement Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000013500 data storage Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000013515 script Methods 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000003321 amplification Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000969 carrier Substances 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000012847 principal component analysis method Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Images
Landscapes
- Image Processing (AREA)
Abstract
The embodiment of the application discloses an image noise removing method, an image noise removing device, a storage medium and electronic equipment, and relates to the technical field of image processing, comprising the following steps: inputting the original image into a denoising model to obtain a target image; the denoising model comprises a first part and a second part, wherein the first part is obtained by modeling a noise component based on exponent power mixing, the second part is obtained by modeling a noise-free component based on low-rank tensor singular value decomposition, and the noise-free component and the noise component are obtained by decomposing a plurality of noisy images. According to the method, complex noise in high-dimensional noise data is modeled by utilizing exponent power mixed distribution, and low-rank components in modeling noise are decomposed by utilizing low-rank tensor singular values, so that the model can learn the distribution of a noisy image, and an original image is accurately decomposed; the noise component can be learned, the real low-rank tensor can be better recovered from the multidimensional data of noise interference under the detailed condition, and the noise removal quality of the noisy image is effectively improved.
Description
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to an image noise removing method, an image noise removing device, a storage medium, and an electronic device.
Background
The high-dimensional data, such as noise-containing images, contains complex noise due to equipment failure, measurement resource limitation, environmental interference or human factors in the acquisition process, which seriously hinders the subsequent application of the high-dimensional data. As a useful subspace learning tool, low rank tensor decomposition (Low-rank Tensor Decomposition, LRTD) can extract efficient Low-dimensional structures in data.
Most LRTD techniques are currently based on minimization of the L2 norm measurement error, however, these methods are sensitive to the type of noise because the L2 norm tends to cause an amplification of the reconstruction error during use. To alleviate this problem, a viable strategy is to replace the L2 norm of LRTD with the L1 norm, which can be attributed to the L1 norm being more robust to heavy tail noise or outliers. However, the LRTD algorithm of the L1 norm causes a non-smooth and non-convex optimization problem, which makes it difficult to be widely applied to the problem of tensor data noise reduction. That is, the prior art has low denoising quality for noisy images.
Disclosure of Invention
The main purpose of the application is to provide an image noise removing method, an image noise removing device, a storage medium and electronic equipment, and aims to solve the problem that in the prior art, the denoising quality of a noisy image is low.
In order to achieve the above purpose, the technical solution adopted in the embodiments of the present application is as follows:
in a first aspect, an embodiment of the present application provides an image noise removing method, including the steps of:
acquiring an original image;
inputting the original image into a denoising model to obtain a target image; the denoising model comprises a first part and a second part, wherein the first part is obtained by modeling a noise component based on exponent power mixing, the second part is obtained by modeling a noise-free component based on low-rank tensor singular value decomposition, and the noise-free component and the noise component are obtained by decomposing a plurality of noisy images.
In a possible implementation manner of the first aspect, before acquiring the original image, the image noise removing method further includes:
decomposing a plurality of noisy images to obtain a noiseless component and a noise component;
modeling noise components based on exponential power mixing to obtain a first portion;
modeling the noiseless component based on low-rank tensor singular value decomposition to obtain a second part;
and obtaining a denoising model according to the first part and the second part.
In a possible implementation manner of the first aspect, after obtaining the denoising model according to the first portion and the second portion, the image noise removal method further includes:
initializing a noise-free component and a noise component in a denoising model;
estimating noise components in the denoising model;
and obtaining a noise-free component according to the estimated noise component, and returning to the step of estimating the noise component in the denoising model until the iteration termination condition is met, so as to obtain the target denoising model.
In one possible implementation manner of the first aspect, estimating a noise component in the denoising model includes:
based on the expectation maximization algorithm, noise components in the denoising model are estimated.
In one possible implementation manner of the first aspect, estimating a noise component in the denoising model based on a expectation maximization algorithm includes:
based on Bayesian rules, obtaining conditional expectations of noise components;
constructing a desired function according to the condition desire;
the noise component in the denoising model is estimated according to the maximized expectation function.
In one possible implementation manner of the first aspect, obtaining a noise-free component according to the estimated noise component includes:
based on the alternate direction multiplier method, a noise-free component is obtained from the estimated noise component.
In a possible implementation manner of the first aspect, based on the alternating direction multiplier method, obtaining a noise-free component from the estimated noise component includes:
obtaining an intermediate function according to the noiseless component after tensor singular value decomposition and the estimated noise component;
updating parameters of the noiseless component after tensor singular value decomposition according to the intermediate function and the augmented Lagrangian function corresponding to the intermediate function;
and obtaining the noiseless component according to the updated parameters of the noiseless component subjected to tensor singular value decomposition.
In a second aspect, an embodiment of the present application provides an image noise removing apparatus, including:
the acquisition module is used for acquiring an original image;
the denoising module is used for inputting the original image into a denoising model to obtain a target image; the denoising model comprises a first part and a second part, wherein the first part is obtained by modeling a noise component based on exponent power mixing, the second part is obtained by modeling a noise-free component based on low-rank tensor singular value decomposition, and the noise-free component and the noise component are obtained by decomposing a plurality of noisy images.
In a third aspect, an embodiment of the present application provides a computer readable storage medium storing a computer program, where the computer program when loaded and executed by a processor implements the image noise removal method provided in any one of the first aspects above.
In a fourth aspect, embodiments of the present application provide an electronic device comprising a processor and a memory, wherein,
the memory is used for storing a computer program;
the processor is configured to load and execute a computer program to cause the electronic device to perform the image noise removal method as provided in any one of the first aspects above.
Compared with the prior art, the beneficial effects of this application are:
the embodiment of the application provides an image noise removing method, an image noise removing device, a storage medium and electronic equipment, wherein the method comprises the following steps: acquiring an original image; inputting the original image into a denoising model to obtain a target image; the denoising model comprises a first part and a second part, wherein the first part is obtained by modeling a noise component based on exponent power mixing, the second part is obtained by modeling a noise-free component based on low-rank tensor singular value decomposition, and the noise-free component and the noise component are obtained by decomposing a plurality of noisy images. According to the method, by modeling the noise component and the noiseless component distribution, complex noise in high-dimensional noise data is modeled by utilizing exponential power mixed distribution, low-rank components, namely noiseless components, in the modeling noise are decomposed by utilizing low-rank tensor singular values, and the noiseless components and the noise components come from the decomposition of the noisy image, so that on one hand, the model can learn the distribution of the noisy image, and the original image can be rapidly and accurately decomposed; on the other hand, the noise component can be learned, so that the real low-rank tensor is better recovered from the multidimensional data of noise interference, and the detail information of the image is reserved, namely, the noiseless component is reserved as much as possible for outputting after denoising the noisy image, and the quality of removing the noise on the noisy image is effectively improved.
Drawings
FIG. 1 is a schematic diagram of an electronic device in a hardware operating environment according to an embodiment of the present application;
fig. 2 is a flow chart of an image noise removal method according to an embodiment of the present application;
fig. 3 is an original image in the image noise removing method provided in the embodiment of the present application;
FIG. 4 is an image of the noise free component corresponding to FIG. 3;
FIG. 5 is a view of an image obtained by denoising the original image shown in FIG. 3 using the image noise removal method according to the embodiment of the present application;
FIG. 6 is a view of the original image of FIG. 3 denoised using BM 4D;
FIG. 7 is a view showing the original image shown in FIG. 3 denoised using TDL;
FIG. 8 is a view of the original image of FIG. 3 denoised using TRPCA;
fig. 9 is a schematic block diagram of an image noise removing device according to an embodiment of the present application;
the marks in the figure: 101-processor, 102-communication bus, 103-network interface, 104-user interface, 105-memory.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
The main solutions of the embodiments of the present application are: provided are an image noise removal method, an image noise removal device, a storage medium and an electronic device, wherein the method comprises the following steps: acquiring an original image; inputting the original image into a denoising model to obtain a target image; the denoising model comprises a first part and a second part, wherein the first part is obtained by modeling a noise component based on exponent power mixing, the second part is obtained by modeling a noise-free component based on low-rank tensor singular value decomposition, and the noise-free component and the noise component are obtained by decomposing a plurality of noisy images.
The purpose of Low-rank tensor approximation (Low-rank Tensor Approximation, LRTA) is to recover its inherent Low-dimensional subspace from degraded high-dimensional data, and the extracted subspace is often capable of delivering a potentially Low-rank prior of the data, so that high-dimensional data processing based on tensor Low-rank constraint attracts more and more attention of students in the fields of computer vision, machine learning, pattern recognition, statistical analysis and the like. However, the high-dimensional data contains complex noise due to equipment failure, measurement resource limitation, environmental interference or human factors in the acquisition process, which seriously hinders the subsequent application of the high-dimensional data.
As a useful subspace learning tool, low rank tensor decomposition (Low-rank Tensor Decomposition, LRTD) can extract efficient Low-dimensional structures in data. The challenge faced in the low-rank tensor decomposition process is how to design a reasonably suitable decomposition model to represent the low-rank structure in high-dimensional data that is disturbed by noise. Most LRTD techniques are currently based on minimization of the L2 norm measurement error. However, these methods are sensitive to the type of noise because the L2 norm tends to cause reconstruction errors to be amplified during use. To alleviate this problem, a viable strategy is to replace the L2 norm of LRTD with the L1 norm, which can be attributed to the L1 norm being more robust to heavy tail noise or outliers. However, the LRTD algorithm of the L1 norm causes a non-smooth and non-convex optimization problem, which makes it difficult to be widely applied to the problem of tensor data noise reduction. In order to fit unknown noise in real high-dimensional data, which is neither subject to laplace distribution based on L1-norm nor to gaussian distribution based on L2-norm, the prior art proposes modeling the mixed noise in data with a gaussian mixture (Mixture of Gaussians) distribution in the LRTD framework, the key idea behind which is to assume that the real noise of any distribution can be better approximated with a more complex gaussian mixture.
Although the LRTD based MoG noise model can overcome the disadvantage of losing structural information caused by matrixing high-dimensional data, and at the same time, the sensitivity of the LRTD model to outliers or noise is avoided compared to a single noise distribution hypothesis. However, there are three major problems with MoG-based LRTD models. First, similar to a matrix rank, a tensor rank has an important meaning for separating signal components and noise components, but in practical applications, it is difficult to accurately estimate the rank of tensor decomposition due to lack of reasonably efficient a priori knowledge. Secondly, mogs based on noise models can theoretically approximate any distributed noise according to the central limit theorem, but provided that the components of mogs tend to infinity, but in practical applications only a limited number of components can be specified. Finally, there is a lack of rational methods to reasonably infer the number of gaussian mixture components from the noise actually mixed in the multidimensional data. It is therefore important to provide a better distribution modeling strategy with greater adaptation to data noise to replace MoG.
Therefore, the method and the device provide a solution, by modeling the noise component and the noiseless component distribution, modeling the complex noise in the high-dimensional noise data by using the exponential power mixed distribution, and decomposing the low-rank component, namely the noiseless component, in the modeling noise by using the low-rank tensor singular value, wherein the noiseless component and the noise component come from the decomposition of the noisy image, so that on one hand, the model can learn the distribution of the noisy image, and rapidly and accurately decompose the original image; on the other hand, the noise component can be learned, so that the real low-rank tensor is better recovered from the multidimensional data of noise interference, and the detail information of the image is reserved, namely, the noiseless component is reserved as much as possible for outputting after denoising the noisy image, and the quality of removing the noise on the noisy image is effectively improved.
Referring to fig. 1, fig. 1 is a schematic structural diagram of an electronic device of a hardware operating environment according to an embodiment of the present application, where the electronic device may include: a processor 101, such as a central processing unit (Central Processing Unit, CPU), a communication bus 102, a user interface 104, a network interface 103, a memory 105. Wherein the communication bus 102 is used to enable connected communication between these components. The user interface 104 may include a Display, an input unit such as a Keyboard (Keyboard), and the optional user interface 104 may also include standard wired, wireless interfaces. The network interface 103 may alternatively comprise a standard wired interface, a WIreless interface (e.g., a WIreless-FIdelity (WI-FI) interface). The Memory 105 may alternatively be a storage device independent of the foregoing processor 101, where the Memory 105 may be a high-speed random access Memory (Random Access Memory, RAM) Memory or may be a stable Non-Volatile Memory (NVM), such as at least one magnetic disk Memory; the processor 101 may be a general purpose processor including a central processing unit, a network processor, etc., as well as a digital signal processor, an application specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.
It will be appreciated by those skilled in the art that the structure shown in fig. 1 is not limiting of the electronic device and may include more or fewer components than shown, or may combine certain components, or may be arranged in different components.
As shown in fig. 1, the memory 105, which is a storage medium, may include an operating system, a data storage module, a network communication module, a user interface module, and an electronic program.
In the electronic device shown in fig. 1, the network interface 103 is mainly used for data communication with a network server; the user interface 104 is mainly used for data interaction with a user; the processor 101 and the memory 105 in the present application may be provided in an electronic device, and the electronic device invokes the image noise removing device stored in the memory 105 through the processor 101 and executes the image noise removing method provided in the embodiment of the present application.
Referring to fig. 2, based on the hardware device of the foregoing embodiment, an embodiment of the present application provides an image noise removing method, including the following steps:
s10: and decomposing the plurality of noisy images to obtain a noiseless component and a noise component.
In the specific implementation process, the noise high-dimensional data is modeled for matching the distribution proposed by the application
Decomposition into noise-free components->
And noise component->
The sum, i.e.)>
The method comprises the steps of carrying out a first treatment on the surface of the Wherein n is 1 And n 2 Representing the number of rows and columns, n, of each band image 3 Representing the total number of spectra of the remote sensing image. />
S20: the noise component is modeled based on the exponential power mixture to obtain a first portion.
In a specific implementation process, modeling a noise component, namely a first part, in data by using an exponential power mixed distribution, and approximating real noise by using the modeled mixed noise, wherein the specific steps are as follows:
modeling complex noise in high-dimensional noise data using exponential power hybrid (Mixture of Exponential Power, moEP) distribution
；
Wherein,,
is a mixed proportion and->
K is the number of mixed components, +.>
The expression parameter is->
And->
The k-th exponential power distribution of (c). Let->
By setting each +.>
An exponential power distribution with an average value of 0 can be obtained>
Density function of (2), namely:
wherein,,
as the precision parameter, p is the shape parameter, Γ (·) is the gamma function, exp represents the exponential distribution describing the kurtosis (0) by varying the shape parameter p<p<2) And low kurtosis (p)>2) Distribution. In particular, when p=1,
is a Laplace distribution; when p=2, _>
Is Gaussian distribution; when p → ≡infinity>
Is uniformly distributed.
Each noise is assumed
All corresponding to an indicator variable +.>
Wherein->
And is also provided with
。
Representing noise->
From the kth exponential power distribution, and which obeys a polynomial distribution +.>
Wherein->
. Then, there are:
wherein the method comprises the steps of
Representing the continuous multiplication between K variables.
Then the full likelihood function can be written as:
wherein the method comprises the steps of
Is->
Index set of non-missing elements in +.>
And
. The log likelihood function is then: />
Then the full log likelihood function can be written as:
combining the penalty likelihood method with the MoEP distribution, a penalty MoEP (peaked MoEP, PMoEP) model is proposed, i.e. PMoEP-LRTSVD:
wherein,,
wherein,,
represents a very small positive number (+)>
），
Represents a tuning parameter (/ -A)>
），
The number of free parameters representing the kth component, log (-) represents log. Because of adding the punishment item, the model error is minimized by the original solving parameter, and the model error is minimized by changing the original solving parameter into the solving parameter, and meanwhile, the model complexity is minimized.
S30: and modeling the noiseless component based on the low-rank tensor singular value decomposition to obtain a second part.
In the specific implementation process, the singular value decomposition is utilized to simplify data, noise is removed, the low-rank tensor decomposition is utilized to model a noiseless part, namely a second part, so that effective low-dimensional structure is extracted from the data, and the specific steps are that:
modeling Low rank components in noise data using Low rank tensor singular value decomposition (Low-Rank Tensor Singular Value Decomposition, LRTSVD)
I.e.
Wherein,,
and->
Is the orthogonal tensor, < >>
Is the matrix f-diagonal tensor, +.>
Representing the tensor product.
S40: and obtaining a denoising model according to the first part and the second part.
In the specific implementation process, after the distributed modeling, the distributed modeling is carried out through the existing modeling means, and finally the denoising model with multiple capacities is obtained, so that the denoising work can be carried out repeatedly and efficiently.
S50: acquiring an original image; the original image with noise is obtained by an image acquisition device, such as a camera, or by recalling images from an existing gallery.
S60: inputting the original image into a denoising model to obtain a target image; the denoising model comprises a first part and a second part, wherein the first part is obtained by modeling a noise component based on exponent power mixing, the second part is obtained by modeling a noise-free component based on low-rank tensor singular value decomposition, and the noise-free component and the noise component are obtained by decomposing a plurality of noisy images.
In the specific implementation process, as different parts of the model respectively learn different capacities under the distributed modeling, after the original image, namely the image to be denoised, is input into the denoising model, the model can learn the decomposing capacity of the image at first, decompose the image into a noise-free component and a noise component, and then respectively process the noise component and the noise component by the different parts, on one hand, the mixed noise component is self-adaptively fitted with complex real noise, and on the other hand, the proper noise component is learned from the mixed noise component, so that the real low-rank tensor is better recovered from multi-dimensional data of noise interference.
In the embodiment, by modeling the noise component and the noise-free component distribution, complex noise in high-dimensional noise data is modeled by using exponential power mixed distribution, low-rank components, namely noise-free components, in the modeling noise are decomposed by using low-rank tensor singular values, and the noise-free components and the noise components come from the decomposition of the noise-containing image, so that on one hand, the model can learn the distribution of the noise-containing image, and rapidly and accurately decompose the original image; on the other hand, the noise component can be learned, so that the real low-rank tensor is better recovered from the multidimensional data of noise interference, and the detail information of the image is reserved, namely, the noiseless component is reserved as much as possible for outputting after the denoising of the noisy image, the quality of removing the noise on the noisy image is effectively improved, and the running efficiency, accuracy and robustness are improved.
In one embodiment, after obtaining the denoising model according to the first part and the second part, the image noise removal method further comprises:
initializing a noise-free component and a noise component in the denoising model.
In the specific implementation process, in order to optimize the model parameters, the validity of the model parameters is verified, the model parameters are continuously optimized in an iterative mode, so that the denoising capability of the model is further improved, the noise-free components are updated by estimating the noise components and reversely, and the noise-free components are initialized
Complex noise->
Is 0.
The noise component in the denoising model is estimated.
In a specific implementation process, a noise estimation method, such as minimum tracking noise estimation, histogram noise estimation, etc., is provided in this embodiment, and a noise component in a denoising model is estimated based on an expectation maximization algorithm. Specific: estimating noise components in the denoising model based on a expectation maximization algorithm, comprising:
based on bayesian rules, conditional expectations of noise components are obtained.
Depending on the conditional expectation, the expectation function is constructed.
In a specific implementation process, the expectation-maximization algorithm is abbreviated as EM, and generally includes two steps, namely an E step and an M step, in which the E step is calculated by using bayesian rules
The conditions of (2) are as follows:
then, a desired function Q function can be constructed:
the noise component in the denoising model is estimated according to the maximized expectation function.
In a specific implementation, Θ is updated by maximizing the Q function in step M. To update pi, lagrange multipliers are introduced
To restrict->
The following Lagrangian function is then maximized:
by varying the variables
Taking the first derivative and making it zero to obtain
Wherein,,
，
indicating the number of non-missing elements. To get updated->
The function Q is modified>
Is the first derivative of (2) to obtain the zero point:
wherein,,
。
and obtaining a noise-free component according to the estimated noise component, and returning to the step of estimating the noise component in the denoising model until the iteration termination condition is met, so as to obtain the target denoising model.
In a specific implementation process, after estimating the noise component, the noise-free component is updated next, that is, an iterative method such as a least square method, an alternate direction multiplier method, etc. is needed, and the embodiment of the present application provides a method adopting the alternate direction multiplier method, specifically: obtaining a noise-free component from the estimated noise component, comprising:
based on the alternate direction multiplier method, a noise-free component is obtained from the estimated noise component.
The alternate direction multiplier method (Alternating Direction Method of Multipliers, ADMM) is a computational framework for solving the convex optimization problem with separability, and as the combination of the dual decomposition method and the augmented Lagrangian multiplier method, the algorithm has the advantages of decomposability and high processing speed, and meanwhile, good convergence is ensured.
Specific: based on the alternate direction multiplier method, obtaining a noise-free component from the estimated noise component, comprising:
and obtaining an intermediate function according to the noiseless component after tensor singular value decomposition and the estimated noise component.
In the concrete implementation process, in order to update
It is desirable to maximize the following function:
equivalent to solving the following equation:
wherein,,
to solve the above, tensor singular value decomposition is introduced
Equivalent thereof can be rewritten as
I.e. an intermediate function.
And updating parameters of the noise-free component after tensor singular value decomposition according to the intermediate function and the extended Lagrange function corresponding to the intermediate function.
In a specific implementation process, the corresponding augmented lagrangian function corresponding to the intermediate function is:
wherein the method comprises the steps of
Is Lagrangian multiplier->
Is a positive scalar.
Updating parameters of noise-free components after tensor singular value decomposition, i.e
：
And obtaining the noiseless component according to the updated parameters of the noiseless component subjected to tensor singular value decomposition.
In the specific implementation process, the update is performed
：
Considering its non-convexity and non-smoothness, it is divided into independent scalar optimization problems as follows:
wherein,,
。
correspondingly, the Lagrangian multiplier and the step size need to be updated
And finally, returning to the step of estimating the noise component in the denoising model until the iteration termination condition is met, and obtaining the target denoising model. The iteration termination condition can be the maximum iteration number or the convergence condition, the maximum iteration number itermax, the convergence criterion epsilon, the regularization parameter
，
The method comprises the steps of carrying out a first treatment on the surface of the If the iteration number is n<itermax, or relative error (relative error)>Epsilon, n=n+1, otherwise outputting the noise-free component +.>
。
With reference to fig. 3 to 8, the denoising method of the present application is compared with the existing denoising method:
the original image shown in fig. 3 is selected, the corresponding noiseless component of the image is shown in fig. 4, and fig. 5 is an image after denoising by adopting the method provided by the application.
Fig. 6 is an image obtained by denoising the original image shown in fig. 3 by using BM4D, fig. 7 is an image obtained by denoising the original image shown in fig. 3 by using TDL, BM4D is an image sequence denoising method based on a three-dimensional block matching algorithm, and both TDL and BM4D belong to a gaussian denoising method.
Fig. 8 is an image obtained by denoising the original image shown in fig. 3 using TRPCA, which is a tensor robust principal component analysis method.
It is obvious that the BM4D can hardly remove noise, and the TDL and the TRPCA can remove partial noise but seriously damage the image structure.
The PSNR is used as an evaluation index, and the noise reduction results of the method are compared as follows:
psnr is an abbreviation of "Peak Signal to Noise Ratio", i.e., peak signal-to-noise ratio, which is an objective criterion for evaluating an image, and a larger value indicates a better denoising effect. According to the table, different images are called in the data storage positions Chart-and-buffered-toy, and the method can obtain larger PSNR values for different data and different noise conditions, so that the better the imperceptibility of the images, namely the better the image quality, the better the robustness of the method is proved.
Referring to fig. 9, an embodiment of the present application further provides an image noise removing apparatus based on the same inventive concept as in the previous embodiment, the apparatus including:
the acquisition module is used for acquiring an original image;
the denoising module is used for inputting the original image into a denoising model to obtain a target image; the denoising model comprises a first part and a second part, wherein the first part is obtained by modeling a noise component based on exponent power mixing, the second part is obtained by modeling a noise-free component based on low-rank tensor singular value decomposition, and the noise-free component and the noise component are obtained by decomposing a plurality of noisy images.
It should be understood by those skilled in the art that the division of each module in the embodiment is merely a division of a logic function, and may be fully or partially integrated onto one or more actual carriers in practical application, and the modules may be fully implemented in a form of software called by a processing unit, or may be fully implemented in a form of hardware, or may be implemented in a form of combination of software and hardware, and it should be noted that each module in the image noise removing apparatus in this embodiment is in one-to-one correspondence with each step in the image noise removing method in the foregoing embodiment, so that a specific implementation of this embodiment may refer to an implementation of the foregoing image noise removing method, and will not be repeated herein.
Based on the same inventive concept as in the previous embodiments, embodiments of the present application further provide a computer readable storage medium storing a computer program, which when loaded and executed by a processor, implements an image noise removal method as provided in the embodiments of the present application.
Based on the same inventive concepts as in the previous embodiments, embodiments of the present application also provide an electronic device, including a processor and a memory, wherein,
the memory is used for storing a computer program;
the processor is configured to load and execute the computer program to cause the electronic device to execute the image noise removal method as provided in the embodiments of the present application.
In some embodiments, the computer readable storage medium may be FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; but may be a variety of devices including one or any combination of the above memories. The computer may be a variety of computing devices including smart terminals and servers.
In some embodiments, the executable instructions may be in the form of programs, software modules, scripts, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and they may be deployed in any form, including as stand-alone programs or as modules, components, subroutines, or other units suitable for use in a computing environment.
As an example, the executable instructions may, but need not, correspond to files in a file system, may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a hypertext markup language (HTML, hyper Text Markup Language) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code).
As an example, executable instructions may be deployed to be executed on one computing device or on multiple computing devices located at one site or, alternatively, distributed across multiple sites and interconnected by a communication network.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The foregoing embodiment numbers of the present application are merely for describing, and do not represent advantages or disadvantages of the embodiments.
From the above description of embodiments, it will be clear to a person skilled in the art that the above embodiment method may be implemented by means of software plus a necessary general hardware platform, but may of course also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. read-only memory/random-access memory, magnetic disk, optical disk), comprising several instructions for causing a multimedia terminal device (which may be a mobile phone, a computer, a television receiver, or a network device, etc.) to perform the method described in the embodiments of the present application.
In summary, the present application provides an image noise removing method, an apparatus, a storage medium, and an electronic device, where the method includes: acquiring an original image; inputting the original image into a denoising model to obtain a target image; the denoising model comprises a first part and a second part, wherein the first part is obtained by modeling a noise component based on exponent power mixing, the second part is obtained by modeling a noise-free component based on low-rank tensor singular value decomposition, and the noise-free component and the noise component are obtained by decomposing a plurality of noisy images. According to the method, by modeling the noise component and the noiseless component distribution, complex noise in high-dimensional noise data is modeled by utilizing exponential power mixed distribution, low-rank components, namely noiseless components, in the modeling noise are decomposed by utilizing low-rank tensor singular values, and the noiseless components and the noise components come from the decomposition of the noisy image, so that on one hand, the model can learn the distribution of the noisy image, and the original image can be rapidly and accurately decomposed; on the other hand, the noise component can be learned, so that the real low-rank tensor is better recovered from the multidimensional data of noise interference, and the detail information of the image is reserved, namely, the noiseless component is reserved as much as possible for outputting after denoising the noisy image, and the quality of removing the noise on the noisy image is effectively improved.
The foregoing description of the preferred embodiments of the present application is not intended to limit the invention to the particular embodiments of the present application, but to limit the scope of the invention to the particular embodiments of the present application.
Claims (10)
1. An image noise removal method, characterized by comprising the steps of:
acquiring an original image;
inputting the original image into a denoising model to obtain a target image; the denoising model comprises a first part and a second part, wherein the first part is obtained by modeling a noise component based on exponential power mixing, the second part is obtained by modeling a noise-free component based on low-rank tensor singular value decomposition, and the noise-free component and the noise component are obtained by decomposing a plurality of noisy images.
2. The image noise removal method according to claim 1, wherein before the original image is acquired, the image noise removal method further comprises:
decomposing a plurality of the noisy images to obtain the noiseless component and the noise component;
modeling the noise component based on the exponential power mixture to obtain the first portion;
modeling the noise-free component based on low-rank tensor singular value decomposition to obtain the second portion;
and obtaining the denoising model according to the first part and the second part.
3. The image noise removal method according to claim 2, wherein after the obtaining the denoising model from the first portion and the second portion, the image noise removal method further comprises:
initializing the noise-free component and the noise component in the denoising model;
estimating the noise component in the denoising model;
and obtaining the noise-free component according to the estimated noise component, and returning to the step of estimating the noise component in the denoising model until the iteration termination condition is met, so as to obtain a target denoising model.
4. The image noise removal method of claim 3, wherein said estimating said noise component in said denoising model comprises:
the noise component in the denoising model is estimated based on a expectation maximization algorithm.
5. The image noise removal method of claim 4, wherein said estimating said noise component in said denoising model based on a expectation maximization algorithm comprises:
based on Bayesian rules, obtaining conditional expectations of the noise components;
constructing a desired function according to the condition desire;
estimating the noise component in the denoising model according to the maximized expectation function.
6. The image noise removal method according to claim 3, wherein said obtaining said noise-free component from said estimated noise component comprises:
based on an alternate direction multiplier method, the noise-free component is obtained from the estimated noise component.
7. The image noise removal method according to claim 6, wherein the obtaining the noise-free component from the estimated noise component based on an alternate direction multiplier method includes:
obtaining an intermediate function according to the noiseless component after tensor singular value decomposition and the estimated noise component;
updating parameters of the noiseless component after tensor singular value decomposition according to the intermediate function and an augmented Lagrangian function corresponding to the intermediate function;
and obtaining the noiseless component according to the updated parameters of the noiseless component subjected to tensor singular value decomposition.
8. An image noise removing apparatus, comprising:
the acquisition module is used for acquiring an original image;
the denoising module is used for inputting the original image into a denoising model to obtain a target image; the denoising model comprises a first part and a second part, wherein the first part is obtained by modeling a noise component based on exponential power mixing, the second part is obtained by modeling a noise-free component based on low-rank tensor singular value decomposition, and the noise-free component and the noise component are obtained by decomposing a plurality of noisy images.
9. A computer readable storage medium storing a computer program, wherein the computer program, when loaded and executed by a processor, implements the image noise removal method according to any one of claims 1-7.
10. An electronic device comprising a processor and a memory, wherein,
the memory is used for storing a computer program;
the processor is configured to load and execute the computer program to cause the electronic device to perform the image noise removal method according to any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310301759.2A CN116012263A (en) | 2023-03-27 | 2023-03-27 | Image noise removing method and device, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310301759.2A CN116012263A (en) | 2023-03-27 | 2023-03-27 | Image noise removing method and device, storage medium and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116012263A true CN116012263A (en) | 2023-04-25 |
Family
ID=86033970
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310301759.2A Pending CN116012263A (en) | 2023-03-27 | 2023-03-27 | Image noise removing method and device, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116012263A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116402726A (en) * | 2023-06-08 | 2023-07-07 | 四川工程职业技术学院 | Denoising fusion method of hyperspectral-multispectral image |
| CN116433534A (en) * | 2023-06-09 | 2023-07-14 | 四川工程职业技术学院 | Hyperspectral image restoration method and device, storage medium and electronic equipment |
| CN117152023A (en) * | 2023-10-30 | 2023-12-01 | 浪潮软件科技有限公司 | Compressed image denoising method and device based on structure-texture decomposition |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107909548A (en) * | 2017-05-09 | 2018-04-13 | 西安交通大学 | A kind of video and removes rain method based on noise modeling |
| CN107967674A (en) * | 2017-12-20 | 2018-04-27 | 成都信息工程大学 | Nuclear magnetic resonance image denoising method based on image block self-similarity priori |
| CN110458777A (en) * | 2019-08-05 | 2019-11-15 | 湖南大学 | A kind of high spectrum image denoising method, system and medium based on the correction of adaptive order |
| CN113112506A (en) * | 2021-03-18 | 2021-07-13 | 西北工业大学 | Online moving target detection method based on exponential power distribution and matrix decomposition |
| CN113409307A (en) * | 2021-07-15 | 2021-09-17 | 中南大学 | Image denoising method, device and medium based on heterogeneous noise characteristics |
-
2023
- 2023-03-27 CN CN202310301759.2A patent/CN116012263A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107909548A (en) * | 2017-05-09 | 2018-04-13 | 西安交通大学 | A kind of video and removes rain method based on noise modeling |
| CN107967674A (en) * | 2017-12-20 | 2018-04-27 | 成都信息工程大学 | Nuclear magnetic resonance image denoising method based on image block self-similarity priori |
| CN110458777A (en) * | 2019-08-05 | 2019-11-15 | 湖南大学 | A kind of high spectrum image denoising method, system and medium based on the correction of adaptive order |
| CN113112506A (en) * | 2021-03-18 | 2021-07-13 | 西北工业大学 | Online moving target detection method based on exponential power distribution and matrix decomposition |
| CN113409307A (en) * | 2021-07-15 | 2021-09-17 | 中南大学 | Image denoising method, device and medium based on heterogeneous noise characteristics |
Non-Patent Citations (2)
| Title |
|---|
| XIANGYONG CAO: "Low-rank Matrix Factorization under General Mixture Noise Distributions", 《PROCEEDINGS OF THE IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION 2015》 * |
| XIANGYONG CAO: "Robust Low-Rank Matrix Factorization Under General Mixture Noise Distributions", 《 IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116402726A (en) * | 2023-06-08 | 2023-07-07 | 四川工程职业技术学院 | Denoising fusion method of hyperspectral-multispectral image |
| CN116402726B (en) * | 2023-06-08 | 2023-08-22 | 四川工程职业技术学院 | Denoising fusion method of hyperspectral-multispectral image |
| CN116433534A (en) * | 2023-06-09 | 2023-07-14 | 四川工程职业技术学院 | Hyperspectral image restoration method and device, storage medium and electronic equipment |
| CN116433534B (en) * | 2023-06-09 | 2023-08-22 | 四川工程职业技术学院 | Hyperspectral image restoration method and device, storage medium and electronic equipment |
| CN117152023A (en) * | 2023-10-30 | 2023-12-01 | 浪潮软件科技有限公司 | Compressed image denoising method and device based on structure-texture decomposition |
| CN117152023B (en) * | 2023-10-30 | 2024-02-02 | 浪潮软件科技有限公司 | Compressed image denoising method and device based on structure-texture decomposition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116012263A (en) | Image noise removing method and device, storage medium and electronic equipment | |
| CN112748483B (en) | Air temperature forecast deviation correction method and device based on deep learning | |
| CN110032704B (en) | Data processing method, device, terminal and storage medium | |
| Tan et al. | Pixelwise Estimation of Signal‐Dependent Image Noise Using Deep Residual Learning | |
| CN110765843A (en) | Face verification method and device, computer equipment and storage medium | |
| CN110807428B (en) | Coal sample identification method, device, server and storage medium | |
| CN107341768B (en) | Grid noise reduction | |
| CN118446939A (en) | Color image complement method, device, equipment and medium | |
| CN113077384B (en) | Data spatial resolution improving method, device, medium and terminal equipment | |
| CN113409307A (en) | Image denoising method, device and medium based on heterogeneous noise characteristics | |
| CN109388784A (en) | Minimum entropy Density Estimator device generation method, device and computer readable storage medium | |
| CN117278049A (en) | OMP signal reconstruction method and device based on weighted QR decomposition | |
| CN115760670B (en) | Unsupervised hyperspectral fusion method and device based on network implicit priori | |
| CN116959003A (en) | Character recognition method and device, electronic equipment and storage medium | |
| CN116087435A (en) | Air quality monitoring method, electronic equipment and storage medium | |
| CN116663388A (en) | Grain pile temperature prediction method, device, equipment and storage medium | |
| Cho et al. | An iterative algorithm for singular value decomposition on noisy incomplete matrices | |
| Li et al. | Image denoising via bidirectional low rank representation with cluster adaptive dictionary | |
| CN110147819B (en) | Video feature extraction method and device, readable storage medium and terminal equipment | |
| CN112152633A (en) | Vibration signal compression method and device | |
| CN113807517A (en) | Pruning parameter searching method, pruning method, device, equipment and medium | |
| CN116091367B (en) | Blind deblurring method, device, equipment and medium for optical remote sensing image | |
| Leong et al. | Multiple 3D far-field/near-field moving target localization using wideband echo chirp signals | |
| CN114332607B (en) | Incremental learning method and system for multi-frame image spectrum dictionary construction | |
| CN118411305A (en) | Underwater image enhancement method and system based on denoising and multi-order sparse combined prior |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20230425 |
|
| RJ01 | Rejection of invention patent application after publication |