Image noise reduction method and device, equipment and medium
By employing wavelet decomposition and multi-channel joint denoising techniques, and utilizing high-frequency similarity vector matrix and covariance matrix parameters, the efficiency and effectiveness of noise removal in multi-channel images are addressed, achieving more efficient image denoising.
Patent Information
- Application Number
- CN202511053590.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-28
- Publication Date
- 2025-11-21
AI Technical Summary
Existing technologies struggle to effectively remove noise from images, especially in multi-channel image processing. Single-channel noise reduction methods fail to fully utilize the correlation between channels, resulting in poor noise reduction performance.
By decomposing the image using wavelets, multiple high-frequency channel vectors are determined, a high-frequency similarity vector matrix is constructed, and parameters are determined using the covariance matrix and covariance matrix to achieve multi-channel joint noise reduction. The noise-reduced image is then generated using wavelet reconstruction technology.
It improves the accuracy of noise estimation, enables fast joint denoising of multi-channel images, and enhances the denoising effect and efficiency.
Smart Images

Figure CN120997517A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of image technology, and in particular to an image denoising method and device, equipment and medium. BACKGROUND
[0002] The application field of images is wide, involving biomedical, machine vision and other fields. The quality of images directly affects the effectiveness of their application in these fields. However, images are inevitably disturbed by noise in the process of acquisition, processing and transmission, so it is extremely important to filter out noise in images. SUMMARY
[0003] The embodiments of the present application provide an image denoising method, device, equipment and medium, to realize multi-channel wavelet high-frequency denoising, not only improve the noise estimation accuracy, but also quickly realize the joint denoising of multi-channel, so as to make the image denoising effect better and the efficiency higher.
[0004] The embodiments of the present application provide an image denoising method, device, equipment and medium, to realize multi-channel wavelet high-frequency denoising, not only improve the noise estimation accuracy, but also quickly realize the joint denoising of multi-channel, so as to make the image denoising effect better and the efficiency higher.
[0005] Obtain a to-be-denoised image, and perform wavelet decomposition on the to-be-denoised image to obtain a plurality of high-frequency components;
[0006] Based on the plurality of high-frequency components, a plurality of high-frequency channel vectors are determined; wherein each high-frequency channel vector is obtained by combining high-frequency components of the same image coordinate position of a plurality of channels of the to-be-denoised image.
[0007] For each high-frequency channel vector:
[0008] Determine a similar vector of the high-frequency channel vector, and generate a high-frequency similar vector matrix using the high-frequency channel vector and the similar vector thereof;
[0009] Based on the high-frequency similar vector matrix, determine a parameter for denoising the high-frequency channel vector;
[0010] Use the parameter for denoising the high-frequency channel vector to perform denoising processing on the high-frequency channel vector to obtain a denoised high-frequency channel vector;
[0011] Based on each denoised high-frequency channel vector, perform wavelet reconstruction in a manner corresponding to the wavelet decomposition to obtain a denoised image.
[0012] The image denoising method provided in the embodiments of the present application comprises the following steps: obtaining a to-be-denoised image, and performing wavelet decomposition on the to-be-denoised image to obtain a plurality of high-frequency components; then, based on the plurality of high-frequency components, a plurality of high-frequency channel vectors are determined; wherein each high-frequency channel vector is obtained by combining high-frequency components of the same image coordinate position of a plurality of channels of the to-be-denoised image; thus, the multi-channel joint of the image is realized. Subsequently, for each high-frequency channel vector, a similar vector of the high-frequency channel vector is determined, and a high-frequency similar vector matrix is generated by using the high-frequency channel vector and the similar vector thereof; based on the high-frequency similar vector matrix, parameters for denoising the high-frequency channel vector are determined; the high-frequency channel vector is denoised by using the parameters for denoising the high-frequency channel vector, to obtain a denoised high-frequency channel vector, so that the joint denoising of the multi-channel image is realized, and wherein the similar vector of the high-frequency channel vector is found, the high-frequency similar vector matrix is generated by using the high-frequency channel vector and the similar vector thereof, the parameters for denoising the high-frequency channel vector are determined based on the high-frequency similar vector matrix, so that more accurate noise estimation can be realized, and the denoising effect is better. Finally, based on each denoised high-frequency channel vector, wavelet reconstruction is performed in a manner corresponding to the wavelet decomposition, to obtain a denoised image. It can be seen that the embodiments of the present application realize multi-channel wavelet high-frequency denoising, which not only improves the accuracy of noise estimation, but also can quickly realize the joint denoising of multi-channels, so that the image denoising effect is better and the efficiency is higher.
[0013] In some embodiments, the determining the similar vector of the high-frequency channel vector comprises:
[0014] The high-frequency channel vector is extended to obtain a tensor corresponding to the high-frequency channel vector;
[0015] Based on the tensor corresponding to the high-frequency channel vector, a target tensor satisfying a preset condition is found in a preset range;
[0016] The high-frequency channel vector of an image position corresponding to the target tensor is determined as the similar vector.
[0017] In some embodiments, the preset condition comprises:
[0018] The F-norm of the difference between the target tensor and the tensor corresponding to the high-frequency channel vector is less than a preset threshold.
[0019] In some embodiments, the determining the parameters for denoising the high-frequency channel vector based on the high-frequency similar vector matrix comprises:
[0020] determining a covariance matrix of the noisy signal based on the high-frequency similar vector matrix, and a covariance matrix of the noise;
[0021] determining a covariance matrix of the noiseless signal based on the covariance matrix of the noisy signal and the covariance matrix of the noise.
[0022] In some embodiments, determining the covariance matrix of the noisy signal based on the high-frequency similar vector matrix comprises:
[0023] HM×HM T The obtained matrix is determined as the covariance matrix of the noisy signal, where HM represents the high-frequency similar vector matrix.
[0024] In some embodiments, determining the covariance matrix of the noise based on the high-frequency similar vector matrix comprises:
[0025] HM×HM T performing PCA decomposition on the obtained matrix to obtain an eigenvector matrix Q and diagonal elements where HM represents the high-frequency similar vector matrix; subscript B is the number of channels of the image to be denoised;
[0026] The covariance matrix of the noise is determined by the following formula:
[0027]
[0028] wherein, Σ n is determined by the following formula:
[0029]
[0030] wherein, ra is a preset constant, and satisfies 1=ra1<ra2<…ra B ;
[0031] thr σ is a value determined by the following formula:
[0032]
[0033] wherein, Era is a preset constant.
[0034] In some embodiments, the parameters for denoising the high-frequency channel vector comprise a covariance matrix of the noiseless signal and a covariance matrix of the noisy signal;
[0035] denoising the high-frequency channel vector using the parameters for denoising the high-frequency channel vector to obtain a denoised high-frequency channel vector, comprising:
[0036] The high-frequency channel vector is denoised by using the following formula to obtain a denoised high-frequency channel vector:
[0037] DHV(k)=Γ x ×(Γ y +I×ε) -1 ×HV(k);
[0038] wherein:
[0039] DHV(k) represents the denoised high-frequency channel vector, wherein k represents the position of the high-frequency channel vector on the image;
[0040] Γ x represents the covariance matrix of the noise-free signal;
[0041] Γ y represents the covariance matrix of the noisy signal;
[0042] I represents a unit matrix;
[0043] ε represents a preset constant;
[0044] HV(k) represents the high-frequency channel vector.
[0045] An image denoising device provided by an embodiment of the present application comprises:
[0046] a wavelet decomposition unit configured to obtain a to-be-denoised image and perform wavelet decomposition on the to-be-denoised image to obtain a plurality of high-frequency components;
[0047] a multi-channel high-frequency joint unit configured to determine a plurality of high-frequency channel vectors based on the plurality of high-frequency components; wherein each high-frequency channel vector is obtained by combining high-frequency components of the same image coordinate position of a plurality of channels of the to-be-denoised image;
[0048] a joint denoising unit configured to, for each high-frequency channel vector:
[0049] determine a similar vector of the high-frequency channel vector, and generate a high-frequency similar vector matrix using the high-frequency channel vector and the similar vector thereof;
[0050] determine a parameter for denoising the high-frequency channel vector based on the high-frequency similar vector matrix;
[0051] perform denoising processing on the high-frequency channel vector using the parameter for denoising the high-frequency channel vector to obtain a denoised high-frequency channel vector;
[0052] A wavelet reconstruction unit is configured to perform wavelet reconstruction on each of the denoised high-frequency channel vectors in a manner corresponding to the wavelet decomposition to obtain a denoised image.
[0053] Another embodiment of the present application provides an electronic device, comprising a memory and a processor, wherein the memory is configured to store program instructions, and the processor is configured to invoke the program instructions stored in the memory to execute any of the above methods.
[0054] Another embodiment of the present application provides a computer readable storage medium, which stores computer executable instructions, and the computer executable instructions are configured to cause the computer to execute any of the above methods. BRIEF DESCRIPTION OF DRAWINGS
[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort.
[0056] Figure 1 A flowchart of an image denoising method provided by an embodiment of the present application;
[0057] Figure 2 A wavelet decomposition schematic diagram provided by an embodiment of the present application;
[0058] Figure 3 A wavelet decomposition result schematic diagram of multiple channels provided by an embodiment of the present application;
[0059] Figure 4 A high-frequency channel vector determination schematic diagram of the same image position k provided by an embodiment of the present application;
[0060] Figure 5 A high-frequency channel vector expansion schematic diagram provided by an embodiment of the present application;
[0061] Figure 6 A denoising weight rate curve schematic diagram provided by an embodiment of the present application;
[0062] Figure 7 A wavelet reconstruction schematic diagram provided by an embodiment of the present application;
[0063] Figure 8 A structure schematic diagram of an electronic device provided by an embodiment of the present application;
[0064] Figure 9 A structure schematic diagram of an image denoising device provided by an embodiment of the present application. DETAILED DESCRIPTION
[0065] The technical solutions in the embodiments of the present application will be clearly and completely described in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.
[0066] The embodiments of the present application provide an image denoising method, device, equipment and medium, to realize multi-channel wavelet high-frequency denoising, not only improve the noise estimation accuracy, but also quickly realize the joint denoising of multi-channel, so as to make the image denoising effect better and the efficiency higher.
[0067] Among them, the method and the device, equipment and medium are based on the same application concept, and since the principles of the method and the device, equipment and medium for solving problems are similar, the implementation of the device, equipment and medium and the method can be mutually referred to, and the repeated parts will not be described again.
[0068] The terms "first", "second", and the like (if any) in the specification and claims of the embodiments of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.
[0069] The following examples and embodiments will only be understood as illustrative examples. Although the present specification may refer to "a", "one", or "some" example or embodiment in several places, this does not mean that each such reference relates to the same example or embodiment, nor does it mean that the feature only applies to a single example or embodiment. Individual features of different embodiments can also be combined to provide other embodiments. In addition, terms such as "include" and "contain" should be understood as not limiting the described embodiments to only those features mentioned; such examples and embodiments can also include features, structures, units, modules, etc. that are not specifically mentioned.
[0070] The various embodiments of the present application are described in detail below with reference to the accompanying drawings. It should be noted that the sequence of the embodiments of the present application only represents the sequence of the embodiments, and does not represent the advantages and disadvantages of the technical solutions provided by the embodiments.
[0071] Referring to Figure 1 The image denoising method provided by the embodiments of the present application can, for example, denoise the noise existing in the snapshot image of a camera, and the method comprises the following steps.
[0072] S101, obtaining an image to be denoised, and performing wavelet decomposition on the image to be denoised to obtain a plurality of high-frequency components;
[0073] For example, referring to Figure 2 If 2-level wavelet decomposition is performed, the following high-frequency components are obtained: HH1, LH1, HL1, HH2, LH2, and HL2.
[0074] S102, determining a plurality of high-frequency channel vectors based on the plurality of high-frequency components; wherein each high-frequency channel vector is obtained by combining the high-frequency components of the same image coordinate position of a plurality of channels of the image to be denoised.
[0075] In the embodiments of the present application, the correlation between the plurality of channels of the image is considered, so that the plurality of high-frequency channel vectors are determined based on the plurality of high-frequency components, so that the denoising effect is better and the efficiency is higher.
[0076] For example, for the high-frequency component HH of the image coordinate position k, the high-frequency components of the image coordinate position k of the plurality of channels of the image to be denoised can be combined to obtain the high-frequency channel vector HV(k).
[0077] S103, for each high-frequency channel vector: determining a similar vector of the high-frequency channel vector, and generating a high-frequency similar vector matrix using the high-frequency channel vector and its similar vector; determining parameters for denoising the high-frequency channel vector based on the high-frequency similar vector matrix; and performing denoising processing on the high-frequency channel vector using the parameters for denoising the high-frequency channel vector to obtain a denoised high-frequency channel vector.
[0078] It can be seen that in the embodiments of the present application, the noise can be estimated for a plurality of channels together, and then the denoising of the plurality of channels can be realized together, that is, for the same image to be denoised, the multi-channel wavelet high-frequency denoising can be realized at one time.
[0079] And, a high-frequency similar vector matrix is generated by finding a similar vector of the high-frequency channel vector and using the high-frequency channel vector and the similar vector thereof, and parameters for denoising the high-frequency channel vector are determined based on the high-frequency similar vector matrix, so that more accurate noise estimation can be achieved, and the denoising effect is better.
[0080] For generating the high-frequency similar vector matrix, for example, for the high-frequency channel vector HV(k), a high-frequency similar vector matrix [HV1, HV2, …, HV N ] corresponding to the high-frequency channel vector HV(k) is generated using the high-frequency channel vector HV(k) and the similar vector thereof HV(i).
[0081] The parameters for denoising the high-frequency channel vector HV(k) include, for example, at least any two of the covariance matrix Γ y of the signal with noise, the covariance matrix Γ n of the noise, and the covariance matrix Γ x of the signal without noise, and all three can also be included, depending on the denoising manner, which is not limited in the embodiments of the present application. The denoised high-frequency channel vector corresponding to the high-frequency channel vector HV(k) is denoted as DHV(k), for example.
[0082] S104, based on each of the denoised high-frequency channel vectors, wavelet reconstruction is performed in a manner corresponding to the wavelet decomposition to obtain a denoised image.
[0083] In some embodiments, the determination of the similar vector of the high-frequency channel vector includes:
[0084] The high-frequency channel vector is expanded, for example, to a three-dimensional tensor, to obtain a tensor corresponding to the high-frequency channel vector.
[0085] Based on the tensor corresponding to the high-frequency channel vector, a target tensor satisfying a preset condition is found within a preset range (for example, a 7x7 field).
[0086] The high-frequency channel vector of the image position corresponding to the target tensor is determined as the similar vector.
[0087] In some embodiments, the preset condition includes:
[0088] The F-norm of the difference between the target tensor and the tensor corresponding to the high-frequency channel vector is less than a preset threshold.
[0089] In some embodiments, determining the parameters for denoising the high-frequency channel vector based on the high-frequency similar vector matrix comprises:
[0090] determining a covariance matrix of the noisy signal based on the high-frequency similar vector matrix, and a covariance matrix of the noise;
[0091] determining a covariance matrix of the noiseless signal based on the covariance matrix of the noisy signal and the covariance matrix of the noise (e.g. x = Γ y - Γ n ).
[0092] In some embodiments, determining the covariance matrix of the noisy signal based on the high-frequency similar vector matrix comprises:
[0093] HM×HM T The resulting matrix is determined as the covariance matrix of the noisy signal, where HM represents the high-frequency similar vector matrix.
[0094] In some embodiments, determining the covariance matrix of the noise based on the high-frequency similar vector matrix comprises:
[0095] HM×HM T performing PCA decomposition on the resulting matrix to obtain an eigenvector matrix Q and diagonal elements where HM represents the high-frequency similar vector matrix; subscript B is the number of channels of the image to be denoised;
[0096] The covariance matrix of the noise is determined by the following formula:
[0097] Γ n = Q×Σ n ×Q T ;
[0098] where Σ n is determined by the following formula:
[0099]
[0100] where ra is a preset constant, and satisfies 1 = ra1 < ra2 < … ra B ;
[0101] thr σ is a value determined by the following formula:
[0102]
[0103] where Era is a preset constant.
[0104] In some embodiments, the parameters used for noise reduction of the high-frequency channel vector include a covariance matrix of a noise-free signal and a covariance matrix of a noisy signal.
[0105] The high-frequency channel vector is processed by noise reduction using the parameters used for noise reduction of the high-frequency channel vector, to obtain a noise-reduced high-frequency channel vector, including:
[0106] The high-frequency channel vector is processed by noise reduction using the parameters used for noise reduction of the high-frequency channel vector, to obtain a noise-reduced high-frequency channel vector, including:
[0107] DHV(k)=Γ x ×(Γ y +I×ε) -1 ×HV(k);
[0108] Wherein:
[0109] DHV(k) represents a noise-reduced high-frequency channel vector, wherein k represents the position of the high-frequency channel vector on the image;
[0110] Γ x represents the covariance matrix of the noise-free signal;
[0111] Γ y represents the covariance matrix of the noisy signal;
[0112] I represents the unit matrix;
[0113] ε represents a predetermined constant;
[0114] HV(k) represents the high-frequency channel vector.
[0115] Another embodiment is illustrated below.
[0116] In related art, the noise reduction model is generally assumed to be:
[0117] y=x+n Formula one
[0118] Wherein, y is an image (i.e. a to-be-noise-reduced image) obtained by photographing by the embodiment of the application, which contains noise. x is a noise-free image, which is an image (i.e. a noise-reduced image) that the embodiment of the application hopes to obtain. n represents noise, which is subject to an independent Gaussian distribution N(0,σ n ). σ n represents noise variance, the smaller the value, the smaller the noise, the larger the value, the larger the noise.
[0119] The mathematical model of wavelet Bayesian noise reduction is:
[0120]
[0121] Given y, the maximum possible x can be found.
[0122] Expanding equation two above gives:
[0123]
[0124] Where p(y / x) is the denoising model assumed in the embodiments of the present application, which can be expressed as:
[0125]
[0126] It can be seen that the expression of p(x) needs to be determined.
[0127] The high frequency information after wavelet decomposition of an image obeys a Gaussian distribution, a Laplace distribution or other more complex distributions, etc. The embodiments of the present application adopt a Gaussian distribution, i.e. p(x) is expressed as:
[0128]
[0129] By maximizing p(x / y), we can get:
[0130]
[0131] In order to improve the quality of the denoised image and improve the accuracy of noise estimation of wavelet denoising, the image denoising method provided by the embodiments of the present application can be divided into the following four steps, for example:
[0132] Step one: wavelet decomposition;
[0133] Step two: multi-channel high frequency joint;
[0134] Step three: joint denoising;
[0135] Among them, step three can include:
[0136] Step a: constructing a high frequency similar vector matrix;
[0137] Step b: determining the covariance matrix of the noisy signal and the covariance matrix of the noise;
[0138] Step c: wavelet high frequency denoising;
[0139] Step four: wavelet reconstruction.
[0140] The above steps will be specifically illustrated below.
[0141] Regarding step one, wavelet decomposition:
[0142] The wavelet is used to decompose the image to be denoised, for example, a 2-level wavelet transform can be expressed as Figure 2As shown. Among them, LL is a low-frequency signal, HL, LH and HH are horizontal detail components, vertical detail components and diagonal detail components, also known as horizontal high-frequency components, vertical high-frequency components and diagonal high-frequency components, and the subscripts 1 and 2 represent 1st and 2nd levels, respectively. Then HH1 is a 1st diagonal high-frequency component, and HH2 is a 2nd diagonal high-frequency component.
[0143] There are many kinds of wavelets that can be used for noise reduction, and the embodiments of the present application do not limit specific wavelets. The method of wavelet transform can be expressed as follows using a matrix:
[0144] C = w x I x w T Formula seven
[0145] Among them, I is the original image, C is the wavelet after transformation, including low frequency and high frequency. w is the wavelet transform matrix. For the low frequency in I, using the above formula seven again, the 2nd wavelet transform is realized.
[0146] The embodiments of the present application extend the single-spectrum image noise reduction to multi-spectrum images, and support joint noise reduction of multiple channels. Therefore, the embodiments of the present application can perform wavelet transformation on all channels in the same way. The embodiments of the present application assume that the image to be denoised has B channels, and then the original image can be represented as I1, I2, …, IB. B The image after wavelet transform is represented as C1, C2, …, CB. B The corresponding high-frequency information is marked with the corresponding channel number, for example, the HH2 of the 3rd channel can be represented as HH23. The transformed three-channel image is as shown in Figure 3 .
[0147] Regarding the multi-channel high-frequency joint of step two:
[0148] In related technologies, wavelet noise reduction often only supports single-channel image noise reduction, and for multi-channel images, independent noise reduction is required after being split into single channels. The embodiments of the present application realize joint noise reduction of multi-channel images. Therefore, the basic unit of noise reduction in the embodiments of the present application is a high-frequency channel vector. The combination method is as shown in Figure 4 . Figure 4 Taking the HH high-frequency channel vector combination method of 3 channels (for example, R, G, B channels) as an example, the high-frequency components at the corresponding positions (that is, the same image coordinate position k) are combined into a high-frequency channel vector, which can be represented as (HH1(k), HH2(k), HH3(k)) T , where k represents the two-dimensional coordinates of the image, that is, the image coordinates of HH, and different high-frequency components correspond to different image coordinates k.
[0149] When the image has B channels, then the HH high-frequency channel vector can be represented as:
[0150] (HH1(k), HH2(k),..., HH B (k)) T .
[0151] For simplicity, embodiments of the present application represent the high frequency channel vector as HV(k). Since the subsequent processing is independent for different high frequencies, this representation HV(k) does not distinguish between high frequencies of different levels, nor does it distinguish between HL, LH and HH.
[0152] Regarding different high frequency components, for example, if a 2-level wavelet decomposition is performed, there are 6 different high frequency components, HH1, LH1, HL1, HH2, LH2, HL2, and thus 6 high frequency channel vectors can be obtained. If a 3-level wavelet decomposition is performed, there are 3 more high frequency components, and correspondingly, 3 more high frequency channel vectors.
[0153] Regarding step three, joint noise reduction:
[0154] Step a, constructing a high frequency similar vector matrix:
[0155] Find the similar vectors of the high frequency channel vector HV(k), and then combine them into a high frequency similar vector matrix.
[0156] The high frequency channel vector is a one-dimensional column vector with a length of B. In some embodiments, for any high frequency channel vector HV(k), when finding its similar vector, the high frequency channel vector HV(k) can be expanded, for example, into a 3-dimensional tensor, denoted as SxSxB, and named as THV(k). The effect of expansion is shown in Figure 5 THV(k) is a tensor expanded to 3x3x3.
[0157] Embodiments of the present application take THV(k) as a unit to find similar tensors THV(i) around position k. The way to find is to calculate the F-norm of the difference between THV(k) and THV(i) as a unit, which can be calculated by the following formula:
[0158]
[0159] The calculation method of F-norm is square sum and then square root. For example, if the search range is 7x7, then 49 values of THV thr can be calculated by the above formula. When the value of THV thr is less than the set threshold value, record the value of i at this time, and determine HV(i) as the similar vector of HV(k). The threshold value is an external configuration parameter, which can be pre-set according to actual application.
[0160] In this embodiment, all HV(i) values less than a set threshold are determined and combined into a matrix in a columnar manner, represented as [HV1, HV2, ..., HV]. N This yields a B×N matrix. B represents the number of channels, and N represents the number of channel vectors that satisfy the threshold, i.e., the number of high-frequency vectors in the high-frequency similarity vector matrix. The high-frequency similarity vector matrix at position K is named HM(k), thus establishing the high-frequency similarity vector matrix at position K. The subsequent processing method for the high-frequency similarity vector matrix at each position is consistent, therefore it can be abbreviated as HM.
[0161] In this embodiment of the application, a high-frequency similarity vector matrix is established for each of the high-frequency components obtained by the wavelet decomposition. For example, if the wavelet decomposition yields 6 high-frequency components, then 6 high-frequency similarity vector matrices can be established.
[0162] Step b: Determine the covariance matrix of the noisy signal and the covariance matrix of the noise.
[0163] There are many ways to estimate single-channel noise. Common methods include finding a flat region, calculating its variance, and using this variance as the noise estimate. Another method is to obtain a noise estimate from the high frequencies of a statistical wavelet. These methods all assume that the noise magnitude is uniform. Figure 1 In practice, it has been found that the noise level fluctuates in different regions. Therefore, if the embodiments of this application can estimate the noise value of each pixel, the noise reduction effect will inevitably be improved.
[0164] Wavelet denoising techniques in related technologies are all single-channel denoising. To achieve multi-channel denoising, wavelets must be used for multiple independent single-channel denoising operations. This type of method essentially sacrifices the correlation between channels. To effectively utilize the correlation between signals between channels, this application proposes multi-channel denoising. Therefore, in some embodiments, the noise covariance matrix can be further estimated.
[0165] In practical use, this application embodiment has found that when a matrix is constructed using similar blocks, and then PCA (Principal Component Analysis) or singular value decomposition is performed, the signal shrinks to a few eigenvalues, while the noise is evenly distributed across the eigenvalues. Therefore, this application embodiment calculates the noise covariance matrix as follows:
[0166] HM×HM T The resulting matrix is determined to be the covariance matrix Γ of the noisy signal (the signal carrying noise). y ;
[0167] The following formula is used to calculate HM×HMT The obtained matrix is decomposed by PCA to determine a characteristic vector matrix Q and diagonal elements
[0168]
[0169] In the embodiment of the application, Σ can be used to represent the diagonal elements: The diagonal elements are arranged in descending order, and Q represents the characteristic vector matrix, and the arrangement of the characteristic vectors is in the order of , that is, the characteristic vector corresponding to σ1 is arranged first, and so on.
[0170] By splitting Σ y , a noise covariance matrix is determined:
[0171] In which, Σ y is split into Σ x + Σ n , and the following can be obtained:
[0172] Q×Σ y ×Q T is the covariance matrix Γ y of the noisy signal (the signal carrying noise).
[0173] Q×Σ x ×Q T is the covariance matrix Γ x of the noise-free signal (the signal without noise).
[0174] Q×Σ n ×Q T is the covariance matrix Γ n of the noise.
[0175] That is, by formula nine, the covariance Γ y of the noisy signal, that is, HM×HM T , of the obtained matrix can be determined.
[0176] And the covariance matrix Γ x of the noise-free signal is equal to the covariance matrix Γ y of the noisy signal minus the covariance matrix Γ n of the noise.
[0177] Therefore, after Γ y and Γ n are determined, Γ x can be obtained.
[0178] In which, Σ y is split into Σ x + Σ n The formula based on which includes:
[0179]
[0180] Era is an external configuration parameter, i.e., a preset constant. It should be configured according to the actual application; the larger the setting, the stronger the noise reduction.
[0181] parameter thr σ The value is determined by the above formula, for example, thr σ Starting from 0, the value is taken and it is checked whether the above formula is satisfied. If not, the value is increased by thr. σ The value of is determined, and the current thr is further evaluated. σ The value of is determined by whether it satisfies Formula 10 above, and so on, until Formula 10 is satisfied. At this point, the increment stops, and the first value to satisfy Formula 10 is thr. σ The value of thr σ The final value.
[0182] In other words, Formula 10 is used to determine the parameter thr. σ The value of .
[0183] ra can be determined using a noise reduction weighting curve, such as Figure 6 As shown, 1 = ra1 < ra2 < ... ra B From a PCA mathematical perspective, the eigenvector corresponding to σ1 is biased towards low frequencies, primarily representing the contour information of the image. Meanwhile, σ... B The corresponding feature vectors are biased towards high frequencies, mainly representing the image's detailed information. The feature vectors between these two types are biased towards contours and details. During noise reduction, low frequencies are generally prioritized for preservation. Figure 6 The curve shown represents making ra1, corresponding to eigenvalue 1 (σ1), smaller. And it represents making eigenvalue B (σ1) smaller... B The corresponding raB is larger. This is done to make noise reduction more inclined to reduce σ. B The value of .
[0184] The value of ra can be pre-configured or learned from training samples.
[0185] So Σ n It can be determined using the following formula:
[0186]
[0187] Therefore, based on Σ obtained from Formula 11 n The noise covariance matrix Γ can then be determined using the following formula. n :
[0188] Γ n- =Q×Σ n ×QT Formula Twelve
[0189] Step c: wavelet high frequency noise reduction
[0190] The p(y / x) and p(x) are extended from one dimension to multi-dimensions in the embodiment of the present application to adapt to the multi-channel noise reduction. Then the p(y / x) and p(x) can be respectively expressed as:
[0191]
[0192]
[0193] Then, the wavelet Bayesian noise reduction model is optimized The final noise reduction formula can be obtained as:
[0194]
[0195] The high frequency channel vector HV of all coordinate positions is brought into the above formula fifteen in the embodiment of the present application, and the channel vector after noise reduction can be obtained. In this way, the wavelet high frequency multi-channel noise reduction function is realized. For example, the specific noise reduction formula of the high frequency channel vector of position k can be expressed as:
[0196] DHV(k) = Γ x ×(Γ y + I × ε) -1 × HV(k) Formula Sixteen
[0197] Wherein, I is a unit matrix, and ε is a very small preset constant, for example, it can be set to 0.0001, mainly to prevent the irreversible case of Γ y . K represents the coordinate position corresponding to the high frequency component on the image, and Γ y is determined by the above formula nine, Γ x = Γ y - Γ n , and Γ n is determined by the above formula twelve.
[0198] In this step, the calculation operation of the above formula sixteen needs to be performed on all high frequency components obtained by the above wavelet decomposition, so as to obtain the high frequency components after noise reduction, and realize the noise reduction processing of the image.
[0199] Regarding the wavelet reconstruction of step four:
[0200] The wavelet reconstruction of step four corresponds to the wavelet decomposition of step one, that is, the same wavelet is used.
[0201] The wavelet reconstruction can be expressed in the form of a matrix, for example:
[0202]
[0203] wherein, is the 2-level wavelet coefficient, is the 1-level low frequency after denoising. W is the wavelet transform matrix. The above formula seventeen completes the 2-level wavelet reconstruction, and the 1-level wavelet reconstruction is completed by using the same method, so as to realize the complete wavelet reconstruction, as shown in the following formula eighteen. Figure 7
[0204] It should be noted that the wavelet decomposition and reconstruction exemplified in the embodiments of the present application are 2-level decomposition and 2-level reconstruction respectively. However, the technical solutions provided in the embodiments of the present application can be applied to wavelet decomposition and reconstruction of any level. The symbols used in the above process only play a representative role and can be replaced by any other mathematical symbol.
[0205] The following describes the device or apparatus provided in the embodiments of the present application. The explanation or exemplification of the same or corresponding technical features in the above method will not be repeated hereinafter.
[0206] The electronic device provided in the embodiments of the present application can be any network-side server device or terminal device, for example. Referring to Figure 8 , the electronic device provided in the embodiments of the present application can include, for example:
[0207] The processor 600 is configured to read the program in the memory 620 and perform the following process.
[0208] obtaining a to-be-denoised image, and performing wavelet decomposition on the to-be-denoised image to obtain a plurality of high-frequency components;
[0209] determining a plurality of high-frequency channel vectors based on the plurality of high-frequency components; wherein each high-frequency channel vector is obtained by combining the high-frequency components of the same image coordinate position of a plurality of channels of the to-be-denoised image.
[0210] For each high-frequency channel vector, the following steps are performed.
[0211] determining a similar vector of the high-frequency channel vector, and generating a high-frequency similar vector matrix by using the high-frequency channel vector and the similar vector thereof;
[0212] determining a parameter for denoising the high-frequency channel vector based on the high-frequency similar vector matrix;
[0213] performing denoising processing on the high-frequency channel vector by using the parameter for denoising the high-frequency channel vector to obtain a denoised high-frequency channel vector;
[0214] Based on each of the denoised high-frequency channel vectors, wavelet reconstruction is performed using a method corresponding to the wavelet decomposition to obtain the denoised image.
[0215] In some implementations, determining the similarity vector of the high-frequency channel vector includes:
[0216] The high-frequency channel vector is expanded to obtain the tensor corresponding to the high-frequency channel vector;
[0217] Based on the tensor corresponding to the high-frequency channel vector, find the target tensor that meets the preset conditions within the preset range;
[0218] The high-frequency channel vector of the image position corresponding to the target tensor is determined as the similarity vector.
[0219] In some implementations, the preset conditions include:
[0220] The F-norm of the difference between the target tensor and the tensor corresponding to the high-frequency channel vector is less than a preset threshold.
[0221] In some implementations, parameters for denoising the high-frequency channel vector are determined based on the high-frequency similarity vector matrix, including:
[0222] Based on the high-frequency similarity vector matrix, the covariance matrix of the noisy signal and the covariance matrix of the noise are determined.
[0223] The covariance matrix of the noiseless signal is determined using the covariance matrix of the noisy signal and the covariance matrix of the noise.
[0224] In some implementations, determining the covariance matrix of the noisy signal based on the high-frequency similarity vector matrix includes:
[0225] HM×HM T The resulting matrix is determined to be the covariance matrix of the noisy signal, where HM represents the high-frequency similarity vector matrix.
[0226] In some implementations, determining the noise covariance matrix based on the high-frequency similarity vector matrix includes:
[0227] HM×HM T The obtained matrix is decomposed using PCA to obtain the eigenvector matrix Q and the diagonal elements. Where HM represents the high-frequency similarity vector matrix; the subscript B is the number of channels in the image to be denoised;
[0228] The covariance matrix of the noise is determined by the following formula:
[0229] Γn = Q x Σ n x Q T ;
[0230] wherein, Σn is determined by the following formula:
[0231]
[0232] wherein, ra is a preset constant, and satisfies 1 = ra1 < ra2 < … < ra B ;
[0233] thr σ is a value determined by the following formula:
[0234]
[0235] wherein, Era is a preset constant.
[0236] In some embodiments, the parameter used for noise reduction of the high-frequency channel vector comprises a covariance matrix of a noise-free signal and a covariance matrix of a noisy signal.
[0237] The parameter used for noise reduction of the high-frequency channel vector is used to perform noise reduction processing on the high-frequency channel vector to obtain a noise-reduced high-frequency channel vector, comprising:
[0238] The high-frequency channel vector is processed by the following formula to obtain a noise-reduced high-frequency channel vector:
[0239] DHV(k) = Γ x x (Γ y + I x ε) -1 x HV(k);
[0240] wherein:
[0241] DHV(k) represents a noise-reduced high-frequency channel vector, wherein k represents the position of the high-frequency channel vector on the image;
[0242] Γ x represents the covariance matrix of the noise-free signal;
[0243] Γ y represents the covariance matrix of the noisy signal;
[0244] I represents the unit matrix;
[0245] ε represents a preset constant;
[0246] HV(k) represents the high-frequency channel vector.
[0247] The transceiver 610 is configured to receive and transmit data under the control of the processor 600.
[0248] In the above method, the image is obtained by the image acquisition unit 10. Figure 8 The bus architecture can include any number of interconnected buses and bridges, which are used to link various circuits of the processor 600, which is represented by one or more processors, and the memory 620, which is represented by various circuits of the memory. The bus architecture can also link various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art, and thus, will not be further described herein. The bus interface provides an interface. The transceiver 610 can be a plurality of elements, i.e., including a transmitter and a receiver, which provides a unit for communicating with various other devices on a transmission medium, including a wireless channel, a wired channel, an optical cable, and the like. The user interface 630 can also be an interface that can be externally connected to the required device, including but not limited to a keypad, a display, a speaker, a microphone, a joystick, and the like.
[0249] The processor 600 is responsible for managing the bus architecture and general processing, and the memory 620 can store data used by the processor 600 when performing operations.
[0250] In some embodiments, the processor 600 can be a CPU (Central Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a CPLD (Complex Programmable Logic Device), and the processor can also adopt a multi-core architecture.
[0251] The processor executes any of the methods provided by the embodiments of the present application by calling the computer program stored in the memory. The processor and the memory can also be physically arranged separately.
[0252] It should be noted that the above-mentioned device provided by the embodiments of the present application can realize all the method steps realized by the above-mentioned method embodiments, and can achieve the same technical effects. Therefore, the same parts and beneficial effects of the method embodiments will not be described in detail.
[0253] Referring to Figure 9 The image denoising device provided by the embodiments of the present application comprises:
[0254] The wavelet decomposition unit 11 is configured to obtain a to-be-denoised image, and perform wavelet decomposition on the to-be-denoised image to obtain a plurality of high-frequency components.
[0255] a multi-channel high-frequency joint unit 12, configured to determine a plurality of high-frequency channel vectors based on the plurality of high-frequency components, wherein each high-frequency channel vector is obtained by combining high-frequency components of the same image coordinate position of a plurality of channels of the image to be de-noised;
[0256] a joint de-noising unit 13, configured to, for each high-frequency channel vector:
[0257] determine a similar vector of the high-frequency channel vector, and generate a high-frequency similar vector matrix using the high-frequency channel vector and the similar vector of the high-frequency channel vector;
[0258] determine a parameter for de-noising the high-frequency channel vector based on the high-frequency similar vector matrix;
[0259] de-noise the high-frequency channel vector using the parameter for de-noising the high-frequency channel vector to obtain a de-noised high-frequency channel vector;
[0260] a wavelet reconstruction unit 14, configured to perform wavelet reconstruction on each de-noised high-frequency channel vector in a manner corresponding to the wavelet decomposition to obtain a de-noised image.
[0261] In some embodiments, the determination of the similar vector of the high-frequency channel vector comprises:
[0262] extending the high-frequency channel vector to obtain a tensor corresponding to the high-frequency channel vector;
[0263] finding a target tensor satisfying a preset condition within a preset range based on the tensor corresponding to the high-frequency channel vector;
[0264] determining a high-frequency channel vector of an image position corresponding to the target tensor as the similar vector.
[0265] In some embodiments, the preset condition comprises:
[0266] an F-norm of a difference between the target tensor and the tensor corresponding to the high-frequency channel vector is less than a preset threshold.
[0267] In some embodiments, the determination of the parameter for de-noising the high-frequency channel vector based on the high-frequency similar vector matrix comprises:
[0268] determining a covariance matrix of a noisy signal and a covariance matrix of noise based on the high-frequency similar vector matrix;
[0269] determining a covariance matrix of a noise-free signal based on the covariance matrix of the noisy signal and the covariance matrix of the noise.
[0270] In some embodiments, determining the covariance matrix of the noisy signal based on the high-frequency similar vector matrix comprises:
[0271] HM×HM T The obtained matrix is determined as the covariance matrix of the noisy signal, wherein HM represents the high-frequency similar vector matrix.
[0272] In some embodiments, determining the covariance matrix of the noise based on the high-frequency similar vector matrix comprises:
[0273] HM×HM T The obtained matrix is decomposed by PCA to obtain an eigenvector matrix Q and diagonal elements wherein HM represents the high-frequency similar vector matrix; subscript B is the number of channels of the image to be denoised;
[0274] The covariance matrix of the noise is determined by the following formula:
[0275] Γ n = Q x Σ n x Q T ;
[0276] wherein Σ n is determined by the following formula:
[0277]
[0278] wherein ra is a preset constant, and satisfies 1 = ra1 < ra2 < … ra B ;
[0279] thr σ is a value determined by the following formula:
[0280]
[0281] wherein Era is a preset constant.
[0282] In some embodiments, the parameters for denoising the high-frequency channel vector comprise: the covariance matrix of the noiseless signal and the covariance matrix of the noisy signal.
[0283] Denoising the high-frequency channel vector using the parameters for denoising the high-frequency channel vector to obtain a denoised high-frequency channel vector comprises:
[0284] Denoising the high-frequency channel vector using the following formula to obtain a denoised high-frequency channel vector:
[0285] DHV(k) = Γx x (I y + I x e) -1 x HV(k) ;
[0286] wherein:
[0287] DHV(k) denotes a high frequency channel vector after noise reduction, wherein k denotes the position of the high frequency channel vector on the image;
[0288] x denotes the covariance matrix of the noise-free signal;
[0289] y denotes the covariance matrix of the noisy signal;
[0290] I denotes a unit matrix;
[0291] e denotes a preset constant;
[0292] HV(k) denotes the high frequency channel vector.
[0293] It should be noted that the division of units in the embodiments of the present application is illustrative, and is only a logical functional division. In actual implementation, another division manner can be used. In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present alone, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.
[0294] When the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.
[0295] Any of the devices or apparatuses provided in the embodiments of the present application can specifically be a desktop computer, a portable computer, a smart phone, a tablet computer, a personal digital assistant (PDA), and the like. The devices or apparatuses can include a central processing unit (CPU), a memory, an input / output device, and the like. The input device can include a keyboard, a mouse, a touch screen, and the like. The output device can include a display device, such as a liquid crystal display (LCD), a cathode ray tube (CRT), and the like.
[0296] The memory can include a read-only memory (ROM) and a random access memory (RAM), and provide the processor with program instructions and data stored in the memory. In the embodiments of the present application, the memory can be used to store the programs of any of the methods provided in the embodiments of the present application.
[0297] The processor calls the program instructions stored in the memory, and the processor is configured to perform any of the methods provided in the embodiments of the present application according to the obtained program instructions.
[0298] The embodiments of the present application also provide a computer program product or a computer program, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to enable the computer device to perform any of the methods in the above embodiments. The program product can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, be but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any combination of the above. More specific examples (a non-exhaustive list) of the readable storage medium include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or a flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.
[0299] The embodiments of the present application provide a computer readable storage medium for storing computer program instructions for the apparatus provided in the above embodiments of the present application, which includes programs for executing any of the methods provided in the above embodiments of the present application. The computer readable storage medium can be a non-transitory computer readable medium.
[0300] The computer-readable storage medium can be any available media or data storage device that can be accessed by a computer, including but not limited to magnetic storage (e.g. floppy disks, hard disks, tape, MO, etc.), optical storage (e.g. CD, DVD, BD, HVD, etc.), and semiconductor storage (e.g. ROM, EPROM, EEPROM, NAND FLASH, SSD, etc.), etc.
[0301] It should be understood that:
[0302] The access technology via which the entities in the communication network transfer traffic to and from each other can be any suitable current or future technology, such as WLAN (Wireless Local Access Network), WiMAX (Worldwide Interoperability for Microwave Access), LTE, LTE-A, 5G, Bluetooth, infrared, etc. can be used; in addition, embodiments can also apply wired technologies, for example, IP-based access technologies, such as wired networks or fixed lines.
[0303] Embodiments suitable to be implemented as software code or part thereof and run using a processor or processing functionality are independent of the software code and can be specified using any known or future developed programming language, such as a high-level programming language, such as objective-C, C, C++, C#, Java, Python, Javascript, other scripting languages, etc., or a low-level programming language, such as a machine language or assembly.
[0304] The implementation of embodiments is independent of hardware and can be implemented using any known or future developed hardware technology or any mixture thereof, such as microprocessors or CPUs (Central Processing Units), MOS (Metal Oxide Semiconductor), CMOS (Complementary MOS), BiMOS (Bipolar MOS), BiCMOS (Bipolar CMOS), ECL (Emitter Coupled Logic), and / or TTL (Transistor-Transistor Logic).
[0305] Embodiments can be implemented as separate devices, apparatuses, units, components or functions, or in a distributed manner, for example, one or more processors or processing functionality can be used or shared in processing, or one or more processing segments or processing portions can be used and shared in processing, wherein one physical processor or more than one physical processor can be used to implement one or more processing portions dedicated to a specific processing as described.
[0306] The apparatuses can be implemented by semiconductor chips, chip sets, or (hardware) modules including such chips or chip sets.
[0307] The embodiments can also be implemented as any combination of hardware and software, such as ASIC (Application Specific IC (Integrated Circuit)) components, FPGA (Field-Programmable Gate Array) or CPLD (Complex Programmable Logic Device) components or DSP (Digital Signal Processor) components.
[0308] The embodiments can also be implemented as a computer program product including a computer readable program code embodied in a computer available medium, the computer readable program code adapted to carry out processes as described in the embodiments, wherein the computer available medium can be a non-transitory medium.
[0309] Those skilled in the art will appreciate that embodiments of the present application can be supplied as a method, a system, or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer available storage medium (including, but not limited to, disk storage and optical storage) having computer usable program code embodied in the medium.
[0310] The present application is described with reference to the flowchart illustrations and / or block diagrams of the methods, apparatus (systems) and computer program products according to embodiments of the present application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, an embedded processor or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 means for (an apparatus with) one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks.
[0311] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions means that implement the function specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 means for (an apparatus with) one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks.
[0312] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams block or blocks.Figure 1 the steps of the functions specified in the one or more blocks.
[0313] It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the spirit or scope of the application. Thus, it is intended that the present application cover the modifications and variations of this application provided they come within the scope of the appended claims and their equivalents.
Claims
1. A method of image denoising, characterized in that, The method comprises: obtaining an image to be de-noised, and performing wavelet decomposition on the image to be de-noised to obtain a plurality of high-frequency components; based on the plurality of high-frequency components, determining a plurality of high-frequency channel vectors; wherein each high-frequency channel vector is obtained by combining high-frequency components of the same image coordinate position of a plurality of channels of the image to be de-noised; for each high-frequency channel vector: determine the similar vector of the high-frequency channel vector, and generate a high-frequency similar vector matrix using the high-frequency channel vector and its similar vector; based on the high-frequency similar vector matrix, determine the parameter for de-noising the high-frequency channel vector; using the parameter for de-noising the high-frequency channel vector, de-noising the high-frequency channel vector to obtain a de-noised high-frequency channel vector; based on each de-noised high-frequency channel vector, performing wavelet reconstruction in a manner corresponding to the wavelet decomposition to obtain a de-noised image.
2. The method of claim 1, wherein, The determination of the similar vector of the high-frequency channel vector comprises: extending the high-frequency channel vector to obtain a tensor corresponding to the high-frequency channel vector; based on the tensor corresponding to the high-frequency channel vector, finding a target tensor satisfying a preset condition within a preset range; determining the high-frequency channel vector of the image position corresponding to the target tensor as the similar vector.
3. The method of claim 2, wherein, The preset condition comprises: the F-norm of the difference between the target tensor and the tensor corresponding to the high-frequency channel vector is less than a preset threshold.
4. The method of claim 1, wherein, Based on the high-frequency similar vector matrix, determining the parameter for de-noising the high-frequency channel vector comprises: based on the high-frequency similar vector matrix, determining a covariance matrix carrying noise signals and a noise covariance matrix; determining a noise-free signal covariance matrix through the covariance matrix carrying noise signals and the noise covariance matrix.
5. The method of claim 4, wherein, Based on the high-frequency similar vector matrix, determining the covariance matrix carrying noise signals comprises: HMxHM T The resulting matrix is determined as a covariance matrix carrying noise signals, wherein HM denotes the high-frequency similarity vector matrix.
6. The method of claim 4, wherein, based on the high-frequency similar vector matrix, determining a noise covariance matrix comprises: HMxHM T The resulting matrix is decomposed by PCA to obtain an eigenvector matrix Q and diagonal elements where HM denotes the high-frequency similar vector matrix; subscript B is the channel number of the image to be denoised; determining the noise covariance matrix through the following formula: Γ n = Q x Σ n x Q T ; wherein Σ n is determined by the formula: wherein ra is a preset constant, and satisfies 1 = ra1 < ra2 <... ra B ; thr σ is a value determined by the following equation: wherein Era is a preset constant.
7. The method of claim 4, wherein, The parameter for de-noising the high-frequency channel vector comprises a noise-free signal covariance matrix and a covariance matrix carrying noise signals. Using the parameter for de-noising the high-frequency channel vector to de-noise the high-frequency channel vector to obtain a de-noised high-frequency channel vector comprises: using the following formula to de-noise the high-frequency channel vector to obtain a de-noised high-frequency channel vector: DHV(k) = Γ x × (Γ y + I x ε) -1 × HV(k); wherein: DHV(k) represents the de-noised high-frequency channel vector, wherein k represents the position of the high-frequency channel vector on the image; Γ x denotes the covariance matrix of the noiseless signal; Γ y denotes the covariance matrix of the noise signal; I represents a unit matrix; ε represents a preset constant; HV(k) represents the high-frequency channel vector.
8. An image noise reduction apparatus, characterized by comprising: The device comprises: a wavelet decomposition unit configured to obtain an image to be de-noised, and perform wavelet decomposition on the image to be de-noised to obtain a plurality of high-frequency components; The multi-channel high-frequency joint unit is configured to determine a plurality of high-frequency channel vectors based on the plurality of high-frequency components, wherein each high-frequency channel vector is obtained by combining high-frequency components of the same image coordinate position of a plurality of channels of the image to be de-noised. The joint de-noising unit is configured to, for each high-frequency channel vector: determine a similar vector of the high-frequency channel vector, and generate a high-frequency similar vector matrix using the high-frequency channel vector and the similar vector of the high-frequency channel vector; determine a parameter for de-noising the high-frequency channel vector based on the high-frequency similar vector matrix; de-noise the high-frequency channel vector using the parameter for de-noising the high-frequency channel vector to obtain a de-noised high-frequency channel vector; and The wavelet reconstruction unit is configured to perform wavelet reconstruction on each de-noised high-frequency channel vector in a manner corresponding to the wavelet decomposition to obtain a de-noised image.
9. An electronic device, comprising: The computer readable storage medium stores computer executable instructions for causing the computer to perform the method of any one of claims 1 to 7. The computer readable storage medium stores computer executable instructions for causing the computer to perform the method of any one of claims 1 to 7. 10. A computer-readable storage medium, characterized in that,
