Image two-dimensional feature selection method and system

Abstract

The present invention relates to the field of image processing and proposes a method and system for selecting two-dimensional features of an image, comprising acquiring image data; extracting a two-dimensional matrix of the image data; performing low-rank recovery on noise data of the two-dimensional matrix to obtain a recovery matrix; performing regression learning on the two-dimensional matrix using the recovery matrix to obtain a left projection matrix and a right projection matrix of the two-dimensional matrix; obtaining a feature selection matrix based on the left projection matrix and the right projection matrix, and adding a sparse constraint to the feature selection matrix; performing feature selection on the image data using the feature selection matrix with the sparse constraint added to obtain a two-dimensional feature subset of the image. The present invention can perform feature selection on image data while retaining the spatial position information of the image data without increasing the dimension of the image data, and the selected features can more accurately represent the original image information.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and more particularly, to a method and system for selecting two-dimensional image features. Background Art

[0002] With the advent of image information technology and image information collection tools, our daily lives are now surrounded by a variety of image information. Faced with this massive amount of image information, human effort alone cannot guarantee the real-time and efficient processing of this information. Currently, artificial intelligence and machine learning methods are often used to perform image classification and segmentation, replacing manual effort in learning and processing massive amounts of image data.

[0003] A compensation-based feature selection method for image steganalysis is currently available. The method uses a steganography algorithm to generate steganalysis images with different embedding rates from a collected carrier image, and extracts one-dimensional gait steganalysis detection features from the steganalysis images. The one-dimensional gait steganalysis detection features are used to construct candidate feature vectors, and the separability of each candidate feature vector is measured. The candidate feature vector with the largest separability value is selected as the local optimal feature vector, and a supplementary selection strategy is used to add feature components to the local optimal feature vector to obtain the final selected features.

[0004] However, when extracting image features, the above method chooses to pull the image into a vector, which increases the dimension of the image data and increases the computational complexity, and destroys the original spatial position information of the image data, resulting in the selected features being unable to accurately represent the original image information. Summary of the Invention

[0005] In order to overcome the defects of existing image feature selection methods, such as high feature dimension and inability to accurately represent original image information, the present invention provides a two-dimensional image feature selection method and system.

[0006] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0007] In a first aspect, the present invention provides a method for selecting two-dimensional features of an image, comprising the following steps:

[0008] S1: Acquire image data;

[0009] S2: extract the two-dimensional matrix of image data;

[0010] S3: Perform low-rank recovery on the noise data of the two-dimensional matrix to obtain a recovery matrix;

[0011] S4: performing regression learning on the two-dimensional matrix using the restoration matrix to obtain a left projection matrix and a right projection matrix of the two-dimensional matrix;

[0012] S5: performing matrix reorganization on the left projection matrix and the right projection matrix to obtain a feature selection matrix, and adding a sparse constraint to the feature selection matrix;

[0013] S6: Use the feature selection matrix with sparse constraints added to perform feature selection on the image data to obtain a two-dimensional feature subset of the image.

[0014] As a preferred solution, in S3, the matrix recovery model is used to perform low-rank recovery on the noise data of the two-dimensional matrix to obtain the recovery matrix , whose expression is as follows:

[0015]

[0016] in, For vector vec ( X i ), It means pulling the two-dimensional matrix into a column vector by column. Indicates the i Two-dimensional matrix samples ( i =1,2,... t ), m Represents a two-dimensional matrix the number of rows, n Represents a two-dimensional matrix The number of columns, t Represents the number of samples of a two-dimensional matrix; for The recovery matrix, E for The noise matrix, Indicates a priori constraints on noise errors; for The recovery matrix, Indicates restoring the vector into a matrix by column; and are different weight parameters respectively.

[0017] As a preferred solution, in S4, the recovery matrix Perform regression learning on the two-dimensional matrix to obtain the left projection matrix of the two-dimensional matrix and the right projection matrix , whose expression is as follows:

[0018]

[0019] in, c Indicates the total number of label categories, r Indicates the label category, l Represents the number of labeled two-dimensional matrix samples, Indicates the i The recovery matrix of a two-dimensional matrix, is the left projection matrix, is the right projection matrix, y is a configurable hyperparameter, For the r A 2D matrix-like offset vector, Tr (∙) represents the trace of the matrix, Indicates the i The recovery matrix of a two-dimensional matrix About r The cluster label of the class is indicated; if Belong to r Class recovery matrix, then ;otherwise, .

[0020] As a preferred solution, in S5, matrix reorganization is performed on the left projection matrix and the right projection matrix, specifically including:

[0021] The mn left projection matrices Multiply by mn right projection matrices respectively Transpose to obtain mn intermediate matrices; pull the mn intermediate matrices into mn column vectors by column , and then mn column vectors Arrange by rows and get the feature selection matrix , whose expression is as follows:

[0022]

[0023]

[0024] The expression for adding a sparse constraint to the feature selection matrix is as follows:

[0025]

[0026] in, is the feature selection matrix, is the feature selection matrix The k vectors, , p is a configurable parameter. is the weight parameter, m Represents the recovery matrix the number of rows, n Represents the recovery matrix The number of columns, h represents the feature selection matrix P h OK, h =k , j represents the feature selection matrix P j List; Represents the feature selection matrix Perform row sparse constraints, reflect The feature importance in the two-dimensional matrix, each With a two-dimensional matrix The two-dimensional features in are in one-to-one correspondence.

[0027] As a preferred solution, the method further comprises:

[0028] In S4, a non-negative spectral clustering model is introduced for unsupervised learning. According to the recovery matrix and the non-negative spectral clustering model, a clustering label matrix of a two-dimensional matrix is obtained, and its expression is as follows:

[0029]

[0030] in, is the clustering class label matrix, L represents the recovery matrix The resulting Laplacian matrix, I c represents the identity matrix;

[0031] Using the clustering class label matrix, regression learning is performed on the two-dimensional matrix to obtain the left projection matrix and the right projection matrix of the two-dimensional matrix, which are expressed as follows:

[0032]

[0033] As a preferred solution, in S6, for unsupervised learning, the feature selection matrix with sparse constraints is used to select features of the image data to obtain a two-dimensional feature subset of the image. , specifically including:

[0034] The objective function of two-dimensional feature selection is established, and its expression is as follows:

[0035]

[0036] The alternating direction multiplier method is used to iteratively update the objective function to obtain the feature selection matrix The optimal value of

[0037] According to the matrix Calculate the optimal value of The value of The values are sorted from largest to smallest;

[0038] Select the first S The corresponding S two-dimensional matrix features constitute the two-dimensional feature subset of the image , S is a positive integer.

[0039] As a preferred solution, the method further comprises:

[0040] In S4, a manifold smoothing model is introduced for semi-supervised learning. According to the recovery matrix, the manifold smoothing model and part of the two-dimensional matrix with labels, a clustering class label matrix of the unlabeled two-dimensional matrix is obtained, where:

[0041]

[0042] in, is the class label matrix of the two-dimensional matrix sample with known labels, for the ) two-dimensional matrix sample with known labels, its class label is )hour, No. The element in row r column is 1, No. All elements except the rth column in the row are 0; is a zero matrix, corresponding to Unknown label two-dimensional matrix samples, m Represents the total number of two-dimensional matrix samples, c represents the total number of class labels, is the weight parameter, is a diagonal matrix with diagonal elements , q is a configurable parameter, u represents the number of rows and columns of the diagonal matrix; is the clustering class label matrix, L represents the recovery matrix The resulting Laplacian matrix, I c represents the identity matrix;

[0043] Using the clustering class label matrix, regression learning is performed on the two-dimensional matrix to obtain the left projection matrix and the right projection matrix of the two-dimensional matrix, which are expressed as follows:

[0044]

[0045] As a preferred solution, in S6, for semi-supervised learning, the feature selection matrix with sparse constraints is used to select features of the image data to obtain a two-dimensional feature subset of the image. , specifically including:

[0046] The objective function of two-dimensional feature selection is established, and its expression is as follows:

[0047]

[0048] The alternating direction multiplier method is used to iteratively update the objective function to obtain the feature selection matrix The optimal value of

[0049] According to the matrix Calculate the optimal value of The value of The values are sorted from largest to smallest;

[0050] Select the first S The corresponding S two-dimensional matrix features constitute the two-dimensional feature subset of the image , S is a positive integer.

[0051] As a preferred solution, in step S2, a grayscale feature extractor is used to identify and extract the grayscale value of each pixel of the image data, and the extraction result is used to construct a two-dimensional matrix of the image data.

[0052] In a second aspect, the present invention further provides a two-dimensional image feature selection system, comprising:

[0053] An image acquisition module, used for acquiring image data;

[0054] A two-dimensional matrix extraction module, used to extract the two-dimensional matrix of image data;

[0055] A matrix recovery module, configured to perform low-rank recovery on the noise data of the two-dimensional matrix to obtain a recovery matrix;

[0056] A regression learning module, configured to perform regression learning on the two-dimensional matrix using the restoration matrix to obtain a left projection matrix and a right projection matrix of the two-dimensional matrix;

[0057] A feature selection matrix acquisition module is used to perform matrix reorganization on the left projection matrix and the right projection matrix to obtain a feature selection matrix, and add a sparse constraint to the feature selection matrix;

[0058] The feature selection module is used to perform feature selection on image data using the feature selection matrix with sparse constraints added to obtain a two-dimensional feature subset of the image.

[0059] Compared with the prior art, the beneficial effect of the technical solution of the present invention is: the present invention extracts the two-dimensional matrix of image data, performs low-rank recovery and regression learning on the two-dimensional matrix, obtains the left projection matrix and the right projection matrix of the two-dimensional matrix, and obtains the feature selection matrix based on the left projection matrix and the right projection matrix, and uses the feature selection matrix to perform feature selection on the image to obtain a two-dimensional feature subset of the image. It can perform feature selection on the image data while retaining the spatial position information of the image data without increasing the dimension of the image data, and the selected features can more accurately represent the original image information. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 Flowchart of the two-dimensional feature selection method for images.

[0061] Figure 2 Architecture diagram of the image 2D feature selection system. DETAILED DESCRIPTION

[0062] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0063] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0064] Example 1

[0065] See also Figure 1 This embodiment proposes a method for selecting two-dimensional features of an image, comprising the following steps:

[0066] S1: Acquire image data.

[0067] S2: Extract the two-dimensional matrix of image data.

[0068] S3: Perform low-rank recovery on the noise data of the two-dimensional matrix to obtain a recovery matrix.

[0069] S4: Using the restoration matrix to perform regression learning on the two-dimensional matrix to obtain a left projection matrix and a right projection matrix of the two-dimensional matrix.

[0070] S5: Reorganize the left projection matrix and the right projection matrix to obtain a feature selection matrix, and add a sparse constraint to the feature selection matrix.

[0071] S6: Use the feature selection matrix with sparse constraints added to perform feature selection on the image data to obtain a two-dimensional feature subset of the image.

[0072] In the specific implementation process, by extracting the two-dimensional matrix of the image data, low-rank recovery and regression learning are performed on the two-dimensional matrix to obtain the left projection matrix and the right projection matrix of the two-dimensional matrix, and a feature selection matrix is obtained based on the left projection matrix and the right projection matrix. The feature selection matrix is used to perform feature selection on the image to obtain a two-dimensional feature subset of the image. It is possible to perform feature selection on the image data while retaining the spatial position information of the image data without increasing the dimension of the image data, and the selected features can more accurately represent the original image information.

[0073] Example 2

[0074] S1: Acquire image data.

[0075] In one specific embodiment, the image data is acquired through manual on-site photography; in another specific embodiment, the image data uploaded by the user is acquired from a background server.

[0076] S2: Extract the two-dimensional matrix of image data.

[0077] In this embodiment, a grayscale feature extractor is used to identify and extract the grayscale value of each pixel of the image data, and the extraction result is used to construct a two-dimensional matrix of the image data.

[0078] In practice, a grayscale image is viewed in a computer system as a collection of pixels with values between 0 and 255. 0 represents black, 255 represents white, and there are 256 grayscale levels between black and white. The grayscale feature extractor treats the image as a collection of pixels, identifies and extracts the grayscale value of each pixel, and uses the extracted value to convert the image into a two-dimensional matrix.

[0079] S3: Perform low-rank recovery on the noise data of the two-dimensional matrix to obtain a recovery matrix.

[0080] In this embodiment, a matrix recovery model is used to perform low-rank recovery on the noise data of the two-dimensional matrix to obtain the recovery matrix , whose expression is as follows:

[0081]

[0082] in, For vector vec ( X i ), It means pulling the two-dimensional matrix into a column vector by column. Indicates the i Two-dimensional matrix samples ( i =1,2,... t ), m Represents a two-dimensional matrix the number of rows, n Represents a two-dimensional matrix The number of columns, t Represents the number of samples of a two-dimensional matrix; for The recovery matrix, E for The noise matrix, Indicates a priori constraints on noise errors; for The recovery matrix, Indicates restoring the vector into a matrix by column; and are different weight parameters respectively.

[0083] S4: Using the recovery matrix Perform regression learning on the two-dimensional matrix to obtain the left projection matrix of the two-dimensional matrix and the right projection matrix , whose expression is as follows:

[0084]

[0085] in, c Indicates the total number of label categories, r Indicates the label category, l Represents the number of labeled two-dimensional matrix samples, Indicates the i The recovery matrix of a two-dimensional matrix, is the left projection matrix, is the right projection matrix, y is a configurable hyperparameter, For the r A 2D matrix-like offset vector, Tr (∙) represents the trace of the matrix, Indicates the i The recovery matrix of a two-dimensional matrix About r The cluster label of the class is indicated; if Belong to r Class recovery matrix, then ;otherwise, .

[0086] In unsupervised learning, since regression learning requires constructing a regression relationship between two-dimensional matrix samples and their class labels, it is necessary to perform non-negative spectral clustering on the two-dimensional matrix samples without label information to obtain the clustering class label matrix of the two-dimensional matrix samples.

[0087] In this embodiment, a non-negative spectral clustering model is introduced for unsupervised learning. According to the recovery matrix and the non-negative spectral clustering model, a clustering label matrix of a two-dimensional matrix is obtained, and its expression is as follows:

[0088]

[0089] in, is the clustering class label matrix, L represents the recovery matrix The resulting Laplacian matrix, I c Represents the identity matrix; here we add the non-negative constraint To limit Each row contains only one non-negative element, thereby improving the clustering class label matrix ability to discern.

[0090] Using the clustering class label matrix, regression learning is performed on the two-dimensional matrix to obtain the left projection matrix and the right projection matrix of the two-dimensional matrix, which are expressed as follows:

[0091]

[0092] In semi-supervised learning, it is necessary to use the manifold smoothing model to transfer the label information to the unlabeled two-dimensional matrix samples using the labeled two-dimensional matrix samples, under the premise that only some of the two-dimensional matrix samples have label information.

[0093] In this embodiment, a manifold smoothing model is introduced for semi-supervised learning. Based on the recovery matrix, the manifold smoothing model, and some labeled two-dimensional matrices, a clustering class label matrix of the unlabeled two-dimensional matrix is obtained, where:

[0094]

[0095] in, is the class label matrix of the two-dimensional matrix sample with known labels, for the ) two-dimensional matrix sample with known labels, its class label is )hour, No. The element in row r column is 1, No. All elements except the rth column in the row are 0; is a zero matrix, corresponding to Unknown label two-dimensional matrix samples, m Represents the total number of two-dimensional matrix samples, c represents the total number of class labels, is the weight parameter, is a diagonal matrix with diagonal elements , q It is a configurable parameter, and u represents the number of rows and columns of the diagonal matrix. is the clustering class label matrix, L represents the recovery matrix The resulting Laplacian matrix, I c Represents the identity matrix.

[0096] Using the clustering class label matrix, regression learning is performed on the two-dimensional matrix to obtain the left projection matrix and the right projection matrix of the two-dimensional matrix, which are expressed as follows:

[0097]

[0098] S5: Reorganize the left projection matrix and the right projection matrix to obtain a feature selection matrix, and add a sparse constraint to the feature selection matrix.

[0099] In this embodiment, matrix reorganization is performed on the left projection matrix and the right projection matrix, specifically including:

[0100] The mn left projection matrices Multiply by mn right projection matrices respectively Transpose to obtain mn intermediate matrices; pull the mn intermediate matrices into mn column vectors by column , and then mn column vectors Arrange by rows and get the feature selection matrix , whose expression is as follows:

[0101]

[0102]

[0103] The expression for adding a sparse constraint to the feature selection matrix is as follows:

[0104]

[0105] in, is the feature selection matrix, is the feature selection matrix The k vectors, , p is a configurable parameter. is the weight parameter, m Represents the recovery matrix the number of rows, n Represents the recovery matrix The number of columns, h represents the feature selection matrix P h OK, h = k , j represents the feature selection matrix P j List; Represents the feature selection matrix Perform row sparse constraints, reflect The feature importance in the two-dimensional matrix, each With a two-dimensional matrix The two-dimensional features in are in one-to-one correspondence.

[0106] Feature selection matrix The elements in can be regarded as the importance of the corresponding features in the two-dimensional matrix of the image. Adding sparse constraints will make the feature selection matrix Some elements in are equal to or close to 0, which means that the feature corresponding to the element is not important.

[0107] S6: Use the feature selection matrix with sparse constraints added to perform feature selection on the image data to obtain a two-dimensional feature subset of the image.

[0108] In this embodiment, for unsupervised learning, the feature selection matrix with sparse constraints is used to select features of the image data to obtain a two-dimensional feature subset of the image. , specifically including:

[0109] The objective function of two-dimensional feature selection is established, and its expression is as follows:

[0110]

[0111] The alternating direction multiplier method is used to iteratively update the objective function to obtain the feature selection matrix The optimal value of

[0112] According to the matrix Calculate the optimal value of The value of The values are sorted from largest to smallest;

[0113] Select the first S The corresponding S two-dimensional matrix features constitute the two-dimensional feature subset of the image , S is a positive integer.

[0114] In this embodiment, for semi-supervised learning, the feature selection matrix with sparse constraints is used to select features of the image data to obtain a two-dimensional feature subset of the image. , specifically including:

[0115] The objective function of two-dimensional feature selection is established, and its expression is as follows:

[0116]

[0117] The alternating direction multiplier method is used to iteratively update the objective function to obtain the feature selection matrix The optimal value of

[0118] According to the matrix Calculate the optimal value of The value of The values are sorted from largest to smallest;

[0119] Select the first S The corresponding S two-dimensional matrix features constitute the two-dimensional feature subset of the image , S is a positive integer.

[0120] In this embodiment, for supervised learning, the feature selection matrix with sparse constraints is used to select features of the image data to obtain a two-dimensional feature subset of the image. , specifically including:

[0121] The objective function for two-dimensional feature selection in supervised learning is established, and its expression is as follows:

[0122]

[0123] According to the objective function, the feature selection matrix is calculated The optimal value of .

[0124] According to the matrix Calculate the optimal value of The value of Sort the values from large to small and select the first S .

[0125] According to the first S Perform feature selection on image data to obtain a two-dimensional feature subset of the image .

[0126] The present invention improves the accuracy of two-dimensional feature selection by integrating regression learning and matrix recovery. In addition, by obtaining the clustering class label matrix Different objective functions are designed for supervised learning, unsupervised learning and semi-supervised learning to construct the objective function. Overfitting occurs during the feature selection training process, and the problem of directly using a biased matrix for regression learning that reduces learning ability is overcome. The influence of data disturbance, noise and outliers is reduced, and it has good robustness.

[0127] Example 3

[0128] See also Figure 2 This embodiment proposes a two-dimensional image feature selection system, including:

[0129] The image acquisition module is used to acquire image data.

[0130] The two-dimensional matrix extraction module is used to extract the two-dimensional matrix of image data.

[0131] In this embodiment, the two-dimensional matrix extraction module includes a grayscale feature extractor; the grayscale feature extractor identifies and extracts the grayscale value of each pixel of the image data, and constructs a two-dimensional matrix of the image data using the extraction result.

[0132] The matrix recovery module is used to perform low-rank recovery on the noise data of the two-dimensional matrix to obtain a recovery matrix.

[0133] A regression learning module is used to perform regression learning on the two-dimensional matrix using the restoration matrix to obtain a left projection matrix and a right projection matrix of the two-dimensional matrix.

[0134] The feature selection matrix acquisition module is used to perform matrix reorganization on the left projection matrix and the right projection matrix to obtain a feature selection matrix, and add a sparse constraint to the feature selection matrix.

[0135] The feature selection module is used to perform feature selection on image data using the feature selection matrix with sparse constraints added to obtain a two-dimensional feature subset of the image.

[0136] By extracting the two-dimensional matrix of image data, performing low-rank recovery and regression learning on the two-dimensional matrix, a left projection matrix and a right projection matrix of the two-dimensional matrix are obtained, and a feature selection matrix is obtained based on the left projection matrix and the right projection matrix. The feature selection matrix is used to perform feature selection on the image to obtain a two-dimensional feature subset of the image. Feature selection can be performed on the image data while retaining the spatial position information of the image data without increasing the dimension of the image data, and the selected features can more accurately represent the original image information.

[0137] The terms used in the drawings to describe positional relationships are for illustrative purposes only and should not be construed as limiting this patent;

[0138] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A two-dimensional feature selection method for an image, characterized in that: include: S1: Acquire image data; S2: extract the two-dimensional matrix of image data; S3: Perform low-rank recovery on the noise data of the two-dimensional matrix to obtain a recovery matrix; S4: performing regression learning on the two-dimensional matrix using the restoration matrix to obtain a left projection matrix and a right projection matrix of the two-dimensional matrix; S5: performing matrix reorganization on the left projection matrix and the right projection matrix to obtain a feature selection matrix, and adding a sparse constraint to the feature selection matrix; S6: Use the feature selection matrix with sparse constraints added to perform feature selection on the image data to obtain a two-dimensional feature subset of the image.

2. The image two-dimensional feature selection method according to claim 1, characterized in that: In S3, the matrix recovery model is used to perform low-rank recovery on the noise data of the two-dimensional matrix to obtain the recovery matrix , whose expression is as follows: in, For vector vec ( X i ), It means pulling the two-dimensional matrix into a column vector by column. Indicates the i Two-dimensional matrix samples ( i =1,2,... t ), m Represents a two-dimensional matrix the number of rows, n Represents a two-dimensional matrix The number of columns, t Represents the number of samples of a two-dimensional matrix; for The recovery matrix, E for The noise matrix, Indicates a priori constraints on noise errors; for The recovery matrix, Indicates restoring the vector into a matrix by column; and are different weight parameters respectively.

3. The image two-dimensional feature selection method according to claim 2, characterized in that: In S4, the recovery matrix Perform regression learning on the two-dimensional matrix to obtain the left projection matrix of the two-dimensional matrix and the right projection matrix , whose expression is as follows: in, c Indicates the total number of label categories, r Indicates the label category, l Represents the number of labeled two-dimensional matrix samples, Indicates the i The recovery matrix of a two-dimensional matrix, is the left projection matrix, is the right projection matrix, y is a configurable hyperparameter, For the r A 2D matrix-like offset vector, Tr (∙) represents the trace of the matrix, Indicates the i The recovery matrix of a two-dimensional matrix About r The cluster label of the class is indicated; if Belong to r Class recovery matrix, then ;otherwise, .

4. The method for selecting two-dimensional features of an image according to claim 3, wherein: In S5, matrix reorganization is performed on the left projection matrix and the right projection matrix, specifically including: The mn left projection matrices Multiply by mn right projection matrices respectively Transpose to obtain mn intermediate matrices; pull the mn intermediate matrices into mn column vectors by column , and then mn column vectors Arrange by rows and get the feature selection matrix , whose expression is as follows: The expression for adding a sparse constraint to the feature selection matrix is as follows: in, is the feature selection matrix, is the feature selection matrix The k vectors, , p is a configurable parameter. is the weight parameter, m Represents the recovery matrix the number of rows, n Represents the recovery matrix The number of columns, h Represents the feature selection matrix P h OK, h = k , j represents the feature selection matrix P j List; Represents the feature selection matrix Perform row sparse constraints, reflect The feature importance in the two-dimensional matrix, each With a two-dimensional matrix The two-dimensional features in are in one-to-one correspondence.

5. The method for selecting two-dimensional features of an image according to claim 4, wherein: The method further comprises: In S4, a non-negative spectral clustering model is introduced for unsupervised learning. According to the recovery matrix and the non-negative spectral clustering model, a clustering label matrix of a two-dimensional matrix is obtained, and its expression is as follows: in, is the clustering class label matrix, L represents the recovery matrix The resulting Laplacian matrix, I c represents the identity matrix; Using the clustering class label matrix, regression learning is performed on the two-dimensional matrix to obtain the left projection matrix and the right projection matrix of the two-dimensional matrix, which are expressed as follows:

6. The method for selecting two-dimensional features of an image according to claim 5, wherein: In S6, for unsupervised learning, the feature selection matrix with sparse constraints is used to select features of the image data to obtain a two-dimensional feature subset of the image. , specifically including: The objective function of two-dimensional feature selection is established, and its expression is as follows: The alternating direction multiplier method is used to iteratively update the objective function to obtain the feature selection matrix The optimal value of According to the matrix Calculate the optimal value of The value of The values are sorted from largest to smallest; Select the first S The corresponding S two-dimensional matrix features constitute the two-dimensional feature subset of the image , S is a positive integer.

7. The method for selecting two-dimensional features of an image according to claim 4, wherein: The method further comprises: In S4, a manifold smoothing model is introduced for semi-supervised learning. According to the recovery matrix, the manifold smoothing model and part of the two-dimensional matrix with labels, a clustering class label matrix of the unlabeled two-dimensional matrix is obtained, where: in, is the class label matrix of the two-dimensional matrix sample with known labels, for the ) two-dimensional matrix sample with known labels, its class label is )hour, No. The element in row r column is 1, No. All elements except the rth column in the row are 0; is a zero matrix, corresponding to Unknown label two-dimensional matrix samples, m Represents the total number of two-dimensional matrix samples, c represents the total number of class labels, is the weight parameter, is a diagonal matrix with diagonal elements , q is a configurable parameter, u represents the number of rows and columns of the diagonal matrix; is the clustering class label matrix, L represents the recovery matrix The resulting Laplacian matrix, I c represents the identity matrix; Using the clustering class label matrix, regression learning is performed on the two-dimensional matrix to obtain the left projection matrix and the right projection matrix of the two-dimensional matrix, which are expressed as follows:

8. The method for selecting two-dimensional image features according to claim 7, wherein: In S6, for semi-supervised learning, the feature selection matrix with sparse constraints is used to select features of the image data to obtain a two-dimensional feature subset of the image. , specifically including: The objective function of two-dimensional feature selection is established, and its expression is as follows: The alternating direction multiplier method is used to iteratively update the objective function to obtain the feature selection matrix The optimal value of According to the matrix Calculate the optimal value of The value of The values are sorted from largest to smallest; Select the first S The corresponding S two-dimensional matrix features constitute the two-dimensional feature subset of the image , S is a positive integer.

9. The method for selecting two-dimensional image features according to any one of claims 1 to 8, wherein: In step S2, a grayscale feature extractor is used to identify and extract the grayscale value of each pixel of the image data, and a two-dimensional matrix of the image data is constructed using the extraction results.

10. An image two-dimensional matrix selection system, characterized in that: include: An image acquisition module, used for acquiring image data; A two-dimensional matrix extraction module, used to extract the two-dimensional matrix of image data; A matrix recovery module, configured to perform low-rank recovery on the noise data of the two-dimensional matrix to obtain a recovery matrix; A regression learning module, configured to perform regression learning on the two-dimensional matrix using the restoration matrix to obtain a left projection matrix and a right projection matrix of the two-dimensional matrix; A feature selection matrix acquisition module is used to perform matrix reorganization on the left projection matrix and the right projection matrix to obtain a feature selection matrix, and add a sparse constraint to the feature selection matrix; The feature selection module is used to perform feature selection on image data using the feature selection matrix with sparse constraints added to obtain a two-dimensional feature subset of the image.

Images