Small-size image recognition method, device, equipment, medium and product

Abstract

The invention discloses a small-size image recognition method, device and equipment, a medium and a product in the technical field of computer vision, and the method comprises the steps: obtaining a small-size training image, and carrying out the preprocessing of the small-size training image, and generating a training image matrix; splicing all the training image matrixes to generate an image augmentation matrix; constructing a low-rank and low-dimension combined 2DERPCA optimization model of enhanced robust principal component analysis, and inputting the image augmentation matrix into the 2DERPCA optimization model to calculate an optimal projection matrix; obtaining a to-be-recognized small-size image, and preprocessing the to-be-recognized small-size image to generate a to-be-recognized image matrix; performing feature extraction on the to-be-recognized image matrix and the training image matrix by using the optimal projection matrix to obtain corresponding low-dimensional features; inputting the low-dimensional features into a nearest neighbor classifier for identification; according to the method, feature extraction can be carried out on a picture containing noise, the influence caused by the noise is effectively suppressed, and the robustness of recognition is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a small-size image recognition method, device, equipment, medium and product. Background Art

[0002] With the rapid development of artificial intelligence technology and its wide application in various fields, the demand for image data processing has shown explosive growth. However, the complexity and diversity of image data itself and the various interferences it faces during acquisition, storage and dissemination have brought huge challenges to computer vision tasks. The presence of image noise not only reduces the quality of the image, but also causes the computer to be unable to accurately extract key information during analysis and recognition, which seriously affects the reliability and accuracy of the recognition results. Computers are highly dependent on the support of algorithms and models in feature extraction and pattern recognition. Therefore, many feature extraction methods for noisy images have been proposed, and it is hoped to find an efficient, robust and adaptable feature extraction method.

[0003] PCA is a classic feature extraction method, but it is not very robust. In the process of improving it, the main thing is to change the distance metric to make it robust, such as PCA-L1, HQ-PCA, L2P-PCA, etc. These methods can reduce the impact of noise on the image but cannot achieve the purpose of removing noise. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a small-size image recognition method, device, equipment, medium and product that can extract features from noisy images, effectively suppress the impact of noise, and improve the robustness of small-size image recognition.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions: In a first aspect, the present invention provides a small-size image recognition method, comprising: Acquire a small-size training image, and preprocess the small-size training image to generate a training image matrix; splicing all the training image matrices to generate an image augmentation matrix while retaining the original spatial structure; Constructing a 2DERPCA optimization model of enhanced robust principal component analysis combining low rank and low dimension, and inputting the image augmentation matrix into the 2DERPCA optimization model to calculate the optimal projection matrix; Acquire a small-sized image to be identified, and preprocess the small-sized image to be identified to generate an image matrix to be identified; Using the optimal projection matrix, extracting features from the to-be-recognized image matrix and the training image matrix to obtain corresponding low-dimensional features; The low-dimensional features are input into a nearest neighbor classifier for small-size image recognition.

[0006] Optionally, the preprocessing of the small-size training image to generate a training image matrix includes: A set noise is added to the small-size training image, and grayscale and normalization processing is performed to generate a training image matrix.

[0007] Optionally, the objective function of the 2DERPCA optimization model is: ; ; In the formula, is the input image augmentation matrix, is a sparse matrix, is a low-rank matrix, , , is a low-rank matrix , image augmentation matrix And sparse matrices No. Column vector, is a low-rank matrix No. Column vector, is a low-rank matrix The number of columns, , is the nuclear norm, are the 1-norm and 2-norm, is the projection matrix, for dimensional identity matrix, is a constant parameter.

[0008] Optionally, inputting the image augmentation matrix into the 2DERPCA optimization model to calculate the optimal projection matrix includes: Formulate the solution equation using the augmented Lagrange multiplier method : ; In the formula, is the Lagrange multiplier, is the inner product, is the F-norm, is a constant parameter; Solving the equation Simplifying, we get: ; Define intermediate variables: ; ; ; In the formula, For the matrix Middle OK The elements of the column, For the matrix Middle OK Elements of columns, matrix Divide elements The rest of the elements are 0; Based on intermediate variables The simplified solution equation By transformation we can get: ; In the formula, is the trace operation of the matrix; Use alternating direction multiplication and division to solve the transformed equation Iteration operation is performed until the objective function of the 2DERPCA optimization model converges, and the optimal projection matrix is ​​output.

[0009] Optionally, the transformed solution equation is solved by using alternating direction multiplication and division The iterative operations include: Fixed sparse matrix , low-rank matrix , intermediate variables , projection matrix , Lagrange multipliers , iteratively update the low-rank matrix Get the updated low-rank matrix : ; Define intermediate variables: ; ; Based on intermediate variables For low-rank matrices By transformation we can get: ; Fixed sparse matrix , low-rank matrix , intermediate variables , projection matrix , Lagrange multipliers , iteratively update the low-rank matrix Get the updated low-rank matrix : ; Fixed low-rank matrix , intermediate variables , projection matrix , Lagrange multipliers , iteratively update the sparse matrix Get the updated sparse matrix : ; Fixed low-rank matrix , sparse matrix , intermediate variables , Lagrange multipliers , iteratively update the projection matrix Get the updated projection matrix : ; Fixed low-rank matrix , sparse matrix , projection matrix , Lagrange multipliers , iteratively update the intermediate variables Get the updated projection matrix : ; ; ; In the formula, For the matrix Middle OK The elements of the column, For the matrix Middle OK Elements of columns, matrix Divide elements The rest of the elements are 0; is a low-rank matrix No. Column vector; Fixed low-rank matrix , sparse matrix , intermediate variables , projection matrix , iteratively update the Lagrange multiplier Get the updated Lagrange multiplier : ; .

[0010] In a second aspect, the present invention provides a small-size image recognition device, characterized in that it includes: A training data processing module is configured to obtain a small-sized training image and pre-process the small-sized training image to generate a training image matrix; A training data augmentation module is configured to concatenate all the training image matrices to generate an image augmentation matrix under the condition of retaining the original spatial structure; A model building and solving module is configured to build a 2DERPCA optimization model of enhanced robust principal component analysis combining low rank and low dimension, and input the image augmentation matrix into the 2DERPCA optimization model to calculate the optimal projection matrix; The to-be-recognized data processing module is configured to obtain the to-be-recognized small-size image, and pre-process the to-be-recognized small-size image to generate an to-be-recognized image matrix; A low-dimensional feature extraction module is configured to perform feature extraction on the to-be-recognized image matrix and the training image matrix using the optimal projection matrix to obtain corresponding low-dimensional features; The low-dimensional feature recognition module is configured to input the low-dimensional features into a nearest neighbor classifier for small-size image recognition.

[0011] Optionally, the objective function of the 2DERPCA optimization model is: ; ; In the formula, is the input image augmentation matrix, is a sparse matrix, is a low-rank matrix, , , is a low-rank matrix , image augmentation matrix And sparse matrices No. Column vector, is a low-rank matrix No. Column vector, is a low-rank matrix The number of columns, , is the nuclear norm, are the 1-norm and 2-norm, is the projection matrix, for dimensional identity matrix, is a constant parameter.

[0012] In a third aspect, the present invention provides an electronic device, including a processor and a storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps according to the above method.

[0013] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0014] In a fifth aspect, the present invention provides a computer program product, comprising a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a small-size image recognition method, device, equipment, medium and product, which introduces the 2D concept, pre-processes the small-size training images, and expands all the small-size training images into augmented matrices as input. The input augmented matrix is ​​subjected to low-rank and low-dimensional enhanced robust principal component analysis (2DERPCA) to obtain the optimal projection matrix. Finally, the optimal projection matrix is ​​used to extract features of the small-size image to be identified to obtain low-dimensional features, and the low-dimensional features are used to perform small-size image recognition prediction to improve the robustness of small-size image recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of a small-size image recognition method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0018] Embodiment 1:

[0019] like Figure 1 As shown, an embodiment of the present invention provides a small-size image recognition method, comprising the following steps: Step S1: Obtain a small-size training image, and preprocess the small-size training image to generate a training image matrix.

[0020] Specifically in this embodiment, the preprocessing includes: adding set noise to the small-size training image, graying and normalizing it, and generating a training image matrix. By adding noise, the influence of the noise can be learned and suppressed in the subsequent training process.

[0021] Step S2: All training image matrices are concatenated to generate an image augmentation matrix while retaining the original spatial structure.

[0022] Specifically in this implementation, all the selected small-size training images have the same size, so their original spatial structures are also the same, and the image augmentation matrix can be generated by splicing them in sequence.

[0023] Step S3, construct a 2DERPCA optimization model of enhanced robust principal component analysis combining low rank and low dimension, and input the image augmentation matrix into the 2DERPCA optimization model to calculate the optimal projection matrix.

[0024] PCA is a classic feature extraction method, but this method does not have good robustness. ERPCA is obtained by combining low rank and low dimension with enhanced robust principal component analysis, and then expanded to 2D to finally obtain 2DERPCA.

[0025] Specifically in this embodiment, the objective function of the 2DERPCA optimization model is: ; ; In the formula, is the input image augmentation matrix, is a sparse matrix, is a low-rank matrix, , , is a low-rank matrix , image augmentation matrix And sparse matrices No. Column vector, is a low-rank matrix No. Column vector, is a low-rank matrix The number of columns, , is the nuclear norm, are the 1-norm and 2-norm, is the projection matrix, for dimensional identity matrix, is a constant parameter.

[0026] Specifically in this embodiment, the image augmentation matrix is ​​input into the 2DERPCA optimization model to calculate the optimal projection matrix, including: (1) Use the augmented Lagrange multiplier method to construct the solution equation : ; In the formula, is the Lagrange multiplier, is the inner product, for F -norm, is a constant parameter; (2) Solving the equation To simplify: ; ; After simplification, we can get: ; (3) Define intermediate variables: ; ; ; In the formula, For the matrix Middle OK The elements of the column, For the matrix Middle OK Elements of columns, matrix Divide elements The rest of the elements are 0; (4) Based on intermediate variables The simplified solution equation By transformation we can get: ; In the formula, is the trace operation of the matrix; (5) Use alternating direction multiplication and division to solve the transformed equation Iterate until the objective function of the 2DERPCA optimization model converges and outputs the optimal projection matrix.

[0027] Among them, the transformed solution equation is solved by alternating direction multiplication and division The iterative operations include: (5.1) Fixed sparse matrix , low-rank matrix , intermediate variables , projection matrix , Lagrange multipliers , iteratively update the low-rank matrix Get the updated low-rank matrix : ; Define intermediate variables: ; ; Based on intermediate variables For low-rank matrices By transformation we can get: ; The push process is as follows: ; (5.2) Fixed sparse matrix , low-rank matrix , intermediate variables , projection matrix , Lagrange multipliers , iteratively update the low-rank matrix Get the updated low-rank matrix : ; The push process is as follows: ; The above formula satisfies the Sylvester equation: ; (5.3) Fixed low-rank matrix , intermediate variables , projection matrix , Lagrange multipliers , iteratively update the sparse matrix Get the updated sparse matrix : ; The push process is as follows: ; (5.4) Fixed low-rank matrix , sparse matrix , intermediate variables , Lagrange multipliers , iteratively update the projection matrix Get the updated projection matrix : ; (5.5) Fixed low-rank matrix , sparse matrix , projection matrix , Lagrange multipliers , iteratively update the intermediate variables Get the updated projection matrix : ; ; ; In the formula, For the matrix Middle OK The elements of the column, For the matrix Middle OK Elements of columns, matrix Divide elements The rest of the elements are 0; is a low-rank matrix No. Column vector; (5.6) Fixed low-rank matrix , sparse matrix , intermediate variables , projection matrix , iteratively update the Lagrange multiplier Get the updated Lagrange multiplier : ; .

[0028] Step S4: obtaining a small-sized image to be identified, and preprocessing the small-sized image to be identified to generate an image matrix to be identified.

[0029] Compared with the preprocessing in step S2, the preprocessing here does not need to add the set noise, and only grayscale and normalization processing is performed. Grayscale and normalization processing facilitates subsequent feature extraction operations.

[0030] Step S5: Use the optimal projection matrix to perform feature extraction on the image matrix to be identified and the training image matrix to obtain corresponding low-dimensional features.

[0031] Step S6: input the low-dimensional features into a nearest neighbor classifier for small-size image recognition.

[0032] The nearest neighbor classifier uses the label of the small-sized training image with high similarity as the recognition result based on the similarity between the low-dimensional features corresponding to the image matrix to be recognized and the training image matrix. This embodiment applies the solved optimal projection matrix to feature extraction, effectively maps the image matrix to a low-dimensional space, and ensures that important features of the image are extracted, thereby improving the performance of the model.

[0033] In the embodiment of the present invention, it is considered to test in two sets of data sets, Extended Yale B and AR. At the same time, it is compared with models such as PCA, PCA_L1, 2DPCA, 2DPCA_L1, 2DPCA, ERPCA, etc. to analyze whether the overall expressiveness of the model in this paper is better than other models. In the Yale B data set, 50% of the pictures in the database are randomly selected to add salt and pepper noise, and the noise ratio varies between 2-20 (%) of the image pixels and is randomly distributed. 50% of the images are randomly selected as training data, and the rest are used as test data. In the AR database, 50% of the pictures in the data set are randomly selected to add Gaussian noise, and the noise ratio varies between 2-20 (%) of the image pixels and is randomly distributed. 50% of the images are randomly selected as training data, and the rest are used as test data. The test results are shown in Tables 1, 2, 3 and 4.

[0034] Table 1: Recognition rates at different dimensions in the Yale B face database

[0035] Table 2: Average recognition accuracy and reconstruction error in the Yale B face database

[0036] It can be seen from Table 1 and Table 2 that the recognition accuracy of the 2DERPCA method proposed in this embodiment in the Yale B data set is higher than the average recognition accuracy of other methods, and the reconstruction error of the 2DERPCA method is relatively small.

[0037] Table 3: Recognition rates at different dimensions in the AR face database

[0038] Table 4: Average recognition accuracy and reconstruction error in the Yale B face database

[0039] It can be seen from Tables 3 and 4 that the recognition accuracy of the 2DERPCA method proposed in this embodiment in the AR data set is higher than the average recognition accuracy of other methods, and the reconstruction error of the 2DERPCA method is relatively small.

[0040] Embodiment 2: An embodiment of the present invention provides a small-size image recognition device, comprising: A training data processing module is configured to obtain a small-sized training image and pre-process the small-sized training image to generate a training image matrix; The training data augmentation module is configured to concatenate all training image matrices to generate an image augmentation matrix while retaining the original spatial structure; The model building and solving module is configured to build a 2DERPCA optimization model of enhanced robust principal component analysis combining low rank and low dimension, and input the image augmentation matrix into the 2DERPCA optimization model to calculate the optimal projection matrix; The to-be-recognized data processing module is configured to obtain the to-be-recognized small-size image, and pre-process the to-be-recognized small-size image to generate an to-be-recognized image matrix; A low-dimensional feature extraction module is configured to use an optimal projection matrix to perform feature extraction on the to-be-recognized image matrix and the training image matrix to obtain corresponding low-dimensional features; The low-dimensional feature recognition module is configured to input the low-dimensional features into a nearest neighbor classifier for small-size image recognition.

[0041] Among them, the objective function of the 2DERPCA optimization model is: ; ; In the formula, is the input image augmentation matrix, is a sparse matrix, is a low-rank matrix, , , is a low-rank matrix , image augmentation matrix And sparse matrices No. Column vector, is a low-rank matrix No. Column vector, is a low-rank matrix The number of columns, , is the nuclear norm, are the 1-norm and 2-norm, is the projection matrix, for dimensional identity matrix, is a constant parameter.

[0042] Embodiment three:

[0043] Based on the small-size image recognition method proposed in the first embodiment, an embodiment of the present invention provides an electronic device, including a processor and a storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps according to the above method.

[0044] Embodiment 4:

[0045] Based on the small-size image recognition method proposed in the first embodiment, the embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the above method are implemented.

[0046] Embodiment five:

[0047] Based on the small-size image recognition method proposed in the first embodiment, a computer program product in the embodiment of the present invention includes a computer program / instruction, and the steps of the above method are implemented when the computer program / instruction is executed by a processor.

[0048] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0049] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0050] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0051] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0052] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A small-size image recognition method, characterized in that: include: Acquire a small-size training image, and preprocess the small-size training image to generate a training image matrix; splicing all the training image matrices to generate an image augmentation matrix while retaining the original spatial structure; Constructing a 2DERPCA optimization model of enhanced robust principal component analysis combining low rank and low dimension, and inputting the image augmentation matrix into the 2DERPCA optimization model to calculate the optimal projection matrix; Acquire a small-sized image to be identified, and preprocess the small-sized image to be identified to generate an image matrix to be identified; Using the optimal projection matrix, extracting features from the to-be-recognized image matrix and the training image matrix to obtain corresponding low-dimensional features; The low-dimensional features are input into a nearest neighbor classifier for small-size image recognition.

2. The small-size image recognition method according to claim 1, characterized in that: The preprocessing of the small-size training image to generate a training image matrix comprises: A set noise is added to the small-size training image, and grayscale and normalization processing is performed to generate a training image matrix.

3. The small-size image recognition method according to claim 1, characterized in that: The objective function of the 2DERPCA optimization model is: ； ； In the formula, is the input image augmentation matrix, is a sparse matrix, is a low-rank matrix, , , is a low-rank matrix , image augmentation matrix And sparse matrices No. Column vector, is a low-rank matrix No. Column vector, is a low-rank matrix The number of columns, , is the nuclear norm, are the 1-norm and 2-norm, is the projection matrix, for dimensional identity matrix, is a constant parameter.

4. The small-size image recognition method according to claim 3, characterized in that: The step of inputting the image augmentation matrix into the 2DERPCA optimization model to calculate the optimal projection matrix comprises: Formulate the solution equation using the augmented Lagrange multiplier method : ； In the formula, is the Lagrange multiplier, is the inner product, is the F-norm, is a constant parameter; Solving the equation Simplifying, we get: ； Define intermediate variables: ； ； ； In the formula, For the matrix Middle OK The elements of the column, For the matrix Middle OK Elements of columns, matrix Divide elements The rest of the elements are 0; Based on intermediate variables The simplified solution equation By transformation we can get: ； In the formula, is the trace operation of the matrix; Use alternating direction multiplication and division to solve the transformed equation Iteration operation is performed until the objective function of the 2DERPCA optimization model converges, and the optimal projection matrix is ​​output.

5. The small-size image recognition method according to claim 4, characterized in that: The alternating direction multiplication and division method is used to solve the transformed equation The iterative operations include: Fixed sparse matrix , low-rank matrix , intermediate variables , projection matrix , Lagrange multipliers , iteratively update the low-rank matrix Get the updated low-rank matrix : ； Define intermediate variables: ； ； Based on intermediate variables For low-rank matrices By transformation we can get: ； Fixed sparse matrix , low-rank matrix , intermediate variables , projection matrix , Lagrange multipliers , iteratively update the low-rank matrix Get the updated low-rank matrix : ； Fixed low-rank matrix , intermediate variables , projection matrix , Lagrange multipliers , iteratively update the sparse matrix Get the updated sparse matrix : ； Fixed low-rank matrix , sparse matrix , intermediate variables , Lagrange multipliers , iteratively update the projection matrix Get the updated projection matrix : ； Fixed low-rank matrix , sparse matrix , projection matrix , Lagrange multipliers , iteratively update the intermediate variables Get the updated projection matrix : ； ； ； In the formula, For the matrix Middle OK The elements of the column, For the matrix Middle OK Elements of columns, matrix Divide elements The rest of the elements are 0; is a low-rank matrix No. Column vector; Fixed low-rank matrix , sparse matrix , intermediate variables , projection matrix , iteratively update the Lagrange multiplier Get the updated Lagrange multiplier : ； 。 6. A small-size image recognition device, characterized in that: include: A training data processing module is configured to obtain a small-sized training image and pre-process the small-sized training image to generate a training image matrix; A training data augmentation module is configured to concatenate all the training image matrices to generate an image augmentation matrix under the condition of retaining the original spatial structure; A model building and solving module is configured to build a 2DERPCA optimization model of enhanced robust principal component analysis combining low rank and low dimension, and input the image augmentation matrix into the 2DERPCA optimization model to calculate the optimal projection matrix; The to-be-recognized data processing module is configured to obtain the to-be-recognized small-size image, and pre-process the to-be-recognized small-size image to generate an to-be-recognized image matrix; A low-dimensional feature extraction module is configured to perform feature extraction on the to-be-recognized image matrix and the training image matrix using the optimal projection matrix to obtain corresponding low-dimensional features; The low-dimensional feature recognition module is configured to input the low-dimensional features into a nearest neighbor classifier for small-size image recognition.

7. The small-size image recognition device according to claim 6, characterized in that: The objective function of the 2DERPCA optimization model is: ； ； In the formula, is the input image augmentation matrix, is a sparse matrix, is a low-rank matrix, , , is a low-rank matrix , image augmentation matrix And sparse matrices No. Column vector, is a low-rank matrix No. Column vector, is a low-rank matrix The number of columns, , is the nuclear norm, are the 1-norm and 2-norm, is the projection matrix, for dimensional identity matrix, is a constant parameter.

8. An electronic device, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images