Palm print multi-feature fusion identification method and device based on multi-view learning

Abstract

The invention provides a palmprint multi-feature fusion recognition method and device based on multi-view learning. The method comprises the steps that multiple palmprint view features are extracted from a palmprint region-of-interest image; for any palmprint view feature, using a linear discriminant analysis method to maximize the variance between different categories, and minimizing the variance between different samples of the same category to obtain the palmprint view feature after projection dimensionality reduction; using a depth generalized canonical correlation analysis method to extract correlation features among the palmprint view features after projection dimension reduction, and using the correlation features as palmprint fusion features; and performing classification and identification according to the palmprint fusion feature by using a classification method of collaborative representation. The palmprint features of various different views can be fused for recognition, and the stability and accuracy of palmprint feature recognition can be improved.

Description

Technical Field

[0001] The present application belongs to the technical field of palmprint recognition, and in particular, relates to a palmprint multi-feature fusion recognition method and device based on multi-view learning. Background Art

[0002] Biometric technology uses biological or behavioral characteristics to automatically identify identity and is considered one of the most reliable and effective solutions for identity verification, such as face recognition, fingerprint recognition, iris recognition, and palm print recognition. Biometric authentication has become a key component of secure access control systems. Palm print features are difficult to forge and have unique and basically lifelong characteristics, which has attracted more and more research attention.

[0003] Palmprints contain stable features, such as ridges, textures, and main lines, which make palmprint recognition have a high recognition accuracy. However, when performing recognition in different environments or with a small number of samples, the recognition accuracy will decrease. How to achieve a stable and high recognition accuracy by integrating multiple palmprint features in a complex environment is a problem currently faced in the application of palmprint recognition. Summary of the invention

[0004] The main purpose of the embodiments of the present invention is to provide a palmprint multi-feature fusion recognition method and device based on multi-view learning, which can fuse palmprint features from multiple different views for recognition, improve the stability and accuracy of palmprint feature recognition, and solve the problem of unstable and low palmprint recognition accuracy.

[0005] In a first aspect, a palmprint multi-feature fusion recognition method based on multi-view learning is provided, the method comprising: extracting a plurality of palmprint view features from a palmprint region of interest image; for any of the palmprint view features, using a linear discriminant analysis method to maximize the variance between different categories and minimize the variance between different samples of the same category, to obtain the palmprint view feature after projection dimensionality reduction; using a deep generalized canonical correlation analysis method to extract correlation features between the palmprint view features after projection dimensionality reduction as palmprint fusion features; and performing classification and recognition based on the palmprint fusion features using a collaborative representation classification method.

[0006] In a possible implementation, the palmprint view features include palmprint texture features, palmprint direction features and palmprint depth convolution features. The multiple palmprint view features are extracted from the palmprint region of interest image, including: uniformly dividing each subspace of the palmprint region of interest image to obtain regional pixel blocks, applying an improved local binary pattern algorithm to extract texture information and position information of all regional pixel blocks, splicing and fusing the texture information and the position information to construct palmprint texture features; applying 2D Gabor filters in 10 directions to obtain convolution responses with the palmprint region of interest image in 10 directions, calculating the convolution response differences between adjacent directions by weighted fusion, and combining to form palmprint direction features; extracting palmprint depth features in the palmprint region of interest image by a Vanillanet lightweight neural network to obtain a feature matrix, and forming the palmprint depth convolution features after matrix transformation.

[0007] In another possible implementation, the application of the improved local binary pattern algorithm to extract texture information and position information of all regional pixel blocks includes: for any regional pixel block, taking the regional pixel block as the central regional pixel block, calculating the absolute value of the difference between the total average grayscale value of the adjacent 8 regional pixel blocks and the average grayscale value of the central regional pixel block, and obtaining a first threshold; applying the comparison of the average grayscale value of the regional pixel blocks to replace the comparison of single pixel points in the original local binary pattern algorithm, and extracting the texture information of the central regional pixel block based on the following relationship: in, is the average gray value of the pixel block in the central area, is the average gray value of the adjacent regional pixel blocks, P is the number of adjacent regional pixel blocks, P=8, R is the radius of each regional pixel block, and LBP represents the local binary pattern algorithm; in the average gray value of the adjacent regional pixel blocks, the maximum and minimum values ​​are determined, and the coordinate axis is established in the central regional pixel block. The span of each regional pixel block is regarded as 1, and the distance between the two is calculated as the position information of the central regional pixel block.

[0008] In another possible implementation, calculating the distance between the two includes: applying the following relationship to calculate the distance between the two: in, is the coordinate of the central area block, is the coordinate value of the pixel block in the area with the largest average gray value in the adjacent area. It is the coordinate value of the pixel block in the area with the smallest average gray value in the adjacent area. LBP represents the local binary pattern algorithm.

[0009] In another possible implementation, the linear discriminant analysis method is used to maximize the variance between different categories and minimize the variance between different samples of the same category to obtain the palmprint view feature after projection dimensionality reduction, including: performing projection mapping on each sample in the palmprint view feature; calculating the mean of data samples of different categories after projection mapping and the mean of all samples after projection mapping; calculating the inter-class scatter matrix and the intra-class scatter matrix after projection mapping according to the data sample mean and the mean of all samples; learning the projection matrix in the palmprint view based on the optimization objective function according to the inter-class scatter matrix and the intra-class scatter matrix, performing projection dimensionality reduction on the palmprint view feature, and obtaining the palmprint view feature after projection dimensionality reduction.

[0010] In another possible implementation, the extracting of correlation features between the palmprint view features after projection dimensionality reduction using a deep generalized canonical correlation analysis method as palmprint fusion features includes: applying a multi-layer neural network to process the palmprint view features after projection dimensionality reduction respectively to obtain nonlinear mapping features of the palmprint view features; and applying a deep generalized canonical correlation analysis model to maximize the correlation features between the nonlinear mapping features to obtain palmprint fusion features.

[0011] In another possible implementation, the classification and identification is performed by applying the collaborative representation classification method according to the palmprint fusion feature, including: constructing a training sample set A: , Indicates No. The fusion features of palmprint samples, , , is the total number of palmprint sample categories, is the first The number of class samples, represents the dimension of each palmprint fusion feature vector, Represented as the total number of training feature samples; taking the palmprint fusion feature as the test sample, the palmprint fusion feature in the training sample set is used to linearly represent the fusion feature vector of a test sample : , , is the coefficient matrix, Indicates The coefficient vector of the class; the distance between the fused feature vector of the test sample and the fused feature vector of each class of training samples is used as the weight to calculate the coefficient matrix Calculate the fusion feature vector of the test sample With The matching error between the classes is minimized, and the fusion feature vector of the test sample is determined. For the kind.

[0012] In a second aspect, a palmprint multi-feature fusion recognition device based on multi-view learning is provided, and the device includes: a data acquisition module, used to extract multiple palmprint view features from the palmprint area of ​​interest image; a projection dimension reduction module, used to maximize the variance between different categories and minimize the variance between different samples of the same category for any of the palmprint view features, so as to obtain the palmprint view features after projection dimension reduction; a feature fusion module, used to extract the correlation features between the palmprint view features after projection dimension reduction by using the deep generalized canonical correlation analysis method as the palmprint fusion feature; a classification recognition module, used to perform classification recognition according to the palmprint fusion feature by applying the classification method of collaborative representation.

[0013] In a third aspect, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, a palmprint multi-feature fusion recognition method based on multi-view learning as provided in the first aspect is implemented.

[0014] In a fourth aspect, a non-transitory computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a palmprint multi-feature fusion recognition method based on multi-view learning as provided in the first aspect is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in describing the embodiments of the present application are briefly introduced below.

[0016] Figure 1 A flowchart of a palmprint multi-feature fusion recognition method based on multi-view learning provided by an embodiment of the present invention; Figure 2 A structural diagram of a palmprint multi-feature fusion recognition device based on multi-view learning provided by one embodiment of the present invention; Figure 3 A schematic diagram of the physical structure of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0017] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar modules or modules with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present invention.

[0018] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, modules and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, modules, components and / or groups thereof. It should be understood that when we refer to a module as being "connected" or "coupled" to another module, it may be directly connected or coupled to the other modules, or there may be intermediate modules. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any modules and all combinations of one or more associated listed items.

[0019] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation method of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0020] The technical solution of the present application and how to solve the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0021] like Figure 1 FIG. 1 is a flowchart of a palmprint multi-feature fusion recognition method based on multi-view learning provided by an embodiment of the present invention, wherein the method comprises: Step S11, extracting a variety of palmprint view features from the palmprint region of interest image.

[0022] Specifically, three view features are extracted from the palmprint region of interest image at the same time. The palmprint texture feature is extracted from the palmprint region of interest image by an improved local binary pattern (LBP) algorithm, the palmprint direction feature is extracted from the palmprint region of interest image by an improved Gabor filter, and the palmprint depth convolution feature is extracted from the palmprint region of interest image by a Vanillanet lightweight neural network.

[0023] Step S12: for any of the palmprint view features, a linear discriminant analysis method is used to maximize the variance between different categories and minimize the variance between different samples of the same category, so as to obtain the palmprint view feature after projection dimensionality reduction.

[0024] Specifically, the Linear Discriminant Analysis (LDA) method is used for each view feature to maximize the variance between different categories and minimize the variance between different samples of the same category, so as to effectively distinguish samples of different categories.

[0025] Step S13, using a deep generalized canonical correlation analysis method to extract correlation features between the palmprint view features after projection dimensionality reduction as palmprint fusion features.

[0026] The deep generalized canonical correlation analysis method is used to extract the correlation features between the three palmprint view features as discriminant features, eliminate redundant information, and obtain more discriminative and representative palmprint fusion features.

[0027] Step S14, performing classification and identification using a collaborative representation classification method according to the palmprint fusion features.

[0028] Recognition is performed through a collaborative representation classification method, that is, the palmprint fusion feature vector is used to construct a training sample set to represent the relationship between samples and capture the intrinsic structure of the data. The distance between the fusion feature vector of the test sample and the fusion feature vector of each type of training sample is used as the weight, and the linear representation of the test sample on the training sample set is learned for classification and recognition.

[0029] The embodiment of the present invention performs palmprint multi-feature fusion recognition based on multi-view learning, which can fuse palmprint features from multiple different views for recognition, thereby improving the stability and accuracy of palmprint feature recognition and solving the problem of unstable and low palmprint recognition accuracy.

[0030] In the embodiment of the present invention, in step S11, each subspace of the palmprint region of interest image is evenly divided into blocks to obtain regional pixel blocks, and an improved local binary pattern algorithm is applied to extract texture information and position information of all regional pixel blocks, and the texture information and the position information are spliced ​​and fused to construct the palmprint texture feature (view1).

[0031] Optionally, first, for any regional pixel block, take the regional pixel block as the central regional pixel block, calculate the absolute value of the difference between the total average grayscale value of the adjacent 8 regional pixel blocks and the average grayscale value of the central regional pixel block, and obtain the first threshold. When extracting features, evenly divide each subspace into blocks, and after block division, use the comparison of the average grayscale value of the regional block to replace the comparison of the single pixel points in the original LBP. Calculate the absolute value of the difference between the total average grayscale value of the surrounding 8 adjacent regional pixel blocks and the average grayscale value of the central regional block pixels, and use it as the first threshold : (1) Then, the comparison of the average grayscale value of the regional pixel block is applied to replace the comparison of the single pixel point in the original local binary pattern algorithm, and the texture information of the central regional pixel block is extracted based on the following relationship: (2) (3) in, is the average gray value of the pixel block in the central area, is the average gray value of the adjacent pixel blocks, P is the number of adjacent pixel blocks, P=8, R is the radius of each pixel block, LBP represents the local binary pattern algorithm, , is the first threshold. Each regional pixel block adopts a 3×3 square area, R=3.

[0032] The maximum and minimum values ​​are determined in the average grayscale values ​​of the adjacent pixel blocks, and a coordinate axis is established in the central pixel block. The span of each pixel block is regarded as 1, and the distance between the two is calculated as the position information of the central pixel block. The following relationship is used to calculate the distance between the two: (4) in, is the coordinate of the central area block, is the coordinate value of the pixel block in the area with the largest average gray value in the adjacent area. It is the coordinate value of the regional pixel block with the smallest average gray value in the adjacent region. LBP stands for Local Binary Pattern Algorithm. If the pixel values ​​of all regional pixel blocks are equal, it is defined as 1.

[0033] The texture information of all pixels in the palmprint region of interest is extracted using formula (2). The position information is extracted using formula (4), and the two pieces of information are combined to construct the palmprint texture feature.

[0034] In the embodiment of the present invention, 2D Gabor filters in 10 directions are applied to obtain the convolution responses of the palmprint region of interest image in 10 directions, and the convolution response differences between adjacent directions are calculated by weighted fusion, and combined to form the palmprint direction feature (view2). The 2D Gabor filter is shown in formula (5).

[0035] (5) in, represents the standard deviation of the Gaussian function, , is the frequency of the sine wave, represents the direction of the Gabor filter. As shown in formula (6), 2D Gabor filters in 10 directions are selected and the responses are obtained, which are recorded as , , the corresponding direction is , It is the convolution response of the Gabor template with the palmprint region of interest image in 10 directions, and the information in multiple directions is obtained. I (x, y) represents the palmprint region of interest image.

[0036] (6) in," ” is the convolution operator.

[0037] The convolution response differences between adjacent directions are calculated by weighted fusion, as shown in formula (7): (7) According to formula (8), the palmprint direction feature (view2) is formed: (8) in, , represents the corresponding weight, and and The sum is 1.

[0038] The palm print depth features in the palm print region of interest image are extracted by the Vanillanet lightweight neural network to obtain a feature matrix, which is transformed into the palm print depth convolution feature (view3). Specifically, the cosine annealing algorithm is used to train the Vanillanet lightweight network for feature extraction, and then the trained Vanillanet network is used to extract the palm print depth features in the palm print region of interest image to obtain a feature matrix, which is transformed into the palm print depth convolution feature.

[0039] In step S12, optionally, projection mapping is performed on each sample in the palmprint view feature; the mean values ​​of data samples of different categories after projection mapping and the mean value of all samples after projection mapping are calculated; the inter-class scatter matrix and the intra-class scatter matrix after projection mapping are calculated according to the data sample mean and the mean value of all samples; the projection matrix in the palmprint view is learned based on the optimization objective function according to the inter-class scatter matrix and the intra-class scatter matrix, and projection dimensionality reduction is performed on the palmprint view feature to obtain the palmprint view feature after projection dimensionality reduction.

[0040] Taking palm print texture features as an example, The texture view feature Class samples, through formula (9) Projection mapping is performed, and the mean values ​​of data samples of different categories after mapping are calculated according to formula (10), and the mean values ​​of all samples after mapping are calculated according to formula (11); the inter-class scatter matrix after mapping is calculated according to formula (12), and the intra-class scatter matrix after mapping is calculated according to formula (13).

[0041] (9) (10) (11) (12) (13) is the projection matrix in the palmprint view. The optimization objective function is designed as shown in formula (14), the projection matrix in the view is learned, and the palmprint texture features are projected and reduced to obtain the palmprint texture features after projection dimensionality reduction. Then, the palmprint direction features and palmprint depth convolution features are projected and reduced to obtain the palmprint direction features and palmprint depth convolution features after projection dimensionality reduction.

[0042] (14) in, Represents trace operation.

[0043] In step S13, optionally, a multi-layer neural network is first applied to process each of the palmprint view features after projection dimension reduction to obtain nonlinear mapping features of each of the palmprint view features. Each palmprint view feature after projection dimension reduction is passed through a multi-layer neural network. represents the input view features, and , . Let the variable represents the data sample size, Indicates The dimension of the view feature. The network of views has Layer, View Layer neurons, and the output layer has Neuron, View Output of the layer As shown in formula (15), is a non-linear activation function (ReLU), is the weight matrix, is the bias vector. The output of the last layer of the network is expressed as (16) to obtain the nonlinear mapping feature of the view , It is a summary of the overall function transformation of multi-layer neural networks.

[0044] (15) , (16) Then, a deep generalized canonical correlation analysis model is applied to maximize the correlation characteristics between the nonlinear mapping features to obtain the palmprint fusion features. To fuse features, calculate With fusion features The reconstruction error is taken as the objective function to construct the deep generalized canonical correlation analysis model according to formula (17). represents the data sample size, For the The projection change of the output of the network, variable express The dimension of and is the penalty parameter.

[0045] (17) in, is an identity matrix. Under these conditions, we aim to find Palmprint fusion features from different views .

[0046] In step S14, a training sample set A is constructed: (18) Indicates No. The palmprint fusion features of palmprint samples, , , is the total number of palmprint sample categories, is the first The number of class samples, Represents the dimension of each palmprint fusion feature, Represented as the total number of training feature samples.

[0047] The palmprint fusion feature is used as a test sample, and the palmprint fusion feature in the training sample set is used to linearly represent a fusion feature vector of a test sample. : , , is the coefficient matrix, Expressed as The coefficient vector of the class. Represents a fused feature vector of a test sample, then the fused feature vector in the training sample set can be linearly represented The coefficient matrix is ​​calculated by using the distance between the fused feature vector of the test sample and the fused feature vector of each type of training sample as the weight. The estimated value of . The weight matrix Determined by formula (19) (19) in, is the weight adjustment parameter. Then according to formula (20), we establish The objective function is: where λ is the regularization parameter, y is the feature of a given test sample, make sure The sparsity of .

[0048] (20) in, By differentiating formula (20), we can obtain: .

[0049] Calculate the fusion feature vector of the test sample With The matching error between the classes is minimized, and the fusion feature vector of the test sample is determined. For the Class. Specifically, the category residual can be calculated using formula (21): .in For vector With The matching error between classes, In and Category The associated items are not zero, and the others are zero. Use formula (22) to calculate the category of the test sample.

[0050] (twenty one) (twenty two) in, is the training sample set About The partial set vector of class samples. When the minimum value is reached, the corresponding The represented category is taken as the category of the test sample and is denoted by z.

[0051] In summary, the embodiment of the present invention extracts a variety of palmprint view features from the palmprint region of interest image; for any of the palmprint view features, a linear discriminant analysis method is used to maximize the variance between different categories and minimize the variance between different samples of the same category to obtain the palmprint view features after projection dimensionality reduction; a deep generalized canonical correlation analysis method is used to extract the correlation features between the palmprint view features after projection dimensionality reduction as palmprint fusion features; and a collaborative representation classification method is applied to perform classification and recognition according to the palmprint fusion features, so that palmprint features of multiple different views can be fused for recognition, thereby improving the stability and accuracy of palmprint feature recognition.

[0052] like Figure 2 FIG. 1 is a structural diagram of a palmprint multi-feature fusion recognition device based on multi-view learning provided by an embodiment of the present invention, wherein the device comprises: A data acquisition module, used for extracting various palmprint view features from the palmprint region of interest image; A projection dimension reduction module, for maximizing the variance between different categories and minimizing the variance between different samples of the same category for any of the palmprint view features, to obtain the palmprint view features after projection dimension reduction; A feature fusion module, used for extracting correlation features between the palmprint view features after projection dimensionality reduction by using a deep generalized canonical correlation analysis method as palmprint fusion features; The classification and recognition module is used to perform classification and recognition based on the palmprint fusion features using a collaborative representation classification method.

[0053] The device of the above embodiment is applied to the corresponding method in the above embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0054] Figure 3 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 3As shown, the electronic device may include: a processor 301, a communication interface 302, a memory 303 and a communication bus 304, wherein the processor, the communication interface and the memory communicate with each other through the communication bus. The processor may call the logic instructions in the memory to execute a palmprint multi-feature fusion recognition method based on multi-view learning, the method comprising: extracting a plurality of palmprint view features from the palmprint region of interest image; for any of the palmprint view features, using a linear discriminant analysis method to maximize the variance between different categories and minimize the variance between different samples of the same category, to obtain the palmprint view feature after projection dimensionality reduction; using a deep generalized canonical correlation analysis method to extract the correlation features between the palmprint view features after projection dimensionality reduction as palmprint fusion features; and applying a collaborative representation classification method to classify and identify the palmprint fusion features.

[0055] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0056] On the other hand, an embodiment of the present invention further provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute a palmprint multi-feature fusion recognition method based on multi-view learning provided by the above-mentioned method embodiments, the method comprising: extracting multiple palmprint view features from a palmprint region of interest image; for any of the palmprint view features, using a linear discriminant analysis method to maximize the variance between different categories and minimize the variance between different samples of the same category to obtain the palmprint view feature after projection dimensionality reduction; using a deep generalized canonical correlation analysis method to extract correlation features between the palmprint view features after projection dimensionality reduction as palmprint fusion features; and applying a collaborative representation classification method to perform classification and recognition according to the palmprint fusion features.

[0057] On the other hand, an embodiment of the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute a palmprint multi-feature fusion recognition method based on multi-view learning provided by the above embodiments, the method comprising: extracting a plurality of palmprint view features from a palmprint region of interest image; for any of the palmprint view features, using a linear discriminant analysis method to maximize the variance between different categories and minimize the variance between different samples of the same category to obtain the palmprint view feature after projection dimensionality reduction; using a deep generalized canonical correlation analysis method to extract correlation features between the palmprint view features after projection dimensionality reduction as palmprint fusion features; and performing classification and recognition based on the palmprint fusion features by applying a collaborative representation classification method.

[0058] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0059] The above is only a partial implementation 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 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 palmprint multi-feature fusion recognition method based on multi-view learning, characterized in that: The method comprises: Extracting multiple palmprint view features from the palmprint region of interest image; For any of the palmprint view features, a linear discriminant analysis method is used to maximize the variance between different categories and minimize the variance between different samples of the same category, so as to obtain the palmprint view feature after projection dimensionality reduction; A deep generalized canonical correlation analysis method is used to extract correlation features between the palmprint view features after projection dimensionality reduction as palmprint fusion features; Classification and recognition are performed using a collaborative representation classification method based on the palmprint fusion features.

2. The method according to claim 1, characterized in that The palmprint view features include palmprint texture features, palmprint direction features and palmprint depth convolution features. The multiple palmprint view features are extracted from the palmprint region of interest image, including: Each subspace of the palmprint region of interest image is uniformly divided into blocks to obtain regional pixel blocks, and the texture information and position information of all regional pixel blocks are extracted by applying an improved local binary pattern algorithm, and the texture information and the position information are spliced ​​and fused to construct palmprint texture features; Apply 2D Gabor filters in 10 directions to obtain the convolution responses of the palmprint region of interest image in 10 directions, and calculate the convolution response differences between adjacent directions by weighted fusion, and combine them to form the palmprint direction features; The palmprint depth features in the palmprint region of interest image are extracted by using a Vanillanet lightweight neural network to obtain a feature matrix, which is then transformed into the palmprint depth convolution feature.

3. The method according to claim 2, characterized in that The improved local binary pattern algorithm is used to extract texture information and position information of all regional pixel blocks, including: For any regional pixel block, taking the regional pixel block as the central regional pixel block, calculating the absolute value of the difference between the total average grayscale value of the adjacent eight regional pixel blocks and the average grayscale value of the central regional pixel block, to obtain a first threshold; The comparison of the average grayscale value of the regional pixel block is used instead of the comparison of the single pixel point in the local binary pattern algorithm, and the texture information of the central regional pixel block is extracted based on the following relationship: in, is the average gray value of the pixel block in the central area, is the average gray value of the adjacent pixel blocks, P is the number of adjacent pixel blocks, P=8, R is the radius of each pixel block, LBP represents the local binary pattern algorithm, , is the first threshold; The maximum and minimum values ​​are determined in the average grayscale values ​​of the adjacent area pixel blocks, a coordinate axis is established in the central area pixel block, the span of each area pixel block is regarded as 1, and the distance between the two is calculated as the position information of the central area pixel block.

4. The method according to claim 3, characterized in that The calculating the distance between the two comprises: The distance between the two is calculated using the following relationship: in, is the coordinate of the central area block, is the coordinate value of the pixel block in the area with the largest average gray value in the adjacent area. It is the coordinate value of the pixel block in the area with the smallest average gray value in the adjacent area. LBP represents the local binary pattern algorithm.

5. The method according to claim 1, characterized in that The linear discriminant analysis method is used to maximize the variance between different categories and minimize the variance between different samples of the same category to obtain the palmprint view features after projection dimensionality reduction, including: Performing projection mapping on each sample in the palmprint view feature; Calculate the mean of data samples of different categories after projection mapping and the mean of all samples after projection mapping; Calculating the inter-class scatter matrix and the intra-class scatter matrix after projection mapping according to the data sample mean and the all sample mean; The projection matrix in the palmprint view is learned based on the optimization objective function according to the inter-class scatter matrix and the intra-class scatter matrix, and the palmprint view features are projected and dimensionally reduced to obtain the palmprint view features after projection and dimension reduction.

6. The method according to claim 1, characterized in that The method of using the deep generalized canonical correlation analysis method to extract the correlation features between the palmprint view features after projection dimension reduction as the palmprint fusion features includes: Applying a multi-layer neural network to process the palmprint view features after projection dimensionality reduction respectively, to obtain nonlinear mapping features of the palmprint view features; A deep generalized canonical correlation analysis model is applied to maximize the correlation characteristics between the nonlinear mapping features to obtain palmprint fusion features.

7. The method according to claim 1, characterized in that The classification and identification is performed by applying a classification method based on collaborative representation according to the palmprint fusion features, including: Construct training sample set A: , Indicates No. The palmprint fusion features of palmprint samples, , , is the total number of palmprint sample categories, is the first The number of class samples, Represents the dimension of each palmprint fusion feature, Expressed as the total number of training samples; The palmprint fusion feature is used as a test sample, and the palmprint fusion feature in the training sample set is used to linearly represent a fusion feature vector of a test sample. : , , is the coefficient matrix, Indicates coefficient vector of classes; The coefficient matrix is ​​calculated by using the distance between the fused feature vector of the test sample and the fused feature vector of each type of training sample as the weight. An estimated value of Calculate the fusion feature vector of the test sample With The matching error between the classes is minimized, and the fusion feature vector of the test sample is determined. For the kind.

8. A palmprint multi-feature fusion recognition device based on multi-view learning, characterized in that: The device comprises: A data acquisition module, used for extracting various palmprint view features from the palmprint region of interest image; A projection dimension reduction module, for maximizing the variance between different categories and minimizing the variance between different samples of the same category for any of the palmprint view features, to obtain the palmprint view features after projection dimension reduction; A feature fusion module, used for extracting correlation features between the palmprint view features after projection dimensionality reduction by using a deep generalized canonical correlation analysis method as palmprint fusion features; The classification and recognition module is used to perform classification and recognition based on the palmprint fusion features using a collaborative representation classification method.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, a palmprint multi-feature fusion recognition method based on multi-view learning is implemented as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, a palmprint multi-feature fusion recognition method based on multi-view learning is implemented as described in any one of claims 1 to 7.

Images