Palm vein living body identification method, identification device and readable storage medium

Abstract

This application discloses a palm vein liveness detection method, which includes extracting the effective region of the palm vein from a palm vein image and performing orientation rotation correction on the effective region; using the first feature extraction layer of the backbone network to extract palm vein texture features from the orientation rotation-corrected effective region to obtain a texture feature map f1; performing texture enhancement processing on the texture feature map f1 to obtain a texture feature map T; using the second feature extraction layer of the backbone network to extract features from the texture feature map f1 to obtain a feature map f2; inputting the feature map f2 into an attention module that focuses on local texture regions of the palm vein to obtain an attention feature map D; performing average pooling on the obtained feature map f2 and attention feature map D to obtain a texture feature matrix P; and inputting the texture feature matrix P into the fully connected layer of the backbone network for palm vein liveness detection. This application also provides a storage medium and a detection device using the palm vein liveness detection method.

Description

Technical Field

[0001] This application relates to the field of image processing, and more specifically, to a method, device and readable storage medium for palm vein liveness recognition. Background Technology

[0002] Currently, fingerprint recognition and facial recognition are two common and mature identification technologies, widely used in identity verification, security checks, access control, and medical research. Palm vein recognition, as a more secure biometric feature, has been studied for decades. Its non-contact nature makes it highly promising for identification in public places. To ensure the security and accuracy of palm vein recognition, it is necessary to defend against spoofing attacks using non-living palm veins, such as using printed palm vein images for authentication. Therefore, the palm vein identification process must first verify whether the palm vein is a living vein. Summary of the Invention

[0003] In view of the prior art, the technical problem solved by this application is to provide a palm vein liveness identification method, identification device and computer-readable storage medium that can identify genuine and fake palm veins.

[0004] To address the aforementioned technical problems, in a first aspect, this application provides a method for palm vein liveness detection, comprising:

[0005] Extract the effective area of ​​the palm vein from the palm vein image, and perform orientation rotation correction on the effective area of ​​the palm vein so that the point between the base of the index finger and the base of the middle finger and the point between the base of the middle finger and the base of the ring finger are located on the same horizontal line parallel to the X-axis.

[0006] The first feature extraction layer of the backbone network is used to extract the palm vein texture features in the effective area of ​​the palm vein after orientation rotation correction, and the texture feature map f1 is obtained.

[0007] The texture feature map f1 is subjected to texture enhancement processing to obtain the texture feature map T;

[0008] The second feature extraction layer of the backbone network is used to extract features from the texture feature map f1 to obtain the feature map f2.

[0009] Input feature map f2 into the attention module that focuses on the local region of palm vein texture to obtain attention feature map D; and,

[0010] The obtained feature map f2 and attention feature map D are averaged to obtain the texture feature matrix P. The texture feature matrix P is then input into the fully connected layer of the backbone network to determine the authenticity of palm vein liveness.

[0011] The first and second feature extraction layers are both multiple convolutional layers distributed in the order of the backbone network.

[0012] In one possible implementation, the steps of performing texture enhancement processing on texture feature map f1 to obtain texture feature map T include:

[0013] The texture feature map f1 is downsampled to obtain the feature map t after average pooling;

[0014] Upsample and augment the feature map t to the same dimension as the texture feature map f1;

[0015] Aggregate the texture feature map f1 with the augmented feature map t;

[0016] The aggregation result is subjected to dense convolution to obtain the texture feature map T after texture enhancement.

[0017] In one possible implementation, the palm vein liveness detection method further includes:

[0018] The third feature extraction layer of the backbone network is used to extract features from feature map f2 to obtain deep feature map f3;

[0019] The deep feature map f3 is subjected to average pooling to obtain the global feature map G;

[0020] The texture feature matrix P and the global feature map G are input into the fully connected layer of the backbone network to determine the authenticity of palm vein liveness.

[0021] The third feature extraction layer consists of multiple convolutional layers, and both the second and third feature extraction layers are multiple convolutional layers distributed in the order of the backbone network.

[0022] In one possible implementation, the backbone network is a ResNet50 network, the first feature extraction layer is a convolutional layer from the first to the seventh layer of the ResNet50 network, the second feature extraction layer is a convolutional layer from the eighth to the fifteenth layer of the ResNet50 network, and the third feature extraction layer is a convolutional layer from the sixteenth to the forty-ninth layer of the ResNet50 network.

[0023] In one possible implementation, the step of performing orientational rotation correction on the effective area of ​​the palmar vein so that the point between the base of the index finger and the base of the middle finger and the point between the base of the middle finger and the base of the ring finger are located on the same horizontal line parallel to the X-axis includes:

[0024] Four finger root points were located within the effective area of ​​the palm vein. These four finger root points include P1, between the base of the thumb and the base of the index finger; P2, between the base of the index finger and the base of the middle finger; P3, between the base of the middle finger and the base of the ring finger; and P4, between the base of the ring finger and the base of the little finger.

[0025] Determine whether the effective area of ​​the palmar vein belongs to the left or right hand based on the four finger base points;

[0026] Adjust the effective area of ​​the palm vein so that the fingers are pointing upwards, and perform orientation rotation correction on the effective area of ​​the palm vein according to the structural characteristics of the left and right hands.

[0027] In one possible implementation, the steps for determining whether the effective area of ​​the detected palmar vein belongs to the left or right hand based on the four finger root points include:

[0028] Establish a two-dimensional rectangular coordinate system with key point P1 as the origin, and calculate the deflection angle α of the lines connecting P2, P3, and P4 to P1 relative to the X-axis direction. n1 , where n∈(2,3,4);

[0029] Determine whether P2, P3, and P4 are located in the same quadrant of a two-dimensional rectangular coordinate system with the key point P1 as the origin;

[0030] If so, based on the deflection angle α corresponding to P2, P3, and P4 21 α 31 and α 41 To determine whether the effective area of ​​the palmar vein belongs to the left or right hand;

[0031] If not, select two key points P located in the same quadrant. i and P j According to P i and P j The corresponding deflection angle α i1 and α j1 To determine whether the effective area of ​​the palm vein belongs to the left or right hand, i∈(2,3,4), j∈(2,3,4) and i<j.

[0032] In one possible implementation, adjusting the effective area of ​​the palmar veins so that the fingers point upwards, and performing orientational rotation correction of the effective area of ​​the palmar veins based on the structural characteristics of the left and right hands, includes the following steps:

[0033] Select key points P2 and P4;

[0034] For palmar veins belonging to the right hand, calculate the orientation rotation angle β = arctan((y4'-y2') / (x4'-x2')). For palmar veins belonging to the left hand, calculate the orientation rotation angle β = arctan((y2'-y4') / (x2'-x4')).

[0035] If the orientation rotation angle β is greater than 0, then the effective area of ​​the palm vein will be rotated counterclockwise around the center point of the image by angle β.

[0036] If the orientation rotation angle β is less than 0, then the effective area of ​​the palm vein will be rotated clockwise around the center point of the image by angle β.

[0037] Where x2' and y2' are the x and y coordinates of the second key point P2, respectively, and x4' and y4' are the x and y coordinates of the fourth key point P4, respectively.

[0038] Secondly, this application also provides a computer-readable storage medium storing program code that, when executed on a computer, causes the computer to perform the palm vein liveness recognition method.

[0039] Thirdly, this application also provides a palm vein liveness detection device, including a processor and a memory connected to the processor, the memory being used to store executable code and the processor being used to execute the executable code in the memory, wherein when the processor executes the code, the palm vein liveness detection method is implemented.

[0040] The beneficial effects of the palm vein liveness detection method provided in this application are as follows: the difference between live and non-live palm veins is reflected in the local texture area of ​​the superficial layer. The palm vein liveness detection method extracts the effective area of ​​the palm vein through the backbone network to perform palm vein texture features, and enhances the extracted texture features. At the same time, it focuses on the local texture area through the attention module. Finally, the texture features and the local features of interest are merged into the fully connected layer to determine the authenticity of the palm vein liveness detection. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart of the palm vein liveness detection method according to an embodiment of this application;

[0043] Figure 2This is a schematic diagram of the flow structure of the palm vein liveness recognition method according to an embodiment of this application;

[0044] Figure 3 This is a schematic diagram illustrating the left palmar vein ROI and the four finger root points in an embodiment of this application.

[0045] Figure 4 This is a schematic diagram showing the right palmar vein ROI and the four finger root points in an embodiment of this application;

[0046] Figure 5 This is a schematic diagram showing the result of the SCRFD network detecting veins in a right palm image according to an embodiment of this application.

[0047] Figure 6 for Figure 5 The final effective area of ​​the palm vein and the results of four finger root points within the effective area of ​​the palm vein obtained from the palm vein image;

[0048] Figure 7 This is a schematic diagram showing the deflection angles of the left and right key points P2, P3, and P4 in each quadrant of an embodiment of this application.

[0049] Figure 8 This is a schematic diagram showing the result of azimuth rotation correction of the effective area of ​​the left palm vein in an embodiment of this application. Detailed Implementation

[0050] To make the technical problems, technical solutions, and beneficial effects to be solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.

[0051] It should be noted that when a component is referred to as being "fixed to" or "set on" another component, it can be directly on or indirectly on that other component. When a component is referred to as being "connected to" another component, it can be directly connected to or indirectly connected to that other component.

[0052] It should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0053] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0054] The palm vein liveness identification method, identification device, and computer-readable storage medium of this application will now be described in detail with reference to the accompanying drawings.

[0055] Please refer to the reference. Figure 1 and Figure 2 The palm vein liveness detection method provided in this application includes:

[0056] Step S100: Extract the effective area of ​​the palm vein image, and perform orientation rotation correction on the effective area of ​​the palm vein so that the point between the base of the index finger and the base of the middle finger and the point between the base of the middle finger and the base of the ring finger are located on the same horizontal line parallel to the X-axis.

[0057] Step S200: Use the first feature extraction layer of the backbone network to extract the palm vein texture features of the effective area of ​​the palm vein after orientation rotation correction, and obtain the texture feature map f1;

[0058] Step S300: Perform texture enhancement processing on texture feature map f1 to obtain texture feature map T;

[0059] Step S400: Use the second feature extraction layer of the backbone network to extract features from the texture feature map f1 to obtain feature map f2;

[0060] Step S500: Input feature map f2 into the attention module that focuses on the local region of palm vein texture to obtain attention feature map D;

[0061] Step S600: Perform average pooling on the obtained feature map f2 and attention feature map D to obtain the texture feature matrix P, and input the texture feature matrix P into the fully connected layer of the backbone network to determine the authenticity of palm vein liveness.

[0062] Furthermore, after step S400, the palm vein liveness detection method further includes:

[0063] The third feature extraction layer of the backbone network is used to extract features from feature map f2 to obtain deep feature map f3;

[0064] The deep feature map f3 is subjected to average pooling to obtain the global feature map G;

[0065] The texture feature matrix P and the global feature map G are input into the fully connected layer of the backbone network to determine the authenticity of palm vein liveness.

[0066] In the aforementioned palm vein liveness detection method, the first, second, and third feature extraction layers are multiple convolutional layers, distributed sequentially according to the backbone network. The difference between live and non-live palm veins lies in the superficial local texture regions. The palm vein liveness detection method extracts effective palm vein texture features from the backbone network, enhances these features, and simultaneously focuses on local texture regions through an attention module. Finally, the texture features and the focused local features are merged into a fully connected layer for palm vein liveness detection.

[0067] In one embodiment, since the difference between true and false palm veins usually occurs in local regions, the attention module of this invention is a lightweight weighted attention module. This attention module consists of a 1×1 convolutional layer, a batch normalization layer, and a non-linear activation function ReLU. The feature map f2 is input into this attention module (CBAM) to obtain the attention feature map D. The attention module effectively enhances useful local region information and weakens useless local region information, thereby enhancing the model's ability to distinguish between true and false palm vein images.

[0068] It is worth noting that, in Figure 2 In the diagram, backbone layer 1 represents the first feature extraction layer, backbone layer 2 represents the second feature extraction layer, and backbone layers 3 and 4 represent the third feature extraction layers. The texture enhancement module implements step S300.

[0069] In one embodiment, the backbone network is a ResNet50 network, the first feature extraction layer is a convolutional layer consisting of layers 1 to 7 of the ResNet50 network, the second feature extraction layer is a convolutional layer consisting of layers 8 to 15 of the ResNet50 network, and the third feature extraction layer is a convolutional layer consisting of layers 16 to 49 of the ResNet50 network; the fully connected layer of the backbone network is layer 50 of the ResNet50 network, and the fully connected layer serves as... Figure 2 The classifier in the network. In another embodiment, the backbone network is not limited to the ResNet50 network.

[0070] In one embodiment, during the training of the backbone network, a combination of cross entropy loss and arcface loss (Additive Angular Margin Loss) is used as the loss function. Generated palm vein data, original palm vein data, and forged palm vein data are merged to form a new training set to train the palm vein recognition model ResNet50. The advantages of arcface loss are high performance, ease of programming implementation, low complexity, and for performance stability, it does not require joint supervision with other loss functions and can easily converge to any dataset. The loss function calculation expression is: L1 = λ1 * L ce +λ2*L arc ; among which, L ce and L arc λ1 and λ2 represent the cross entropy loss and arcface loss, respectively, and the weights of the two loss functions. If the classifier identifies a palm vein image as a non-living image (e.g., the palm vein image was taken from a printed palm vein image), then the palm vein image will not be used for subsequent vein authentication. If the classifier identifies a palm vein image as a living image, then subsequent vein authentication will be performed.

[0071] In one embodiment, step S300, which involves performing texture enhancement processing on texture feature map f1 to obtain texture feature map T, includes:

[0072] Step S310: Downsample the texture feature map f1 to obtain the feature map t after average pooling;

[0073] Step S320: Upsample and expand the feature map t to the same dimension as the texture feature map f1;

[0074] Step S300: Aggregate the texture feature map f1 with the augmented feature map t;

[0075] Step S330: Perform dense convolution on the aggregation result to obtain the texture feature map T after texture enhancement.

[0076] Specifically, the texture feature map f1 is aggregated by weighted averaging with the augmented feature map t; the above dense convolution includes multiple convolutional layers, which may include three convolutional layers, specifically conv1, conv2 and conv3.

[0077] In one embodiment, the step of performing orientational rotation correction on the effective area of ​​the palm vein in step S100, so that the point between the base of the index finger and the base of the middle finger and the point between the base of the middle finger and the base of the ring finger are located on the same horizontal line parallel to the X-axis, includes:

[0078] Step S110: Locate four finger root points in the effective area of ​​the palm vein, including point P1 between the base of the thumb and the base of the index finger, point P2 between the base of the index finger and the base of the middle finger, point P3 between the base of the middle finger and the base of the ring finger, and point P4 between the base of the ring finger and the base of the little finger.

[0079] Step S120: Determine whether the effective area of ​​the palm vein belongs to the left or right hand based on the four finger root points;

[0080] Step S130: Adjust the effective area of ​​the palm vein so that the fingers are pointing upwards, and perform orientation rotation correction on the effective area of ​​the palm vein according to the structural characteristics of the left and right hands.

[0081] It is worth noting that the effective area of ​​the palm vein is the vein region in the center of the palm. By establishing an image coordinate system with the upper left corner of the image as the origin, and adjusting the effective area of ​​the palm vein so that the fingers point upwards, we can uniformly align the effective area of ​​the palm vein with the direction that the human eye is accustomed to viewing. Unifying the finger orientation of the effective area of ​​the palm vein facilitates subsequent orientation and rotation correction.

[0082] In step S100, a lightweight detection network based on SCRFD is used to detect the palm vein ROI (effective region) and four finger root points. During the training phase, near-infrared palm vein images at different rotation angles are first acquired using a camera, and the effective ROIs and four finger root keypoints of the left and right palm veins are labeled respectively. The ROI label includes the coordinates and category of the bounding box, where the category refers to the effective palm vein region. The four finger root keypoints are labeled with their two-dimensional coordinate values, and the labeling order is fixed from thumb to little finger. The points between the thumb and index finger roots, the index and middle finger roots, the middle and ring finger roots, and the ring and little finger roots are labeled as keypoints P1, P2, P3, and P4, respectively. Then, the palm vein images and the labeled content are input to train SCRFD. During the inference phase, palm vein images at any angle are input, and the SCRFD network can quickly output the coordinates of the palm vein ROI detection box and the coordinates of the four finger root points. Figure 3 A schematic diagram showing the ROI of the left palmar vein and the markings of the four finger bases; Figure 4 A schematic diagram showing the ROI of the right palmar vein and the four finger root points;

[0083] Figure 5This is a schematic diagram showing the results of SCRFD network detecting veins in a right palm image. Figure 6 for Figure 5 The final effective area of ​​the palm vein and the results of four finger root points within the effective area of ​​the palm vein obtained from the palm vein image.

[0084] Further, in step S100, the coordinates of the four finger root points detected by the SCRFD network are the original coordinates in the original palm vein image, and the four finger root points in the effective area of ​​the palm vein are located as: x n '=x n -x src ;y n '=y n -y src .

[0085] Where, x n 'and y n 'Locate the nth key point P respectively n x and y coordinates; n and y n These are the nth keypoints P in the original palm vein image. n The x and y coordinates, i.e., the nth keypoint P in the palm vein image detected using the SCRFD network. n x and y coordinates; src and y src The x and y coordinates are the upper left corner of the effective area of ​​the detected rectangular palm vein.

[0086] In one embodiment, step S120, which involves determining whether the effective area of ​​the detected palmar vein belongs to the left or right hand based on the four finger root points, includes:

[0087] Establish a two-dimensional rectangular coordinate system with the key point P1 as the origin;

[0088] Calculate the deflection angle α of the lines connecting P2, P3, P4 and P1 relative to the X-axis. n1 , where n∈(2,3,4);

[0089] Determine whether P2, P3, and P4 are located in the same quadrant of a two-dimensional rectangular coordinate system with the key point P1 as the origin;

[0090] If so, based on the deflection angle α corresponding to P2, P3, and P4 21 α 31 and α 41 To determine whether the effective area of ​​the palmar vein belongs to the left or right hand;

[0091] If not, select two key points P located in the same quadrant. i and P jAccording to P i and P j The corresponding deflection angle α i1 and α j1 To determine whether the effective area of ​​the palm vein belongs to the left or right hand, i∈(2,3,4), j∈(2,3,4) and i<j.

[0092] Wherein, the deflection angle α n1 The formula for calculating α is: n1 =arctan((y n '-y1') / (x n '-x1'));

[0093] Where, x n 'and y n 'These are the nth key points P' n The x and y coordinates are given, and x1' and y1' are the x and y coordinates of the first keypoint P1, respectively. n '、y n x1' and y1' can be calculated using the method in step S100.

[0094] Among them, such as Figure 7 The diagram shows the deflection angles relative to the X-axis of the lines connecting key points P2, P3, and P4 of the left and right hands in the same quadrant to P1 in a two-dimensional rectangular coordinate system with key point P1 as the origin. In the above steps, the deflection angles α corresponding to P2, P3, and P4 are... 21 α 31 and α 41 To determine whether the effective area of ​​the palmar vein belongs to the left or right hand, specifically: if it is α 21 >α 31 >α 41 The effective area for palmar veins is from the left hand; if it is α 21 <α 31 <α 41 The effective area of ​​the palmar vein comes from the right hand.

[0095] In the above steps, two key points P located in the same quadrant are selected. i and P j According to P i and P j The corresponding deflection angle α i1 and α j1 To determine whether the effective area of ​​the palmar vein belongs to the left or right hand, specifically: calculate the difference in deflection angle Δα = α i1 -α j1 If Δα > 0, the effective area of ​​the palmar vein originates from the left hand; if Δα < 0, the effective area of ​​the palmar vein originates from the right hand.

[0096] It is worth noting that in a two-dimensional rectangular coordinate system established with P1 as the origin, key points P2, P3, and P4 are either in the same quadrant or not in the same quadrant. When determining whether the effective palm rest area originates from the right hand or left / right hand, there are four finger root points that may or may not be in the same quadrant. Specifically, when dealing with the case where they are not in the same quadrant, for key points P2, P3, and P4 that are not in the same quadrant, according to the structural characteristics of the palm, there will be two key points in the same quadrant; that is, either key points P2 and P3 are in the same quadrant, or key points P3 and P4 are in the same quadrant. Taking key points P2 and P3 as being in the same quadrant as an example, when △α=α... 21 -α 31 When Δα is greater than 0, the effective area of ​​the palmar vein originates from the right hand. 21 -α 31 When the value is less than 0, the effective area of ​​the palm vein originates from the left hand.

[0097] In one embodiment, step S130, adjusting the effective area of ​​the palm vein so that the fingers face upwards, and performing orientation rotation correction on the effective area of ​​the palm vein according to the structural characteristics of the left and right hands, includes:

[0098] Step S131: Select key points P2 and P4;

[0099] Step S132: For the palmar veins belonging to the right hand, calculate the orientation rotation angle β = arctan((y4'-y2') / (x4'-x2'));

[0100] Step S133: For the palmar veins belonging to the left hand, calculate the orientation rotation angle β = arctan((y2'-y4') / (x2'-x4'));

[0101] Step S134: If the orientation rotation angle β is greater than 0, then rotate the effective area of ​​the palm vein counterclockwise around the center point of the image by angle β.

[0102] Step S135: If the orientation rotation angle β is less than 0, then rotate the effective area of ​​the palm vein clockwise around the center point of the image by angle β.

[0103] In the above steps, x2' and y2' are the x-coordinate and y-coordinate of the second key point P2, respectively, and x4' and y4' are the x-coordinate and y-coordinate of the fourth key point P4, respectively.

[0104] It is worth noting that the structural characteristics of the left and right hands have the following features: the effective area of ​​the palmar veins is uniformly adjusted so that the fingers point upwards; the arrangement order of key points P2 and P4 in the left palmar veins is exactly the opposite of that in the right palm. For the same rotation angle β calculation method, the starting point is the second key point of P2 and P4 in the positive X-axis direction, and the ending point is the first key point of P2 and P4 in the positive X-axis direction. Figure 8 A schematic diagram showing the result of azimuth rotation correction of an effective area of ​​the left palmar vein.

[0105] In step S100, the effective area of ​​the palm vein in the palm vein image is extracted to exclude interference areas other than the palm region. It is then determined whether the extracted effective area belongs to the left or right hand, and the effective area is adjusted so that the fingers point upwards. Based on the structural characteristics of both hands, the points between the bases of the index and middle fingers and between the bases of the middle and ring fingers in the effective area of ​​the palm vein in both hands are located on the same horizontal line parallel to the X-axis. This unifies the palm orientation angle within the effective area of ​​the palm vein. Therefore, this method of extracting and correcting the effective area of ​​the palm vein effectively eliminates interference areas for subsequent palm vein recognition and removes the influence of palm angle issues on palm vein recognition.

[0106] This application also provides a computer-readable storage medium storing program code that, when executed on a computer, causes the computer to perform the palm vein liveness recognition method.

[0107] This application also provides a palm vein liveness detection device, including a processor and a memory connected to the processor. The memory is used to store executable code and the processor is used to execute the executable code in the memory. When the processor executes the code, it implements the palm vein liveness detection method.

[0108] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for palm vein liveness detection, characterized in that, include: Extract the effective area of ​​the palm vein from the image, and perform orientational rotation correction on the effective area so that the points between the bases of the index and middle fingers and between the bases of the middle and ring fingers are located on the same horizontal line parallel to the X-axis. The steps include: locating four finger root points in the effective area of ​​the palm vein, including point P1 between the bases of the thumb and index finger, point P2 between the bases of the index and middle fingers, point P3 between the bases of the middle and ring fingers, and point P4 between the bases of the ring and little fingers; determining whether the effective area of ​​the palm vein belongs to the left or right hand based on the four finger root points; adjusting the effective area of ​​the palm vein so that the fingers face upwards, and performing orientational rotation correction on the effective area of ​​the palm vein according to the structural characteristics of the left and right hands; The steps for determining whether the effective area of ​​the detected palmar vein belongs to the left or right hand based on the four finger root points include: establishing a two-dimensional rectangular coordinate system with the key point P1 as the origin, and calculating the deflection angle α of the lines connecting P2, P3, and P4 to P1 relative to the X-axis direction. n1 Where n∈(2,3,4); determine whether P2, P3, and P4 are located in the same quadrant of a two-dimensional rectangular coordinate system with the key point P1 as the origin; if so, determine the deflection angle α corresponding to P2, P3, and P4. 21 α 31 and α 41 To determine whether the effective area of ​​the palmar vein belongs to the left or right hand; if not, select two key points P located in the same quadrant. i and P j According to P i and P j The corresponding deflection angle α i1 and α j1 To determine whether the effective area of ​​the palm vein belongs to the left or right hand, where i∈(2,3,4), j∈(2,3,4) and i<j; The steps for adjusting the effective area of ​​the palm veins so that the fingers are pointing upwards and correcting the orientation rotation of the effective area of ​​the palm veins based on the structural characteristics of the left and right hands include: selecting key points P2 and P4; for palm veins belonging to the right hand, calculating the orientation rotation angle β=arctan((y4'-y2') / (x4'-x2')), and for palm veins belonging to the left hand, calculating the orientation rotation angle β=arctan((y2'-y4') / (x2'-x4')); if the orientation rotation angle β is greater than 0, then the effective area of ​​the palm veins is rotated counterclockwise around the image center point by angle β; if the orientation rotation angle β is less than 0, then the effective area of ​​the palm veins is rotated clockwise around the image center point by angle β; where x2' and y2' are the x and y coordinates of the second key point P2, respectively, and x4' and y4' are the x and y coordinates of the fourth key point P4, respectively. The first feature extraction layer of the backbone network is used to extract the palm vein texture features in the effective area of ​​the palm vein after orientation rotation correction, and the texture feature map f1 is obtained. The texture feature map f1 is subjected to texture enhancement processing to obtain the texture feature map T; The second feature extraction layer of the backbone network is used to extract features from the texture feature map f1 to obtain the feature map f2. Input the feature map f2 into the attention module that focuses on the local region of the palm vein texture to obtain the attention feature map D; The texture feature map T and attention feature map D are obtained by average pooling to obtain the texture feature matrix P. The texture feature matrix P is then input into the fully connected layer of the backbone network to determine the authenticity of palm vein liveness. The first and second feature extraction layers are both multiple convolutional layers distributed in the order of the backbone network.

2. The palm vein liveness detection method as described in claim 1, characterized in that, The steps to obtain texture feature map T by performing texture enhancement processing on texture feature map f1 include: The texture feature map f1 is downsampled to obtain the feature map t after average pooling; Upsample and augment the feature map t to the same dimension as the texture feature map f1; Aggregate the texture feature map f1 with the augmented feature map t; The aggregation result is subjected to dense convolution to obtain the texture feature map T after texture enhancement.

3. The palm vein liveness detection method as described in claim 1, characterized in that, The palm vein liveness detection method also includes: The third feature extraction layer of the backbone network is used to extract features from feature map f2 to obtain deep feature map f3; The deep feature map f3 is subjected to average pooling to obtain the global feature map G; The texture feature matrix P and the global feature map G are input into the fully connected layer of the backbone network to determine the authenticity of palm vein liveness. The third feature extraction layer consists of multiple convolutional layers, and both the second and third feature extraction layers are multiple convolutional layers distributed in the order of the backbone network.

4. The palm vein liveness detection method as described in claim 3, characterized in that, The backbone network is a ResNet50 network. The first feature extraction layer is a convolutional layer consisting of layers 1 to 7 of the ResNet50 network. The second feature extraction layer is a convolutional layer consisting of layers 8 to 15 of the ResNet50 network. The third feature extraction layer is a convolutional layer consisting of layers 16 to 49 of the ResNet50 network.

5. A computer-readable storage medium, characterized in that, The computer-readable medium stores program code that, when executed on a computer, causes the computer to perform the palm vein liveness detection method as described in any one of claims 1 to 4.

6. A palm vein liveness detection device, characterized in that, The method includes a processor and a memory connected to the processor, the memory being used to store executable code and the processor being used to execute the executable code in the memory, wherein when the processor attempts to execute the code, it implements the palm vein liveness recognition method as described in any one of claims 1 to 4.

Images