A method and device for binocular finger vein identification

Abstract

The present invention discloses a method and device for binocular finger vein identity recognition. The method includes: calibrating the parameters of the left and right cameras; calculating two-dimensional left and right finger vein texture images; performing stereo rectification on the left and right finger vein texture images; obtaining finger vein texture point cloud data with a three-dimensional spatial topological structure; performing noise reduction processing on the finger vein texture point cloud data; constructing a model for cross-modal finger vein identity information representation with the collaboration of two-dimensional images and three-dimensional point clouds; calculating a combined representation vector of the fusion feature corresponding to the maximum correlation and the binocular finger vein feature attributes; constructing a finger vein feature database with identity identification attributes. According to the matching relationship between binocular infrared finger vein texture images, the present invention establishes a convolutional neural network model, converts two-dimensional finger vein texture images and three-dimensional finger vein texture point clouds into corresponding feature vectors, and constructs a combined representation of cross-modal finger vein feature information, thereby improving the identification ability of identity recognition.

Description

Technical Field

[0001] The present invention relates to a method for binocular finger vein identification and also to a device for implementing this method, belonging to the technical field of biometric identification. Background Art

[0002] The importance of information security in today's society has become increasingly prominent with the rapid development of computer and information technologies, especially in key fields such as finance, security, and online transactions. As an important means of protecting information and property security, identity recognition and authentication have shifted from traditional methods based on passwords or certificates to biometric identification technologies that utilize inherent human characteristics such as faces and fingerprints. Due to their uniqueness, immutability, and portability, these biometric characteristics have been widely applied in daily life.

[0003] Finger vein recognition technology, as an emerging biometric identification method, has received extensive attention in research and applications in recent years. This technology captures near-infrared images of finger veins inside the finger, extracts vein features therefrom, and performs pattern matching with an existing feature library to achieve identity recognition. Since finger vein information is protected by the human skin tissue, it is not easily damaged or contaminated by external factors, and it is difficult to steal the vein information. Therefore, it has an inherent anti-counterfeiting advantage. Compared with biometric identification technologies based on fingerprints and faces, finger vein recognition technology has a higher security level and anti-counterfeiting ability, becoming a research hotspot in the field of biometric identification.

[0004] However, despite the many advantages of finger vein recognition technology, it also faces some challenges. Currently, most finger vein recognition research is based on two-dimensional vein images, which ignores the problem of information loss that may occur when the actual three-dimensional finger vein information of the human body is projected onto a plane for imaging. This information loss limits the accuracy of finger vein recognition technology in large-scale user identification. In addition, existing recognition methods based on two-dimensional finger vein texture images are easily affected by factors such as environmental light changes, finger placement changes, and finger perturbations, resulting in relatively high false rejection rates and false acceptance rates in the finger vein recognition system. These methods have relatively strict requirements for the finger placement position of users and insufficient environmental adaptability. Therefore, in order to promote and apply finger vein recognition technology, it is particularly important to construct a more robust finger vein feature representation and identity recognition method. Summary of the Invention

[0005] The primary technical problem to be solved by the present invention is to provide a method for binocular finger vein identification.

[0006] Another technical problem to be solved by the present invention is to provide a device for binocular finger vein identification.

[0007] To achieve the above technical objectives, the present invention adopts the following technical solutions:

[0008] According to the first aspect of the embodiments of the present invention, a binocular finger vein identity recognition method is provided, including the following steps:

[0009] S1: Set the positions of the left and right cameras using the planar checkerboard calibration method, and calibrate the parameters of the left and right cameras;

[0010] S2: Calculate the two-dimensional left and right finger vein texture images according to the calibrated parameters;

[0011] S3: Stereo-correct the left and right finger vein texture images according to the calibrated parameters, and transform the left and right finger vein texture images onto the same plane with parallel optical axes to each other;

[0012] S4: According to the template matching method, construct the corresponding matching relationship between the left and right finger vein texture images, perform three-dimensional reconstruction of the finger vein texture images, and obtain the finger vein texture point cloud data with a three-dimensional spatial topological structure;

[0013] S5: Perform noise reduction processing on the finger vein texture point cloud data;

[0014] S6: According to the left and right finger vein texture information and the finger vein texture point cloud data, construct a model for cross-modal finger vein identity information representation with the collaboration of two-dimensional images and three-dimensional point clouds;

[0015] S7: Using a deep convolutional neural network as the backbone network, combine two two-dimensional left and right finger vein texture images of the same moment and equal size to form a multi-channel input finger vein texture image;

[0016] S8: According to the multi-channel input finger vein texture image, obtain the feature vectors of the corresponding two-dimensional finger vein texture image and the feature vectors of the corresponding three-dimensional finger vein point cloud data through a convolutional neural network;

[0017] S9: Through the canonical correlation analysis method, calculate the correlation between the feature vectors of the two-dimensional finger vein texture image and the feature vectors of the corresponding three-dimensional finger vein point cloud data, and obtain the fusion feature corresponding to the maximum correlation;

[0018] S10: Concatenate the two-dimensional finger vein texture features corresponding to the maximum correlation and the three-dimensional finger vein point cloud data features corresponding to the maximum correlation to obtain a joint representation vector of the binocular finger vein feature attributes;

[0019] S11: Based on the joint representation vectors of the binocular finger vein feature attributes obtained from different individuals, construct a finger vein feature database with identity identification attributes.

[0020] Preferably, the parameters include: the internal parameter K of the left camera l ; the distortion coefficient D of the left camera l; The internal parameters K of the right camera r ; The distortion coefficients D of the right camera r ; The rotation transformation R between the left and right cameras; the translation transformation T between the left and right cameras; the relative pose relationship [R, T] of the left and right cameras;

[0021] Among them, the internal parameters of the camera include: the principal point coordinates of the finger vein texture image and the focal lengths between the left and right cameras.

[0022] Preferably, the method for calculating the two-dimensional left and right finger vein texture information includes the following sub-steps:

[0023] S21: Collect the left finger vein texture image obtained by the left camera and the right finger vein texture image obtained by the right camera;

[0024] S22: Perform distortion correction on the left finger vein texture image obtained by the left camera and the right finger vein texture image obtained by the right camera through the internal parameters of the left camera, the distortion coefficients of the left camera, the internal parameters of the right camera, and the distortion coefficients of the right camera obtained by calibration;

[0025] S23: Preprocess the corrected left finger vein texture image and right finger vein texture image by the non-linear median filtering method;

[0026] S24: Extract the binary texture information of the preprocessed left finger vein texture image and right finger vein texture image by the local dynamic threshold segmentation method;

[0027] S25: Obtain the two-dimensional left finger vein texture image L v (x l , y l ) and the two-dimensional right finger vein texture image R v (x r , y r );

[0028] Among them, (x l , y l ) are the two-dimensional pixel coordinates of the left finger vein texture in the left camera image; (x r , y r ) are the two-dimensional pixel coordinates of the right finger vein texture in the right camera image.

[0029] Preferably, the method for stereo rectification includes the following sub-steps:

[0030] S31: Calculate the disparity of the left and right finger vein texture images on the same horizontal line;

[0031] S32: Calculate the corresponding three-dimensional space coordinates P of the left and right finger vein texture points by the triangulation method i(X i , Y i , Z i );

[0032] S33: Transform the left and right finger vein texture images onto the same plane with parallel optical axes according to the three-dimensional space coordinates corresponding to the left and right finger vein texture points.

[0033] Preferably, the formula for calculating the parallax of the left and right finger vein texture images on the same horizontal line is:

[0034] d i = x il - x ir

[0035] where x il is the abscissa value in the left finger vein texture image; x ir is the abscissa value in the right finger vein texture image.

[0036] Preferably, the calculation formula for each coordinate in the space coordinates is:

[0037]

[0038] where T0 is the baseline distance between the left and right cameras, and its value is the modulus of the translational exchange between the calibrated left and right cameras; (u0, v0) is the principal point coordinates of the left and right finger vein texture images; f is the focal length between the left and right cameras.

[0039] Preferably, the method for denoising the finger vein texture point cloud data is:

[0040] Calculate the average Euclidean distance between each finger vein texture point and its surrounding k nearest neighbor points, and statistically obtain the corresponding average distance and variance value from the single-frame finger vein texture point cloud data, and accordingly set the pseudo finger vein texture point cloud judgment threshold;

[0041] If the average distance of a certain finger vein texture point is greater than the pseudo finger vein texture point cloud judgment threshold, then this point is defined as an outlier and deleted from the finger vein texture point cloud data; if the average distance of a certain finger vein texture point is less than or equal to the pseudo finger vein texture point cloud judgment threshold, then this point is retained.

[0042] Preferably, the expression of the cross-modal finger vein identity information representation model is:

[0043] T vein = Ω{L v (x l , y l ), R v (x r , y r),P fv}

[0044] Among them, P fv is the finger vein texture point cloud data; Ω{g} is the cross-modal finger vein identity information expression function.

[0045] Preferably, the calculation method of the maximum correlation is as follows:

[0046]

[0047]

[0048] Among them, M is the feature vector of the two-dimensional finger vein texture image; N is the feature vector of the three-dimensional finger vein point cloud data; M * is the two-dimensional finger vein texture feature corresponding to the maximum correlation; N * is the three-dimensional finger vein point cloud data feature corresponding to the maximum correlation; S mm is the autocovariance matrix of the feature vector of the two-dimensional finger vein texture image; S nn is the autocovariance matrix of the feature vector of the three-dimensional finger vein point cloud data; S mn is the cross-covariance matrix between the feature vector of the two-dimensional finger vein texture image and the feature vector of the three-dimensional finger vein point cloud data; W m is the transfer matrix of the two-dimensional finger vein texture image; W n is the transfer matrix of the three-dimensional finger vein point cloud data; is the transpose matrix of W m ; is the transpose matrix of W n ;

[0049] Preferably, the expression of the joint representation vector of the binocular finger vein feature attributes is:

[0050] Z * =(M * ,N * ).

[0051] According to the second aspect of the embodiments of the present invention, a binocular finger vein identity recognition device is provided, including a processor and a memory; wherein, the memory is coupled to the processor and is used to store a computer program, and when the computer program is executed by the processor, the processor implements the above method.

[0052] Compared with the prior art, the present invention realizes the three-dimensional spatial representation of finger vein texture information by establishing the matching relationship between binocular infrared finger vein texture images. On this basis, a convolutional neural network model is introduced to convert the two-dimensional finger vein texture images and the three-dimensional finger vein texture point clouds into corresponding feature vectors. Then, the canonical correlation analysis method is used to align the finger vein feature vectors in different modalities, and a joint representation of cross-modal finger vein feature information is constructed to improve its identification ability and authentication effect in the identity recognition process. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flowchart of a binocular finger vein identity recognition method provided by an embodiment of the present invention;

[0054] Figure 2 It is a schematic structural diagram of the camera placement in an embodiment of the present invention;

[0055] Figure 3 It is a schematic diagram of a binocular finger vein identity recognition device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] The technical content of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] First Embodiment

[0058] As Figure 1 shown, a binocular finger vein identity recognition method provided by the first embodiment of the present invention includes:

[0059] S1: Set the positions of the left and right cameras using the planar chessboard calibration method and calibrate the parameters of the left and right cameras.

[0060] It should be noted that the Planar Chessboard Calibration method is a camera calibration method proposed by Zhang Zhengyou. This method mainly realizes camera calibration by shooting planar images containing chessboard patterns, that is, determining the internal parameters (such as focal length, principal point position, etc.) and external parameters (such as the position and attitude of the camera relative to the chessboard) of the camera.

[0061] The specific steps of the planar chessboard calibration method are as follows:

[0062] Prepare the chessboard pattern: Make or print a chessboard pattern, usually a black and white checkerboard pattern, and the size of each grid is known.

[0063] Take images: Place the chessboard in different positions and orientations and take multiple images containing the chessboard. These images will be used for subsequent calculations.

[0064] Feature point detection: In each image, computer vision algorithms are used to detect the corner points of the checkerboard. The positions of these corner points in the image are known.

[0065] Model establishment: The detected corner points are associated with the actual physical positions of the checkerboard to establish a mathematical model. This model will be used to calculate the internal and external parameters of the camera.

[0066] Parameter optimization: Optimization algorithms (such as the least squares method) are used to adjust the model parameters to minimize the error between the corner point positions predicted by the model and the actually detected corner point positions.

[0067] Calibration result: The internal and external parameters of the camera are finally obtained. These parameters can be used to correct the images captured by the camera or for subsequent computer vision tasks such as 3D reconstruction.

[0068] Through the above planar checkerboard calibration method, the accuracy of image processing and analysis can be improved, laying a foundation for subsequent tasks such as image recognition and object tracking.

[0069] Figure 2 It is a schematic structural diagram of the camera placement in an embodiment of the present invention. This figure shows the relative position relationship between the two cameras and the finger in the binocular system, which is crucial for understanding the entire identity recognition process.

[0070] In Figure 2 , two rectangles can be seen, marked as "1" and "2" respectively, which represent the left and right cameras in the system. These two cameras are placed on both sides of the finger to facilitate capturing the finger vein texture images from different angles. This dual-camera layout is the basis for realizing stereoscopic vision and 3D information acquisition. The oval in the center of the figure, marked as "3", represents the finger. The finger is the key object in the identity recognition process, and the internal vein texture will be captured by the two cameras and used for identity verification. The position of the finger is within the common field of view of the two cameras, ensuring that images can be obtained from two different perspectives. In addition, the placement angles and positions of the two cameras are carefully designed to achieve the best image capture effect. This layout allows the system to reconstruct the 3D finger vein structure of the finger by comparing the images on both sides, thereby improving the accuracy and robustness of identity recognition.

[0071] Figure 2 The structural relationship shown is the key to achieving high-precision identity recognition, which ensures that the system can effectively capture and process finger vein information.

[0072] In an embodiment of the present invention, the parameters of the left and right cameras include: the internal parameter K of the left camera l , the distortion coefficient D of the left camera l , the internal parameter K of the right camera r, the distortion coefficient D of the right camera r , the rotation transformation R between the left and right cameras, the translation transformation T between the left and right cameras, and the relative pose relationship [R, T] between the left and right cameras.

[0073] Among them, the internal parameters of the camera include: the principal point coordinates of the finger vein texture image and the focal lengths between the left and right cameras.

[0074] S2: Calculate the two-dimensional left and right finger vein texture images according to the calibrated parameters.

[0075] The method for calculating the two-dimensional left and right finger vein texture information includes the following sub-steps:

[0076] S21: Collect the left finger vein texture image obtained by the left camera and the right finger vein texture image obtained by the right camera.

[0077] S22: Perform distortion correction on the left finger vein texture image obtained by the left camera and the right finger vein texture image obtained by the right camera through the internal parameters of the left camera, the distortion coefficient of the left camera, the internal parameters of the right camera, and the distortion coefficient of the right camera obtained by calibration.

[0078] S23: Preprocess the corrected left finger vein texture image and right finger vein texture image by the non-linear median filtering method.

[0079] The function of the preprocessing is to ensure the vein texture information while reducing the image noise.

[0080] S24: Extract the binary texture information of the preprocessed left finger vein texture image and right finger vein texture image by the local dynamic threshold segmentation method.

[0081] S25: Obtain the two-dimensional left finger vein texture image L v (x l , y l ) and the two-dimensional right finger vein texture image R v (x r , y r ).

[0082] Among them, (x l , y l ) is the two-dimensional pixel coordinate of the left finger vein texture in the left camera image. (x r , y r ) is the two-dimensional pixel coordinate of the right finger vein texture in the right camera image.

[0083] S3: Perform stereo correction on the left and right finger vein texture images according to the calibrated parameters, and transform the left and right finger vein texture images onto the same plane with parallel optical axes to each other.

[0084] The method for three-dimensional correction includes the following sub-steps:

[0085] S31: Calculate the parallax of the left and right finger vein texture images at the same horizontal line.

[0086] The formula for calculating the parallax of the left and right finger vein texture images at the same horizontal line is:

[0087] d i = x il - x ir

[0088] where x il is the abscissa value in the left finger vein texture image. x ir is the abscissa value in the right finger vein texture image.

[0089] S32: Calculate the three-dimensional space coordinates P i (X i , Y i , Z i ) corresponding to the left and right finger vein texture points through the triangulation method.

[0090] where the calculation formulas for the coordinates in this space coordinate are:

[0091]

[0092] where T0 is the baseline distance between the left and right cameras, and its value is the modulus of the translational transformation between the calibrated left and right cameras. (u0, v0) are the principal point coordinates of the left and right finger vein texture images. f is the focal length between the left and right cameras.

[0093] S33: According to the three-dimensional space coordinates corresponding to the left and right finger vein texture points, transform the left and right finger vein texture images onto the same plane with parallel optical axes to each other.

[0094] S4: According to the template matching method, construct the corresponding matching relationship between the left and right finger vein texture images, perform three-dimensional reconstruction of the finger vein texture images, and obtain the finger vein texture point cloud data with a three-dimensional space topological structure.

[0095] S5: Perform noise reduction processing on the finger vein texture point cloud data.

[0096] Calculate the average Euclidean distance between each finger vein texture point and its surrounding k nearest neighbor points, and statistically obtain the corresponding average distance and variance value from the single-frame finger vein texture point cloud data, and set the pseudo finger vein texture point cloud judgment threshold accordingly.

[0097] If the average distance of a certain finger vein texture point is greater than the pseudo-finger vein texture point cloud judgment threshold, then this point is defined as an outlier and is deleted from the finger vein texture point cloud data. If the average distance of a certain finger vein texture point is less than or equal to the pseudo-finger vein texture point cloud judgment threshold, then this point is retained.

[0098] Considering that binary finger vein texture images are prone to phenomena such as vein burrs or broken pseudo-veins, which in turn lead to noise or pseudo-vein point clouds in the finger vein texture point cloud data obtained by 3D reconstruction. Therefore, based on the assumption of the spatial continuity of ideal finger vein textures, the present invention constructs a method for 3D filtering of finger vein texture point clouds based on a sliding window.

[0099] S6: According to the left and right finger vein texture information and the finger vein texture point cloud data, construct a model T for cross-modal finger vein identity information representation with the coordination of 2D images and 3D point clouds vein .

[0100] The expression of the cross-modal finger vein identity information representation model is:

[0101] T vein =Ω{L v (x l ,y l ),R v (x r ,y r ),P fv}

[0102] Among them, P fv is the finger vein texture point cloud data. Ω{g} is the cross-modal finger vein identity information expression function.

[0103] S7: Using a deep convolutional neural network (Residual Network, ResNet) as the backbone network, combine two 2D left and right finger vein texture images of the same moment and equal size to form a multi-channel input finger vein texture image.

[0104] S8: According to the multi-channel input finger vein texture image, obtain the feature vectors of the corresponding 2D finger vein texture image and the feature vectors of the corresponding 3D finger vein point cloud data through a convolutional neural network (ConvolutionalNeural Networks, CNN).

[0105] S9: Through the Canonical Correlation Analysis (CCA), calculate the correlation between the feature vectors of the 2D finger vein texture image and the feature vectors of the corresponding 3D finger vein point cloud data, and obtain the fusion feature corresponding to the maximum correlation.

[0106] The calculation method of the maximum correlation is as follows:

[0107]

[0108] Among them, M is the feature vector of the two-dimensional finger vein texture image. N is the feature vector of the three-dimensional finger vein point cloud data. M * is the two-dimensional finger vein texture feature corresponding to the maximum correlation. N * is the three-dimensional finger vein point cloud data feature corresponding to the maximum correlation. S mm is the autocovariance matrix of the feature vector of the two-dimensional finger vein texture image. S nn is the autocovariance matrix of the feature vector of the three-dimensional finger vein point cloud data. S mn is the cross-covariance matrix between the feature vector of the two-dimensional finger vein texture image and the feature vector of the three-dimensional finger vein point cloud data. W m is the transfer matrix of the two-dimensional finger vein texture image. W n is the transfer matrix of the three-dimensional finger vein point cloud data. is W m 's transpose matrix. is W n 's transpose matrix.

[0109] Based on the attributes of the two-dimensional finger vein texture image, this model has the characteristic of rotational invariance of the three-dimensional finger vein texture image, and has better adaptability to different finger postures and orientations during the recognition process. However, due to the large distribution difference in the physical space scale between the left and right finger vein texture information and the finger vein texture point cloud data in this model, a reliable alignment method is required during specific recognition and matching to map the finger vein feature information in different dimensions into the same state vector.

[0110] S10: Concatenate the two-dimensional finger vein texture feature corresponding to the maximum correlation and the three-dimensional finger vein point cloud data feature corresponding to the maximum correlation to obtain a joint representation vector of the binocular finger vein feature attributes.

[0111] The expression of the joint representation vector of the binocular finger vein feature attributes is:

[0112] Z * =(M * , N * )

[0113] S11: Based on the joint representation vectors of the binocular finger vein feature attributes obtained from different individuals, construct a finger vein feature database with identity identification attributes.

[0114] When identifying an individual, through the method provided by the embodiments of the present invention, the corresponding multi-modal joint representation attribute feature vector can be obtained and matched with the prior finger vein feature database to achieve accurate identification of the target identity.

[0115] Second Embodiment

[0116] Based on the above method, the second embodiment of the present invention further provides a binocular finger vein identity recognition device. As Figure 3 shown, the device 400 includes one or more processors 41 and a memory 42. The memory 42 is used to store one or more programs, and when the one or more programs are executed by the one or more processors 41, the method in the above embodiments is implemented.

[0117] Among them, the processor is used to control the overall operation of the device to complete all or part of the steps of the above method. The processor can be a central processing unit (CPU), a graphics processing unit (GPU), a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a digital signal processing (DSP) chip, etc. The memory is used to store various types of data to support the operation of the device. These data can include, for example, instructions for any application program or method operating on the device, and data related to the application program. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, etc.

[0118] In an exemplary embodiment, the device can be specifically implemented by a computer or a microprocessor entity, or by a product with a certain function, for executing the above method and achieving the same technical effect as the above method. Specifically, the computer can be, for example, a personal computer, a laptop computer, an in-vehicle human-machine interaction device, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0119] In another exemplary embodiment, the present invention also provides a computer-readable storage medium including program instructions, and when the program instructions are executed by a processor, the steps of the method in any one of the above embodiments are implemented. For example, the computer-readable storage medium can be the above memory including program instructions, and the above program instructions can be executed by the processor to complete the above method and achieve the same technical effect as the above method.

[0120] It should be noted that the above-mentioned multiple embodiments are only examples. The technical solutions of each embodiment can be combined, and all are within the protection scope of the present invention.

[0121] In addition, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality of" means two or more, unless otherwise specifically defined.

[0122] The above has provided a detailed description of the method and device for binocular finger vein identification of the present invention. For those of ordinary skill in the art, any obvious changes made to it without departing from the essence of the present invention will constitute an infringement of the patent right of the present invention and will bear corresponding legal responsibilities.

Claims

1. A method for binocular finger vein identification, characterized in that Including: S1: Set the positions of the left and right cameras using a planar chessboard calibration method and calibrate the parameters of the left and right cameras; S2: Calculate the two-dimensional left and right finger vein texture images based on the calibrated parameters; S3: Stereo-correct the left and right finger vein texture images according to the calibrated parameters, and transform the left and right finger vein texture images onto the same plane with parallel optical axes to each other; S4: According to the template matching method, construct the corresponding matching relationship between the left and right finger vein texture images, perform three-dimensional reconstruction of the finger vein texture images, and obtain the finger vein texture point cloud data with a three-dimensional spatial topological structure; S5: Denoise the finger vein texture point cloud data; S6: Construct a model with cross-modal finger vein identity information representation that combines two-dimensional images and three-dimensional point clouds based on the left and right finger vein texture information and the finger vein texture point cloud data; S7: Using a deep convolutional neural network as the backbone network, combine two two-dimensional left and right finger vein texture images of the same size at the same moment to form a multi-channel input finger vein texture image; S8: According to the multi-channel input finger vein texture image, obtain the feature vectors of the corresponding two-dimensional finger vein texture image and the feature vectors of the corresponding three-dimensional finger vein point cloud data through a convolutional neural network; S9: Calculate the correlation between the feature vectors of the two-dimensional finger vein texture image and the feature vectors of the corresponding three-dimensional finger vein point cloud data through canonical correlation analysis, and obtain the fusion feature corresponding to the maximum correlation; S10: Concatenate the two-dimensional finger vein texture feature corresponding to the maximum correlation and the three-dimensional finger vein point cloud data feature corresponding to the maximum correlation to obtain a joint representation vector of the binocular finger vein feature attributes; S11: Based on the joint representation vectors of the binocular finger vein feature attributes obtained from different individuals, construct a finger vein feature database with identity identification attributes.

2. The method according to claim 1, characterized in that The parameters include: the internal parameter K of the left camera l ; the distortion coefficient D of the left camera l ; the internal parameter K of the right camera r ; the distortion coefficient D of the right camera r ; the rotation transformation R between the left and right cameras; the translation transformation T between the left and right cameras; the relative pose relationship [R, T] between the left and right cameras; Among them, the internal parameters of the camera include: the principal point coordinates of the finger vein texture image and the focal lengths between the left and right cameras.

3. The method according to claim 1, wherein The method for calculating the two-dimensional left and right finger vein texture information includes the following sub-steps: S21: Collect the left finger vein texture image obtained by the left camera and the right finger vein texture image obtained by the right camera; S22: Perform distortion correction on the left finger vein texture image obtained by the left camera and the right finger vein texture image obtained by the right camera through the internal parameters of the left camera, the distortion coefficient of the left camera, the internal parameters of the right camera, and the distortion coefficient of the right camera obtained by calibration; S23: Preprocess the corrected left finger vein texture image and right finger vein texture image through a non-linear median filtering method; S24: Extract the binary texture information of the preprocessed left finger vein texture image and right finger vein texture image through a local dynamic threshold segmentation method; S25: Obtain the two-dimensional left finger vein texture image L through the skeleton extraction algorithm v (x l , y l ) and the two-dimensional right finger vein texture image R v (x r , y r ); Among them, (x l , y l ) is the two-dimensional pixel coordinates of the left finger vein texture in the left camera image; (x r , y r ) is the two-dimensional pixel coordinates of the right finger vein texture in the right camera image.

4. The method according to claim 1, wherein The method for stereo correction includes the following sub-steps: S31: Calculate the disparity of the left and right finger vein texture images on the same horizontal line; S32: Calculate the three-dimensional spatial coordinates P corresponding to the left and right finger vein texture points through the triangulation method i (X i , Y i , Z i ); S33: According to the three-dimensional spatial coordinates corresponding to the left and right finger vein texture points, transform the left and right finger vein texture images onto the same plane with parallel optical axes to each other.

5. The method according to claim 4, wherein The formula for calculating the disparity of the left and right finger vein texture images on the same horizontal line is: d i = x il - x ir where x il is the abscissa value in the left finger vein texture image; x ir is the abscissa value in the right finger vein texture image.

6. The method according to claim 4, wherein The calculation formulas for each coordinate in the spatial coordinates are as follows: Among them, T0 is the baseline distance between the left and right cameras, and its value is the modulus of the translational transformation between the calibrated left and right cameras; (u0, v0) are the principal point coordinates of the left and right finger vein texture images; f is the focal length between the left and right cameras.

7. The method according to claim 1, wherein The method for denoising the finger vein texture point cloud data includes the following sub-steps: Calculate the average Euclidean distance between each finger vein texture point and its surrounding k nearest neighbor points, and statistically obtain the corresponding average distance and variance value from the single-frame finger vein texture point cloud data, and accordingly set the pseudo finger vein texture point cloud judgment threshold; If the average distance of a certain finger vein texture point is greater than the pseudo finger vein texture point cloud judgment threshold, then this point is defined as an outlier and deleted from the finger vein texture point cloud data; if the average distance of a certain finger vein texture point is less than or equal to the pseudo finger vein texture point cloud judgment threshold, then this point is retained.

8. The method according to claim 1, wherein The expression of the model for cross-modal finger vein identity information representation is: T vein = Ω{L v (x l ,y l ),R v (x r ,y r ),P fv} Among them, P fv is the finger vein texture point cloud data; Ω{g} is the cross-modal finger vein identity information expression function.

9. The method according to claim 1, wherein The calculation method of the maximum correlation is: Among them, M is the feature vector of the two-dimensional finger vein texture image; N is the feature vector of the three-dimensional finger vein point cloud data; M * is the two-dimensional finger vein texture feature corresponding to the maximum correlation; N * is the three-dimensional finger vein point cloud data feature corresponding to the maximum correlation; S mm is the autocovariance matrix of the feature vector of the two-dimensional finger vein texture image; S nn is the autocovariance matrix of the feature vector of the three-dimensional finger vein point cloud data; S mn is the cross-covariance matrix between the feature vector of the two-dimensional finger vein texture image and the feature vector of the three-dimensional finger vein point cloud data; W m is the transfer matrix of the two-dimensional finger vein texture image; W n is the transfer matrix of the three-dimensional finger vein point cloud data; is the transpose matrix of W m ; is the transpose matrix of W n ; 10. A binocular finger vein identity recognition device, characterized in that It includes a processor and a memory; among them, the memory is coupled to the processor and is used to store a computer program. When the computer program is executed by the processor, the processor implements the method described in any one of claims 1 to 9.

Images