A Finger Vein Recognition Method Based on Feature Fusion

Abstract

The present invention discloses a finger vein recognition method based on feature fusion. First, the input finger vein image is successively subjected to finger edge extraction, finger region segmentation, finger rotation correction, and extraction of the region of interest of finger veins. Then, the histogram of oriented gradients features of the region of interest of finger veins is extracted, and the local phase quantization features of the region of interest of finger veins are extracted. Finally, on this basis, the principal component analysis algorithm is used to perform feature selection on the above two kinds of features respectively, and the joint sparse representation classification method is used to fuse and recognize the above histogram of oriented gradients features and local phase quantization features, so as to achieve the final identity recognition. This solution further improves the security and effectiveness of identity recognition, has the advantages of good reliability and high use value, and can be widely applied to fields such as access control systems and security.

Description

Technical Field

[0001] The present invention relates to the technical field of pattern recognition, and in particular, to a finger vein recognition method based on feature fusion. Background Art

[0002] With the rapid development of science and technology, information security plays an increasingly important role in people's daily life and work. Especially in the aspect of identity recognition, the traditional identity recognition methods can no longer meet the growing security requirements. Compared with the traditional identity recognition methods, the biometric-based identity recognition methods have gradually replaced the traditional ones due to their high security, good credibility and ease of use. Among them, finger vein, as an internal feature of the human body and with the property of liveness, has received extensive attention. However, the existing finger vein recognition methods often use only a single feature to identify the identity, which to a certain extent leads to the loss of finger vein features and thus affects the final recognition effect. Summary of the Invention

[0003] The present invention mainly solves the technical problems existing in the prior art, such as single feature, easy loss and affecting the recognition effect, and provides a finger vein recognition method based on feature fusion that is oriented to multiple features, has high reliability and strong security.

[0004] The present invention mainly solves the above technical problems through the following technical solutions: A finger vein recognition method based on feature fusion, comprising the following steps:

[0005] S1. Preprocess the registered finger vein image and the finger vein image to be recognized collected by the finger vein acquisition device to obtain the region of interest of each finger vein image;

[0006] S2. Extract the pyramid histogram of oriented gradients features and local phase quantization features from the regions of interest of the two finger vein images;

[0007] S3. Combine the pyramid histogram of oriented gradients features and local phase quantization features of the registered finger vein image to form a joint dictionary, and use the joint sparse representation classification algorithm to implement identity recognition.

[0008] The processing of the two finger vein images does not require simultaneous execution. Generally, the registered finger vein image is processed first in the registration stage until the joint dictionary is formed, and there can be multiple registered finger vein images; in the recognition stage, the finger vein image to be recognized is processed. When there are multiple registered finger vein images during recognition, the joint sparse representation algorithm is used sequentially in step S3 for recognition.

[0009] Preferably, the specific steps for extracting the pyramid histogram of oriented gradients features are:

[0010] S201. Perform a three-level pyramid segmentation on the finger vein region of interest, that is, divide it into two segments according to both the horizontal and vertical coordinates of the image, and each region segmented at the previous level is divided into four regions at the next level; thus, a set of finger vein images at different resolutions is obtained and arranged in sequence to form an image pyramid model;

[0011] S202. Calculate the histogram of oriented gradients features of the finger vein images at each resolution (i.e., the different resolutions obtained from the previous segmentation) and perform histogram normalization on them;

[0012] S203. In the order of the image pyramid model, cascade the histogram of oriented gradients features after each normalization process to obtain the final pyramid histogram of oriented gradients features.

[0013] Preferably, the specific steps for extracting local phase quantization features are as follows:

[0014] S211. In the finger vein region of interest f(x), take a neighborhood interval N of size M×M for each pixel point λ , and then perform a short-time Fourier transform, and its calculation formula is as follows:

[0015]

[0016] where u is the frequency, y represents the corresponding integer value in the neighborhood interval N λ , F(u,x) is the finger vein region of interest after Fourier transform, T is the vector transpose operation, and f(x - y) are the pixel values in the M×M neighborhood interval N λ ; the neighborhood interval N λ is of size M×M, that is, M is the number of pixels on one side of the current pixel point;

[0017] In the local phase quantization operator, only four subsets of frequency points need to be considered to calculate the local Fourier coefficients, that is, u1 = [a,0] T , u2 = [0,a] T , u3 = [a,a] T , u4 = [a,-a] T , where

[0018] For the pixel points in the finger vein region of interest f(x), its vector representation F x is:

[0019] F x = [F(u1,x), F(u2,x).F(u3,x), F(u4,x)]

[0020] Let F xIf the elements in

[0021] G(x) = [Re{F x}, Im{F x}]

[0022] G(x) is the vector composed of the real and imaginary parts of each element in F x ;

[0023] S212. According to the value g j of each element in G(x), perform quantization processing on the phase, that is

[0024]

[0025] where q j is the j-th component of G(x);

[0026] S213. According to the quantization result, convert the binary sequence into a decimal number and use it as the local phase quantization value f C (x) of the local neighborhood center point, and its calculation formula is as follows:

[0027]

[0028] S214. Statistically analyze the histogram distribution of all local phase quantization values and use this distribution as the final local phase quantization feature.

[0029] Preferably, in step S2, after extracting the histogram of oriented gradients features and local phase quantization features of the pyramid, the principal component analysis algorithm is used to perform feature selection on the two features respectively as the basis for constructing the combined dictionary. After feature selection by the principal component analysis algorithm, the number of elements of the two features is the same.

[0030] The histogram of oriented gradients features of the pyramid has a good description of the gradient information of the finger vein image, but only the contour information in the image is used in the calculation process, ignoring a large amount of non-edge information. The local phase quantization feature has good feature extraction performance for both non-blurred and blurred images, but it is not very sensitive to the texture direction of the finger vein image. Therefore, in order to improve the accuracy and security of finger vein recognition, the present invention combines the histogram of oriented gradients features of the pyramid and the local phase quantization features to form a combined dictionary, and thus uses the combined sparse representation classification algorithm to implement identity recognition.

[0031] Preferably, step S3 is specifically as follows:

[0032] S301. The histogram of oriented gradients features of the registered user's finger vein image after feature selection is D1 = [d 11 , d12 ,..., d 1n , the local phase quantization feature of the registered user's finger vein image is D2 = [d 21 , d 22 ,..., d 2n . The combined dictionary is represented as D = [D1, D2];

[0033] S302. The combined sparse coefficient in the combined sparse representation classification algorithm is expressed as:

[0034]

[0035] where, when i = 1, Y i represents the pyramid gradient histogram feature of the finger vein image of the user to be identified. When i = 2, Y i represents the local phase quantization feature of the finger vein image of the user to be identified; β i represents the corresponding sparse coefficient (i.e., β1 represents the sparse coefficient of the pyramid gradient histogram feature, β2 represents the sparse coefficient of the local phase quantization feature, and the sparse coefficient is obtained by convex optimization). β = [β1, β2] is the combined sparse coefficient matrix. argmin represents finding the parameter value that minimizes the following expression, and λ is the balance parameter; the double vertical bars represent finding the distance between two vectors; the subscripts 2 and 1, 2 of the double vertical bars represent the norm;

[0036] S303. After obtaining the combined sparse coefficient matrix β, identity recognition is performed using the principle of minimum reconstruction error, that is:

[0037]

[0038] identity(Y) represents the identity of the user to be identified, where represents the c-th registered sample in the i-th feature matrix, is the sparse coefficient vector corresponding to ; c is a positive integer less than the number of registered users. The identity has no threshold limit and is the user information in the registered template corresponding to the minimum difference. However, in actual applications, considering the influence of unregistered users, a threshold is still required, otherwise unregistered users will still be recognized; the registered samples are the finger vein image features of registered users; that is, for an identification library, it contains several combined dictionaries D, and the combined dictionary of the c-th registered user is represented as while is the sparse coefficient vector corresponding to , is the sparse coefficient vector corresponding to .

[0039] Preferably, step S1 specifically includes:

[0040] S101. Edge extraction;

[0041] S102. Region segmentation;

[0042] S103. Rotation correction;

[0043] S104. Extraction of the region of interest of the finger vein image;

[0044] Specifically, step S101 is as follows: The Sobel edge detection operator is mainly used to calculate the horizontal and vertical gradient information of the finger vein image. The calculation formula is as follows:

[0045]

[0046] Where G x and G y are the finger vein images after horizontal and vertical edge detections respectively, A and B are the Sobel horizontal and vertical edge detection operators respectively, I is the original finger vein image; A and B are preset values;

[0047] After obtaining the horizontal and vertical gradient information of the finger vein image, the gradient magnitude G and direction θ of each pixel point are calculated by integrating the gradient information in these two directions. The calculation formula is as follows:

[0048]

[0049] The calculated gradient values have certain repetitive information. Therefore, non-maximum suppression is used to refine the finger edge and suppress non-edge points; the specific operation of non-maximum suppression is to set all gradient values outside the locally maximum gradient value in the image to 0, that is, to suppress all gradient values except the local maximum;

[0050] Finally, the gradient information after non-maximum processing is binarized by a selected threshold to obtain the finger edge map. The threshold here needs to be determined according to the actual situation.

[0051] Preferably, step S102 is specifically as follows:

[0052] All pixel values between the finger edges are set to 255, and morphological operations such as erosion first and then dilation are performed on the filled image to obtain a mask map of the finger region; the mask map is subjected to a logical AND operation with the original finger vein image to obtain the finger region image, realizing finger region segmentation.

[0053] Preferably, step S103 is specifically as follows: First, calculate the coordinates of each pixel point on the upper and lower edges of the finger along the finger direction in the finger edge diagram, and calculate the midpoints of the upper and lower edges of the finger, so as to obtain a finger center curve;

[0054] After obtaining the finger center curve, according to its coordinate values, use the least squares method to fit it into a straight line L(x):

[0055] L(x) = kx + b

[0056] Where k is the slope of the fitted straight line and b is the intercept of the fitted straight line. The calculation formulas for both are as follows:

[0057]

[0058]

[0059] Where is the mean value of x, and m is the number of pixels on the finger center curve;

[0060] According to the fitted finger midline, calculate the deflection angle of the finger Its calculation formula is as follows:

[0061]

[0062] Rotate the finger with the midpoint of the fitted midline and the deflection angle to achieve the rotation correction of the finger.

[0063] Preferably, step S104 is specifically as follows: First, perform a longitudinal projection summation on the finger vein image after rotation correction to determine the straight line where the maximum pixel sum is located; then use the intersection point of this straight line and the rotated fitted midline as the reference point of the region of interest of the finger vein image; finally, expand 60 pixels up and down, 180 pixels to the left, and 60 pixels to the right with this reference point as the center, so as to obtain the region of interest of the finger vein image.

[0064] The substantial effect brought by the present invention is that it fully considers the characteristics of the finger vein image in the spatial domain and frequency domain, and fuses and identifies the pyramid direction gradient histogram feature and local phase quantization feature of the finger vein through the joint sparse representation classification method, with high reliability, strong security, and good usability. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is an overall flowchart of the present invention;

[0066] Figure 2 is a schematic diagram of the pyramid segmentation of the finger vein image of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0067] The technical solution of the present invention will be further specifically described below through embodiments in conjunction with the accompanying drawings.

[0068] Embodiment: A finger vein recognition method based on feature fusion in this embodiment includes the following steps:

[0069] S1. Preprocess the registered finger vein image and the finger vein image to be recognized collected by the finger vein acquisition device to obtain the region of interest of each finger vein image.

[0070] S2. Extract the histogram of oriented gradients (HOG) features and local phase quantization (LPQ) features from the regions of interest of the two finger vein images.

[0071] S3. Combine the HOG features and LPQ features of the registered finger vein image to form a joint dictionary, and use the joint sparse representation classification algorithm to achieve identity recognition.

[0072] Figure 1 It is an overall flowchart of the present invention. The processing of the two finger vein images does not require simultaneous execution. Generally, the registered finger vein image is processed first in the registration stage until the joint dictionary is formed, and there can be multiple registered finger vein images; in the recognition stage, the finger vein image to be recognized is processed. If there are multiple registered finger vein images during recognition, the joint sparse representation algorithm is used sequentially for recognition in step S3.

[0073] The specific steps for extracting the HOG features are as follows:

[0074] S201. Perform a 3-level pyramid segmentation on the region of interest of the finger vein, that is, divide it into two segments according to the horizontal and vertical coordinates of the image, and each region segmented in the previous level is divided into 4 regions in the next level; thus, a group of finger vein images with different resolutions is obtained and arranged in sequence to form an image pyramid model. The schematic diagram of the finger vein image pyramid segmentation is as Figure 2 shown;

[0075] S202. Calculate the HOG features of the finger vein images at each resolution (i.e., the different resolutions obtained from the previous segmentation) and perform histogram normalization on them.

[0076] S203. Concatenate the normalized HOG features in sequence according to the order of the image pyramid model to obtain the final HOG features.

[0077] The specific steps for extracting the LPQ features are as follows:

[0078] S211. In the finger vein region of interest f(x), a neighborhood interval N of size M×M is taken for each pixel point λ , and then short-time Fourier transform is performed. The calculation formula is as follows:

[0079]

[0080] where u is the frequency, y represents the corresponding integer value in the neighborhood interval N λ , F(u,x) is the finger vein region of interest after Fourier transform, T is the vector transpose operation, and f(x - y) are the pixel values in the M×M neighborhood interval N λ ; the neighborhood interval N λ is of size M×M, that is, M is the number of pixels on one side of the current pixel point;

[0081] In the local phase quantization operator, only the frequency points of four subsets need to be considered to calculate the local Fourier coefficients, that is, u1 = [a,0] T , u2 = [0,a] T , u3 = [a,a] T , u4 = [a,-a] T , where

[0082] For the pixel points in the finger vein region of interest f(x), its vector representation F x is:

[0083] F x = [F(u1,x), F(u2,x), F(u3,x), F(u4,x)]

[0084] Express each element in F x in complex form, then the real part and imaginary part of each component can be marked respectively as:

[0085] G(x) = [Re{F x}, Im{F x}]

[0086] G(x) is the vector composed of the real part and imaginary part of each element in F x ;

[0087] S212. According to the value g j of each element in G(x), perform quantization processing on the phase, that is

[0088]

[0089] where q j is the j-th component of G(x);

[0090] S213. According to the quantized result, convert the binary sequence into a decimal number, and use it as the local phase quantization value f C (x) of the local neighborhood center point, and its calculation formula is as follows:

[0091]

[0092] S214. Statistically analyze the histogram distribution of all local phase quantization values, and use this distribution as the final local phase quantization feature.

[0093] In step S2, after extracting the histogram of oriented gradients features and local phase quantization features of the finger vein image, the principal component analysis algorithm is used to perform feature selection on the two features respectively, which serves as the basis for constructing the combined dictionary. After feature selection by the principal component analysis algorithm, the number of elements of the two features is the same.

[0094] The histogram of oriented gradients features can well describe the gradient information of the finger vein image, but only the contour information in the image is used in the calculation process, ignoring a large amount of non-edge information. The local phase quantization features have good feature extraction performance for both non-blurred and blurred images, but they are not very sensitive to the texture direction of the finger vein image. Therefore, in order to improve the accuracy and security of finger vein recognition, the present invention combines the histogram of oriented gradients features and local phase quantization features to form a combined dictionary, and thus uses the joint sparse representation classification algorithm to achieve identity recognition.

[0095] Step S3 is specifically as follows:

[0096] S301. The histogram of oriented gradients features of the registered user's finger vein image after feature selection is D1 = [d 11 , d 12, ..., d 1n , and the local phase quantization features of the registered user's finger vein image are D2 = [d 21 , d 22 ,..., d 2n . The combined dictionary is expressed as D = [D1, D2];

[0097] S302. The combined sparse coefficients in the joint sparse representation classification algorithm are expressed as:

[0098]

[0099] Among them, when i = 1, Y i represents the histogram of oriented gradients features of the finger vein image of the user to be recognized. When i = 2, Y i represents the local phase quantization features of the finger vein image of the user to be recognized; β irepresent the corresponding sparse coefficients (i.e., β1 represents the sparse coefficient of the pyramid histogram of oriented gradients feature, β2 represents the sparse coefficient of the local phase quantization feature, and the sparse coefficients are obtained by solving convex optimization), β = [β1, β2] is the joint sparse coefficient matrix, argmin represents finding the parameter value that minimizes the following expression, and λ is the balance parameter; the double vertical bars represent calculating the distance between two vectors; the subscripts 2 and 1,2 of the double vertical bars represent norms;

[0100] S303. After obtaining the joint sparse coefficient matrix β, perform identity recognition using the principle of minimizing the reconstruction error, that is:

[0101]

[0102] identity(Y) represents the identity of the user to be recognized, where represents the c-th class of registered samples in the i-th feature matrix, is the sparse coefficient vector corresponding to ; c is a positive integer less than the number of registered users. There is no threshold limit for identity, and it is the user information in the registered template corresponding to the minimum difference. However, in actual applications, considering the influence of unregistered users, a threshold is still required, otherwise unregistered users will still be recognized; the registered samples are the finger vein image features of registered users; that is, for a recognition library, it contains several joint dictionaries D, and the joint dictionary of the c-th registered user is represented as while is the sparse coefficient vector corresponding to ; is the sparse coefficient vector corresponding to ;

[0103] Step S1 specifically includes:

[0104] S101. Edge extraction;

[0105] S102. Region segmentation;

[0106] S103. Rotation correction;

[0107] S104. Extraction of the region of interest of the finger vein image;

[0108] Step S101 is specifically as follows: Mainly use the Sobel edge detection operator to calculate the horizontal and vertical gradient information of the finger vein image, and its calculation formula is as follows:

[0109]

[0110] where G x and G yThey are finger vein images after horizontal and vertical edge detection respectively. A and B are Sobel horizontal and vertical edge detection operators respectively, and I is the original finger vein image; A and B are preset values;

[0111] After obtaining the horizontal and vertical gradient information of the finger vein image, the gradient magnitude G and direction θ of each pixel point are calculated by integrating the gradient information in these two directions. The calculation formulas are as follows:

[0112]

[0113] The calculated gradient values have certain repetitive information. Therefore, non-maximum suppression is used to refine the finger edge and suppress non-edge points; the specific operation of non-maximum suppression is to set all gradient values other than the local maximum gradient value in the image to 0, that is, to suppress all gradient values other than the local maximum value;

[0114] Finally, the gradient information after non-maximum processing is binarized by a selected threshold to obtain the finger edge map. The threshold here needs to be determined according to the actual situation.

[0115] The specific step S102 is as follows:

[0116] All pixel values between the finger edges are set to 255, and morphological operations such as erosion first and then dilation are performed on the filled image to obtain a mask map of the finger region; the mask map is subjected to a logical AND operation with the original finger vein image to obtain the finger region image, realizing finger region segmentation.

[0117] The specific step S103 is as follows: First, calculate the coordinates of each pixel point on the upper and lower edges of the finger along the finger direction in the finger edge map, and calculate the midpoint of the upper and lower edges of the finger, so as to obtain a finger center curve;

[0118] After obtaining the finger center curve, according to its coordinate values, use the least squares method to fit it into a straight line L(x):

[0119] L(x) = kx + b

[0120] Where k is the slope of the fitted straight line and b is the intercept of the fitted straight line. The calculation formulas for the two are as follows:

[0121]

[0122]

[0123] Where is the mean value of x, and m is the number of pixels on the finger center curve;

[0124] According to the fitted finger center line, calculate the deflection angle of the finger Its calculation formula is as follows:

[0125]

[0126] The finger is rotated by using the midpoint of the fitting center line and the deflection angle, so as to realize the rotational correction of the finger.

[0127] The specific content of step S104 is as follows: First, perform a vertical projection summation on the finger vein image after rotational correction to determine the straight line where the maximum pixel sum is located; then use the intersection point of this straight line and the fitting center line after rotation as the reference point of the region of interest of the finger vein image; finally, expand 60 pixels up and down, 180 pixels to the left, and 60 pixels to the right with this reference point as the center, so as to obtain the region of interest of the finger vein image.

[0128] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Those skilled in the art to which the present invention pertains can make various modifications or supplements to the described specific embodiments or use similar methods for substitution, but will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

[0129] Although terms such as histogram of oriented gradients pyramid features and local phase quantization features are used more frequently herein, the possibility of using other terms is not excluded. The use of these terms is only for more conveniently describing and explaining the essence of the present invention; interpreting them as any additional limitation is contrary to the spirit of the present invention.

Claims

1. A finger vein recognition method based on feature fusion, characterized in that, It includes the following steps: S1. Preprocess the registered finger vein image and the finger vein image to be recognized collected by the finger vein acquisition device to obtain the region of interest of each finger vein image; S2. Extract the histogram of oriented gradients (HOG) features and local phase quantization (LPQ) features for the regions of interest of the two finger vein images; S3. Combine the HOG features and LPQ features of the registered finger vein image to form a joint dictionary, and use the joint sparse representation classification algorithm to achieve identity recognition.

2. The finger vein recognition method based on feature fusion according to claim 1, wherein In step S2, the specific steps for extracting the HOG features are as follows: S201. Perform a 3-level pyramid segmentation on the region of interest of the finger vein, that is, divide it into two segments according to the horizontal and vertical coordinates of the image, and each region segmented in the previous level is divided into 4 regions in the next level; thus, a group of finger vein images at different resolutions is obtained and arranged in sequence to form an image pyramid model; S202. Calculate the HOG features of the finger vein images at each resolution and perform histogram normalization on them; S203. Concatenate the normalized HOG features in sequence according to the order of the image pyramid model to obtain the final HOG features.

3. The finger vein recognition method based on feature fusion according to claim 2, wherein, In step S2, the specific steps for extracting the LPQ features are as follows: S211. In the finger vein region of interest f(x), a neighborhood interval N of size M×M is taken for each pixel point λ , and then a short-time Fourier transform is performed. The calculation formula is as follows: where \(u\) is the frequency, \(y\) represents the corresponding integer value in the neighborhood interval \(N\) λ , \(F(u, x)\) is the region of interest of the finger vein after Fourier transform, \(T\) is the vector transpose operation, and \(f(x - y)\) is the pixel value of each pixel in the \(M\times M\) neighborhood interval \(N\) λ ; the neighborhood interval \(N\) λ has a size of \(M\times M\), that is, \(M\) is the number of pixels on one side of the current pixel point; In the local phase quantization operator, only the frequency points of four subsets need to be considered to calculate the local Fourier coefficients, that is, u1 = [a, 0] T , u2 = [0, a] T , u3 = [a, a] T , u4 = [a, -a] T , where For a pixel point in the region of interest f(x) of finger vein, its vector representation F x is as follows: F x = [F(u1, x), F(u2, x), F(u3, x), F(u4, x)] Express each element in F x in the plural form, and then the real part and the imaginary part of each component can be respectively labeled as: G(x) = [Re{F x}, Im{F x}] G(x) is the vector composed of the real part and the imaginary part of each element in F x ; S212. According to the value g of each element in G(x), perform quantization processing on the phase, that is j , where q j is the j-th component of G(x); S213. According to the quantized result, convert the binary sequence into a decimal number, which is used as the local phase quantization value f C (x) of the local neighborhood center point, and its calculation formula is as follows: S214. Statistically analyze the histogram distribution of all local phase quantization values and use this distribution as the final LPQ feature.

4. A finger vein recognition method based on feature fusion according to claim 3, characterized in that In step S2, after extracting the HOG features and LPQ features, the principal component analysis algorithm is used to perform feature selection on the two features respectively as the basis for forming the joint dictionary.

5. A finger vein recognition method based on feature fusion according to claim 1, characterized in that, Step S3 is specifically as follows: S301. The pyramid gradient histogram features of the registered user's finger vein image after feature selection are D1 = [d 11 , d 12 ,..., d 1n , and the local phase quantization features of the registered user's finger vein image are D2 = [d 21 , d 22 ,..., d 2n . The joint dictionary is represented as D = [D1, D2]; S302. The joint sparse coefficient in the joint sparse representation classification algorithm is expressed as: where, when i = 1, Y i represents the pyramid gradient histogram feature of the finger vein image of the user to be recognized. When i = 2, Y i represents the local phase quantization feature of the finger vein image of the user to be recognized; β i represents the corresponding sparse coefficient. β = [β1, β2] is the joint sparse coefficient matrix. argmin represents finding the parameter value that minimizes the following expression. λ is the balance parameter; the double vertical bars represent calculating the distance between two vectors; the subscripts 2 and 1, 2 of the double vertical bars represent norms; S303. After obtaining the joint sparse coefficient matrix β, use the principle of minimum reconstruction error for identity recognition, that is: identity(Y) represents the identity of the user to be recognized, where represents the c-th category of registered samples in the i-th feature matrix, is for the corresponding sparse coefficient vector; c is a positive integer less than the number of registered users.

6. A finger vein recognition method based on feature fusion according to claim 1 or 3 or 5, characterized in that, Step S1 specifically includes: S101. Edge extraction; S102. Region segmentation; S103. Rotation correction; S104. Extraction of the region of interest of the finger vein image; Step S101 is specifically as follows: Mainly use the Sobel edge detection operator to calculate the horizontal and vertical gradient information of the finger vein image, and its calculation formula is as follows: where G x and G y are the finger vein images after horizontal and vertical edge detection respectively, A and B are the Sobel horizontal and vertical edge detection operators respectively, and I is the original finger vein image; After obtaining the horizontal and vertical gradient information of the finger vein image, calculate the gradient magnitude G and direction θ of each pixel point by combining the gradient information in these two directions, and its calculation formula is as follows: Refine the finger edge and suppress non-edge points through non-maximum suppression; Finally, binarize the gradient information after non-maximum processing through a selected threshold to obtain the finger edge map.

7. A finger vein recognition method based on feature fusion according to claim 6, characterized in that, The specific step of step S102 is as follows: Set all pixel values between the finger edges to 255, and perform erosion followed by dilation on the filled image to obtain a mask image of the finger region; perform a logical AND operation on this mask image and the original finger vein image to obtain the finger region image, and achieve finger region segmentation.

8. A finger vein recognition method based on feature fusion according to claim 7, characterized in that, The specific steps of S103 are as follows: First, calculate the coordinates of each pixel point on the upper and lower edges of the finger along the finger direction in the finger edge diagram, and calculate the midpoint of the upper and lower edges of the finger, so as to obtain a finger center curve; After obtaining the finger center curve, according to its coordinate values, use the least squares method to fit it into a straight line L(x): L(x) = kx + b where k is the slope of the fitted straight line and b is the intercept of the fitted straight line. The calculation formulas for both are as follows: where is the mean value of x, and m is the number of pixels on the finger center curve; Calculate the deflection angle of the finger according to the fitted center line of the finger The calculation formula is as follows: Rotate the finger with the midpoint of the fitted median line and the deflection angle to achieve the rotation correction of the finger.

9. A finger vein recognition method based on feature fusion according to claim 8, wherein The specific steps of S104 are as follows: First, perform a longitudinal projection summation on the rotated and corrected finger vein image to determine the straight line where the maximum pixel sum is located; then use the intersection point of this straight line and the rotated fitted median line as the reference point of the region of interest of the finger vein image; finally, expand 60 pixels up and down, 180 pixels to the left, and 60 pixels to the right with this reference point as the center, so as to obtain the region of interest of the finger vein image.

Images