A finger vein image recognition method based on multi-index fusion pre-evaluation

Abstract

The application discloses a kind of finger vein image recognition methods based on multi-index fusion pre-evaluation, comprising: obtaining image quality evaluation index, carrying out normalization processing to evaluation index, generating sample vector, determining classifier core parameter and generating SVM classifier, evaluating and screening picture to generate database to image, carrying out HOG feature extraction and gray level co-occurrence matrix feature extraction to the image processed, fusion two kinds of features form fusion vector, identity matching.The application can effectively reduce the problems such as recognition rate reduction caused by uneven image quality.

Description

Technical Field

[0001] This invention belongs to the field of biometric identification technology in information security, and specifically designs a finger vein image recognition method. Technical Background

[0002] People encounter many biometric devices in their daily lives, and the security and accuracy of these devices have received widespread attention. Currently, biometric features used in the field of biometrics include finger vein recognition, fingerprint recognition, facial recognition, palm print recognition, and iris recognition. Among these, finger vein recognition, as one of many biometric technologies, is gradually becoming a research hotspot.

[0003] Finger vein image acquisition utilizes the principle that hemoglobin in veins can absorb infrared light of a specific wavelength of 700nm to 1000nm. Non-contact imaging is achieved under near-infrared light within this specific wavelength range. This identification method is highly secure, convenient, and more hygienic due to its non-contact operation, making it more acceptable to users.

[0004] During the image acquisition process of a finger vein device, factors such as the pressure applied by the subject's finger, finger placement, changes in ambient lighting, and surrounding temperature can all lead to low-quality finger vein images, or even failure to identify and match them. When there are too many low-quality images in the finger database, it not only reduces the recognition success rate but also increases the comparison time for finger veins, resulting in lower overall device performance. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a finger vein image recognition method based on multi-index fusion pre-evaluation.

[0006] Because the quality of acquired finger vein images varies due to changes in environmental and user factors during the acquisition and recognition process, this invention proposes a relatively efficient and accurate finger vein recognition algorithm based on image quality evaluation to improve the accuracy of subsequent recognition. This invention combines multiple finger vein evaluation indicators. First, these indicators are used as input and fused using an SVM algorithm for pre-evaluation of image acquisition. Then, features are extracted and matched based on a filtered finger vein image database to obtain the final result. Compared with traditional finger vein recognition algorithms, the method proposed in this invention has higher accuracy and can effectively improve the performance of finger vein recognition devices to a certain extent.

[0007] To achieve the above objectives, the specific steps of the technical solution provided by this invention are as follows:

[0008] Step S1: Select four evaluation indicators—equivalent number of views, information entropy, grayscale standard deviation, and contrast—based on the characteristics of finger vein images in the database, and evaluate the quality of the image under test using these four evaluation indicators to obtain four initial quality evaluation indicator scores.

[0009] Step S2 involves normalizing the scores of the four initial quality evaluation indicators to unify their dimensions.

[0010] Step S3: The scores of the four normalized quality evaluation indicators are used as one component of the input sample vector of the SVM classifier.

[0011] Step S4: Determine the kernel function and other core parameters of the SVM classifier, and generate the SVM classifier based on these parameters using training samples.

[0012] Step S5: Input the acquired finger vein images to be detected into the SVM classifier to obtain the image quality evaluation results, retain high-quality finger vein images as the recognition database, and delete low-quality finger vein images.

[0013] Step S6: Preprocess the collected finger veins.

[0014] Step S7: Use the Histogram of Oriented Gradients (HOG) feature extraction algorithm to extract features from the finger vein image. At the same time, use the Gray-Level Co-occurrence Matrix (GLDM) to calculate the contrast, entropy, and correlation of the finger vein image, and calculate the standard deviation and mean of these feature quantities to form another six-dimensional feature vector.

[0015] Step S8: Use PCA to reduce the dimensionality of the extracted HOG feature vector, and then perform weighted fusion with the feature vector extracted by the gray-level co-occurrence matrix to form the fused feature vector HOG-G.

[0016] Step S9: Perform feature matching. If the match is successful, the identity authentication is successful; otherwise, the identity authentication fails.

[0017] The finger vein image recognition method based on multi-index fusion pre-evaluation of the present invention has the following beneficial effects:

[0018] 1. This invention combines a multi-index fusion method to propose a new finger vein recognition method, thereby improving the performance of finger vein recognition devices.

[0019] 2. This invention evaluates the quality of finger vein images by calculating relevant evaluation indicators and performing binary classification of the quality of finger vein images using SVM.

[0020] 3. This invention filters and retains the collected images in the database through image quality evaluation, which can optimize some steps of subsequent feature extraction, reduce the amount of computation, and improve the overall performance of the recognition algorithm.

[0021] 4. Traditional recognition methods may experience increased recognition time and errors due to excessively large or small features such as image information entropy and standard deviation in the database. The pre-evaluation method used in this invention can effectively reduce the problem of decreased recognition rate caused by uneven image quality. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method of the present invention.

[0023] Figure 2 This is a schematic diagram showing the grayscale co-occurrence matrix of the present invention moving away from the frame and in the direction of movement.

[0024] Figure 3 This is a flowchart of the ten-fold cross-validation method of the present invention. Detailed Implementation

[0025] The embodiments of the present invention will be described below.

[0026] The flowchart of the finger vein image recognition method based on multi-index fusion pre-evaluation proposed in this embodiment is as follows: Figure 1 As shown, the method includes the following steps:

[0027] A finger vein image recognition method based on multi-index fusion pre-evaluation comprises eight parts: obtaining image quality evaluation indicators, normalizing the evaluation indicators, generating sample vectors, determining the core parameters of the classifier and generating an SVM classifier, evaluating images and selecting images to generate a database, extracting HOG features and gray-level co-occurrence matrix features from the processed images, fusing the two features to form a fusion vector, and identity matching. The specific steps are as follows:

[0028] S1. Based on the characteristics of finger vein images in the database, four evaluation indicators are selected: equivalent number of views, information entropy, grayscale standard deviation, and contrast. The quality of the image under test is evaluated using these four evaluation indicators to obtain four initial quality evaluation indicator scores.

[0029] S2, normalizes the scores of the four initial quality evaluation indicators to unify their dimensions.

[0030] S3 uses the normalized scores of the four quality evaluation indicators as one component of the input sample vector of the SVM classifier.

[0031] S4. Determine the kernel function and other core parameters of the SVM classifier, and generate the SVM classifier based on these parameters using training samples.

[0032] S5. Input the acquired finger vein images to be detected into the SVM classifier to obtain the image quality evaluation results. High-quality finger vein images are retained as the recognition database, and low-quality finger vein images are deleted.

[0033] S6, Preprocessing of the collected finger veins.

[0034] S7 uses the Histogram of Oriented Gradients (HOG) feature extraction algorithm to extract features from the finger vein image. At the same time, the contrast, entropy and correlation of the finger vein image are calculated using the Gray Co-occurrence Matrix (GLDM), and the standard deviation and mean of these feature quantities are calculated to form another six-dimensional feature vector.

[0035] S8 uses PCA to reduce the dimensionality of the extracted HOG feature vector, and then performs weighted fusion with the feature vector extracted by the gray-level co-occurrence matrix to form the fused feature vector HOG-G.

[0036] S9 performs feature matching; if the match passes, the identity authentication is successful; otherwise, the identity authentication fails.

[0037] Step S1 specifically includes:

[0038] S11. Obtain the original image of the finger vein and obtain the grayscale image P1 of the original image of the finger vein; arbitrarily select two points (x, y) and (x+a, y+b) in the grayscale image of the finger vein, set the value of the point pair to (i, j), and let the point (x, y) scan on the entire image, which will form various (i, j) values; record the number of occurrences of each (i, j), and then arrange them into a grayscale co-occurrence matrix.

[0039] S12. The equivalent visual number is expressed by the following formula:

[0040]

[0041] Where μ is the mean gray value of the vein image, and σ is the standard deviation of the gray value of the vein image.

[0042] The equivalent number of views can effectively reflect the clarity of vein image texture and the grayscale contrast of the entire image. The higher the equivalent number of views, the lower the contrast, the blurrier the image, and the worse the image quality.

[0043] S13. The information entropy is expressed by the following formula:

[0044]

[0045] Where p(i,j) refers to the normalized gray-level co-occurrence matrix.

[0046] Information entropy is a measure of the amount of information contained in an image. It represents the complexity of the texture in the image. The higher the information entropy value, the more complex the texture of the image and the better the image quality.

[0047] S14. The grayscale standard deviation is expressed by the following formula:

[0048]

[0049] Here, mean refers to the average gray level obtained using the normalized gray-level co-occurrence matrix.

[0050] A larger grayscale standard deviation indicates a greater variation in grayscale values ​​in the image. For finger veins, the texture variation in the image is even richer.

[0051] S15. The contrast ratio is expressed by the following formula:

[0052]

[0053] Contrast reflects the total amount of local grayscale change. In an image, the greater the grayscale difference of local pixels, the greater the contrast, the clearer the visual effect, and the better the image quality.

[0054] Step S2 specifically includes:

[0055] S21. The normalization process mentioned above is used to normalize the four quality evaluation indicators when the calculation results are not completely on the same order of magnitude, by using the max-min normalization method.

[0056] S3. Use the normalized scores of the four quality evaluation indicators as one component of the input sample vector to the SVM classifier. Then, the following formula is the expression for the sample vector X:

[0057] X = (Q) d Q s Q b Q c (3-1)

[0058] Among them, Q d Q represents the equivalent visual value of a finger vein image. s The information entropy value, Q, represents the information entropy value of a finger vein image. b Q represents the standard deviation of gray levels in a finger vein image. c This represents the contrast value of the finger vein image.

[0059] For each image, two results are defined for its category Y: -1 indicates that the image is of low quality and is an unqualified image; 1 indicates that the image is of high quality and is a qualified image.

[0060] Step S4 specifically includes:

[0061] S41. Kernel functions mainly include linear kernel, polynomial kernel, radial basis function kernel, sigmoid kernel, etc. Among them, RBF, as a linear kernel function, requires fewer core parameters to be determined compared to other kernel functions and has lower complexity. Therefore, RBF is chosen as the kernel function of this classifier.

[0062] S42. The expression for the kernel function RBF is as follows:

[0063] k(x,x′)=exp(-γ1‖xx′‖ 2 (4-1)

[0064] Where γ1 is the radius of RBF.

[0065] S43. Calculate the misclassification rate based on the given parameter penalty coefficient C and γ, and use, for example... Figure 3 The ten-fold cross-validation method shown is used to calculate the error rate;

[0066] S44. Using the grid parameter optimization method, find the numerical combination of C and γ at the lowest error rate, and determine the final SVM classifier model.

[0067] Step S6 specifically includes: performing grayscale processing on the acquired finger vein image.

[0068] Step S7 specifically includes:

[0069] S71, Gamma correction, uses a normalized image obtained through Gamma correction to effectively reduce the effects of local shadows and lighting variations in the image. Its formula is as follows:

[0070]

[0071] Where I is the current pixel value and γ2 is the grayscale coefficient.

[0072] S72, calculate the gradients in two different directions of the finger vein image, and then calculate the gradient magnitude and direction for each pixel. Perform a convolution operation on the processed image using the [-1,0,1] gradient operator to obtain the gradient component gr in the x-direction. x Use [1, 0, -1] T The gradient operator performs a convolution operation on the processed image to obtain the gradient component gr in the y-direction. y Then calculate the gradient magnitude and direction of that pixel. The calculation formula is as follows:

[0073] gr x (x,y)=P(x+1,y)-P(x-1,y) (7-2)

[0074] gry (x,y)=P(x,y+1)-P(x,y-1) (7-3)

[0075] Where gr x (x,y) represents the horizontal gradient at a pixel (x,y) in the image, gr y (x,y) represents the vertical gradient of a pixel, and P(x,y) represents the pixel value at that point.

[0076] S73 calculates the gradient magnitude and direction of a pixel. The calculation formula is as follows:

[0077]

[0078]

[0079] Where gr(x,y) represents the gradient magnitude of a pixel, and α(x,y) represents the gradient direction at that point.

[0080] S74, construct the gradient direction histogram of the cell unit.

[0081] S75, Block gradient strength normalization.

[0082] S76, extract the HOG feature vector H1.

[0083] S77. Take any two points (x1, y1) and (x1+a, y1+b) from the image, and set the value of this point pair to (i, j). Let point (x, y) be scanned across the entire image, resulting in various (i, j) values. Record the number of occurrences of each (i, j) value, and then arrange them into a gray-level co-occurrence matrix. Calculate the contrast, information entropy, and correlation of the gray-level co-occurrence matrix. The calculation formula is as follows:

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091] S78 calculates the mean and standard deviation of the relevant features of the gray-level co-occurrence matrix to form a 6-dimensional feature vector G1(g1,g2…g6).

[0092] The calculation formula for the weighted fusion described in step S8 is as follows:

[0093]

[0094]

[0095] Where H represents the mean of the HOG feature and α represents the variance of the HOG feature.

[0096]

[0097]

[0098] Where G represents the mean of the gray-level co-occurrence matrix features, and β represents the variance of the gray-level co-occurrence matrix features.

[0099]

[0100] Where HOGG represents the fused feature vector.

[0101] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for recognizing a finger vein image based on multi-index fusion pre-evaluation, characterized in that, The specific steps include the following: S1. Based on the characteristics of finger vein images in the database, four evaluation indicators are selected: equivalent number of views, information entropy, grayscale standard deviation, and contrast. The quality of the finger vein images is evaluated using the four evaluation indicators to obtain four initial quality evaluation indicator scores. S2, normalize the scores of the four initial quality evaluation indicators to unify the dimensions; S3, the normalized scores of the four quality evaluation indicators are used as one component of the input sample vector of the SVM classifier; S4, determine the kernel function and other core parameters of the SVM classifier, and generate the SVM classifier based on these parameters and training samples; S5. Input the acquired finger vein images into the SVM classifier to obtain the image quality evaluation results, retain high-quality finger vein images as the recognition database, and delete low-quality finger vein images. S6, Preprocess the acquired finger vein images; S7. The Histogram of Oriented Gradients (HOG) feature extraction algorithm is used to extract features from the finger vein image. At the same time, the contrast, information entropy and correlation of the finger vein image are calculated using the Gray-Level Co-occurrence Matrix (GLDM). The standard deviation and mean of these feature quantities are also calculated to form another six-dimensional feature vector. S8. PCA is used to reduce the dimensionality of the extracted HOG feature vector, and then the feature vector extracted by the gray-level co-occurrence matrix is ​​weighted and fused to form the fused feature vector HOG-G. S9 performs feature matching; if successful, identity authentication is successful; otherwise, identity authentication fails.

2. The method of claim 1, wherein the method is based on multi-index fusion pre-evaluation of a finger vein image. Step S1 specifically includes: S11. Obtain the original image of the finger vein and obtain the grayscale image of the finger vein. ; Randomly select two points from the grayscale image of the finger vein and Set the value of this point pair to , command point Scanning the entire image will produce various results. Value; record each one The number of occurrences is counted, and then they are arranged into a gray-level co-occurrence matrix; S12. The equivalent number of views is expressed by the following formula: wherein is the mean gray value of the finger vein gray image, is the standard deviation of the gray value of the finger vein gray image; The equivalent number of views can effectively reflect the clarity of the texture of a finger vein image and the grayscale contrast of the entire image. The higher the equivalent number of views, the lower the contrast, the blurrier the image, and the worse the image quality. S13. The information entropy is expressed by the following formula: wherein denotes the normalized gray level co-occurrence matrix; Information entropy is a measure of the amount of information contained in an image. It represents the complexity of the texture in the image. The larger the information entropy value, the more complex the texture of the image and the better the image quality. S14. The grayscale standard deviation is expressed by the following formula: wherein denotes the mean grey value obtained using the normalized grey level co-occurrence matrix; A larger grayscale standard deviation indicates a greater variation in grayscale values ​​in the image. For finger vein images, the texture variations are even richer. S15. The contrast ratio is expressed by the following formula: Contrast reflects the total amount of local grayscale change. In an image, the greater the grayscale difference of local pixels, the greater the contrast, the clearer the visual effect, and the better the image quality.

3. The method of claim 1, wherein the method is based on multi-index fusion pre-evaluation of a finger vein image. The normalization process described in step S2 is to normalize the scores of the four initial quality evaluation indicators, which are not entirely on the same order of magnitude, using the max-min normalization method.

4. The method of claim 1, wherein the method is based on multi-index fusion pre-evaluation of a finger vein image. The sample vector described in step S3 Expression: in, This represents the equivalent visual value of a finger vein image. This represents the information entropy value of a finger vein image. This represents the standard deviation of grayscale values ​​in a finger vein image. This represents the contrast value of the finger vein image; At the same time, assign a category to each image. Define two results, This indicates that the image is of low quality and is a substandard image; This indicates that the image is of high quality and is a qualified image.

5. The method of claim 1, wherein the method is based on multi-index fusion pre-evaluation of a finger vein image. Step S4 specifically includes: S41. Kernel functions include linear kernel, polynomial kernel, radial basis function kernel, and sigmoid kernel. Among them, RBF, as a linear kernel function, requires fewer core parameters to be determined and has lower complexity compared to other kernel functions. Therefore, RBF is chosen as the kernel function for this classifier. S42. The expression for the kernel function RBF is as follows: wherein R is the radius of the RBF; S43、According to the given parameter penalty coefficient With Calculate the error rate, and calculate the error rate using ten-fold cross-validation method; S44, find the minimum error rate under the grid parameter optimization method and the numerical combination, and determine the final SVM classifier model.

6. The finger vein image recognition method based on multi-index fusion pre-evaluation as described in claim 1, characterized in that: Step S6 specifically includes: performing grayscale processing on the acquired finger vein image.

7. The method of claim 1, wherein the method is based on multi-index fusion pre-evaluation of a finger vein image. Step S7 specifically includes: S71, Gamma correction, uses a normalized image obtained through Gamma correction to effectively reduce the effects of local shadows and lighting variations in the image. Its formula is as follows: wherein is the current pixel value, is the gray scale coefficient; S72, calculate the gradient in two different directions of the finger vein image, and then calculate the gradient magnitude and direction for each pixel; using The gradient operator performs a convolution operation on the processed image to obtain... gradient components of direction ,use The gradient operator performs a convolution operation on the processed image to obtain... gradient components of direction Then calculate the gradient magnitude and direction of that pixel; the calculation formula is as follows: wherein denotes the horizontal gradient at a certain pixel point in the image, denotes the vertical gradient at the pixel point, denotes the pixel value at the point;​ S73 calculates the gradient magnitude and gradient direction of a pixel; the calculation formula is as follows: wherein represents the gradient magnitude of the pixel point, and represents the gradient direction of the point; S74, construct the gradient direction histogram of the cell unit; S75, Block gradient strength normalization; S76, extract HOG feature vector ; S77, arbitrarily select two points from the image. and Set the value of this pixel pair to , command point Scanning the entire image will produce various results. Value; record each one The number of occurrences is counted, and then these are arranged into a gray-level co-occurrence matrix. The contrast, information entropy, and correlation of the gray-level co-occurrence matrix are calculated; the calculation formula is as follows: S78, calculate the mean and standard deviation of the contrast, information entropy and correlation of the gray level co-occurrence matrix to form a 6-dimensional feature vector .

8. The method of claim 1, wherein the method is based on multi-index fusion pre-evaluation of a finger vein image. The calculation formula for the weighted fusion described in step S8 is as follows: wherein denotes the mean of the HOG features, denotes the variance of the HOG features; in The mean value representing the features of the gray-level co-occurrence matrix. The variance representing the features of the gray-level co-occurrence matrix; wherein represents the fused feature vector.

Images