A finger vein recognition method based on compact binary coding learning

Abstract

A finger vein recognition method based on compact binary coding learning relates to the technical field of pattern recognition and image processing. The method comprises the following steps: extracting finger vein texture difference vector, learning mapping function, mapping and feature integration, calculating and splicing block histogram, feature matching, finger vein recognition. The existing algorithm has two main defects: (1) most current methods highly depend on rich prior knowledge base, such dependence not only increases the complexity of implementation, but also is easily affected by adverse factors; (2) the sample quantity of existing finger vein database is small, and the finger vein recognition method based on deep learning is limited in practical application. The present application can effectively improve the recognition accuracy of finger vein recognition under the condition of few samples, and has strong self-adaptability and generalization. Therefore, the present application has certain application value and significance.

Description

Technical Field

[0001] This invention relates to the fields of biometrics and image processing, and in particular to a finger vein recognition technology for situations with a small number of samples. The invention proposes a compact binary encoding learning method that focuses on automatically extracting and encoding binary features from finger vein images to achieve accurate finger vein recognition. Background Technology

[0002] With the acceleration of digitalization, the demand for accuracy and security in identity verification is becoming increasingly urgent. Although traditional biometric identification methods such as face and fingerprint recognition are widely used, issues such as privacy, security, and physical limitations are becoming increasingly prominent. Against this backdrop, finger vein recognition technology is gradually emerging due to its unique advantages. This technology is based on the unique absorption characteristics of infrared light by the veins inside the finger for imaging, and has the following significant highlights: (1) Enhanced security: Subcutaneous vein features are difficult to forge and require live verification. The non-contact collection method also improves hygiene. (2) Improved recognition accuracy: Vein features are stable and not easily affected by external environmental interference. At the same time, individual differences are large, ensuring a high recognition rate. Continuous technological progress further promotes the improvement of accuracy. (3) Enhanced stability: This technology is adaptable to various environments, and the equipment is robust and durable, ensuring long-term reliability. This technology is not only used in traditional access control and financial services, but has also gradually expanded to rapid identity verification under epidemic prevention and control, strict security checks in customs and border management, security of electronic payments, seamless integration of smart offices, and accurate identification in the field of intelligent transportation, fully demonstrating its broad value and application prospects as a future identity authentication technology.

[0003] Currently, existing finger vein recognition algorithms can be broadly categorized into four types:

[0004] Global feature methods, such as Principal Components Analysis (PCA) and Linear Discriminant Analysis (LDA), deeply analyze image data from statistical and mathematical dimensions, extracting key global features to aid in identification. Local feature methods, including Local Binary Pattern (LBP), maximum curvature, and Gabor wavelets, delicately capture the diversity of vein features. Meanwhile, lightweight learning models, such as binary feature learning, sparse coding, and multi-view techniques, demonstrate powerful adaptive learning capabilities, accurately extracting discriminative features. Furthermore, deep learning, relying on advanced networks such as Convolutional Neural Networks (CNNs) or Generative Adversarial Networks (GANs), automatically extracts and constructs efficient finger vein recognition templates, further expanding the boundaries of the technology's applications.

[0005] Traditional finger vein recognition strategies primarily revolve around global and local features, relying on researchers' deep understanding of vein image characteristics and manually extracting key features such as texture and details. While these methods are based on rich experience and design knowledge, they are more suitable for class differentiation in closed environments. On the other hand, lightweight learning methods attempt to transform image data into a feature space through relatively simple mapping rules or adaptive learning algorithms to enhance recognition efficiency; however, they are often limited by the manual design process of pre-extracting features. As for deep learning technology, although it holds great promise, its potential is limited by the scarcity of existing finger vein database samples. Therefore, developing a lightweight finger vein recognition method that combines high accuracy, low sample requirements, and low reliance on prior knowledge has become a core challenge that urgently needs to be addressed in this field. Summary of the Invention

[0006] This invention provides a finger vein recognition method based on compact binary encoding learning, which automatically learns vein features from finger vein images to achieve high-accuracy finger vein recognition.

[0007] To implement the above method, the specific steps are as follows:

[0008] Step 100: First, the input dataset is randomly and evenly divided into a training set and a test set to ensure the independence and effectiveness of model training and evaluation. During the training phase, an optimized feature extraction process is employed. Specifically, the finger vein images in the training set are convolved with a predefined Local Binary Pattern (LBP) template to obtain rich texture response results. Subsequently, by meticulously calculating the difference between the response results of each central pixel and its surrounding neighboring pixels in the LBP template, a unique Texture Local Difference Vector (TLDV) is generated for each finger vein image. This process not only preserves the key texture information of the image but also enhances the discriminative power of features through difference operations, laying a solid foundation for subsequent feature learning and recognition.

[0009] Step 200: Using the TLDVs and corresponding sample labels of these training sets, a Compact Binary Codes Learning (CBCL) model is constructed. The core of this model lies in the introduction of an objective function designed to adaptively learn a mapping matrix and an error matrix, converting the TLDVs of the training set into binary codes. The objective function considers both regularization and sparsity constraints, transforming the feature learning problem in the subspace into an optimization problem of solving the CBCL model.

[0010] Step 300: To optimize the above CBCL model, this technique adopts the efficient Alternating Direction Method of Multipliers (ADMM). This algorithm decomposes the optimization problem into multiple easily solvable variable optimization problems and introduces an augmented Lagrangian function to make the updated variables satisfy the above constraints, thereby achieving the optimization solution of the overall model and finally outputting the optimal mapping matrix and error matrix.

[0011] Step 400: During the testing process, the optimal mapping matrix and error matrix obtained above are used to perform projection operations on the training set and test set data, thereby extracting binary features from these two sets of data for subsequent processing and recognition.

[0012] Step 500: First, convert the binary features mapped from the above training data into real-valued features. Then, divide the real-valued features into blocks and calculate the local block histograms. Finally, stitch them together into a global histogram as the final feature representation.

[0013] Step 600: By calculating the Euclidean distance between the final feature representations of the training set and the test set, the nearest neighbor classifier (1-NN) is used to classify the category of the finger vein image, and further perform finger vein recognition.

[0014] Step 700: Repeat steps 100 to 600 ten times to obtain the recognition accuracy rate for the ten times, and calculate the average recognition accuracy rate.

[0015] The process of randomly dividing the input dataset into a training set and a test set, and generating a texture difference vector (TLDV) for each finger vein image by convolving the finger vein images and local binary pattern (LBP) templates in the training set during the training phase, and then calculating the difference between the response results of the center pixel and the neighboring pixels to generate the TLDV for each finger vein image, includes the following steps:

[0016] First, the input dataset is randomly divided into a training set and a test set. There are two specific approaches: the first is to randomly select two finger vein images from each category as the training set, and the remaining finger vein images of that category as the test set; the second is to randomly select three finger vein images from each category as the training set, and the remaining finger vein images of that category as the test set.

[0017] Secondly, during the training phase, for the finger vein images in the training set, a certain region of adjacent pixels is selected as a finger vein patch, and this patch is convolved with the Local Binary Pattern (LBP) template from "Finger vein recognition with gabor wavelets and local binary patterns" to obtain the response result. Finally, by calculating the difference between the response results of the center pixel and the neighboring pixels, a Texture Local Difference Vector (TLDV) is generated for each local patch of the finger vein image, with a vector dimension of 24.

[0018] The process of constructing a compact binary code learning model (CBCL) using the TLDVs and corresponding sample labels of these training sets is crucial. The core of this model lies in introducing a novel objective function designed to adaptively learn a mapping matrix and an error matrix. Step 200, which converts the TLDVs of the training set into binary codes and transforms the feature learning problem in the subspace into an optimization problem of the CBCL model, includes:

[0019] A unified feature learning model (CBCL) is constructed using the TLDV of the training set and its corresponding labels. An objective function is introduced to adaptively learn a mapping matrix and an error matrix. These two matrices can adaptively convert the training data into the binary code corresponding to each data. Then, considering both regularization constraints and sparsity constraints, a new objective function is proposed, which transforms the feature learning problem in the subspace into an optimization problem of solving the CBCL model.

[0020] The Alternating Directed Multiplier Method (ADMM) is used to solve the optimization problem of the CBCL model. All variables in the model are iteratively updated to gradually approach the global optimum. Finally, step 300, which outputs the optimized mapping matrix and error matrix, includes:

[0021] By introducing Lagrange multipliers within the ADMM framework, the objective function optimization problem of the CBCL model is transformed into a multivariate optimization problem of solving the Lagrange function, thus reducing the difficulty of handling complex optimization problems. Then, an iterative method is employed, fixing other variables each time and minimizing the Lagrange function for only one variable, simplifying the multivariate optimization problem into a series of univariate optimization problems. This strategy significantly improves computational efficiency and simplifies the solution process. Iteration continues until the convergence condition is met, i.e., the objective function no longer changes significantly in consecutive iterations, or the preset maximum number of iterations is reached. At this point, the model is considered to have converged to a stable state, and the current mapping matrix and error matrix are output. These matrices can effectively convert TLDV into binary encoding, supporting subsequent recognition tasks.

[0022] In the testing phase, step 400, which involves projecting the learned mapping matrix onto the training and test sets respectively to obtain the binary features on the training and test sets, includes:

[0023] Using the learned mapping matrix and error matrix, the training set data and test set data are projected according to the initial objective function to obtain the binary features of the training set data and test set data.

[0024] The step 500, which involves first converting the binary features mapped from the training data into real-valued features, then dividing the real-valued features into blocks and calculating local block histograms, and finally concatenating them into a global histogram as the final feature representation, includes:

[0025] First, based on the mapping matrix, the image TLDV is converted into a binary feature vector. Each bit of the vector has its corresponding weight. Each bit value is multiplied by the weight, and finally the results are summed to convert the binary features into real-valued features.

[0026] Then, the real-valued feature is transformed to the same size as the original finger vein image and divided into multiple non-overlapping small blocks of size . For each block, its local histogram is extracted, and these histograms are combined into a feature vector of dimension as the final feature representation.

[0027] The step 600, which involves calculating the Euclidean distance between the final feature representations of the training and test sets and classifying the finger vein images using a nearest neighbor classifier (1-NN), further includes:

[0028] First, calculate the Euclidean distance between the final feature representation of each image in the training set and the final feature representation of a single image in the test set.

[0029] Next, a nearest neighbor classifier (1-NN) is used to classify the finger vein images in the test set based on these distances, thereby obtaining the predicted label for each test image.

[0030] Finally, by comparing the predicted labels with the actual labels in the test data, the recognition accuracy of the CBCL method can be calculated, which serves as the basis for evaluating the model's performance.

[0031] Step 700, which involves repeating steps 100 to 600 ten times to obtain the recognition accuracy rate ten times, includes:

[0032] Repeating steps 100 to 600 ten times yields the recognition accuracy for ten tests. The average recognition accuracy is then calculated as an indicator to evaluate the recognition performance of the method of the present invention.

[0033] The present invention has the following advantages:

[0034] (1) It has strong adaptability. The method proposed in this invention is a data-driven adaptive feature learning algorithm that has little dependence on prior knowledge.

[0035] (2) Improve the average recognition accuracy;

[0036] (3) The present invention does not require a large database size and can be applied to learning with few samples.

[0037] (4) It exhibits excellent generalization potential. By simply adjusting the parameter settings in the objective function, this method can be flexibly transferred to other similar biometric recognition fields, covering multiple application scenarios such as knuckle prints, palm prints, and even palm vein recognition, achieving wide applicability and high flexibility. Attached Figure Description

[0038] Figure 1 This is a flowchart of a finger vein recognition method based on compact binary encoding learning.

[0039] Figure 2 This is a schematic diagram of a TLDV image obtained from a finger vein.

[0040] Figure 3 Schematic diagram of finger vein ROI images from two different databases

[0041] Figure 4 A schematic diagram of sensitivity analysis for parameters λ1 and λ2 on two different databases.

[0042] Figure 5 The relationship curves between the objective function value and the number of iterations on two different databases. Detailed Implementation

[0043] The method will be described in detail below with reference to the accompanying drawings. It should be noted that, in general, this method is applicable to finger vein recognition with different image qualities and in different application scenarios.

[0044] This application provides a finger vein recognition method based on compact binary encoding learning, which adaptively learns compact binary encoded vein features in the case of small samples to improve the accuracy of finger vein recognition.

[0045] Figure 1 A flowchart of a finger vein recognition method based on compact binary encoding learning according to this application is shown.

[0046] This application proposes a finger vein recognition method based on compact binary encoding learning, the specific steps of which are as follows:

[0047] Step 100: Randomly divide the input dataset into a training set and a test set; During the training phase, convolve the finger vein images of the training set with the local binary pattern (LBP) template to obtain the response results, and then generate the texture local difference vector (TLDV) for each finger vein image by calculating the difference between the response results of the center pixel and the neighboring pixels.

[0048] Step 200: Using the TLDVs and corresponding sample labels of these training sets, a Compact Binary Codes Learning (CBCL) model is constructed. The core of this model lies in introducing an objective function designed to adaptively learn a mapping matrix and an error matrix, converting the TLDVs of the training set into binary codes. Regularization and sparsity constraints are also considered, transforming the feature learning problem in the subspace into an optimization problem of solving the CBCL model.

[0049] Step 300: Solve the optimization problem of the above CBCL model using the Alternating Direction Method of Multipliers (ADMM). Iterate and update all variables until the model converges or reaches the preset maximum number of iterations, and output the optimal mapping matrix and error matrix.

[0050] Step 400: During the testing phase, the training set data and test set data are projected using the mapping matrix learned above to obtain the binary features on the training set data and test set data respectively.

[0051] Step 500: First, convert the binary features mapped from the above training data into real-valued features. Then, divide the real-valued features into blocks and calculate the local block histograms. Finally, stitch them together into a global histogram as the final feature representation.

[0052] Step 600: By calculating the Euclidean distance between the final feature representations of the training set and the test set, the nearest neighbor classifier (1-NN) is used to classify the category of the finger vein image, and further perform finger vein recognition.

[0053] Step 700: Repeat steps 100 to 600 ten times to obtain the recognition accuracy rate for the ten times, and calculate the average recognition accuracy rate.

[0054] Step 100 includes the following sub-steps:

[0055] Sub-step 110: Randomly divide the input dataset into a training set and a test set. There are two specific methods: the first is to randomly select two finger vein images from each category as the training set, and the remaining finger vein images of that category as the test set; the second is to randomly select three finger vein images from each category as the training set, and the remaining finger vein images of that category as the test set. Figure 3 This is a schematic diagram of finger vein ROI images from two different databases.

[0056] Sub-step 120: During the training phase, for the finger vein images in the training set, select 5×5 neighboring pixels as a finger vein patch, and convolve it with a Local Binary Pattern (LBP) template to obtain the response result. Finally, by calculating the difference between the response results of the center pixel and the neighboring pixels, a 24-dimensional texture difference vector (TLDV) is generated for each finger vein patch. Figure 2 This is a schematic diagram of obtaining a digital vein image (TLDV).

[0057] Step 200 includes the following sub-steps:

[0058] Sub-step 210: Construct a unified feature learning model (CBCL) using the TLDVs of the training set and their corresponding labels, and introduce the objective function:

[0059]

[0060] Where x n ∈R d×p For the TLDV extracted from the nth finger vein sample, R represents x n All values ​​are real numbers, where d represents the dimension of the TLDV and p represents the image size (rows × columns). n,k ∈{0,1} k It is the kth learned binary code, q n This is the projection regularization of the nth sample; let X = [x1, x2, ..., x n ]∈R d×m (m = np) represents the set of n finger vein samples, Q = [q1, q2, ..., q n ] represents the projection error, Y = [y1, y2, ..., y n ] represents binary encoding, and the objective function in formula (1) can be rewritten in matrix form:

[0061] Y = sgn(P) T X+Q) (2)

[0062] The goal is to adaptively learn a mapping matrix P and an error matrix Q to convert training data X into binary encoded Y.

[0063] Sub-step 220: By introducing constraints in the early stage, the goal of the CBCL model is: (1) to minimize the Euclidean distance of the intra-class sample projections and maximize the Euclidean distance of the inter-class sample projections. (2) The sparse feature selection matrix helps to select more important and compact features, and through l 2,1 Norm constraints can improve the separability of data. In this case, the overall objective function can be expressed as:

[0064]

[0065] Here, Intra-class represents samples of the same class, and Inter-class represents samples of different classes. R(Y,Q) is a regularization term to avoid overfitting, where λ1 and λ2 are two positive weighting parameters used to balance the contribution of the relative constraint term. The parameter λ1 takes any one of the nine values ​​{1e-4, 1e-3, 1e-2, 1e-1, 1e0, 1e1, 1e2, 1e3, 1e4}, and similarly, the parameter λ2 takes any one of the nine values ​​{1e-4, 1e-3, 1e-2, 1e-1, 1e0, 1e1, 1e2, 1e3, 1e4}. Indicate l 2，1 Norm, ||·|| 2 Indicate l F Norm.

[0066] Suppose C is a binary matrix representing the relationships between images within and between classes. C = 1 if two images belong to the same intra-class, and -1 otherwise. Let Y = [y1, y2, ..., y...]. n If ] is the binary matrix of the projection, then formula (3) can be rewritten in matrix form:

[0067]

[0068] in ||·|| 2，1 They represent l respectively F Norm, l 2,1 In Equation (4), the proposed CBCL aims to learn the projection matrix P and the projection error Q. By minimizing the overall objective function, two core objectives are achieved: one is to minimize the error during the projection process, and the other is to minimize the distance between data of the same class and maximize the distance between data of different classes. Therefore, the feature learning problem in the subspace is transformed into the optimization problem of solving the CBCL model;

[0069] Furthermore, step 300 includes the following sub-steps:

[0070] Sub-step 310: Introducing the Lagrange multiplier M, formula (4) can be transformed into the following augmented Lagrange function:

[0071]

[0072] Where M∈R n×d It is an n x d matrix of real numbers; δ = 0.1 is the penalty parameter. Formula (5) is a multivariate optimization problem of the Lagrangian function. By iteratively updating one of the variables through the Lagrangian function, the optimal solutions of variables P, Y, Q, M and δ when the Lagrangian function reaches its minimum value are found.

[0073] Sub-step 320: Before performing the iterative update, the variables need to be initialized, let δ... max =e 5 ,ρ=1.01,δ=0.1, use the randn() function in Matlab to initialize variables P, Y, Q, and use the rand() function to initialize variable M, thereby generating matrix random numbers, where each element in M ​​is between 0 and 1; calculate the initial value of the overall objective function of the sub-formula (5) under the current variable;

[0074] Sub-step 330: Solve the optimization problem of the above CBCL model using the Alternating Directed Multiplier Method (ADMM), pre-setting the maximum number of iterations t = 50, and repeat the following iterative steps:

[0075] Step 1) Fix P, Y, and M, according to Update variable Q;

[0076] Step 2) Fix Q, Y, and M, and according to P = (λ1D + 2XX) T ) -1 X·U T Update variable P, where definition P k Let P be the k-th row vector;

[0077] Step 3) Fix P, Q, and M, according to Update variable Y;

[0078] Step 4) Fix P, Q, and Y, and use M = M + δ(YP) T XQ), δ=min(ρ,δ,δ) max Update variables M and δ, where δ max =e 5 ρ = 1.01, δ = 0.1;

[0079] Step 5) Based on the updated variables, calculate the difference between the current value of the overall objective function (5) and the value of the overall objective function in the previous iteration;

[0080] Sub-step 340: Repeat the above process until the preset maximum number of iterations t = 50 is reached or the difference between the current value of the overall objective function (5) and the value of the overall objective function in the previous iteration is less than 10. -4 Then output variables P and Q.

[0081] Furthermore, step 400 includes:

[0082] Sub-step 410: Using the learned mapping matrix P and error matrix Q, according to Y = sgn(P) T Projecting the training set data X onto the training set data (X+Q) yields the binary feature matrix Y = [y1, y2, ..., y...]. n ]∈R d×m y i ∈R m×1 Let be the feature vector after mapping the i-th image.

[0083] Sub-step 420: Process the test set data in the same way to obtain the binary feature matrix Z = [z1, z2, ..., z2] of the test set data. q ]∈R d×s(s = qp), where d represents the dimension of TLDV, p represents the size of the image (rows × columns), and z i ∈R s×1 Let be the feature vector after mapping the i-th image.

[0084] Furthermore, step 500 includes:

[0085] Sub-step 510: Based on the mapping matrix, convert the image TLDV into an m-bit binary feature vector y. i ∈R m×1 Each bit has a weight of 2. m-1 Each bit value is multiplied by a weight, and the results are summed to convert binary features into real-valued features.

[0086] Sub-step 520: Convert the real-valued feature to the same size as the original finger vein image, and divide it into multiple non-overlapping 16×16 blocks. Extract the local histogram for each block, and combine these histograms into a feature vector with a dimension of 256×p as the final feature representation, where p represents the total number of blocks.

[0087] Step 600 includes:

[0088] Sub-step 610: Calculate the Euclidean distance between the final feature representation of each image in the training set and the final feature representation of an image in the test set after mapping. Then, use the nearest neighbor classifier to classify the finger vein images in the test set and output the labels of the finger vein images in the test set.

[0089] Sub-step 620: Calculate the recognition accuracy by matching the labels of the test set images with the real labels of the data.

[0090] Step 700 includes:

[0091] Repeat steps 100 to 620 ten times to obtain the recognition accuracy rate for the ten times, and calculate the average recognition accuracy rate.

[0092] The original finger vein images used in this invention were obtained from two publicly available finger vein databases, SDUMLA-fv and CAUC-fv. The SDUMLA-fv database, a public finger vein dataset created by Shandong University, comprises 636 categories. This dataset consists of images from the left and right hands of 106 volunteers, with three fingers from each hand grouped into one category (106 × 2 × 3 = 636). The ROI images in this dataset have been cropped to 120 × 233 pixels. The CAUC-fv database, established by the Civil Aviation University of China, contains 5850 images from 585 fingers of 195 individuals. For each person, the index, middle, and ring fingers of one hand were collected. Ten finger vein images were obtained for each finger. The ROI images in this database have been cropped to 91 × 200 pixels. Figure 3 This is a schematic diagram of finger vein ROI images from two different databases.

[0093] The experiment was conducted on a PC using Matlab R2016a.

[0094] First, in this invention, the settings of the two parameters λ1 and λ2 in the objective function affect the recognition performance of finger vein recognition. Therefore, a hierarchical grid search strategy is adopted to select appropriate parameter values ​​for each database, by selecting the two parameters respectively from {1e -4 ,1e -3 ,1e -2 ,1e -1 ,1e 0 ,1e 1 ,1e 2 ,1e 3 ,1e 4}, that is, λ1 or λ2 can be any one of these nine data points each time, so there are a total of 9×9=81 combinations, and then parameter sensitivity analysis was performed. Figure 4 This is a schematic diagram illustrating the sensitivity analysis of parameters λ1 and λ2 on two different databases.

[0095] from Figure 4 It can be observed that when λ1∈[1e -4 ,1e 4 Within the range, λ2∈[1e 1 ,1e 2 Within the range of λ2, the system performance is stable and good. However, when λ2∈[1e... 3 ,1e 4 Within the specified range, the average accuracy drops rapidly. Under other parameter combinations, the system performance is mediocre and not stable enough. Therefore, for the SDUMLA-fv dataset, the recommended optimal parameter combination is λ1 = 1e 3 and λ2=1e -3Similarly, experimental results show that when λ1∈[1e -4 ,1e 4 Within the range, λ2∈[1e -4 ,1e 3 When λ2 ∈ [1e], the system exhibits stable and good recognition performance on the CAUC-fv dataset. However, when λ2 ∈ [1e], the performance is significantly reduced. 3 ,1e 4 When [the parameter] is [value], the average accuracy decreases significantly. For the CAUC-fv dataset, the optimal parameter combination was determined to be λ1 = 1e. 4 , λ2=1e -2 .

[0096] Second, the convergence speed of the objective function of the method of this invention (Compact Binary Encoding Learning Model CBCL) was investigated. Figure 5 The graphs show the relationship between the objective function value and the number of iterations on two different databases. Figure 5 The experiment, conducted over t=50 iterations, revealed that the objective function value initially decreased rapidly before leveling off and stabilizing. Notably, the proposed CBCL objective function converged very quickly on both finger vein datasets, typically within 10 iterations.

[0097] Third, the average recognition accuracy of the method of this invention (Compact Binary Coding Learning Model CBCL) was compared with that of ten common finger vein recognition methods (Global Local Binary Pattern (GLBP), Local Binary Pattern (LLBP), Competitive Coding, Local Linear Discriminant Projection (LLDP), Weighted Local Graph Structure (WLGS), VGG Network (VGG-F), Convolutional Neural Network (CNN-based), Joint Multi-View Feature Learning (JMvFL), Deep Structural Feature Discrimination (DSFD), and Coupled Multimodal Feature Representation (CMrFD)). All comparison methods used the same matching protocol and were run 10 times to calculate the average recognition accuracy. Table 1 shows the average recognition accuracy of each method on the SDUMLA-fv and CAUC-fv finger vein databases. Experimental results show that the method of this invention consistently outperforms other methods under both training and test set allocation methods (Tr=2, Tr=3). The accuracy of the method of this invention on SDUMFL-fv can reach 99.2745% (Tr=3), and the accuracy on CAUCC-fv can reach 100% (Tr=3).

[0098] Table 1 shows the average recognition accuracy (%) of different recognition methods on two databases.

[0099]

[0100]

[0101] Fourth, further, in order to verify the effectiveness of different constraints in the CBCL objective function, the method of the present invention was subjected to ablation experiments. Table 2 lists the average recognition accuracy after removing different constraint terms.

[0102] Table 2. Average recognition accuracy (%) after removing different constraints.

[0103]

[0104] As clearly shown in Table 2, regardless of the two finger vein databases (SDUMLA-fv and CAUC-fv), the performance of CBCL is affected even without applying sparsity constraints (λ1 = 0) or regularization constraints (λ2 = 0), resulting in a decrease in recognition accuracy. This indicates that regularization and sparsity constraints are crucial for improving CBCL recognition performance. Notably, on both datasets, the decrease in recognition accuracy after removing the regularization constraint is more significant, demonstrating that regularization constraints outperform sparsity constraints.

[0105] The finger vein recognition method based on compact binary encoding learning in this invention has the following advantages:

[0106] (1) It has strong adaptability. The method proposed in this invention is a data-driven adaptive feature learning algorithm that has little dependence on prior knowledge.

[0107] (2) Improve the average recognition accuracy;

[0108] (3) The present invention does not require a large database size and can be applied to learning with few samples.

[0109] (4) It has strong generalization ability. By simply adjusting the hyperparameters in the objective function, the extraction method of this invention can be applied to other similar biometric recognition, such as knuckle print recognition, palm print recognition and palm vein recognition.

Claims

1. A finger vein recognition method based on compact binary encoding learning, characterized in that, Includes the following steps: Step 100: Randomly divide the input dataset into a training set and a test set; During the training phase, the finger vein images in the training set are convolved with the local binary pattern (LBP) template to obtain the response results. Then, by calculating the difference between the response results of the center pixel and the neighboring pixels, the texture difference vector (TLDV) of each finger vein image is generated. Step 200: Use the TLDVs of these training sets and the corresponding sample labels to construct a learning model CBCL based on compact binary encoding. The core of this model is to introduce an objective function that aims to adaptively learn a mapping matrix and an error matrix, convert the TLDVs of the training set into binary encoding, and consider regularization constraints and sparsity constraints. The feature learning problem in the subspace is transformed into the optimization problem of solving the CBCL model. Step 300: Solve the optimization problem of the above CBCL model using the Alternating Multiplier Method (ADMM). Iterate and update all variables until the model converges or reaches the preset maximum number of iterations, and output the optimal mapping matrix and error matrix. Step 400: In the testing phase, the training set data and the test set data are projected onto the mapping matrix respectively to obtain the binary features on the training set data and the test set data. Step 500: First, convert the binary features mapped from the training data into real-valued features. Then, divide the real-valued features into blocks and calculate the local block histograms. Finally, stitch them together into a global histogram as the final feature representation. Step 600: By calculating the Euclidean distance between the final feature representations of the training set and the test set, the nearest neighbor classifier is used to classify the category of the finger vein image, and further perform finger vein recognition. Step 700: Repeat steps 100 to 600 ten times to obtain the recognition accuracy rate for the ten times, and calculate the average recognition accuracy rate; Step 200 includes: A unified feature learning model CBCL is constructed using the TLDVs and their corresponding labels from the training set, and an objective function is introduced: (1) in For the first TLDV extracted from individual finger vein samples, represent All values ​​are real numbers. This represents the dimension of the TLDV. The size of the image is represented by... ; It is the first A learned binary code, It is the first Projection regularization of each sample; let Where m=np, representing A collection of finger vein samples, Indicates projection error. The binary matrix representing the projection, and the objective function in formula (1) rewritten in matrix form: (2) The goal is to adaptively learn a mapping matrix. and error matrix training data Convert to binary encoding ; The overall objective function is expressed as: ； in, Indicates samples of the same category, Represents samples of different categories; It is a regularization term to avoid overfitting, where and Two positive weighting parameters are used to balance the contributions of the relative constraint terms. The parameter can be any value selected from the following range: {1e-4, 1e-3, 1e-2, 1e-1, 1e0, 1e1, 1e2, 1e3, 1e4}. Choose any value from the following: {1e-4, 1e-3, 1e-2, 1e-1, 1e0, 1e1, 1e2, 1e3, 1e4}. express Norm, express Norm; set up It is a binary matrix representing the relationships between images within and between classes; where, if two images belong to the same class, then... ,otherwise, ;set up If the binary matrix is ​​the projection, then formula (3) can be rewritten in matrix form: ； in , Represent Norm, The norm, in formula (4), the proposed CBCL aims to learn the projection matrix. Projection error By minimizing the overall objective function, two objectives are achieved: one is to minimize the error during the projection process, and the other is to minimize the distance between data of the same category and maximize the distance between data of different categories. Step 500 includes: Based on the mapping matrix, the image TLDV is transformed into Bit binary feature vector , This is the feature vector mapped from the i-th image sample; each bit has a weight of 2. m-1 Each code value is multiplied by a weight, and the results are summed to convert the binary feature into a real-valued feature. This real-valued feature is then converted to the same size as the original finger vein image and divided into multiple non-overlapping sizes. For each small block, extract its local histogram and combine these histograms into a single dimension. The feature vector of i is used as the final feature representation, where i represents the total number of blocks.

2. The method according to claim 1, characterized in that, Step 100 includes: First, the input dataset is randomly divided into a training set and a test set. There are two specific methods: the first is to randomly select two finger vein images from each category as the training set, and the remaining finger vein images from that category as the test set; the second is to randomly select three finger vein images from each category as the training set, and the remaining finger vein images from that category as the test set. Next, during the training phase, for the finger vein images in the training set, select... Each adjacent pixel is treated as a finger vein patch. The response result is obtained by convolving the adjacent pixel with the local binary pattern (LBP) template. Finally, a 24-dimensional texture difference vector is generated for each finger vein patch by calculating the difference between the response results of the center pixel and the neighboring pixels.

3. The method according to claim 1, characterized in that: Step 300 includes: Introducing Lagrange multipliers Equation (4) is transformed into the following augmented Lagrangian function: ； in For one OK A real matrix of columns; The penalty parameter is used; Formula (5) is a multivariate optimization problem of the Lagrangian function. The Lagrangian function is used to iteratively update one of the variables to find the variable when the Lagrangian function reaches its minimum value. and The optimal solution; then, before performing iterative updates, the variables need to be initialized, let Initialize variables using the randn() function in Matlab. Initialize variables using the rand() function. This generates a matrix of random numbers, where Each element in the formula is between 0 and 1; calculate the initial value of the overall objective function of the sub-formula (5) under the current variable; The optimization problem of the above CBCL model is solved using the Alternating Multiplier Method (ADMM), with a pre-set maximum number of iterations. Repeat the following iterative steps: Step 1) Fix ,according to Update variables ; Step 2) Fix ,according to Update variable P, where ,definition, ; for The Row vectors; Step 3) Fix ,according to Update variables ; Step 4) Fix ,according to , Update variables and ,in , , ; Step 5) Based on the updated variables, calculate the value of the overall objective function of formula (5), and calculate the difference between the current value of the overall objective function (5) and the value of the overall objective function in the previous iteration; Repeat the above process until the preset maximum number of iterations is reached. Or the difference between the current value of the overall objective function (5) and the value of the overall objective function in the previous iteration is less than 10. -4 Then the output variable .

4. The method according to claim 1, characterized in that: Step 400 includes: Using the mapping matrix and error matrix ,according to training set data Projection yields the binary feature matrix of the training set data. , Let be the feature vector after mapping the i-th image sample; process the test set data in the same way to obtain the binary feature matrix of the test set data. ,in This represents the dimension of the TLDV. , The size of the image is represented by... , Let be the feature vector after mapping the i-th image sample.

5. The method according to claim 1, characterized in that: Step 400 includes: Step 600 includes: First, calculate the Euclidean distance between the final feature representation of each image in the training set and the final feature representation of an image in the test set after mapping. Then, use the nearest neighbor classifier to classify the finger vein images in the test set and output the labels of the finger vein images in the test set. Then, the recognition accuracy is calculated by matching the labels of the test set images with the real labels of the data.

Images