A Finger Vein Recognition Method Based on Generative Adversarial Network and Convolutional Neural Network
By using the combination of the dual-channel VGG model and the BP-AdaBoost network in the finger vein recognition technology, the problems of insufficient data sets and low image quality are solved, and the accuracy and application efficiency of finger vein recognition are improved.
Patent Information
- Application Number
- CN202210458700.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-24
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-04-24
AI Technical Summary
The existing venous recognition technology has problems such as overfitting the model, low image quality and information loss caused by multi-module composition, which affects the recognition accuracy.
Using a method based on generative adversarial network and convolutional neural network, feature fusion is performed through a dual-channel VGG model, combining transfer learning and global average pooling layer to reduce engineering complexity, use the BP-AdaBoost network to judge image quality, and use deep learning The model is trained end-to-end.
The effect and accuracy of finger vein recognition are improved, information loss is avoided, the learning ability and generalization ability of the model are enhanced, the insufficient data set and image quality problems are solved, and the efficiency of recognition applications is improved.
Smart Images

Figure CN114743278B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of image processing and biometric feature recognition, and in particular relates to a finger vein recognition method based on a generative adversarial network and a convolutional neural network. Background Art
[0002] Finger vein is an effective living biological information, which is different from biometric features such as fingerprints, palm prints, irises, and faces, which are easy to lose, forget, damage or forge. Finger vein image is the distribution map of veins inside the finger under infrared light. Vein recognition has the characteristics of living body recognition, internal features, non-contact collection, and high security level, which makes it the king in the field of recognition and can be widely used in smart locks, smart security, identity recognition, crime tracking and identification and other fields.
[0003] The research objects of vein recognition technology include finger veins, back of hand veins and palm veins. Compared with palm veins, finger vein hardware equipment is smaller and less expensive; compared with back of hand veins, finger veins will not be interfered by human hair. Therefore, finger veins have more advantages in low-cost recognition.
[0004] In finger vein recognition, the main links include image acquisition, image preprocessing, image amplification, feature extraction, feature matching, etc. Hemoglobin is used to absorb near-infrared light that penetrates the skin to capture vein images. The index finger, ring finger, and middle finger of the left and right hands are taken respectively, and one finger is collected several times, and the collection is carried out in three stages within one year. Insufficient image data sets will lead to overfitting of the model, so the image amplification of the present invention is divided into two steps. The first step is to expand the original vein data through simple rotation, translation, and cropping. The second step is to generate a larger number of finger vein images through conditional generative adversarial network training. It not only avoids the low quality of images generated by the generative network due to too few data sets, but also further expands the finger vein data set on the basis of the first step, making it easier for the model to meet our requirements. The purpose of image preprocessing is to determine the location of the vein area and enhance the vein image. The determined vein area is normalized by grayscale transformation. The finger vein image collected by grayscale transformation normalization is an 8-bit grayscale image, which should have 256 grayscale levels. However, due to the influence of factors such as illumination during acquisition, the grayscale of the image is concentrated in one or several grayscale segments. At this time, the grayscale stretching method can be used to expand the image to 256 grayscale levels. Grayscale stretching reduces the interference of illumination on the vein image. The quality of image preprocessing has a great impact on subsequent links, and the accuracy of the model is often limited to the quality of the image. Common feature extraction methods for finger veins include: repeated line tracking, wide line detector, Gabor filter, maximum curvature point, average curvature, local binary pattern, etc. These feature extraction methods have certain effects, but they will also lead to the loss of some features to varying degrees. The traditional finger vein recognition process is often composed of multiple modules, and the quality of each module will affect the results of the entire training. Summary of the invention
[0005] In order to overcome the shortcomings of the prior art, the present invention provides a finger vein recognition method based on a generative adversarial network and a convolutional neural network. Different from the traditional vein recognition method, the present invention applies the "end-to-end" advantage of the deep learning model in the training process, reduces the complexity of the project, and avoids problems such as information loss caused by multiple modules. The present invention uses an improved dual-channel VGG model to increase feature complementarity and improve feature extraction capabilities through model feature fusion. The dual channels are divided into VGG16 and SimpleVGG. Compared with the conventional dual-channel VGG model, it can effectively save computing resources. The model also integrates the advantages of transfer learning and global average pooling layers, so that the model has stronger learning and generalization capabilities.
[0006] A finger vein recognition method based on a generative adversarial network and a convolutional neural network comprises the following steps:
[0007] S1: Collect and establish a finger vein image database for identification objects, and pre-process all finger vein images in the database;
[0008] S2: amplifying the image samples of the finger vein region of interest of the acquired finger vein database;
[0009] S3: Construct a deep convolutional generative adversarial network, use the amplified finger vein image to train the deep convolutional generative adversarial network, and the generator generates the target finger vein image;
[0010] S4: Construct BP-AdaBoost network and judge the quality of finger vein image by combining multiple parameters including contrast, information entropy, clarity and equivalent visual number;
[0011] S5: The finger vein dataset that meets the quality requirements is divided into a training set and a test set;
[0012] S6: construct an improved dual-channel VGG network for feature extraction, and use the training set to train the improved dual-channel VGG network to obtain a finger vein image classification network;
[0013] S7: After the model parameters are iteratively optimized, the finger vein image in the test set is input for testing, and the finger vein image classification results are output.
[0014] Preferably, S1 comprises the following sub-steps:
[0015] S11: Use a near infrared sensor to obtain a distribution map of finger veins and store a sample;
[0016] S12: establishing a finger vein database through the collected finger vein images;
[0017] S13: performing edge detection using a sobel operator on the finger vein image in the database, and obtaining a finger vein foreground area that does not contain background information through image morphological processing including closing operation and denoising;
[0018] S14: normalization of grayscale transformation of finger vein image.
[0019] Preferably, S2 comprises the following sub-steps:
[0020] S21: obtaining a preprocessed finger vein region of interest database;
[0021] S22: Enlarging the finger vein image samples in the finger vein region of interest database by methods including translation, rotation, gamma transformation, and affine transformation.
[0022] Preferably, S3 includes the following sub-steps:
[0023] S31: Building a deep convolutional generative adversarial network;
[0024] S32: inputting the amplified finger vein image for training;
[0025] S33: The generator generates a finger vein image;
[0026] S34: The discriminator determines whether the finger vein image generated in the above step is a real finger vein image;
[0027] S35: The generator and discriminator are trained alternately in a cycle, so that the data generated by the final generator gradually approaches the real data.
[0028] Preferably, S4 comprises the following sub-steps:
[0029] S41: inputting a finger vein image;
[0030] S42: Select parameter index;
[0031] S43: Obtaining an evaluation index output value;
[0032] S44: Construct sample data set;
[0033] S45: Construct BP-AdaBoost strong classifier;
[0034] S46: Output the classification result of the finger vein image.
[0035] Preferably, S5 includes the following sub-steps:
[0036] S51: establishing a finger vein database that meets the requirements according to the classification results;
[0037] S52: Divided into training set and test set in a ratio of 8:2.
[0038] Preferably, S6 comprises the following sub-steps:
[0039] S61: Performing transfer learning on the finger vein image classification network using Imagenet initialization weights;
[0040] S62: Finger vein feature extraction based on improved dual-channel VGG;
[0041] S63: Select loss function;
[0042] S64: Select an optimizer;
[0043] S65: Perform network training based on the loss function and optimizer.
[0044] Preferably, the improved dual-channel VGG network is constructed by the following method: based on the framework of the VGG convolutional neural network, VGG16 and SimpleVGG are respectively built to form a dual-channel network structure, and the fully connected layer of the original network is deleted and replaced with a custom embedding layer; the front feature extraction layer includes a convolution layer and a maximum pooling layer, the VGG16 network contains 13 convolution layers, each of which includes a convolution and a ReLU activation function, and the network contains 5 maximum pooling layers, wherein the horizontal and vertical steps of the pooling are both 2; the SimpleVGG network contains 6 convolution layers, each of which includes a convolution, a ReLU activation function and batch normalization, and the network contains 3 maximum pooling layers, wherein the horizontal and vertical steps of the pooling are both 2; the features output by the two networks are fused and input into the custom embedding layer; the custom embedding layer includes a global average pooling layer, a LeakyReLU activation function, a batch normalization, a fully connected layer and a discard layer; the dimension of the classification layer is the number of classifications considered during the training process.
[0045] Preferably, S7 includes the following sub-steps:
[0046] S71: calculating the Euclidean distance between the obtained feature vector and the feature vector in the database one by one;
[0047] S72: Select the feature vector category that makes the Euclidean distance the minimum as the matching result. Preferably, the grayscale transformation normalization formula is as follows:
[0048]
[0049] Where N(i,j) represents the grayscale value of the transformed image, I(i,j) represents the grayscale value of the original image, min and max represent the minimum and maximum grayscale values of the original image, respectively.
[0050] The beneficial effects of the present invention are:
[0051] 1. The present invention effectively improves the effect of finger vein recognition, effectively guarantees the applications of intelligent lock control, security, identity identification, etc. based on finger vein information recognition; applies finger vein recognition to the field of criminal investigation, and uses the characteristics of finger vein liveness recognition and internal features to assist the police in better identifying the identity of criminals, which can improve the efficiency of police solving cases to a certain extent;
[0052] 2. After the image is amplified and generated, the present invention adds a BP-AdaBoost network to judge the quality of the finger vein image, solves the problem of uneven quality of images generated by the deep convolutional generative adversarial network, and avoids the influence of image quality problems on the subsequent model recognition accuracy;
[0053] 3. The present invention applies the "end-to-end" advantage of the deep learning model in the training process, reduces the complexity of the project, and avoids problems such as information loss caused by multiple modules;
[0054] 4. The present invention uses the selu activation function instead of the relu activation function in the generative network, which provides richer image features and solves the problem of low network model accuracy caused by insufficient data sets.
[0055] 5. Use the improved dual-channel VGG model to increase feature complementarity and improve feature extraction capabilities through model feature fusion. The dual channels are divided into VGG16 and SimpleVGG. Compared with the conventional dual-channel VGG model, it can effectively save computing resources. The model also integrates the advantages of transfer learning and global average pooling layers, making the model have stronger learning and generalization capabilities. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 is a flow chart of the present invention;
[0057] Figure 2 is the collected original finger vein image;
[0058] Figure 3 It is a flow chart of the invention for extracting the region of interest from the original finger vein image through edge detection;
[0059] Figure 4 It is the flow chart of BP-AdaBoost network algorithm of the present invention;
[0060] Figure 5 is a schematic diagram of the improved dual-channel VGG structure of the present invention;
[0061] Figure 6 This is the improved dual-channel VGG network training result diagram of the present invention. DETAILED DESCRIPTION
[0062] The present invention is further described below in conjunction with specific embodiments, but the scope of the present invention is not limited thereto.
[0063] Reference Figure 1 , a finger vein recognition method based on generative adversarial network and convolutional neural network, comprising the following steps:
[0064] S1: Collect and establish a finger vein image database of the identified object, and pre-process all the finger vein images in the database;
[0065] S1 includes the following sub-steps:
[0066] S11: Use a near infrared sensor to obtain a distribution map of finger veins and store a sample;
[0067] S12: establishing a finger vein image database using the collected finger vein images;
[0068] S13: performing edge detection using a sobel operator on the finger vein image in the finger vein image database, and obtaining a finger vein foreground area that does not contain background information through image morphological processing including closing operation and denoising;
[0069] S14: normalization of grayscale transformation of finger vein image.
[0070] All images in the image database are preprocessed, including ROI region of interest extraction and image grayscale transformation normalization. The collected finger vein image is an 8-bit grayscale image, which should have 256 grayscale levels. However, due to the influence of factors such as illumination during acquisition, the grayscale of the image is concentrated in one or several grayscale segments. In this case, the grayscale stretching method can be used to expand the image to 256 grayscale levels. Grayscale stretching reduces the interference of illumination on the vein image. Obtaining the finger vein region of interest through edge detection allows the convolutional neural network to learn features more easily. The collected original finger vein image is shown in Figure 1. Figure 2 As shown in the figure, the process of extracting the region of interest from the original finger vein image through edge detection is as follows Figure 3 As shown in the figure, the original finger vein image is preprocessed to remove the background, and only the region of interest of the finger vein is retained. For finger region extraction, the method of combining the Sobel operator edge detection and the closing operation and denoising in mathematical morphology processing is adopted. The Sobel operator is a relatively common edge detection method, also known as the weighted average difference method. It is a first-order discrete difference operator used to calculate the gray value approximation of the image brightness function. Because the collected finger vein image is vertical, only the Sobel operator needs to be used for vertical edge detection to obtain the finger contour. In terms of finger vein contour extraction, the effect of the Sobel operator vertical detection is significantly better than other operators such as Canny. The closing operation is to first dilate the image and then erode it, which helps to close the small holes or small black dots on the foreground object, and can achieve the effect of "thinning" the contour.
[0071] The formula for grayscale transformation normalization is as follows:
[0072]
[0073] Where N(i,j) represents the grayscale value of the transformed image, I(i,j) represents the grayscale value of the original image, min and max represent the minimum and maximum grayscale values of the original image, respectively.
[0074] The sobel convolution factors are as follows:
[0075] +1 +2 +1 0 0 0 -1 -2 -1
[0076] The closing operation formula is as follows:
[0077] A·S=(A⊕S)ΘS
[0078] S2: amplifying the image samples of the finger vein region of interest of the acquired finger vein database by translation, rotation, etc.;
[0079] S2 includes the following sub-steps:
[0080] S21: obtaining a preprocessed finger vein region of interest database;
[0081] S22: Enlarging the finger vein image samples in the finger vein region of interest database by methods such as translation, rotation, gamma transformation, and affine transformation.
[0082] S3: Construct a deep convolutional generative adversarial network, use the amplified finger vein image to train the deep convolutional generative adversarial network, and the generator generates the target finger vein image;
[0083] S3 includes the following sub-steps:
[0084] S31: Building a deep convolutional generative adversarial network;
[0085] S32: inputting the amplified finger vein image for training;
[0086] S33: The generator generates a finger vein image;
[0087] S34: The discriminator determines whether the finger vein image generated in the above step is a real finger vein image;
[0088] S35: The generator and discriminator are trained alternately in a cycle, so that the data generated by the final generator gradually approaches the real data.
[0089] The deep convolutional generative adversarial network is trained to generate the required finger vein images, where the activation function in the deep convolutional generative adversarial network is changed from relu to selu, retaining the calculation of input less than 0 and providing richer features.
[0090] S4: Construct a BP-AdaBoost network and judge the quality of the finger vein image by combining multiple parameters including contrast, information entropy, clarity, and equivalent visual number;
[0091] S4 includes the following sub-steps:
[0092] S41: inputting a finger vein image;
[0093] S42: Select parameter index;
[0094] S43: Obtaining an evaluation index output value;
[0095] S44: Construct sample data set;
[0096] S45: Construct BP-AdaBoost strong classifier;
[0097] S46: Output the classification result of the finger vein image.
[0098] S5: The finger vein dataset that meets the quality requirements is divided into a training set and a test set;
[0099] S5 includes the following sub-steps:
[0100] S51: establishing a finger vein database that meets the requirements according to the classification results;
[0101] S52: Divided into training set and test set in a ratio of 8:2.
[0102] For the finger vein images after preprocessing and data enhancement, the output value of the evaluation index is obtained by combining multiple parameters such as contrast, information entropy, and clarity equivalent visual number, a sample data set is constructed, a BP-AdaBoost strong classifier is constructed, and the classification result of the vein image is output. The excellent finger vein samples that meet the standards are retained, an image database for identification is established, and the images in the image database are used to form an experimental data set, which includes a training set and a test set.
[0103] The finger vein image processing process based on BP-AdaBoost network is as follows Figure 4 As shown in the figure, the BP-AdaBoost network uses multiple BP neural networks as weak classifiers. Each BP neural network completes the prediction of the output sample after repeated training, and then generates a strong classifier through the AdaBoost algorithm to obtain the final classification result.
[0104] S6: construct an improved dual-channel VGG network for feature extraction, and use the training set to train the improved dual-channel VGG network to obtain a finger vein image classification network;
[0105] S6 includes the following sub-steps:
[0106] S61: Performing transfer learning on the finger vein image classification network using Imagenet initialization weights;
[0107] S62: Finger vein feature extraction based on improved dual-channel VGG;
[0108] S63: Select loss function;
[0109] S64: Select an optimizer;
[0110] S65: Perform network training according to the loss function.
[0111] Establish an improved dual-channel VGG finger vein image feature extraction learning model. The improved dual-channel VGG structure is as follows Figure 5 As shown in the figure, the specific method is: based on the framework of VGG convolutional neural network, VGG16 and SimpleVGG are built to form a dual-channel network structure, and the fully connected layer of the original network is deleted and replaced with a custom embedding layer. The front feature extraction layer includes a convolution layer and a maximum pooling layer. The VGG16 network contains 13 convolution layers (conv), each of which includes a convolution (Convolution) and a ReLU activation function. The network contains 5 maximum pooling layers, and the horizontal and vertical steps of the pooling are both 2. The SimpleVGG network contains 6 convolution layers, each of which includes a convolution (Convolution), a ReLU activation function and batch normalization (Batch Normalization). The network contains 3 maximum pooling layers, and the horizontal and vertical steps of the pooling are both 2. The features output by the two networks are fused and input into the custom embedding layer. The custom embedding layer includes the Global Average Pooling layer, LeakyReLU activation function, Batch Normalization, Fully Connected Layer, and Dropout layer. The dimension of the classification layer is the number of categories considered during the training process. In the custom embedding layer, the LeakyReLU activation function is selected to replace the original ReLU activation function, which retains some negative axis values so that the information of the negative axis will not be completely lost, solving the shortcoming that the gradient of ReLU is 0 in the negative interval. The use of the global average pooling layer can accept images of any size, better match the category with the feature map of the last convolutional layer, and reduce the number of parameters, which can prevent overfitting in this layer and integrate global spatial information.
[0112] The model optimizer uses the Adam adaptive moment estimation method to quickly optimize parameters. Thanks to its advantage of automatically adjusting the learning rate, the network can converge quickly. The cross entropy method is used to calculate the loss value, and softmax is used as the final classification layer output.
[0113] The improved dual-channel VGG model is used to increase feature complementarity and improve feature extraction capabilities through model feature fusion. The dual channels are divided into VGG16 and SimpleVGG. Compared with the conventional dual-channel VGG model, it can effectively save computing resources. The model also integrates the advantages of transfer learning and global average pooling layers, making the model have stronger learning and generalization capabilities.
[0114] S7: After the model parameters are iteratively optimized, the finger vein image in the test set is input for testing, and the finger vein image classification result is output;
[0115] S7 includes the following sub-steps:
[0116] S71: calculating the Euclidean distance between the obtained feature vector and the feature vector in the database one by one;
[0117] S72: Select the feature vector category that minimizes the Euclidean distance as the matching result.
[0118] The established finger vein image feature extraction learning model is tested using the test set in the database. The softmax function is used for classification according to the Euclidean distance. The feature vector category with the smallest Euclidean distance is selected as the matching result, and the recognition result is output.
[0119] The present invention includes the following steps: using an infrared acquisition device to collect finger vein images of the identification object, and performing edge detection on the collected finger vein images; obtaining the finger vein foreground area that does not contain background information through image morphological processing such as closing operation and denoising; grayscale transformation normalization; randomly using rotation, translation and other operations to amplify the image data of the acquired finger vein data set; building a deep convolution generative adversarial network model, inputting the enhanced image data set to generate a more sufficient number of finger vein images; constructing a BP-AdaBoost network to judge the quality of the finger vein image by combining multiple parameters such as contrast, information entropy, clarity, and equivalent visual number; dividing the finger vein data set that meets the quality into training set and test set in a certain ratio; building an improved dual-channel VGG network for training to obtain a finger vein classification network; and the test set outputs the classification results. The accuracy and loss of model training are shown in Figure 2. Figure 6 As shown, the output test accuracy on the test set reaches 99.16%. The experimental results show that the present invention can accelerate network convergence, save calculation cost, and effectively improve recognition accuracy.
[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A finger vein recognition method based on generative adversarial network and convolutional neural network, characterized in that The steps include: S1: Collect and establish a finger vein image database for identification objects, and pre-process all finger vein images in the database; S2: amplifying the image samples of the finger vein region of interest of the acquired finger vein database; S3: Construct a deep convolutional generative adversarial network, use the amplified finger vein image to train the deep convolutional generative adversarial network, and the generator generates the target finger vein image; S4: Construct BP-AdaBoost network and judge the quality of finger vein image by combining multiple parameters including contrast, information entropy, clarity and equivalent visual number; S5: The finger vein dataset that meets the quality requirements is divided into a training set and a test set; S6: construct an improved dual-channel VGG network for feature extraction, and use the training set to train the improved dual-channel VGG network to obtain a finger vein image classification network; The improved dual-channel VGG network is constructed by the following method: based on the framework of VGG convolutional neural network, VGG16 and SimpleVGG are built to form a dual-channel network structure, and the fully connected layer of the original network is deleted and replaced by a custom embedding layer; the front feature extraction layer includes a convolution layer and a maximum pooling layer. The VGG16 network contains 13 convolution layers, each of which includes convolution and ReLU activation functions. The network contains 5 maximum pooling layers, and the horizontal and vertical steps of the pooling are both 2; the SimpleVGG network contains 6 convolution layers, each of which includes convolution, ReLU activation function and batch normalization. The network contains 3 maximum pooling layers, and the horizontal and vertical steps of the pooling are both 2; the features output by the two networks are fused and input into the custom embedding layer; the custom embedding layer includes a global average pooling layer, a LeakyReLU activation function, a batch normalization, a fully connected layer and a discard layer; the dimension of the classification layer is the number of classifications considered during the training process; S7: After the model parameters are iteratively optimized, the finger vein image in the test set is input for testing, and the finger vein image classification results are output.
2. According to claim 1, the finger vein recognition method based on generative adversarial network and convolutional neural network is characterized in that S1 includes the following sub-steps: S11: Use a near infrared sensor to obtain a distribution map of finger veins and store a sample; S12: establishing a finger vein image database using the collected finger vein images; S13: performing edge detection using a sobel operator on the finger vein image in the finger vein image database, and obtaining a finger vein foreground area that does not contain background information through image morphological processing including closing operation and denoising; S14: normalization of grayscale transformation of finger vein image.
3. According to claim 1, the finger vein recognition method based on generative adversarial network and convolutional neural network is characterized in that S2 includes the following sub-steps: S21: obtaining a preprocessed finger vein region of interest database; S22: Enlarging the finger vein image samples in the finger vein region of interest database by methods including translation, rotation, gamma transformation, and affine transformation.
4. The finger vein recognition method based on generative adversarial network and convolutional neural network according to claim 1 is characterized in that S3 includes the following sub-steps: S31: Building a deep convolutional generative adversarial network; S32: inputting the amplified finger vein image for training; S33: The generator generates a finger vein image; S34: The discriminator determines whether the finger vein image generated in the above step is a real finger vein image; S35: The generator and discriminator are trained alternately in a cycle, so that the data generated by the final generator gradually approaches the real data.
5. The finger vein recognition method based on generative adversarial network and convolutional neural network according to claim 1 is characterized in that S4 includes the following sub-steps: S41: inputting a finger vein image; S42: Select parameter index; S43: Obtaining an evaluation index output value; S44: Construct sample data set; S45: Construct BP-AdaBoost strong classifier; S46: Output the classification result of the finger vein image.
6. The finger vein recognition method based on generative adversarial network and convolutional neural network according to claim 1 is characterized in that S5 includes the following sub-steps: S51: establishing a finger vein database that meets the requirements according to the classification results; S52: Divided into training set and test set in a ratio of 8:
2.
7. The finger vein recognition method based on generative adversarial network and convolutional neural network according to claim 1 is characterized in that S6 includes the following sub-steps: S61: Performing transfer learning on the finger vein image classification network using Imagenet initialization weights; S62: Finger vein feature extraction based on improved dual-channel VGG; S63: Select loss function; S64: Select an optimizer; S65: Perform network training based on the loss function and optimizer.
8. The finger vein recognition method based on generative adversarial network and convolutional neural network according to claim 1 is characterized in that S7 includes the following sub-steps: S71: calculating the Euclidean distance between the obtained feature vector and the feature vector in the database one by one; S72: Select the feature vector category that minimizes the Euclidean distance as the matching result.
9. The finger vein recognition method based on generative adversarial network and convolutional neural network according to claim 2, characterized in that: The grayscale transformation normalization formula is as follows: ; In the formula, N(i,j) Represents the grayscale value of the transformed image, I(i,j) Represents the grayscale value of the original image, min and max They represent the minimum grayscale value and the maximum grayscale value of the original image respectively.
Citation Information
Patent Citations
Radar target detection method based on dual-channel convolutional neural network false alarm controllability
CN110363151A
Convolutional neural network finger vein recognition method based on Densenet
CN113435249A
Training and using a convolutional neural network for person re-identification
US20200226421A1
