User authentication method based on finger vein image, computer equipment and system
By improving the multi-noise recognition model and DNN network, the noise problem in finger vein image acquisition and recognition was solved, achieving high-accuracy user authentication and improving the standardization and compatibility of finger vein image acquisition devices.
Patent Information
- Application Number
- CN202511424930.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-30
- Publication Date
- 2026-02-17
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional finger vein image recognition technology suffers from limitations in data acquisition, image quality issues, and insufficient standards and compatibility, resulting in low authentication accuracy.
A multi-noise recognition model is used to identify image noise types and perform targeted noise reduction. By combining image block serialization and channel spatial attention mechanism, feature extraction and comparison are performed through DNN network to construct a complete multi-noise recognition model.
It effectively identifies and removes noise from finger vein images, improves authentication accuracy, enhances the ability to describe global feature information, and improves the stability and consistency of the authentication system.
Smart Images
Figure CN121543074A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of image processing, and in particular to a user authentication method based on a finger vein image, a computer device and a system. BACKGROUND
[0002] Traditional passwords (such as digital passwords, PIN codes) have the problems of being easy to forget and easy to steal, and authentication technologies based on biological characteristics (fingerprint, face, iris, etc.) gradually become mainstream due to their uniqueness and stability.
[0003] With the progress of optical imaging technology (such as near-infrared imaging) and deep learning, the high-quality collection of finger vein images and more efficient feature extraction are directly promoted, making them have practical value in access control systems, financial payments, medical identity authentication, etc. The user authentication using finger vein image recognition technology has the following defects:
[0004] Collection limitations: finger posture, angle and pressure size will all cause image distortion, and the collection quality varies between different collection devices;
[0005] Image quality issues: finger vein images are affected by uneven lighting, skin thickness and user skin color, etc., and may appear blurry, broken lines, etc. with low contrast;
[0006] Standards and compatibility: there is a lack of unified finger vein feature storage and exchange standards, and different manufacturer systems are difficult to interoperate. SUMMARY
[0007] The technical problem to be solved by the present application is to provide a user authentication method based on a finger vein image, a computer device and a system, which can improve the accuracy of authentication.
[0008] In order to solve the above technical problems, the present application provides a user authentication method based on a finger vein image, comprising: acquiring a finger vein image of a user; preprocessing the finger vein image; inputting the preprocessed finger vein image into a multi-noise recognition model trained in advance to identify the image noise type; performing noise reduction processing on the finger vein image according to the image noise type; performing feature extraction on the finger vein image after noise reduction processing to generate an image embedding feature vector; comparing the image embedding feature vector with a preset feature vector template; if the similarity between the image embedding feature vector and the feature vector template is greater than a preset similarity benchmark, the user is authorized to access; if the similarity between the image embedding feature vector and the feature vector template is less than or equal to the similarity benchmark, the user is denied access.
[0009] As an improvement to the above scheme, the training steps of the multi-noise recognition model include: acquiring finger vein image samples; binarizing the finger vein image samples; performing edge detection processing on the binarized finger vein image samples; performing masking processing on the edge-detected finger vein image samples; dividing the masked finger vein image samples into blocks to form an image block sequence; performing data augmentation processing on the image blocks; adding noise information to the data-augmented image blocks to generate low-quality images; and constructing a training set based on the low-quality images to train the multi-noise recognition model.
[0010] As an improvement to the above scheme, the multi-noise recognition model is a DNN network that incorporates channel spatial attention mechanism and spatial relationship mechanism.
[0011] As an improvement to the above scheme, the multi-noise recognition model includes a multi-path parallel network, a spatial relationship learning module, and a fully connected module. The output of the multi-path parallel network is connected to the input of the spatial relationship learning module, and the output of the spatial relationship learning module is connected to the input of the fully connected module.
[0012] As an improvement to the above scheme, the parallel network includes an input layer, a first convolutional layer, a first CSA module, a second convolutional layer, a second CSA module, a third convolutional layer, and a max pooling layer arranged sequentially.
[0013] As an improvement to the above scheme, both the first CSA module and the second CSA module include a channel attention unit and a spatial attention unit.
[0014] As an improvement to the above scheme, the spatial relationship learning module includes a ConvLSTM module.
[0015] As an improvement to the above scheme, the image noise types include Gaussian noise, salt-and-pepper noise, Poisson noise, fixed-pattern noise, and quantization noise.
[0016] Accordingly, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described user authentication method based on finger vein images.
[0017] Accordingly, the present invention also provides a user authentication system based on finger vein images, comprising: a finger positioning device for placing a user's finger; a near-infrared imaging device for acquiring the user's finger vein image; and the aforementioned computer equipment.
[0018] Implementing this invention has the following beneficial effects:
[0019] The user authentication method based on finger vein images of this invention can effectively identify the noise type in finger vein images and adopt different targeted measures for different image noise types to accurately reduce noise and improve the accuracy of authentication.
[0020] Furthermore, when training the multi-noise recognition model, this invention employs an image block serialization method to increase the image representation capabilities of different regions and enhance the ability to describe global feature information.
[0021] More preferably, this invention performs parallel expansion of the original DNN network using multiple networks to achieve feature representation of different image patches in the same image; simultaneously, this invention enhances the model's ability to represent finger vein patch images by adding a CSA module between adjacent convolutional layers of the original DNN network to perform weighted information from different channels; furthermore, this invention adds a spatial relationship learning module after the parallel network and before the fully connected module of the original DNN network to learn the dependencies between different image patches, thereby achieving the construction of a complete multi-noise recognition model. Attached Figure Description
[0022] Figure 1 This is a flowchart illustrating an embodiment of the user authentication method based on finger vein images of the present invention;
[0023] Figure 2 This is a flowchart of the training process for the multi-noise recognition model in this invention;
[0024] Figure 3 This is a schematic diagram of an embodiment of the multi-noise recognition model in this invention;
[0025] Figure 4 This is a schematic diagram of an embodiment of the channel attention unit in this invention;
[0026] Figure 5 This is a schematic diagram of an embodiment of the spatial attention unit in this invention;
[0027] Figure 6 This is a schematic diagram of an embodiment of the user authentication system based on finger vein images of the present invention. Detailed Implementation
[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. It is hereby declared that the directional terms such as up, down, left, right, front, back, inside, and outside used in this text are based solely on the accompanying drawings and are not intended to specifically limit the invention.
[0029] See Figure 1 , Figure 1 The flowchart illustrating an embodiment of the user authentication method based on finger vein images of the present invention is shown, which includes:
[0030] S101, acquire the user's finger vein image;
[0031] Specifically, finger vein images can be captured in the user's environment.
[0032] S102, Preprocess the finger vein image;
[0033] Specifically, the finger vein image can be binarized, edge detected, and masked.
[0034] S103, input the pre-processed finger vein image into the pre-trained multi-noise recognition model to identify the type of image noise;
[0035] Generally, image noise types include Gaussian noise, salt-and-pepper noise, Poisson noise, fixed-pattern noise, and quantization noise, but these are not limitations and can be set according to the actual situation.
[0036] S104, Denoise the finger vein image according to the type of image noise;
[0037] Because the acquisition process of finger vein images introduces various local noises—meaning a single finger vein image may contain multiple types of noise—leading to low image quality, methods for denoising finger vein images based on different types of image noise include:
[0038] (1) Gaussian noise: Gaussian filtering, Wiener filtering and bilateral filtering can be used to reduce noise;
[0039] (2) Salt and pepper noise: Noise reduction can be achieved by median filtering and bilateral median filtering;
[0040] (3) Poisson noise: Noise reduction can be achieved by nonlocal mean filtering and wavelet transform noise reduction;
[0041] (4) Fixed pattern noise: Noise reduction can be achieved by polynomial fitting correction and Kalman filtering;
[0042] (5) Quantization noise: Noise reduction can be achieved by using dithering algorithms and bilateral grid filtering.
[0043] Therefore, different targeted measures can be used to accurately reduce noise for different types of image noise.
[0044] S105, extract features from the denoised finger vein image to generate an image embedding feature vector;
[0045] In this embodiment, an autoencoder can be used to generate image embedding feature vectors.
[0046] S106, compare the image embedded feature vector with the preset feature vector template;
[0047] S107. If the similarity between the image embedded feature vector and the feature vector template is greater than the preset similarity benchmark, then the user is authorized to access the feature vector.
[0048] S108. If the similarity between the image embedding feature vector and the feature vector template is less than or equal to the similarity benchmark, then user access is denied.
[0049] In this embodiment, the similarity benchmark is 0.99.
[0050] Therefore, the user authentication method based on finger vein images of this invention can effectively identify the noise type in finger vein images and adopt different targeted measures for different image noise types to accurately reduce noise and improve the accuracy of authentication.
[0051] like Figure 2 As shown, the training steps for the multi-noise recognition model include:
[0052] S201, Obtain finger vein image samples;
[0053] Specifically, finger vein image samples of size 128×128 can be acquired in a standard noise-free environment.
[0054] S202, perform binarization processing on the finger vein image sample;
[0055] S203, perform edge detection processing on the binarized finger vein image samples;
[0056] S204, perform masking on the finger vein image sample after edge detection processing;
[0057] It should be noted that after steps S202 to S204, the alignment image can be output.
[0058] S205, The masked finger vein image samples are divided into blocks to form an image block sequence;
[0059] Specifically, the masked finger vein image samples can be divided into blocks of 16×16 to form multiple image block sequences.
[0060] S206, Perform data augmentation processing on the image block;
[0061] S207, add noise information to the image patch after data augmentation to generate a low-quality image;
[0062] Specifically, data augmentation techniques such as Mosaic can be used to augment image patches and add random noise information for annotation to form low-quality images.
[0063] S208, Construct a training set based on low-quality images to train a multi-noise recognition model.
[0064] Therefore, this invention employs an image block serialization method to increase the image representation capability of different regions and enhance the ability to describe global feature information.
[0065] Accordingly, in order to achieve accurate identification of noise types, this invention introduces a unique multi-noise recognition model. Specifically, the multi-noise recognition model in this invention is a DNN network that incorporates channel spatial attention mechanism and spatial relationship mechanism.
[0066] like Figure 3 As shown, the multi-noise recognition model includes a multi-parallel network 31, a spatial relationship learning module 32, and a fully connected module 33. The output of the multi-parallel network 31 is connected to the input of the spatial relationship learning module 32, and the output of the spatial relationship learning module 32 is connected to the input of the fully connected module 33.
[0067] The parallel network 31, the spatial relation learning module 32, and the fully connected module 33 are described in detail below:
[0068] I. Parallel Networks
[0069] This embodiment employs an 8-way parallel network 31, each of which includes, in sequence, an input layer, a first convolutional layer, a first CSA module, a second convolutional layer, a second CSA module, a third convolutional layer, and a max pooling layer. Preferably, the first and second convolutional layers are 5×5 convolutional layers, the first and second CSA modules are preferably 3×3 CSA modules, and the third convolutional layer is preferably a 3×3 convolutional layer.
[0070] Therefore, this invention performs parallel expansion of the original DNN network with 8 channels (128 / 16=8) to achieve feature representation of different image blocks of the same image; at the same time, this invention enhances the model's ability to represent the features of finger vein block images by adding CSA modules between adjacent convolutional layers of the original DNN network to perform weighted information from different channels.
[0071] Furthermore, both the first CSA module and the second CSA module include a channel attention unit and a spatial attention unit.
[0072] like Figure 4As shown, the channel attention unit includes a first global pooling layer, a multi-layer perceptron layer, a second global pooling layer, and a first activation layer arranged sequentially; wherein, the first global pooling layer and the second global pooling layer both include parallel max pooling layer and average pooling layer, and the first activation layer adopts the Sigmoid function.
[0073] like Figure 5 As shown, the spatial attention unit includes a third max pooling layer, a third average pooling layer, a fourth convolutional layer, and a second activation layer arranged sequentially. The second activation layer uses the Sigmoid function.
[0074] II. Spatial Relationship Learning Module
[0075] In this embodiment, the spatial relationship learning module is a ConvLSTM module, but this is not a limitation and can be selected according to the actual situation.
[0076] Therefore, this invention adds a spatial relationship learning module after the parallel network and before the fully connected module of the original DNN network to learn the dependencies between different image patches, thereby realizing the construction of a complete multi-noise recognition model.
[0077] III. Fully Connected Module
[0078] In this embodiment, the fully connected module includes a 3×9 fully connected layer and a 1×1 fully connected layer arranged sequentially.
[0079] In summary, the user authentication method based on finger vein images of this invention improves the network structure of the multi-noise recognition model to achieve targeted enhancement of the quality of the acquired images.
[0080] Accordingly, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described user authentication method based on finger vein images.
[0081] See Figure 6 , Figure 6 The specific structure of the user authentication system 100 based on finger vein images of the present invention is shown, which includes:
[0082] Finger positioning device 1 is used to place the user's finger; specifically, grooves and / or guide rails can be provided on the finger positioning device 1 to guide the user to correctly position their finger and avoid angular deviation.
[0083] Near-infrared imaging device 2 is used to acquire images of the user's finger veins; during operation, near-infrared imaging device 2 focuses on the finger area to ensure clear finger vein images.
[0084] Computer device 3 includes a memory and a processor. The memory stores a computer program, wherein the processor executes the computer program to implement the steps of the user authentication method based on finger vein images described above.
[0085] In summary, implementing this invention has the following beneficial effects:
[0086] This invention can effectively identify noise types in finger vein images and employ different targeted measures for different image noise types to accurately reduce noise, thereby improving the accuracy of authentication.
[0087] Furthermore, when training the multi-noise recognition model, this invention employs an image block serialization method to increase the image representation capabilities of different regions and enhance the ability to describe global feature information.
[0088] More preferably, this invention performs parallel expansion of the original DNN network using multiple networks to achieve feature representation of different image patches in the same image; simultaneously, this invention enhances the model's ability to represent finger vein patch images by adding a CSA module between adjacent convolutional layers of the original DNN network to perform weighted information from different channels; furthermore, this invention adds a spatial relationship learning module after the parallel network and before the fully connected module of the original DNN network to learn the dependencies between different image patches, thereby achieving the construction of a complete multi-noise recognition model.
[0089] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A user authentication method based on finger vein images, characterized in that, include: Obtain the user's finger vein image; The finger vein image is preprocessed; The preprocessed finger vein image is input into a pre-trained multi-noise recognition model to identify the type of image noise. The finger vein image is denoised according to the type of image noise. Feature extraction is performed on the denoised finger vein image to generate an image embedding feature vector; The image embedding feature vector is compared with a preset feature vector template; If the similarity between the image embedded feature vector and the feature vector template is greater than a preset similarity benchmark, then the user is authorized to access the service. If the similarity between the image embedding feature vector and the feature vector template is less than or equal to the similarity benchmark, then the user's access is denied.
2. The user authentication method based on finger vein images as described in claim 1, characterized in that, The training steps of the multi-noise recognition model include: Obtain finger vein image samples; The finger vein image sample is binarized; Edge detection processing is performed on the binarized finger vein image samples; The finger vein image sample after edge detection processing is masked. The masked finger vein image samples are divided into blocks to form an image block sequence; The image block is then augmented with data. Noise information is added to the image blocks after data augmentation to generate a low-quality image; A training set is constructed based on the low-quality images to train the multi-noise recognition model.
3. The user authentication method based on finger vein images as described in claim 1, characterized in that, The multi-noise recognition model is a DNN network that incorporates channel spatial attention mechanism and spatial relationship mechanism.
4. The user authentication method based on finger vein images as described in claim 3, characterized in that, The multi-noise recognition model includes a multi-parallel network, a spatial relationship learning module, and a fully connected module. The output of the multi-parallel network is connected to the input of the spatial relationship learning module, and the output of the spatial relationship learning module is connected to the input of the fully connected module.
5. The user authentication method based on finger vein images as described in claim 4, characterized in that, The parallel network comprises an input layer, a first convolutional layer, a first CSA module, a second convolutional layer, a second CSA module, a third convolutional layer, and a max pooling layer, arranged sequentially.
6. The user authentication method based on finger vein images as described in claim 5, characterized in that, Both the first CSA module and the second CSA module include a channel attention unit and a spatial attention unit.
7. The user authentication method based on finger vein images as described in claim 4, characterized in that, The spatial relationship learning module includes a ConvLSTM module.
8. The user authentication method based on finger vein images as described in claim 1, characterized in that, The image noise types include Gaussian noise, salt and pepper noise, Poisson noise, fixed-pattern noise, and quantization noise.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the user authentication method based on finger vein images as described in any one of claims 1 to 8.
10. A user authentication system based on finger vein images, characterized in that, include: Finger positioning device for placing a user's finger; Near-infrared imaging equipment is used to acquire images of the user's finger veins; The computer device as claimed in claim 9.