Intelligent access control system based on facial recognition

Abstract

The invention relates to the technical field of intelligent access control recognition, and particularly discloses an intelligent access control system based on facial recognition, which comprises the following steps: firstly, capturing a facial image of a user in real time by using a high-resolution camera in combination with a near-infrared imaging technology, and extracting microvascular distribution characteristics and skin texture characteristics around eyes from the facial image; eye capillary features are enhanced and coded through fast Fourier transform and a local binary pattern, and skin texture features are subjected to contrast, correlation, energy, entropy and other statistics through a gray-level co-occurrence matrix, and dimension reduction processing is carried out by applying principal component analysis. The user authenticity analysis module evaluates the authenticity of the user identity by calculating the eye difference feature value; and the user dynamic stability analysis module quantifies the skin texture change amplitude by constructing a covariance matrix, evaluates the dynamic stability of the face, synthesizes the feature vector, inputs the feature vector into a convolutional neural network model, and outputs an identity matching score to decide whether the user is allowed to pass the access control or not.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent access control recognition, and in particular to an intelligent access control system based on facial recognition. Background Art

[0002] As society's demand for security and convenience continues to increase, traditional access control systems such as password locks, magnetic cards, or fingerprint recognition are increasingly unable to meet the demands of modern, complex environments. In particular, in public spaces, corporate offices, and high-end residential areas, the demand for highly secure, low-error authentication methods is growing. Facial recognition, as a contactless, natural, and difficult-to-forge biometric recognition technology, is becoming a core technology for the next generation of access control systems. By analyzing a user's facial features, including details like the microvascular distribution around the eyes and skin texture, highly accurate authentication can be achieved. These unique biometric features are not only highly specific to each individual but also exhibit excellent stability and consistency across different environments. This enables facial recognition-based intelligent access control systems to ensure a high level of security while providing a convenient user experience.

[0003] The existing technology has the following deficiencies:

[0004] Existing facial recognition access control systems still have some significant shortcomings. First, many systems rely on only a single type of facial feature, which makes them vulnerable to lighting conditions, expression changes, and camouflage attacks, resulting in high false recognition and rejection rates. Second, existing systems often lack a mechanism to evaluate the dynamic stability of the user's face, that is, they fail to fully consider the impact of the passage of time or external factors (such as environmental changes and skin aging) on ​​the details of the skin surface texture, thereby affecting the long-term reliability and accuracy of the system. In addition, traditional methods are inefficient when processing large amounts of data and cannot achieve real-time identity authentication, which limits their application in high-traffic scenarios. Finally, existing technologies generally fail to fully utilize the advantages of deep learning and advanced statistical analysis methods, resulting in poor robustness and adaptability of the system when facing complex and changing practical application scenarios. Therefore, developing an intelligent access control system that can combine multiple facial features and has dynamic stability assessment capabilities has become the key to improving security and user experience. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent access control system based on facial recognition to solve the above-mentioned problems.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] Intelligent access control system based on facial recognition, including:

[0008] a data acquisition module, which monitors the user's face in real time through a high-resolution camera installed at the entrance when the user approaches the entrance, and extracts microvascular distribution characteristics and skin texture characteristics around the user's eyes;

[0009] A user authenticity analysis module, which evaluates the authenticity of the user's identity by obtaining the microvascular distribution characteristics around the user's eyes and based on the degree of difference in the microvascular distribution around the eyes;

[0010] A user dynamic stability analysis module, which evaluates the dynamic stability of the user's face by acquiring skin texture features around the user's eyes and based on the magnitude of changes in the user's skin surface texture details;

[0011] The user identity authentication module constructs an identity authentication model by comprehensively analyzing the microvascular distribution characteristics and skin texture characteristics of the user to verify the identity of the user.

[0012] As a further solution of the present invention: the process of extracting the microvessel distribution characteristics around the eye is as follows:

[0013] The user's facial image is captured in real time using a high-resolution camera combined with near-infrared imaging technology, and the eye area is accurately located and segmented using computer vision algorithms. The image of the eye area is converted to the frequency domain using fast Fourier transform technology, and a series of bandpass filters of different scales are applied in the frequency domain. The filtered frequency domain image is restored to an enhanced spatial domain image through inverse fast Fourier transform. A fixed-size neighborhood window is defined on the enhanced image, and the relative brightness of the grayscale values ​​around each pixel relative to the central pixel is calculated to generate a local binary pattern. For each central pixel, if the grayscale value of a point in its neighborhood is greater than or equal to the central pixel, it is marked as 1; otherwise, it is marked as 0, forming an 8-bit binary number, which is then converted to a decimal number. The frequency of occurrence of each value in the entire image is counted to construct a histogram. The histogram is used as a local binary pattern feature vector, encoding the texture information of the microvessels around the eye. The obtained local binary pattern feature vector is used as the distribution feature of the microvessels around the eye.

[0014] As a further solution of the present invention: the skin texture feature extraction process is: using a high-resolution camera to obtain the user's facial image, and taking the cheek area as the analysis object, after selecting the area, calculating the grayscale co-occurrence matrix of the corresponding area, based on the generated grayscale co-occurrence matrix, further calculating multiple statistics that can quantify the image texture properties, including contrast, correlation, energy and entropy, the statistics are constructed into the original multidimensional statistical feature vector, principal component analysis is applied to perform feature dimensionality reduction processing, all original multidimensional statistical feature vectors are standardized, the covariance matrix of these standardized feature vectors is calculated, the eigenvalues ​​and corresponding eigenvectors of the covariance matrix are solved, sorted according to the size of the eigenvalues, and the eigenvectors corresponding to the first m largest eigenvalues ​​are selected as principal components, and the original multidimensional statistical feature vectors are projected into a new coordinate system composed of the selected principal components to generate new low-dimensional skin texture features.

[0015] As a further solution of the present invention, the authenticity of the user's identity is evaluated based on the difference in the distribution of microvessels around the eyes, specifically including:

[0016] Obtain the microvessel distribution characteristics around the user's eyes, calculate the eye difference characteristic value based on the difference degree of the microvessel distribution around the user's eyes, and determine whether the eye difference characteristic value is greater than or equal to a preset threshold. If so, the user identity is not authentic; otherwise, the user identity is authentic.

[0017] As a further solution of the present invention: the process of obtaining the eye difference feature value is:

[0018] Obtaining microvascular distribution features around the user's eyes and microvascular distribution features around the eyes of historical users input into the access control system, and representing them as a first eigenvector and a second eigenvector respectively;

[0019] Normalizing the extracted first eigenvector and the second eigenvector respectively so that their lengths become unit length, specifically comprising: dividing each component in each eigenvector by the Euclidean norm of the eigenvector, and the normalized eigenvectors are respectively referred to as the first unit eigenvector and the second unit eigenvector;

[0020] Calculating the cosine similarity between the first unit eigenvector and the second unit eigenvector;

[0021] The cosine similarity is calculated by multiplying the corresponding components of two unit eigenvectors and summing them up, that is, the dot product operation. Since the unit eigenvectors have been normalized, there is no need to calculate the denominator.

[0022] Based on the calculated cosine similarity, the eye difference feature value is further calculated;

[0023] The eye difference feature value is calculated by subtracting the cosine similarity from the value 1.

[0024] As a further solution of the present invention, the dynamic stability of the user's face is evaluated based on the variation of the surface texture details of the user's skin, specifically including:

[0025] Obtain the skin texture features around the user's eyes, calculate the user's texture detail change characteristic value based on the change amplitude of the user's skin surface texture details, and determine whether the user's texture detail change characteristic value is greater than or equal to a preset threshold. If so, the user's face is dynamically unstable; otherwise, the user's face is dynamically stable.

[0026] As a further solution of the present invention: the process of obtaining the texture detail change feature value is:

[0027] The skin texture features around the user's eyes and the skin texture features around the eyes of historical users input into the access control system are obtained, and are constructed into a third eigenvector and a fourth eigenvector respectively. Based on the extracted third eigenvector and fourth eigenvector, the difference vector between the third eigenvector and the fourth eigenvector is calculated; based on the difference between the third eigenvector and the fourth eigenvector and the mean vector, an outer product matrix is ​​constructed, the average of all deviation outer product matrices is taken to obtain the covariance matrix, the inverse matrix of the covariance matrix is ​​calculated, the difference vector is multiplied by the inverse matrix of the covariance matrix and the square root calculation is performed to obtain the texture detail change eigenvalue.

[0028] As a further solution of the present invention: the comprehensive analysis of the user's microvascular distribution characteristics and skin texture characteristics specifically includes:

[0029] The user's eye difference feature value and texture detail change feature value are obtained, and the user's eye difference feature value and texture detail change feature value are constructed into a comprehensive feature vector as the input of the identity authentication model. The output of the identity authentication model is an identity matching score. The user's identity is verified based on the user's identity matching score. The identity authentication model is a convolutional neural network model.

[0030] As a further solution of the present invention: the construction process of the identity verification model is:

[0031] The comprehensive feature vector is used as the input of the convolutional neural network model. Based on the selected convolutional neural network model, a convolutional neural network model consisting of multiple convolutional layers, pooling layers, and fully connected layers is constructed. The convolutional layer is responsible for extracting high-level features from the input comprehensive feature vector. The pooling layer is used to reduce the size of the feature map to reduce the number of parameters and computational cost. The fully connected layer integrates the extracted feature information and outputs a score indicating the degree of identity matching. The model structure includes the number of layers, the number of neurons in each layer, and the activation function parameters used.

[0032] The training process of the authentication model is as follows:

[0033] Collect and prepare multiple sets of data sets labeled with user identity information. The corresponding data sets contain comprehensive feature vectors composed of the user's eye difference eigenvalues ​​and texture detail change eigenvalues. Use the corresponding data sets to train the constructed convolutional neural network model, minimize the error between the predicted results and the true labels through supervised learning, and monitor the performance of the model on unseen data through cross-validation technology to ensure that the model has excellent generalization ability and high accuracy. The trained model can accept new comprehensive feature vectors as input and output the corresponding identity matching score to determine the authenticity of the user's identity.

[0034] As a further solution of the present invention, the user's identity is verified according to the user identity matching score, specifically including:

[0035] Determine whether the user's identity matching score is greater than or equal to a preset threshold. If so, the user is successfully matched and is allowed to pass. If not, the user is unsuccessful and is not allowed to pass.

[0036] Beneficial effects of the present invention:

[0037] (1) By combining the microvascular distribution characteristics around the eyes and the skin texture characteristics for user identity authentication, the present invention provides a highly accurate and secure identity recognition method. By combining near-infrared imaging technology with fast Fourier transform, the microvascular network structure in the eye area can be effectively highlighted and extracted, and the uniqueness and stability of these features can be further enhanced by local binary patterns. At the same time, the gray-level co-occurrence matrix is ​​used to comprehensively analyze the skin texture features, and statistics such as contrast, correlation, energy and entropy are calculated to capture the complex texture information of the facial skin. In order to further improve the distinguishing ability of the feature vector and reduce redundant information, principal component analysis is applied to perform feature dimensionality reduction processing to ensure the effectiveness and efficiency of the extracted features. This method can not only accurately distinguish between real users and imposters, but also sensitively identify subtle changes caused by the passage of time or external factors, thereby significantly enhancing the robustness and anti-interference ability of the system. In addition, by combining the above-mentioned multiple advanced feature extraction technologies with deep learning models, the present invention constructs a multi-level and multi-dimensional identity authentication system, which greatly improves the overall security protection level and provides users with a safe and reliable intelligent access control solution. This comprehensive approach not only improves the accuracy of identity authentication, but also enhances the system's adaptability to dynamic changes, ensuring long-term stability and reliability.

[0038] (2) The present invention improves the flexibility and adaptability of the identity authentication system by introducing an evaluation mechanism for the dynamic stability of the user's face. By judging the impact of time or external factors (such as environmental changes, skin aging) on ​​the details of the skin surface texture, the system can ensure that it can accurately identify the user even when there are subtle changes in his or her face. Specifically, the present invention uses a method that combines difference vectors with covariance matrices to accurately quantify these changes and evaluates their specific impact on the identity authentication results by calculating the Mahalanobis distance. This advanced analysis method based on statistics can not only capture the direct differences between feature vectors, but also deeply explore the correlation and distribution characteristics within the features, thereby enhancing the robustness and accuracy of the system. In addition, through learning and model optimization of a large amount of historical data, the present invention can maintain high-precision identity recognition capabilities during long-term use and effectively respond to dynamic changes in users. This high adaptability to dynamic changes not only greatly improves the user experience and reduces the inconvenience caused by misidentification, but also significantly improves the reliability and security of the system, providing users with a stable and efficient identity authentication solution. Combining advanced image processing technology and deep learning models, the present invention constructs a comprehensive and sophisticated identity authentication framework to ensure that safe and reliable user identity confirmation can be achieved in various complex scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The present invention will be further described below with reference to the accompanying drawings.

[0040] Figure 1 It is a flow chart of the intelligent access control system based on facial recognition of the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0042] See also Figure 1 As shown, the present invention is an intelligent access control system based on facial recognition, comprising:

[0043] a data acquisition module, which monitors the user's face in real time through a high-resolution camera installed at the entrance when the user approaches the entrance, and extracts microvascular distribution characteristics and skin texture characteristics around the user's eyes;

[0044] A user authenticity analysis module, which evaluates the authenticity of the user's identity by obtaining the microvascular distribution characteristics around the user's eyes and based on the degree of difference in the microvascular distribution around the eyes;

[0045] A user dynamic stability analysis module, which evaluates the dynamic stability of the user's face by acquiring skin texture features around the user's eyes and based on the magnitude of changes in the user's skin surface texture details;

[0046] The user identity authentication module constructs an identity authentication model by comprehensively analyzing the microvascular distribution characteristics and skin texture characteristics of the user to verify the identity of the user.

[0047] In the data acquisition module, when a user approaches the entrance, a high-resolution camera installed at the entrance monitors the user's face in real time and extracts the microvascular distribution characteristics and skin texture characteristics around the user's eyes, including:

[0048] The process of obtaining the microvascular distribution characteristics is as follows:

[0049] To characterize the distribution of microvessels around the eyes from facial images of users, the present invention first uses a high-resolution camera combined with near-infrared imaging technology to capture facial images in real time. Computer vision algorithms are then used to precisely locate and segment the eye region. Fast Fourier transform (FFT) technology is then used to convert the eye region image into the frequency domain. A series of bandpass filters of varying scales are then applied to the frequency domain, optimizing for microvascular structures of varying thicknesses to highlight the subcutaneous microvascular network. The filtered frequency domain image is then restored to an enhanced spatial domain image using an inverse FFT. A fixed-size neighborhood window (e.g., 3x3 pixels) is defined on the enhanced image. The relative brightness of the grayscale values ​​of each pixel relative to the center pixel is calculated, generating a local binary pattern. For each center pixel, if the grayscale value of a point in its neighborhood is greater than or equal to the center pixel, it is marked as 1; otherwise, it is marked as 0, forming an 8-bit binary number that is then converted to a decimal number. The frequency of occurrence of each value in the entire image is counted to construct a histogram. This histogram, as a local binary pattern feature vector, encodes the texture information of the microvessels around the eye. The obtained local binary pattern feature vector serves as a unique identifier of the microvessel distribution around the eye and is used in the subsequent authentication process, completing the complete process of extracting stable and discriminative microvessel distribution features from the original image.

[0050] The process of acquiring the skin texture features is as follows:

[0051] To accurately extract skin texture features from a user's facial image, the present invention uses a high-resolution camera to capture the user's facial image and selects a specific region (such as the cheek) for analysis. After selecting the region, the grayscale co-occurrence matrix (GLCM) is calculated for that region. This method describes image texture characteristics by counting the co-occurrence frequencies of grayscale levels between different pixel pairs. Specifically, for each pixel and its neighboring pixels, the number of occurrences of their grayscale value combinations at a set direction (such as 0°, 45°, 90°, and 135°) and distance is recorded to form a GLCM.

[0052] Based on the generated gray-level co-occurrence matrix, several statistics are further calculated to quantify the image's texture properties, including contrast (a measure of local image variation), correlation (reflecting the degree of linear dependence between pixels), energy (indicating image uniformity and smoothness), and entropy (assessing image complexity and randomness). These statistics form the original multidimensional statistical feature vector. For example, for each combination of direction and distance parameters, a set of vectors containing the above four statistics is generated. Because different direction and distance parameters generate multiple such vectors, the final result is a multidimensional dataset that comprehensively describes the texture characteristics of the selected facial region.

[0053] Principal component analysis (PCA) is applied to perform feature dimensionality reduction. PCA is a technique used to identify the main patterns in data and reduce the dimensionality of the data. First, all the original multidimensional statistical eigenvectors are standardized so that their mean is zero and their variance is one, eliminating the influence of different scales. Then, the covariance matrix of these standardized eigenvectors is calculated, which shows the correlation between the various features. Next, the eigenvalues ​​and corresponding eigenvectors of the covariance matrix are solved. The eigenvalues ​​indicate the degree to which each principal component explains the variation in the data, while the eigenvectors define the direction of the new coordinate system. The eigenvalues ​​are sorted by size, and the eigenvectors corresponding to the first m largest eigenvalues ​​are selected as principal components. These principal components retain the main information of the original data, but the dimensionality is significantly reduced.

[0054] The original multidimensional statistical feature vector is projected onto a new coordinate system composed of the selected principal components to generate a new low-dimensional feature vector. This compact feature vector efficiently encodes the unique information of skin texture, reducing redundant information while enhancing the robustness and computational efficiency of the model. The skin texture features obtained in this way are highly comparable and stable, making them suitable for subsequent identity verification. This completes the extraction process from the original facial image to streamlined and discriminative skin texture features. The entire process progressively progresses from image acquisition, matrix construction, feature extraction, and dimensionality reduction, ensuring that the extracted skin texture features are both accurate and practical.

[0055] In the user authenticity analysis module, the authenticity of the user's identity is assessed by obtaining the microvascular distribution characteristics around the user's eyes and the degree of difference in the microvascular distribution around the eyes. Specifically, the following are performed:

[0056] Obtain the microvessel distribution characteristics around the user's eyes, calculate the eye difference characteristic value based on the difference degree of the microvessel distribution around the user's eyes, and determine whether the eye difference characteristic value is greater than or equal to a preset threshold. If so, the user identity is not authentic; otherwise, the user identity is authentic.

[0057] The process of obtaining the eye difference feature value is as follows:

[0058] Obtaining microvascular distribution features around the user's eyes and microvascular distribution features around the eyes of historical users input into the access control system, and representing them as a first eigenvector and a second eigenvector respectively;

[0059] Normalizing the extracted first eigenvector and the second eigenvector respectively so that their lengths become unit length, including: dividing each component in each eigenvector by the Euclidean norm of the eigenvector, and the normalized eigenvectors are respectively referred to as the first unit eigenvector and the second unit eigenvector;

[0060] Calculating the cosine similarity between the first unit eigenvector and the second unit eigenvector;

[0061] The cosine similarity is calculated by multiplying the corresponding components of two unit eigenvectors and summing them up, that is, the dot product operation. Since the unit eigenvectors have been normalized, there is no need to calculate the denominator.

[0062] Based on the calculated cosine similarity, the eye difference feature value is further calculated;

[0063] The eye difference feature value is calculated by subtracting the cosine similarity from the value 1.

[0064] It should be noted that the technical solution and innovation of this module lies in assessing the authenticity of a user's identity by comparing subtle differences in the distribution of microvessels around the eyes of the current user with historical records. Specifically, this module obtains and represents the user's real-time microvessel distribution features and the corresponding features previously stored in the access control system as first and second eigenvectors. To ensure the accuracy of the comparison, these two eigenvectors are normalized to unit-length vectors, namely the first and second unit eigenvectors. This process is achieved by dividing each eigenvector by its own Euclidean norm. Next, the cosine similarity between these two unit eigenvectors is calculated. This measure measures the directional similarity between vectors. This is obtained by simply performing a dot product operation on the corresponding components of the two vectors. Because the normalization is already done, there is no need to consider the additional denominator. Based on this cosine similarity, an eye difference feature value is further calculated. This value is obtained by subtracting the similarity from 1, reflecting the degree of difference between the two sets of microvessel distribution features. Ultimately, the authenticity of the user's identity is determined by determining whether the difference characteristic value exceeds a preset threshold. If the difference characteristic value is equal to or greater than the set threshold, it is considered a significant difference, indicating that the user's identity may be inauthentic; conversely, if the difference characteristic value is lower than the threshold, it indicates a high degree of consistency, verifying the user's identity. This sophisticated comparison method based on the distribution characteristics of ocular microvessels provides a more accurate and secure means of identity verification.

[0065] In the user dynamic stability analysis module, the dynamic stability of the user's face is evaluated by obtaining the skin texture features around the user's eyes and based on the magnitude of changes in the user's skin surface texture details. Specifically, the following are performed:

[0066] Obtain the skin texture features around the user's eyes, calculate the user's texture detail change characteristic value based on the change amplitude of the user's skin surface texture details, and determine whether the user's texture detail change characteristic value is greater than or equal to a preset threshold. If so, the user's face is dynamically unstable; otherwise, the user's face is dynamically stable.

[0067] The process of obtaining the texture detail change feature value is as follows:

[0068] The skin texture features around the user's eyes and the skin texture features around the eyes of historical users input into the access control system are obtained, and are constructed into a third eigenvector and a fourth eigenvector respectively. Based on the extracted third eigenvector and fourth eigenvector, the difference vector between the third eigenvector and the fourth eigenvector is calculated; based on the difference between the third eigenvector and the fourth eigenvector and the mean vector, an outer product matrix is ​​constructed, the average of all deviation outer product matrices is taken to obtain the covariance matrix, the inverse matrix of the covariance matrix is ​​calculated, the difference vector is multiplied by the inverse matrix of the covariance matrix and the square root calculation is performed to obtain the texture detail change eigenvalue.

[0069] It should be noted that in the user dynamic stability analysis module, we innovatively utilize skin texture features around the eyes to assess the dynamic stability of a user's face. This method provides an efficient and accurate identity verification method by quantifying the magnitude of changes in skin surface texture details. Specifically, we obtain skin texture features around the eyes from current and historical access control systems and construct them into third and fourth eigenvectors. Next, we calculate the difference vector between these two eigenvectors and construct an outer product matrix based on the deviations between these vectors and the mean vector. We then average all the deviation outer product matrices to obtain the covariance matrix, and then calculate its inverse. Finally, we multiply the difference vector by the inverse of the covariance matrix and take the square root to obtain the texture detail change eigenvalue. This method not only considers direct differences between features but also deeply analyzes the correlation and distribution characteristics within the features, resulting in a more accurate assessment of the user's facial dynamic stability. Technically, this process can effectively identify subtle facial changes caused by the passage of time or external factors, thereby improving the reliability and security of the identity verification system and providing users with a more robust authentication mechanism.

[0070] In the user identity authentication module, an identity authentication model is constructed by comprehensively analyzing the user's microvascular distribution characteristics and skin texture characteristics to verify the user's identity. Specifically, the model includes: obtaining the user's eye difference feature value and texture detail change feature value, and constructing the user's eye difference feature value and texture detail change feature value into a comprehensive feature vector as the input of the identity authentication model. The output of the identity authentication model is an identity matching score. The user's identity is verified based on the user's identity matching score. The identity authentication model is a convolutional neural network model.

[0071] The construction process of the authentication model is as follows:

[0072] The comprehensive feature vector is used as the input of the convolutional neural network model. Based on the selected convolutional neural network model, a convolutional neural network model consisting of multiple convolutional layers, pooling layers, and fully connected layers is constructed. The convolutional layers are responsible for extracting high-level features from the input comprehensive feature vector. The pooling layers are used to reduce the size of the feature map to reduce the number of parameters and computational cost. The fully connected layers integrate the extracted feature information and output a score representing the degree of identity matching. The model structure includes the number of layers, the number of neurons in each layer, and the activation function parameters used.

[0073] The training process of the authentication model is as follows:

[0074] Collect and prepare a large-scale dataset labeled with user identity information. This dataset contains a comprehensive feature vector composed of the user's eye difference eigenvalues ​​and texture detail change eigenvalues. Use this dataset to train the constructed convolutional neural network model. Minimize the error between the predicted results and the true labels through supervised learning. Monitor the model's performance on unseen data through cross-validation technology to ensure that the model has excellent generalization ability and high accuracy. The trained model can accept new comprehensive feature vectors as input and output the corresponding identity matching score to determine the authenticity of the user's identity.

[0075] Verifying the user's identity based on the user identity matching score specifically includes:

[0076] Determine whether the user's identity matching score is greater than or equal to a preset threshold. If so, the user is successfully matched and is allowed to pass. If not, the user is unsuccessful and is not allowed to pass.

[0077] The present invention works by using a data acquisition module that utilizes a high-resolution camera combined with near-infrared imaging technology to capture real-time facial images of the user. It then extracts microvascular distribution features and skin texture features around the eyes. For microvascular distribution features, a fast Fourier transform and bandpass filter are used to highlight the subcutaneous microvascular network, subsequently generating a local binary pattern feature vector. For skin texture features, statistics such as contrast, correlation, energy, and entropy are calculated using a gray-level co-occurrence matrix, and principal component analysis is used for feature dimensionality reduction to obtain a concise and discriminative feature vector. The user authenticity analysis module compares the current microvascular distribution features with those in historical records to calculate an eye difference eigenvalue to assess the authenticity of the user's identity. If this value exceeds a preset threshold, the user's identity is deemed inauthentic. The user dynamic stability analysis module constructs the third and fourth eigenvectors, calculates their difference vector, and solves for the texture detail variation eigenvalue based on the covariance matrix to assess the dynamic stability of the user's face. A larger value indicates significant facial change, indicating dynamic instability. The user identity verification module constructs a comprehensive feature vector from these two eigenvalues, which serves as input to a convolutional neural network model and outputs an identity matching score. Based on the comparison of this score with a preset threshold, the system decides whether to allow the user to pass through the gate. This multi-level, multi-dimensional feature analysis method not only improves the robustness and accuracy of the system, but also enhances its sensitivity to subtle changes, providing users with a more secure and reliable identity authentication mechanism.

[0078] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. Intelligent access control system based on facial recognition, characterized by: include: a data acquisition module, which monitors the user's face in real time through a high-resolution camera installed at the entrance when the user approaches the entrance, and extracts microvascular distribution characteristics and skin texture characteristics around the user's eyes; A user authenticity analysis module, which evaluates the authenticity of the user's identity by obtaining the microvascular distribution characteristics around the user's eyes and based on the degree of difference in the microvascular distribution around the eyes; A user dynamic stability analysis module, which evaluates the dynamic stability of the user's face by acquiring skin texture features around the user's eyes and based on the magnitude of changes in the user's skin surface texture details; The user identity authentication module constructs an identity authentication model by comprehensively analyzing the microvascular distribution characteristics and skin texture characteristics of the user to verify the identity of the user.

2. The intelligent access control system based on facial recognition according to claim 1 is characterized in that: The process of extracting the microvascular distribution characteristics around the eye is as follows: The user's facial image is captured in real time using a high-resolution camera combined with near-infrared imaging technology, and the eye area is accurately located and segmented using computer vision algorithms. The image of the eye area is converted to the frequency domain using fast Fourier transform technology, and a series of bandpass filters of different scales are applied in the frequency domain. The filtered frequency domain image is restored to an enhanced spatial domain image through inverse fast Fourier transform. A fixed-size neighborhood window is defined on the enhanced image, and the relative brightness of the grayscale values ​​around each pixel relative to the central pixel is calculated to generate a local binary pattern. For each central pixel, if the grayscale value of a point in its neighborhood is greater than or equal to the central pixel, it is marked as 1; otherwise, it is marked as 0, forming an 8-bit binary number, which is then converted to a decimal number. The frequency of occurrence of each value in the entire image is counted to construct a histogram. The histogram is used as a local binary pattern feature vector, encoding the texture information of the microvessels around the eye. The obtained local binary pattern feature vector is used as the distribution feature of the microvessels around the eye.

3. The intelligent access control system based on facial recognition according to claim 1 is characterized in that: The extraction process of the skin texture features is as follows: A high-resolution camera is used to obtain the user's facial image, and the cheek area is taken as the analysis object. After the area is selected, the grayscale co-occurrence matrix of the corresponding area is calculated. Based on the generated grayscale co-occurrence matrix, multiple statistics that can quantify the image texture properties are further calculated, including contrast, correlation, energy and entropy. The statistics are constructed into original multidimensional statistical eigenvectors, and principal component analysis is applied to perform feature dimensionality reduction. All original multidimensional statistical eigenvectors are standardized, and the covariance matrix of these standardized eigenvectors is calculated. The eigenvalues ​​and corresponding eigenvectors of the covariance matrix are solved, and the eigenvalues ​​are sorted according to their size. The eigenvectors corresponding to the first m largest eigenvalues ​​are selected as principal components. The original multidimensional statistical eigenvectors are projected into a new coordinate system composed of the selected principal components to generate new low-dimensional skin texture features.

4. The intelligent access control system based on facial recognition according to claim 1, characterized in that: The authenticity of the user's identity is assessed based on the difference in the distribution of microvessels around the eyes, specifically including: Obtain the microvessel distribution characteristics around the user's eyes, calculate the eye difference characteristic value based on the difference degree of the microvessel distribution around the user's eyes, and determine whether the eye difference characteristic value is greater than or equal to a preset threshold. If so, the user identity is not authentic; otherwise, the user identity is authentic.

5. The intelligent access control system based on facial recognition according to claim 4 is characterized in that: The process of obtaining the eye difference feature value is as follows: Obtaining microvascular distribution features around the user's eyes and microvascular distribution features around the eyes of historical users input into the access control system, and representing them as a first eigenvector and a second eigenvector respectively; Normalizing the extracted first eigenvector and the second eigenvector respectively so that their lengths become unit length, specifically comprising: dividing each component in each eigenvector by the Euclidean norm of the eigenvector, and the normalized eigenvectors are respectively referred to as the first unit eigenvector and the second unit eigenvector; Calculating the cosine similarity between the first unit eigenvector and the second unit eigenvector; The cosine similarity is calculated by multiplying the corresponding components of two unit eigenvectors and summing them up, that is, the dot product operation. Since the unit eigenvectors have been normalized, there is no need to calculate the denominator. Based on the calculated cosine similarity, the eye difference feature value is further calculated; The eye difference feature value is calculated by subtracting the cosine similarity from the value 1.

6. The intelligent access control system based on facial recognition according to claim 1, characterized in that: The dynamic stability of the user's face is evaluated based on the variation of the user's skin surface texture details, specifically including: Obtain the skin texture features around the user's eyes, calculate the user's texture detail change characteristic value based on the change amplitude of the user's skin surface texture details, and determine whether the user's texture detail change characteristic value is greater than or equal to a preset threshold. If so, the user's face is dynamically unstable; otherwise, the user's face is dynamically stable.

7. The intelligent access control system based on facial recognition according to claim 6, characterized in that: The process of obtaining the texture detail change feature value is as follows: The skin texture features around the user's eyes and the skin texture features around the eyes of historical users input into the access control system are obtained, and are constructed into a third eigenvector and a fourth eigenvector respectively. Based on the extracted third eigenvector and fourth eigenvector, the difference vector between the third eigenvector and the fourth eigenvector is calculated; based on the difference between the third eigenvector and the fourth eigenvector and the mean vector, an outer product matrix is ​​constructed, the average of all deviation outer product matrices is taken to obtain the covariance matrix, the inverse matrix of the covariance matrix is ​​calculated, the difference vector is multiplied by the inverse matrix of the covariance matrix and the square root calculation is performed to obtain the texture detail change eigenvalue.

8. The intelligent access control system based on facial recognition according to claim 1, characterized in that: The comprehensive analysis of the user's microvascular distribution characteristics and skin texture characteristics specifically includes: The user's eye difference feature value and texture detail change feature value are obtained, and the user's eye difference feature value and texture detail change feature value are constructed into a comprehensive feature vector as the input of the identity authentication model. The output of the identity authentication model is an identity matching score. The user's identity is verified based on the user's identity matching score. The identity authentication model is a convolutional neural network model.

9. The intelligent access control system based on facial recognition according to claim 8, characterized in that: The construction process of the authentication model is as follows: The comprehensive feature vector is used as the input of the convolutional neural network model. Based on the selected convolutional neural network model, a convolutional neural network model consisting of multiple convolutional layers, pooling layers, and fully connected layers is constructed. The convolutional layer is responsible for extracting high-level features from the input comprehensive feature vector. The pooling layer is used to reduce the size of the feature map to reduce the number of parameters and computational cost. The fully connected layer integrates the extracted feature information and outputs a score indicating the degree of identity matching. The model structure includes the number of layers, the number of neurons in each layer, and the activation function parameters used. The training process of the authentication model is as follows: Collect and prepare multiple sets of data sets labeled with user identity information. The corresponding data sets contain comprehensive feature vectors composed of the user's eye difference eigenvalues ​​and texture detail change eigenvalues. Use the corresponding data sets to train the constructed convolutional neural network model, minimize the error between the predicted results and the true labels through supervised learning, and monitor the performance of the model on unseen data through cross-validation technology to ensure that the model has excellent generalization ability and high accuracy. The trained model can accept new comprehensive feature vectors as input and output the corresponding identity matching score to determine the authenticity of the user's identity.

10. The intelligent access control system based on facial recognition according to claim 1, characterized in that: Verifying the user's identity based on the user identity matching score specifically includes: Determine whether the user's identity matching score is greater than or equal to a preset threshold. If so, the user is successfully matched and is allowed to pass. If not, the user is unsuccessful and is not allowed to pass.

Images