Safety management system based on face capture

Abstract

The invention relates to the technical field of face recognition, in particular to a safety management system based on face capture, which comprises a multispectral acquisition module, a biological feature extraction module and a space-time modeling module, comprising face image data of a visible light wave band and a near-infrared wave band; the biological feature extraction module performs subcutaneous biological tissue chromatographic analysis based on spectral absorption difference according to the facial image data of the near-infrared band; according to the face image data of the visible light wave band and the near-infrared wave band, surface material reflection characteristic identification is carried out; according to the face image data of the visible light wave band, dynamic micro-expression relevance feature extraction is carried out; the space-time modeling module receives a processing result of the biological feature extraction module and constructs a feature model including a short-term dynamic mode and a long-term evolution trend.

Description

Technical Field

[0001] The present invention relates to the technical field of face recognition, and particularly to a security management system based on face capture. Background Art

[0002] Traditional face recognition systems generally adopt single-spectrum imaging technology, relying on the visible light band to obtain two-dimensional epidermal texture features, lacking the multi-layer perception ability of subcutaneous biological tissues (such as vascular networks), resulting in difficulty in effectively identifying bionic material forgery attacks. Existing solutions are mostly based on static image analysis. Although geometric feature point matching or shallow neural networks are used to extract expression features, a spatio-temporal association model between dynamic micro-expressions and physiological states has not been established, and the recognition robustness significantly decreases in scenarios of expression disguise or long-term facial structure changes. The anti-counterfeiting mechanism is usually limited to the analysis of surface reflection characteristics, unable to fuse multi-spectral data to construct a hierarchical verification system, and lacking the discrimination dimension for the near-infrared absorption characteristics of high-fidelity synthetic materials.

[0003] In addition, traditional classifiers adopt a fixed parameter system and cannot adapt online to the emergence of new forgery materials or the natural evolution of individual physiological characteristics. The parameter solidification leads to a rapid decline in the system's defense ability with technological iteration. At the system architecture level, the feature extraction and decision-making models are mostly independent modules, lacking a multi-modal data fusion and spatio-temporal joint modeling mechanism, and it is difficult to coordinate the contradiction between the transient changes of epidermal features and the long-term evolution of subcutaneous structures, restricting the continuous protection ability in high-security scenarios. Summary of the Invention

[0004] The purpose of the present invention is to provide a security management system based on face capture to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A security management system based on face capture, including a multi-spectral acquisition module, a biometric feature extraction module, and a spatio-temporal modeling module, wherein:

[0006] The multi-spectral acquisition module is configured to capture facial image data in different spectral bands, including facial image data in the visible light band and the near-infrared band;

[0007] The biometric feature extraction module is connected to the multi-spectral acquisition module and performs the following processing:

[0008] Perform tomographic analysis of subcutaneous biological tissues based on spectral absorption differences according to the facial image data in the near-infrared band;

[0009] Identify the surface material reflection characteristics according to the facial image data in the visible light band and the near-infrared band;

[0010] Extract dynamic micro-expression correlation features according to the facial image data in the visible light band;

[0011] The spatio-temporal modeling module receives the processing results of the biometric feature extraction module, constructs a feature model including short-term dynamic patterns and long-term evolution trends, and realizes the biometric recognition function with strong anti-counterfeiting ability and dynamic adaptability through the collaboration of spectral fusion, hierarchical feature extraction, and spatio-temporal modeling.

[0012] As a further improvement of this technical solution, the multispectral acquisition module includes a visible light sensor and a near-infrared sensor with complementary spectral response characteristics.

[0013] As a further improvement of this technical solution, the biometric feature extraction module includes a subcutaneous biological tissue chromatography analysis unit. The process of the subcutaneous biological tissue chromatography analysis unit performing subcutaneous biological tissue chromatography analysis specifically includes:

[0014] Perform weighted average fusion on the multispectral near-infrared image to enhance blood vessel contrast, apply a stereo vision three-dimensional reconstruction algorithm to construct a three-dimensional model of the blood vessel network, extract the blood vessel topology structure through edge detection and region growing algorithms, and generate blood vessel topology data.

[0015] As a further improvement of this technical solution, the biometric feature extraction module includes a reflection characteristic discrimination unit. The process of the reflection characteristic discrimination unit performing surface material reflection characteristic discrimination specifically includes:

[0016] Establish a visible light and near-infrared reflection feature space, and use a support vector machine classifier to output the material classification probability vector and the decision boundary distance value as the material discrimination data.

[0017] As a further improvement of this technical solution, the biometric feature extraction module includes a dynamic expression extraction unit. The process of the dynamic expression extraction unit performing dynamic micro-expression correlation feature extraction includes:

[0018] Apply local binary patterns to extract key facial feature points, track the displacement of the feature points through pixel motion vectors, and use a recurrent neural network to model the spatio-temporal correlation of micro-expressions.

[0019] As a further improvement of this technical solution, the short-term dynamic pattern modeling includes:

[0020] Perform dynamic graph convolutional network modeling on the blood vessel topology data and dynamic expression data, conduct spatio-temporal analysis of the graph neural network, and perform moving average processing on the material discrimination data.

[0021] As a further improvement of this technical solution, the long-term evolution trend modeling includes:

[0022] Apply the hidden Markov model to analyze the vascular topological state transition, establish an observation probability model for the material classification result, and construct a long-term change trend model for micro-expression features.

[0023] As a further improvement of this technical solution, the spatio-temporal modeling module implements multi-task training, including setting a shared layer to extract common features of blood vessels, materials, and expressions, configuring independent output layers for short-term dynamic patterns and long-term evolution trends respectively, and optimizing the multi-task loss function using an adaptive weight allocation mechanism.

[0024] As a further improvement of this technical solution, the feature model integration includes performing weighted average fusion on the outputs of multiple models, using a voting mechanism to comprehensively predict the results of short-term and long-term features, and finally outputting a comprehensive feature vector including three-dimensional vascular topology, material discrimination matrix, and micro-expression coding.

[0025] As a further improvement of this technical solution, the implementation of the support vector machine classifier includes:

[0026] Construct a non-linear classification boundary using a radial basis kernel function, optimize the regularization parameter and kernel width through grid search, apply the Platt scaling method to convert the decision boundary distance value into a material classification probability vector, and establish a dynamic parameter update mechanism to adjust the hyperplane parameters online when a new material category is detected.

[0027] Compared with the prior art, the beneficial effects of the present invention are:

[0028] The security management system based on facial capture achieves triple anti-counterfeiting breakthroughs through multi-spectral fusion and hierarchical feature verification mechanisms. Specifically, it uses visible light and near-infrared collaborative analysis to synchronously capture epidermal texture and subcutaneous vascular network, constructs a biological tissue tomography verification barrier, and effectively identifies highly realistic synthetic materials; dynamic micro-expression tracking combined with spatio-temporal modeling technology captures facial instantaneous micro-changes and long-term physiological evolution laws through a recurrent neural network, significantly improving the adaptability to expression camouflage and natural aging;

[0029] In addition, the cooperation between the dynamic parameter update of the support vector machine and the multi-task learning framework enables the system to have the ability to learn new material features online and autonomously optimize the decision boundary. This "multi-modal perception - hierarchical verification - dynamic evolution" technical architecture not only overcomes the feature drift defect of traditional solutions in complex environments, but also establishes a biometric recognition system with the ability of continuous evolution, providing a comprehensive protection solution with both real-time anti-counterfeiting intensity and long-term stability for high-risk scenarios such as financial security and border control. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the overall module of the present invention;

[0031] Figure 2Schematic diagram of the biometric feature extraction module unit of the present invention.

[0032] In the figure: 100, multispectral acquisition module; 200, biometric feature extraction module; 201, subcutaneous biological tissue chromatography analysis unit; 202, reflection characteristic identification unit; 203, dynamic expression extraction unit; 300, spatio-temporal modeling module. Specific implementation mode

[0033] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0034] Next, please refer to Figure 1 - Figure 2 , the present invention provides a technical solution: a security management system based on face capture, including a multispectral acquisition module 100, a biometric feature extraction module 200, and a spatio-temporal modeling module 300.

[0035] The multispectral acquisition module 100 uses at least two imaging sensors with complementary spectral response characteristics to synchronously capture facial image data in different spectral bands; among them, the imaging sensor group includes a visible light sensor and a near-infrared sensor, where the visible light sensor is used to capture facial image data in the visible light band, which can capture color information visible to the human eye and provide high-resolution color images; the near-infrared sensor is used to capture facial image data in the near-infrared band, and the near-infrared band covers the characteristic absorption spectral lines of hemoglobin; the facial image data in different spectral bands includes facial image data in the visible light band and the near-infrared band.

[0036] The multispectral acquisition module 100 includes a visible light sensor and a near-infrared sensor with complementary spectral response characteristics, where the visible light sensor captures RGB color facial image data with a resolution of not less than 1080p in a synchronous manner, and the near-infrared sensor is configured to collect facial image data in the 780-1100nm band.

[0037] The subcutaneous biological tissue chromatography analysis unit 201 in the biometric feature extraction module 200 performs subcutaneous biological tissue chromatography analysis based on spectral absorption differences according to the facial image data in the near-infrared band to generate vascular topology data, specifically including:

[0038] Fuse the facial image data in the near-infrared band of different spectral bands by using the weighted average method to enhance the contrast of blood vessels;

[0039] Use a stereo vision 3D reconstruction algorithm to construct a 3D model of the subcutaneous blood vessel network; calculate the spatial positions of the blood vessels by matching images from different perspectives; use image processing algorithms such as edge detection, region growing, and morphological operations to extract the contours and branches of the blood vessels. By detecting the edges in the images, the blood vessel contours are highlighted. Starting from seed points, it gradually expands to adjacent similar pixels to form blood vessel regions. Opening and closing operations are used to refine and connect the blood vessel structures;

[0040] Extract features such as the position coordinates, diameters, branch points, and confluence points of the blood vessels, identify and label the connected regions in the images, extract the topological structure of the blood vessels, generate a topological structure diagram of the blood vessels, representing the connection relationships between the blood vessels, as blood vessel topological data.

[0041] The reflection characteristic discrimination unit 202 in the biometric feature extraction module 200 performs surface material reflection characteristic discrimination based on the facial image data in the visible light band and the near-infrared band, and generates material discrimination data, specifically including:

[0042] Extract the reflection intensity and reflection spectrum characteristics of each pixel point in the facial image data in the visible light band and the near-infrared band; combine the reflection characteristics in the visible light and near-infrared bands to form a comprehensive reflection feature space; each point in this reflection feature space represents the reflection characteristic of a pixel;

[0043] Use a support vector machine to train a classifier, classify different materials by finding the optimal hyperplane; map the facial image data to be detected into the reflection feature space; use the trained classifier to classify each pixel point, and output the material classification probability vector and the decision boundary distance value as the material discrimination data; among them, the classification probability vector represents the probability that each pixel belongs to different materials; the decision boundary distance value represents the distance of each pixel to the classification boundary, used to measure the confidence of the classification; the implementation process of the support vector machine classifier specifically includes:

[0044] By adopting a radial basis kernel function to construct a non-linear classification boundary, the support vector machine can handle complex non-linear data distributions and improve the classification accuracy; grid search optimizes the regularization parameter and the kernel width to ensure that the model finds the optimal parameter combination during the training process and further improves the classification performance; applying the Platt scaling method to convert the decision boundary distance value into a material classification probability vector makes the classification result not only contain the class label but also the probability of each class, enhancing the interpretability and confidence of the classification result; in addition, a dynamic parameter update mechanism is established to adjust the hyperplane parameters online when new material categories are detected, enabling the system to adapt to new data distributions, improving the adaptability and robustness of the system, and thus extending the service life of the system.

[0045] The dynamic expression extraction unit 203 in the biometric feature extraction module 200 performs dynamic micro-expression correlation feature extraction based on facial image data in the visible light band to generate dynamic expression data, specifically including:

[0046] Use local binary pattern to extract key feature points (such as eyes, mouth, eyebrows, etc.) of facial image data in the visible light band; extract by comparing the brightness difference between pixels in the facial image data in the visible light band and their neighboring pixels;

[0047] Track the position changes of key feature points in consecutive image frames, and estimate the moving direction and speed of key feature points by calculating the motion vectors of pixels;

[0048] Use a recurrent neural network to model the spatio-temporal correlation of micro-expressions in different regions of the face, capture the dependencies in time series data through a cyclic structure, and is applicable to dynamic expression analysis;

[0049] Extract the motion trajectories of key feature points, synchronous motion features (such as the synchronous motion of eyes and mouth), and minute elevation features (such as the minute elevation of eyebrows) as dynamic expression data, where the motion trajectory records the position changes of key feature points in different frames; the synchronous motion feature analyzes the synchronous motion patterns between different feature points; the minute elevation feature detects the minute elevation of specific feature points, such as the slight upward movement of eyebrows.

[0050] The spatio-temporal modeling module 300 receives the output results of the biometric feature extraction module 200 and constructs a feature model including short-term dynamic patterns and long-term evolution trends, specifically including:

[0051] According to the vascular topology data, use a dynamic graph neural network to model the spatio-temporal correlation of the vascular network, dynamically update the features of nodes and edges through graph convolution operations, and capture the change patterns within a short time; according to the material classification probability vector and decision boundary distance value in the material discrimination data, use a moving average model to capture the short-term dynamic patterns of the material classification results; according to the dynamic expression data, use a dynamic graph neural network to model the spatio-temporal correlation of micro-expressions in different regions of the face, dynamically update the features of nodes and edges through graph convolution operations, and capture the change patterns within a short time; construct short-term dynamic patterns through the analysis of the change patterns within a short time of the vascular topology data, material discrimination data, and dynamic expression data;

[0052] Use a hidden Markov model to perform long-term analysis on the vascular topology data, material discrimination data, and dynamic expression data, capture the changes of hidden states through state transition probabilities and observation probabilities, and extract long-term evolution trends; analyze the vascular topology state transition, establish an observation probability model for the material classification results, and construct a long-term change trend model for micro-expression features;

[0053] Integrate the short-term dynamic pattern and long-term evolution trend to construct a comprehensive feature model, and train multiple tasks simultaneously to capture the short-term dynamic pattern and long-term evolution trend respectively;

[0054] Use a shared layer to extract common features and improve the generalization ability of the model; design specific output layers for each task to optimize short-term and long-term features respectively; integrate the outputs of multiple models to synthesize short-term and long-term features; use a voting mechanism to synthesize the prediction results of multiple models;

[0055] Synthesize the outputs of multiple models through a weighted average method to output a feature model that includes both short-term dynamic patterns and long-term evolution trends. The output results include short-term dynamic features, long-term evolution features, and a comprehensive feature model. Among them, short-term dynamic features capture the change patterns of facial features in a short time, such as the changes in micro-expressions and the fluctuations in blood vessel pulsation frequencies; long-term evolution features capture the change trends of facial features over a long time, such as skin aging and changes in blood vessel distribution; that is, the comprehensive feature vector of three-dimensional blood vessel topology, material discrimination matrix, and micro-expression coding;

[0056] The comprehensive feature model fuses short-term dynamic features and long-term evolution features to form a comprehensive feature model.

[0057] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A security management system based on facial capture, characterized in that: The system comprises a multi-spectral acquisition module (100), a biometric feature extraction module (200) and a spatiotemporal modeling module (300), wherein: The multi-spectral acquisition module (100) is configured to capture facial image data in different spectral bands, including facial image data in a visible light band and a near infrared band; The biometric feature extraction module (200) is connected to the multi-spectral acquisition module (100) and performs the following processing: According to the facial image data in the near-infrared band, subcutaneous biological tissue tomography analysis based on spectral absorption difference is performed; Based on the facial image data in the visible light band and near-infrared band, the surface material reflection characteristics are identified; Extract dynamic micro-expression correlation features based on facial image data in the visible light band; The spatiotemporal modeling module (300) receives the processing result of the biometric feature extraction module (200), constructs a feature model including short-term dynamic patterns and long-term evolution trends, and realizes a biometric feature recognition function with strong anti-counterfeiting ability and dynamic adaptability through the coordination of spectral fusion, hierarchical feature extraction and spatiotemporal modeling.

2. The facial capture-based security management system according to claim 1, characterized in that: The multi-spectral acquisition module (100) comprises a visible light sensor and a near infrared sensor having complementary spectral response characteristics.

3. The facial capture-based security management system according to claim 1, characterized in that: The biological feature extraction module (200) comprises a subcutaneous biological tissue tomography analysis unit (201), and the process of the subcutaneous biological tissue tomography analysis unit (201) performing subcutaneous biological tissue tomography analysis specifically comprises: Multispectral near-infrared images were weighted averaged and fused to enhance vascular contrast. A three-dimensional model of the vascular network was constructed using a stereoscopic vision three-dimensional reconstruction algorithm. The vascular topology structure was extracted through edge detection and region growing algorithms to generate vascular topology data.

4. The facial capture-based security management system according to claim 1, characterized in that: The biometric feature extraction module (200) comprises a reflection characteristic identification unit (202), and the reflection characteristic identification unit (202) performs a process of identifying the reflection characteristics of a surface material, which specifically comprises: The visible light and near-infrared reflectance feature space is established, and the support vector machine classifier is used to output the material classification probability vector and the decision boundary distance value as the material discrimination data.

5. The facial capture-based security management system according to claim 1, characterized in that: The biological feature extraction module (200) comprises a dynamic expression extraction unit (203), and the dynamic expression extraction unit (203) performs a dynamic micro-expression correlation feature extraction process comprising: Local binary patterns are used to extract key facial feature points, the displacement of feature points is tracked by pixel motion vector, and recursive neural networks are used to model the spatiotemporal correlation of micro-expressions.

6. The facial capture-based security management system according to claim 1, characterized in that: The short-term dynamic mode modeling includes: Dynamic graph convolutional network modeling is implemented on vascular topology data and dynamic expression data, graph neural network spatiotemporal analysis is performed, and moving average processing is used for material discrimination data.

7. The facial capture-based security management system according to claim 1, characterized in that: The long-term evolution trend modeling includes: The hidden Markov model is applied to analyze the transition of vascular topological states, an observation probability model of material classification results is established, and a long-term change trend model of micro-expression features is constructed.

8. The facial capture-based security management system according to claim 1, characterized in that: The spatiotemporal modeling module (300) implements multi-task training, including setting a shared layer to extract common features of blood vessels, materials, and expressions, configuring independent output layers for short-term dynamic patterns and long-term evolution trends, and using an adaptive weight allocation mechanism to optimize multi-task loss functions.

9. The facial capture-based security management system according to claim 1, characterized in that: The feature model integration includes weighted average fusion of the outputs of multiple models, using a voting mechanism to integrate short-term and long-term feature prediction results, and finally outputting a comprehensive feature vector including three-dimensional vascular topology, material discrimination matrix and micro-expression coding.

10. The facial capture-based security management system according to claim 4, characterized in that: The implementation of the support vector machine classifier includes: The radial basis kernel function is used to construct the nonlinear classification boundary. The regularization parameter and kernel width are optimized through grid search. The Pratt scaling method is applied to convert the decision boundary distance value into a material classification probability vector. A dynamic parameter updating mechanism is established to adjust the hyperplane parameters online when a new material category is detected.

Images