A method and system for preventing attacks using face video combined with liveness

Abstract

The present invention relates to the field of biometric recognition technology, and specifically to a method and system for preventing attacks using facial videos in combination with living bodies, comprising the following steps: based on video frame data, using an optical flow method and an inter-frame difference method to analyze dynamic changes between consecutive frames, and evaluating the motion vector and change intensity of each frame, and generating an inter-frame dynamic difference report; in the present invention, by combining the optical flow method and the inter-frame difference method, the dynamic changes between consecutive frames can be accurately captured, adaptive resolution optimization improves the processing efficiency of the system, and facial recognition data is converted into a graph structure through the Node2Vec algorithm, which helps to reveal hidden patterns and associations in the data; time series and change pattern recognition combined with time series analysis and wavelet transform can effectively identify time-varying characteristics and dynamic change patterns in facial data; signal integrity and consistency detection improve the system's ability to identify video tampering or simulated attacks.

Description

Technical Field

[0001] The present invention relates to the field of biometric identification technology, and in particular to a method and system for preventing attacks using facial videos combined with live bodies. Background Art

[0002] The field of biometric recognition technology focuses on using human physiological or behavioral characteristics for identity authentication. Biometric recognition technology has a wide range of applications, including fingerprint scanning, iris recognition, facial recognition, etc. In recent years, with the development of technology and the popularization of smart devices, facial recognition technology has received particular attention and is widely used in scenarios such as mobile phone unlocking, access control systems, and payment verification. However, facial recognition systems also face security challenges, especially the problem of liveness detection. Liveness detection aims to prevent attackers from using photos, videos or other forged biometrics to deceive the recognition system.

[0003] Among them, the method of preventing the use of face videos combined with live attacks is a security measure aimed at strengthening the security of facial recognition systems and preventing malicious deception. Its purpose is to ensure that facial recognition systems only respond to real, living faces, rather than photos, videos or facial images forged by other methods. The application of this method helps to improve the overall security of biometric recognition systems and protect users' personal privacy and data security. The method is implemented through a series of algorithms and technical means, including detecting tiny movements in facial images, changes in skin color, blinking or lip movements, etc. to determine whether the scanned object is a real living person. In addition, some advanced methods may also use 3D scanning technology, infrared sensors or other depth perception technologies to increase the recognition ability of non-living attacks. The combination of these technologies makes facial recognition systems more difficult to deceive, thereby improving their security and reliability.

[0004] Traditional biometric recognition technology has the following shortcomings in dealing with complex and sophisticated security challenges. First, existing technologies have difficulty in effectively distinguishing real faces from faces in video replay attacks because they lack advanced analysis capabilities for dynamic changes between frames (such as motion blur and light changes) and sensitive recognition of subtle changes in facial features. In image processing, existing technologies lack an adaptive resolution adjustment mechanism and find it difficult to dynamically adjust the processing flow according to the quality of the input image, resulting in inefficiency when processing low-quality images and waste of resources when processing high-quality images. In addition, existing technologies perform poorly in processing complex data associations, especially when analyzing subtle patterns and associations hidden in complex graph structures. This limitation is particularly evident in the face of This is particularly evident in advanced deception attacks, such as carefully crafted synthetic videos. Similarly, existing technologies have difficulty capturing facial expressions and subtle movements in time series analysis and identifying dynamic change patterns. In terms of signal integrity detection, existing technologies have difficulty effectively identifying and processing unnatural jumps or periodic anomalies in signals, which are key elements in detecting video tampering or simulation attacks. Finally, existing technologies have limited capabilities in detecting abnormal patterns in complex network structures based on facial data. In summary, although traditional biometric recognition technologies have made progress in many aspects, they still have significant room for improvement in handling advanced deception attacks, dynamic change analysis, adaptive image processing, and complex data association and network structure analysis. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a method and system for preventing attacks using face videos combined with live bodies.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for preventing attacks using face videos combined with liveness, comprising the following steps:

[0007] S1: Based on video frame data, it uses optical flow and frame difference methods to analyze the dynamic changes between consecutive frames, evaluate the motion vector and change intensity of each frame, and generate a frame-to-frame dynamic difference report;

[0008] S2: Based on the inter-frame dynamic difference report, the BRISQUE algorithm is used to automatically adjust the resolution of the image processing process, optimize the recognition efficiency, and generate an adaptive resolution optimization data report;

[0009] S3: Based on the adaptive resolution optimization data report, the Node2Vec algorithm is used to convert the data into a graph structure, and the relationship between the vertices is analyzed to generate a graph structure pattern recognition report;

[0010] S4: Based on the graph structure pattern recognition report, time series analysis and wavelet transform are used to analyze the time-varying characteristics of the facial data, and a time-varying characteristic and dynamic change pattern report is generated;

[0011] S5: Based on the time-varying characteristics and dynamic change pattern report, perform spectrum analysis and consistency detection using Fourier transform, evaluate signal continuity and consistency, and generate a signal continuity and consistency analysis report;

[0012] S6: Based on the signal continuity and consistency analysis report, a community detection algorithm and centrality analysis are used to analyze the network formed by the facial data, and a network structure abnormality pattern and attack indicator report is generated;

[0013] S7: Based on the network structure abnormality pattern and attack indicator report, the results of the comprehensive inter-frame dynamic difference report, the adaptive resolution optimization data report, the graph structure pattern recognition report, the time-varying characteristics and dynamic change pattern report, and the signal continuity and consistency analysis report, the data fusion algorithm and the multi-standard decision analysis method are used to perform the final diagnosis and processing and generate a comprehensive diagnosis decision plan.

[0014] As a further solution of the present invention, the inter-frame dynamic difference report is specifically a quantitative data analysis report of motion blur and light changes, the adaptive resolution optimization data report is specifically a parameter and efficiency optimization data analysis report of resolution adjustment, the graph structure pattern recognition report is specifically a relationship and potential correlation pattern analysis report between vertices in the graph, the time-varying characteristics and dynamic change pattern report is specifically a report on the changing characteristics and dynamic patterns of facial data over time, the signal continuity and consistency analysis report is specifically a quantitative index report of signal continuity and consistency, the network structure abnormal pattern and attack index report is specifically a report on abnormal characteristics of the network structure and potential attack indexes, and the comprehensive diagnostic decision-making plan includes comprehensive analysis conclusions and adjustment strategies for identified problems.

[0015] As a further solution of the present invention, based on video frame data, optical flow method and frame difference method are used to analyze the dynamic changes between consecutive frames, and the motion vector and change intensity of each frame are evaluated. The steps of generating the inter-frame dynamic difference report are as follows:

[0016] S101: Based on the video frame data, the Lucas-Kanade optical flow method is used to analyze the motion vectors of pixels in consecutive frames and generate a pixel motion analysis report;

[0017] S102: Based on the pixel motion analysis report, perform inter-frame difference analysis using an absolute difference method to generate a preliminary inter-frame difference analysis report;

[0018] S103: Based on the preliminary inter-frame difference analysis report, a multi-scale image analysis method is used to evaluate the motion changes and background stability of the image, and an inter-frame dynamic difference report is generated.

[0019] As a further solution of the present invention, based on the inter-frame dynamic difference report, the BRISQUE algorithm is used to automatically adjust the resolution of the image processing process and optimize the recognition efficiency. The steps of generating an adaptive resolution optimization data report are as follows:

[0020] S201: Based on the inter-frame dynamic difference report, using the Laplacian operator to quantitatively evaluate the image clarity and generate an image clarity evaluation report;

[0021] S202: Based on the image clarity evaluation report, using image pyramid downsampling technology, analyzing the relationship between image resolution and recognition efficiency, and generating a resolution optimization strategy report;

[0022] S203: Based on the resolution optimization strategy report, an adaptive bilinear interpolation algorithm is used to perform image sampling, and the image resolution is adjusted to generate an adaptive resolution optimization data report.

[0023] As a further solution of the present invention, based on the adaptive resolution optimization data report, the Node2Vec algorithm is used to convert the data into a graph structure and analyze the relationship between vertices to generate a graph structure pattern recognition report. Specifically, the steps are as follows:

[0024] S301: Based on the adaptive resolution optimization data report, the facial recognition data is converted into a graph structure using the Node2Vec algorithm to generate a facial data graph structure report;

[0025] S302: Based on the facial data graph structure report, using a graph centrality analysis method to analyze the relationship between vertices and generate a graph centrality correlation analysis report;

[0026] S303: Based on the graph centrality association analysis report, a subgraph matching algorithm is used to evaluate abnormal patterns in the graph structure and generate a graph structure pattern recognition report.

[0027] As a further solution of the present invention, based on the graph structure pattern recognition report, time series analysis and wavelet transform are used to analyze the time-varying characteristics of facial data, and the steps of generating a time-varying characteristic and dynamic change pattern report are as follows:

[0028] S401: Based on the graph structure pattern recognition report, an autoregressive moving average model is used to analyze the time series of facial data to generate a time series analysis report;

[0029] S402: Based on the time series analysis report, a multi-scale time-frequency analysis is performed using discrete wavelet transform to generate a multi-scale time-varying characteristic report;

[0030] S403: Based on the multi-scale time-varying characteristic report and a statistical time series analysis method, comprehensively evaluate the time-varying characteristics and dynamic change patterns of the facial data, and generate a time-varying characteristic and dynamic change pattern report.

[0031] As a further solution of the present invention, based on the time-varying characteristics and dynamic change pattern report, Fourier transform is used to perform spectrum analysis and consistency detection, and the continuity and consistency of the signal are evaluated. The steps of generating a signal continuity and consistency analysis report are specifically as follows:

[0032] S501: Based on the time-varying characteristics and the dynamic change pattern report, a spectrum analysis is performed using a fast Fourier transform to generate a spectrum analysis report;

[0033] S502: Based on the spectrum analysis report, a phase consistency detection method is used to perform signal consistency evaluation and generate a signal stability analysis report;

[0034] S503: Based on the signal stability analysis report, a signal integrity verification method is used to verify the continuity and consistency of the signal, and a signal continuity and consistency analysis report is generated.

[0035] As a further solution of the present invention, based on the signal continuity and consistency analysis report, a community detection algorithm and centrality analysis are used to analyze the network composed of facial data, and the steps of generating a network structure abnormality pattern and attack indicator report are specifically as follows:

[0036] S601: Based on the signal continuity and consistency analysis report, a modular optimization community detection algorithm is used to analyze the community structure of the face data network and generate a community structure analysis report;

[0037] S602: Based on the community structure analysis report, using a network centrality measurement method, analyzing the importance of key nodes and connections in the network, and generating a network key node analysis report;

[0038] S603: Based on the network key node analysis report, network pattern recognition technology is used to analyze the community structure and centrality, and a network structure abnormality pattern and attack indicator report is generated.

[0039] As a further solution of the present invention, based on the network structure abnormality pattern and attack indicator report, the results of the integrated inter-frame dynamic difference report, the adaptive resolution optimization data report, the graph structure pattern recognition report, the time-varying characteristics and dynamic change pattern report, and the signal continuity and consistency analysis report, a data fusion algorithm and a multi-criteria decision analysis method are used to perform final diagnosis and processing, and the steps of generating a comprehensive diagnosis and decision solution are specifically as follows:

[0040] S701: Based on the network structure anomaly pattern and attack indicator report, linear discriminant analysis is used to extract key information and indicators and generate a preliminary data fusion report;

[0041] S702: Based on the preliminary data fusion report, a high-dimensional data integration method is used to fuse the results of the inter-frame dynamic difference report, the adaptive resolution optimization data report, the graph structure pattern recognition report, the time-varying characteristics and dynamic change pattern report, and the signal continuity and consistency analysis report to generate an optimized data fusion report;

[0042] S703: Based on the optimized data fusion report, the analytic hierarchy process is used to perform final diagnosis and processing while considering the weights and importance of multiple indicators to generate a comprehensive diagnosis decision plan.

[0043] A system for preventing attacks using facial videos in combination with live bodies, the system being used to implement the above-mentioned method for preventing attacks using facial videos in combination with live bodies. The system includes a dynamic change analysis module, an image processing optimization module, a graph structure analysis module, a time-varying characteristic analysis module, a spectrum consistency detection module, a network structure analysis module, a data fusion decision module, and a comprehensive diagnosis decision module.

[0044] The dynamic change analysis module uses optical flow and inter-frame difference methods based on video frame data to analyze the dynamic changes between consecutive frames, evaluate the motion vector and change intensity of each frame, and generate an inter-frame dynamic difference report;

[0045] The image processing optimization module uses the BRISQUE algorithm to automatically adjust the resolution of the image processing process based on the inter-frame dynamic difference report, optimize the recognition efficiency, and generate an adaptive resolution optimization data report;

[0046] The graph structure analysis module optimizes the data report based on adaptive resolution, uses the Node2Vec algorithm to transform the data into a graph structure, analyzes the relationship between vertices, and generates a graph structure pattern recognition report;

[0047] The time-varying characteristics analysis module uses time series analysis and wavelet transform based on the graph structure pattern recognition report to analyze the time-varying characteristics of facial data and generate a time-varying characteristics and dynamic change pattern report;

[0048] The spectrum consistency detection module uses Fourier transform to perform spectrum analysis and consistency detection based on time-varying characteristics and dynamic change pattern reports, evaluates signal continuity and consistency, and generates a signal continuity and consistency analysis report;

[0049] The network structure analysis module analyzes the network composed of face data based on the signal continuity and consistency analysis report, using community detection algorithm and centrality analysis to generate network structure abnormality pattern and attack indicator report;

[0050] The data fusion decision module, based on the network structure anomaly pattern and attack indicator report, adopts data fusion algorithm and multi-criteria decision analysis method to fuse the results of dynamic change analysis module, image processing optimization module, graph structure analysis module, time-varying characteristic analysis module, spectrum consistency detection module and network structure analysis module to generate an optimized data fusion report;

[0051] The comprehensive diagnosis decision module is based on the optimized data fusion report and adopts the hierarchical analysis method to consider the weights and importance of multiple indicators while performing final diagnosis and processing to generate a comprehensive diagnosis decision plan.

[0052] Compared with the prior art, the advantages and positive effects of the present invention are:

[0053] In the present invention, by combining the optical flow method and the inter-frame difference method, the present invention can accurately capture the dynamic changes between consecutive frames, such as motion blur and light changes, which not only enhances the ability to identify the difference between real faces and video replay attacks, but also has excellent perception of subtle changes in facial features. Adaptive resolution optimization improves the processing efficiency of the system, adopts the BRISQUE algorithm to evaluate the image quality, and dynamically adjusts the resolution accordingly, so as to be more efficient in processing low-quality images, while avoiding resource waste in the processing of high-quality images. In terms of data association analysis, the present invention converts face recognition data into a graph structure through the Node2Vec algorithm, deeply analyzes the complex relationship between vertices in the graph, and helps to reveal hidden patterns and associations in the data, especially when countering carefully crafted synthetic video attacks. It is particularly important. Time series and change pattern recognition structures The combination of time series analysis and wavelet transform can effectively identify the time-varying characteristics and dynamic change patterns in facial data, especially in capturing facial expressions and small movements. Signal integrity and consistency detection improve the system's ability to identify video tampering or simulated attacks. Spectral analysis and consistency detection through Fourier transform enhance the evaluation of signal continuity and consistency, thereby more effectively identifying unnatural signal changes. Complex network structure analysis enhances the response capability to complex attack scenarios. Using community detection algorithm and centrality analysis, the present invention can identify abnormal patterns and attack indicators in the network composed of facial data. Finally, the improvement of comprehensive diagnosis and decision-making capabilities is achieved through data fusion algorithm and multi-standard decision analysis method, which can provide comprehensive diagnosis and effective processing solutions, thereby improving the overall security and reliability of the facial recognition system. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0055] Figure 2 This is a detailed flow chart of S1 of the present invention;

[0056] Figure 3 This is a detailed flow chart of S2 of the present invention;

[0057] Figure 4 This is a detailed flow chart of S3 of the present invention;

[0058] Figure 5 This is a detailed flow chart of S4 of the present invention;

[0059] Figure 6 This is a detailed flow chart of S5 of the present invention;

[0060] Figure 7 This is a detailed flow chart of S6 of the present invention;

[0061] Figure 8This is a detailed flow chart of S7 of the present invention;

[0062] Figure 9 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0064] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0065] Example 1

[0066] See also Figure 1 The present invention provides a technical solution: a method for preventing attacks using face videos combined with liveness, comprising the following steps:

[0067] S1: Based on video frame data, it uses optical flow and frame difference methods to analyze the dynamic changes between consecutive frames, evaluate the motion vector and change intensity of each frame, and generate a frame-to-frame dynamic difference report;

[0068] S2: Based on the inter-frame dynamic difference report, the BRISQUE algorithm is used to automatically adjust the resolution of the image processing process, optimize recognition efficiency, and generate an adaptive resolution optimization data report;

[0069] S3: Based on adaptive resolution optimization data report, the Node2Vec algorithm is used to transform the data into a graph structure, analyze the relationship between vertices, and generate a graph structure pattern recognition report;

[0070] S4: Based on the graph structure pattern recognition report, time series analysis and wavelet transform are used to analyze the time-varying characteristics of facial data and generate a report on time-varying characteristics and dynamic change patterns;

[0071] S5: Based on the time-varying characteristics and dynamic change pattern report, Fourier transform is used to perform spectrum analysis and consistency detection, and the continuity and consistency of the signal are evaluated to generate a signal continuity and consistency analysis report;

[0072] S6: Based on the signal continuity and consistency analysis report, the community detection algorithm and centrality analysis are used to analyze the network composed of facial data, and generate a report on network structure anomaly patterns and attack indicators;

[0073] S7: Based on the results of the network structure anomaly pattern and attack indicator report, the comprehensive inter-frame dynamic difference report, the adaptive resolution optimization data report, the graph structure pattern recognition report, the time-varying characteristics and dynamic change pattern report, and the signal continuity and consistency analysis report, the data fusion algorithm and multi-standard decision analysis method are used to perform the final diagnosis and processing and generate a comprehensive diagnosis decision plan.

[0074] The inter-frame dynamic difference report is specifically a quantitative data analysis report of motion blur and light changes, the adaptive resolution optimization data report is specifically a data analysis report of resolution adjustment parameters and efficiency optimization, the graph structure pattern recognition report is specifically an analysis report of the relationship between vertices in the graph and potential correlation patterns, the time-varying characteristics and dynamic change pattern report is specifically a report on the changing characteristics and dynamic patterns of facial data over time, the signal continuity and consistency analysis report is specifically a quantitative indicator report of signal continuity and consistency, the network structure abnormal pattern and attack indicator report is specifically a report on the abnormal characteristics of the network structure and potential attack indicators, and the comprehensive diagnostic decision-making plan includes comprehensive analysis conclusions and adjustment strategies for identified problems.

[0075] By analyzing the dynamic changes between consecutive frames using optical flow and frame difference methods, this method can accurately capture and analyze the motion vector and change intensity of each frame. This in-depth dynamic analysis improves the system's ability to distinguish between real faces and video replay attacks, especially in capturing motion blur and light changes.

[0076] Through the BRISQUE algorithm, this method can automatically adjust the resolution of the image processing process based on the dynamic difference report between frames. This adaptive mechanism makes the system more efficient when processing images of different qualities. It not only improves the processing accuracy of high-quality images, but also optimizes the processing speed of low-quality images, thereby making resource utilization more economical and efficient.

[0077] Using the Node2Vec algorithm, this method converts facial recognition data into a graph structure and deeply analyzes the relationships between vertices in the graph. This method not only improves the ability to understand and analyze complex data patterns, but also enhances the system's accuracy in identifying complex attack patterns.

[0078] Through time series analysis and wavelet transform, this method conducts in-depth analysis of the time-varying characteristics and dynamic change patterns of facial data. This analysis capability is crucial for capturing and understanding facial expressions and subtle movements, and improves the recognition accuracy of complex facial dynamics.

[0079] Spectral analysis and consistency detection using Fourier transform enable this method to comprehensively evaluate the continuity and consistency of the signal. This comprehensive signal analysis is crucial for identifying video tampering or simulation attacks, and greatly improves the system's ability to identify complex attack methods.

[0080] Through community detection algorithms and centrality analysis, this method conducts in-depth analysis of the network structure composed of facial data. This analysis not only reveals abnormal patterns in the network, but also identifies potential attack indicators, enhancing the system's ability to respond to advanced attack scenarios.

[0081] Through data fusion algorithms and multi-criteria decision analysis, this method can integrate all the aforementioned analysis results to provide comprehensive diagnosis and effective treatment solutions. This comprehensive decision-making mechanism not only improves the accuracy of decision-making, but also speeds up the response speed, thereby improving the security and reliability of the entire facial recognition system.

[0082] See also Figure 2 Based on the video frame data, the optical flow method and the frame difference method are used to analyze the dynamic changes between consecutive frames, and the motion vector and change intensity of each frame are evaluated. The specific steps for generating the inter-frame dynamic difference report are as follows:

[0083] S101: Based on the video frame data, the Lucas-Kanade optical flow method is used to analyze the motion vectors of pixels in consecutive frames and generate a pixel motion analysis report;

[0084] S102: Based on the pixel motion analysis report, an absolute difference method is used to perform inter-frame difference analysis to generate a preliminary inter-frame difference analysis report;

[0085] S103: Based on the preliminary analysis report of the inter-frame difference, a multi-scale image analysis method is used to evaluate the motion changes and background stability of the image, and an inter-frame dynamic difference report is generated.

[0086] In step S101, the video frame data is processed and the movement of each pixel in consecutive video frames is accurately tracked by applying the Lucas-Kanade optical flow method. This process involves calculating the motion vector of each pixel between consecutive frames, that is, the change in position of each pixel from one frame to the next. In this way, the system can capture subtle movements in facial images, such as micro-expressions or slight rotations of the head. These motion vectors are summarized and formed into a pixel motion analysis report, providing key motion information for subsequent steps.

[0087] In step S102, based on the pixel motion analysis report, the absolute difference method is used to conduct a more in-depth analysis of the inter-frame differences. This step aims to quantify the degree of difference between each frame by calculating the pixel value difference between consecutive frames. This method can effectively distinguish between real faces and video replay attacks by evaluating the intensity of changes between frames, because the differences in the latter between consecutive frames are usually smaller. The preliminary inter-frame difference analysis report generated at this stage provides detailed data on inter-frame changes, laying the foundation for further dynamic analysis.

[0088] In step S103, based on the preliminary analysis report of inter-frame differences, a multi-scale image analysis method is used to evaluate the motion changes and background stability of the overall image. The key is to identify and distinguish the main motion areas in the image (such as the face area) and the static or slowly changing background areas. By comprehensively analyzing image features at different scales, such as edges, textures and color changes, the system can accurately capture the natural dynamic changes of facial images and compare them with abnormal changes in human operations or replayed videos. This analysis helps the system generate an inter-frame dynamic difference report, which records in detail the natural patterns of facial movement and any abnormal changes, providing a key basis for identifying real living people and fake attacks.

[0089] See also Figure 3 Based on the inter-frame dynamic difference report, the BRISQUE algorithm is used to automatically adjust the resolution of the image processing process and optimize the recognition efficiency. The specific steps for generating the adaptive resolution optimization data report are as follows:

[0090] S201: Based on the inter-frame dynamic difference report, the Laplacian operator is used to quantitatively evaluate the image clarity and generate an image clarity evaluation report;

[0091] S202: Based on the image clarity evaluation report, the image pyramid downsampling technology is used to analyze the relationship between image resolution and recognition efficiency, and a resolution optimization strategy report is generated;

[0092] S203: Based on the resolution optimization strategy report, an adaptive bilinear interpolation algorithm is used to perform image sampling, and the image resolution is adjusted to generate an adaptive resolution optimization data report.

[0093] In step S201, based on the inter-frame dynamic difference report, the Laplacian operator is used to perform a detailed quantitative evaluation of the image clarity. This process determines the overall clarity of the image by analyzing the details and edge sharpness of the image. The Laplacian operator is particularly suitable for capturing detailed changes in the image, such as edge clarity and texture fineness. Through this method, the system can generate a comprehensive image clarity evaluation report, providing key basic information for adaptive resolution optimization.

[0094] In step S202, based on the image clarity assessment report, the image pyramid downsampling technology is used to analyze the relationship between image resolution and recognition efficiency. The image pyramid technology helps the system evaluate the impact of different resolution levels on recognition efficiency by gradually reducing the image resolution. During this process, the system will consider the balance between the information loss that may be caused by reducing the resolution and the improvement of recognition efficiency. Through this method, the system can generate a resolution optimization strategy report to provide guidance for subsequent image processing.

[0095] In step S203, based on the resolution optimization strategy report, an adaptive bilinear interpolation algorithm is used to sample the image and adjust the image resolution accordingly. The key to this step is to dynamically adjust the image resolution according to the image content and the required recognition efficiency. The adaptive bilinear interpolation algorithm allows the system to effectively adjust the image size while maintaining image quality. This flexible resolution adjustment mechanism not only optimizes the recognition efficiency, but also ensures the accuracy of image processing. Through this method, the system can generate an adaptive resolution optimization data report, providing the optimal image resolution setting for the facial recognition system.

[0096] See also Figure 4 Based on the adaptive resolution optimization data report, the Node2Vec algorithm is used to transform the data into a graph structure and analyze the relationship between vertices. The specific steps to generate the graph structure pattern recognition report are as follows:

[0097] S301: Based on the adaptive resolution optimization data report, the Node2Vec algorithm is used to convert the facial recognition data into a graph structure and generate a facial data graph structure report;

[0098] S302: Based on the facial data graph structure report, a graph centrality analysis method is used to analyze the relationship between vertices and generate a graph centrality correlation analysis report;

[0099] S303: Based on the graph centrality correlation analysis report, a subgraph matching algorithm is used to evaluate abnormal patterns in the graph structure and generate a graph structure pattern recognition report.

[0100] In step S301, the facial recognition data obtained from the adaptive resolution optimization data report is converted into a graph structure using the Node2Vec algorithm. The Node2Vec algorithm is a powerful graph embedding technology that can effectively capture and encode the complex relationships between nodes in the graph. This conversion process allows the system to better understand and parse the intrinsic connections of facial data, such as the interactions between different facial features. The converted facial data graph structure report provides a comprehensive view of the relationships and dependencies between facial features.

[0101] In step S302, graph centrality analysis is used to explore the relationships between vertices in the facial data graph structure report. Graph centrality analysis is a method used to identify the most important nodes in a graph. It can reveal the main connections and influences between facial features. Through this analysis, the system can identify key nodes in facial feature data that may play an important role in facial recognition. The generated graph centrality association analysis report provides a basis for further pattern recognition and anomaly detection.

[0102] In step S303, based on the graph centrality association analysis report, a subgraph matching algorithm is used to evaluate abnormal patterns in the graph structure. The subgraph matching algorithm can identify and compare different patterns and structures in the graph, especially those that may indicate abnormal or fraudulent activities. Through this method, the system can identify patterns that are inconsistent with conventional facial recognition data and then generate a graph structure pattern recognition report. This report plays a key role in identifying and preventing facial video replay attacks, providing the system with a powerful tool to identify possible fraudulent activities.

[0103] See also Figure 5 Based on the graph structure pattern recognition report, time series analysis and wavelet transform are used to analyze the time-varying characteristics of face data. The specific steps for generating the time-varying characteristics and dynamic change pattern report are as follows:

[0104] S401: Based on the graph structure pattern recognition report, an autoregressive moving average model is used to analyze the time series of facial data and generate a time series analysis report;

[0105] S402: Based on the time series analysis report, a discrete wavelet transform is used to perform multi-scale time-frequency analysis to generate a multi-scale time-varying characteristic report;

[0106] S403: Based on the multi-scale time-varying characteristic report and the statistical time series analysis method, the time-varying characteristics and dynamic change patterns of the facial data are comprehensively evaluated to generate a time-varying characteristic and dynamic change pattern report.

[0107] In step S401, the time series of facial data extracted from the graph structure pattern recognition report is deeply analyzed using the autoregressive moving average model. This model focuses on analyzing and understanding patterns and trends in time series data. Especially in the context of dynamic changes in facial data, the autoregressive moving average model can effectively reveal implicit time-related features in the data, such as the regularity and periodicity of expression changes. The generated time series analysis report provides in-depth insights into the changes in facial data over time, which is crucial for understanding the dynamic characteristics of facial expressions and subtle movements.

[0108] In step S402, discrete wavelet transform is used to perform multi-scale time-frequency analysis on the time series analysis report. Discrete wavelet transform is a powerful signal processing tool used to extract key frequency and time information from time series data. This step enables the system to analyze facial data at different time scales and capture more subtle dynamic changes. The generated multi-scale time-varying characteristic report reveals the complex patterns of facial expressions and small movements at different time scales, further enhancing the system's understanding and recognition capabilities of complex facial dynamics.

[0109] In step S403, a statistical time series analysis method is used based on the multi-scale time-varying characteristic report to comprehensively evaluate the time-varying characteristics and dynamic change patterns of facial data. This step combines multiple statistical techniques and analysis methods to comprehensively evaluate the time-varying characteristics of facial data, including the natural fluidity and dynamic patterns of facial expressions. The final generated time-varying characteristics and dynamic change pattern report not only captures the basic temporal characteristics of facial data, but also identifies atypical patterns that may indicate fraudulent activities, such as abnormal dynamics in video replay attacks.

[0110] See also Figure 6 Based on the time-varying characteristics and dynamic change pattern report, Fourier transform is used to perform spectrum analysis and consistency detection, and the continuity and consistency of the signal are evaluated. The steps for generating the signal continuity and consistency analysis report are as follows:

[0111] S501: Based on the time-varying characteristics and dynamic change pattern report, fast Fourier transform is used to perform spectrum analysis and generate a spectrum analysis report;

[0112] S502: Based on the spectrum analysis report, a phase consistency detection method is used to perform signal consistency assessment and generate a signal stability analysis report.

[0113] S503: Based on the signal stability analysis report, a signal integrity verification method is used to verify the continuity and consistency of the signal, and a signal continuity and consistency analysis report is generated.

[0114] In step S501, the data extracted from the time-varying characteristics and dynamic change pattern reports are spectrally analyzed by using fast Fourier transform. Fast Fourier transform is an effective method for converting time series data into the frequency domain, thereby revealing periodic and frequency-related features in the data. This step is particularly critical for identifying natural and artificial patterns in facial data. The generated spectrum analysis report provides an important basis for subsequent consistency detection, helping the system better understand the performance of facial data in the frequency domain.

[0115] In step S502, the phase consistency detection method is used based on the spectrum analysis report to evaluate the consistency of the signal. This method focuses on analyzing the phase characteristics of the signal, especially the consistency in frequency distribution. By detecting phase discontinuities and anomalies, the system can identify possible video replay or synthesis attacks. The generated signal stability analysis report provides the system with detailed information about signal consistency, which helps to distinguish real facial data from forged facial images.

[0116] In step S503, a signal integrity verification method is used based on the signal stability analysis report to further verify the continuity and consistency of the signal. This is achieved by analyzing the integrity and continuity of the signal throughout the entire time series. It is especially important in capturing and processing possible signal breaks or abnormal jumps. The generated signal continuity and consistency analysis report provides the system with a comprehensive signal quality assessment, ensuring the accuracy and reliability of the facial recognition process.

[0117] See also Figure 7 Based on the signal continuity and consistency analysis report, the community detection algorithm and centrality analysis are used to analyze the network composed of face data. The specific steps to generate the network structure abnormal pattern and attack indicator report are as follows:

[0118] S601: Based on the signal continuity and consistency analysis report, a modular optimized community detection algorithm is used to analyze the community structure of the face data network and generate a community structure analysis report;

[0119] S602: Based on the community structure analysis report, use the network centrality measurement method to analyze the importance of key nodes and connections in the network and generate a network key node analysis report;

[0120] S603: Based on the network key node analysis report, network pattern recognition technology is used to analyze the community structure and centrality, and generate a network structure abnormal pattern and attack indicator report.

[0121] In step S601, a modular optimized community detection algorithm is used to analyze the community structure of the facial data network based on signal continuity and consistency analysis reports. This algorithm can effectively identify different community groups in the network, that is, divide the network into multiple subgroups based on the similarity and connection strength between data points. Through this analysis, the system can reveal the implicit associations and group structures between facial data, providing a basis for further attack detection. The generated community structure analysis report describes in detail the structural characteristics of each community in the facial data network, which is crucial for understanding the patterns and relationships in the data.

[0122] In step S602, based on the community structure analysis report, the network centrality measurement method is used to analyze the key nodes and connections in the network. The centrality measurement method is an important network analysis tool used to evaluate the influence and importance of each node in the network. By identifying the key nodes in the network, the system can find the most influential parts of the entire facial data network structure and function. The generated network key node analysis report provides in-depth insights and reveals potential key influence points and connection patterns in the data.

[0123] In step S603, based on the network key node analysis report, network pattern recognition technology is used to analyze the community structure and centrality. This step aims to deeply understand the overall structure and dynamic behavior of the network, especially in identifying abnormal patterns and potential attack indicators in the network. Through in-depth analysis of the network structure, the system can detect unconventional patterns and abnormal behaviors, thereby effectively identifying possible attacks or deception attempts. The generated network structure abnormal pattern and attack indicator report provides the system with detailed information about potential security risks, further enhancing the security and reliability of the facial recognition system.

[0124] See also Figure 8 Based on the results of the network structure anomaly pattern and attack indicator report, the comprehensive inter-frame dynamic difference report, the adaptive resolution optimization data report, the graph structure pattern recognition report, the time-varying characteristics and dynamic change pattern report, and the signal continuity and consistency analysis report, the data fusion algorithm and multi-criteria decision analysis method are used to perform the final diagnosis and processing. The specific steps for generating a comprehensive diagnosis and decision plan are as follows:

[0125] S701: Based on the network structure anomaly pattern and attack indicator report, linear discriminant analysis is used to extract key information and indicators and generate a preliminary data fusion report;

[0126] S702: Based on the preliminary data fusion report, a high-dimensional data integration method is used to integrate the results of the inter-frame dynamic difference report, the adaptive resolution optimization data report, the graph structure pattern recognition report, the time-varying characteristics and dynamic change pattern report, and the signal continuity and consistency analysis report to generate an optimized data fusion report;

[0127] S703: Based on the optimized data fusion report, the hierarchical analysis method is used to consider the weights and importance of multiple indicators, and the final diagnosis and processing are performed to generate a comprehensive diagnosis decision plan.

[0128] In step S701, linear discriminant analysis is used to extract key information and indicators based on network structure anomaly patterns and attack indicator reports. This method focuses on identifying the features in the reports that best represent anomaly patterns and potential attack risks, facilitating more accurate classification and judgment of complex data. The generated preliminary data fusion report summarizes this key information, providing an important foundation for subsequent data fusion and decision analysis.

[0129] In step S702, a high-dimensional data integration method is used to conduct in-depth analysis and comprehensive processing of the preliminary data fusion report. This step aims to integrate the results of the inter-frame dynamic difference report, the adaptive resolution optimization data report, the graph structure pattern recognition report, the time-varying characteristics and dynamic change pattern report, and the signal continuity and consistency analysis report to form a comprehensive, multi-dimensional data view. The generated optimized data fusion report contains comprehensive information analyzed from multiple perspectives, which helps to more comprehensively understand and evaluate potential security risks.

[0130] In step S703, based on the optimized data fusion report, the analytic hierarchy process is used for final diagnosis and processing. This method makes the decision-making process more systematic and quantitative by weighing and comparing the relative importance of each indicator. The system will comprehensively consider the weight and importance of each indicator to generate a structured and optimized comprehensive diagnostic decision plan. It not only provides specific solutions to the detected problems, but also covers strategies for how to improve the facial recognition system to prevent future attacks.

[0131] See also Figure 9 A system for preventing attacks using facial videos in combination with live bodies. The system for preventing attacks using facial videos in combination with live bodies is used to execute the above-mentioned method for preventing attacks using facial videos in combination with live bodies. The system includes a dynamic change analysis module, an image processing optimization module, a graph structure analysis module, a time-varying characteristic analysis module, a spectrum consistency detection module, a network structure analysis module, a data fusion decision module, and a comprehensive diagnosis decision module.

[0132] The dynamic change analysis module uses optical flow and inter-frame difference methods based on video frame data to analyze the dynamic changes between consecutive frames, evaluate the motion vector and change intensity of each frame, and generate an inter-frame dynamic difference report;

[0133] The image processing optimization module uses the BRISQUE algorithm to automatically adjust the resolution of the image processing process based on the inter-frame dynamic difference report, optimize the recognition efficiency, and generate an adaptive resolution optimization data report;

[0134] The graph structure analysis module optimizes data reports based on adaptive resolution, uses the Node2Vec algorithm to transform data into graph structures, analyzes the relationships between vertices, and generates graph structure pattern recognition reports;

[0135] The time-varying characteristics analysis module is based on the graph structure pattern recognition report, uses time series analysis and wavelet transform to analyze the time-varying characteristics of facial data, and generates a report on time-varying characteristics and dynamic change patterns;

[0136] The spectrum consistency detection module uses Fourier transform to perform spectrum analysis and consistency detection based on time-varying characteristics and dynamic change pattern reports, evaluates signal continuity and consistency, and generates signal continuity and consistency analysis reports;

[0137] The network structure analysis module uses community detection algorithms and centrality analysis based on signal continuity and consistency analysis reports to analyze the network composed of face data and generate network structure anomaly patterns and attack indicator reports;

[0138] The data fusion decision module uses data fusion algorithms and multi-criteria decision analysis methods based on network structure anomaly patterns and attack indicator reports. It integrates the results of the dynamic change analysis module, image processing optimization module, graph structure analysis module, time-varying characteristics analysis module, spectrum consistency detection module, and network structure analysis module to generate an optimized data fusion report.

[0139] The comprehensive diagnosis and decision-making module is based on the optimized data fusion report and adopts the hierarchical analysis method. It considers the weights and importance of multiple indicators, performs final diagnosis and processing, and generates a comprehensive diagnosis and decision-making plan.

[0140] The introduction of the dynamic change analysis module, especially through the application of optical flow and inter-frame difference methods, improves the system's ability to recognize subtle facial movements and environmental changes. This means that the system can more effectively distinguish between real faces and video replay attacks, thereby reducing the risk of being deceived.

[0141] The image processing optimization module automatically adjusts the image resolution through the BRISQUE algorithm, enabling the system to perform adaptive processing based on input images of different qualities. This not only improves recognition efficiency but also ensures high accuracy, especially in the processing of low-quality images, reducing the possibility of misjudgment.

[0142] The introduction of the graph structure analysis module strengthens the recognition of complex data patterns. In-depth analysis after converting data into a graph structure using the Node2Vec algorithm enables the system to identify more subtle and complex deception patterns, such as carefully crafted synthetic videos.

[0143] The time-varying characteristic analysis module combines time series analysis and wavelet transform to enable the system to more accurately identify and analyze the changing characteristics of facial data over time. This is particularly critical for identifying natural dynamic changes in biometric features, further enhancing the system's ability to resist complex attacks.

[0144] The application of the spectrum consistency detection module provides a powerful tool for the system to identify unnatural signal changes through Fourier transform spectrum analysis and consistency detection of signals. This method can effectively identify unnatural jumps or periodic anomalies in the signal, which is key to detecting video tampering or simulation attacks.

[0145] The network structure analysis module uses community detection algorithms and centrality analysis to enable the system to analyze the network structure of facial data at a higher level and further identify potential abnormal patterns and attack indicators.

[0146] The design of the data fusion decision module enables the system to comprehensively consider the analysis results of multiple modules and achieve more comprehensive and in-depth analysis through data fusion algorithms and multi-criteria decision analysis methods.

[0147] The comprehensive diagnosis decision module uses the hierarchical analysis method to enable the system to comprehensively consider the weights and importance of multiple indicators when making the final diagnosis and treatment plan, thereby providing more accurate and reliable diagnostic decisions.

[0148] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of protection of the technical solution of the present invention.

Claims

1. A method for preventing liveness attacks using face videos, characterized in that: The following steps are involved: Based on video frame data, the optical flow method and inter-frame difference method are used to analyze the dynamic changes between consecutive frames, evaluate the motion vector and change intensity of each frame, and generate an inter-frame dynamic difference report; Based on the inter-frame dynamic difference report, the BRISQUE algorithm is used to automatically adjust the resolution of the image processing process, optimize the recognition efficiency, and generate an adaptive resolution optimization data report; Based on the adaptive resolution optimization data report, the Node2Vec algorithm is used to convert the data into a graph structure, and the relationship between vertices is analyzed to generate a graph structure pattern recognition report; Based on the graph structure pattern recognition report, time series analysis and wavelet transform are used to analyze the time-varying characteristics of the facial data and generate a time-varying characteristic and dynamic change pattern report; Based on the time-varying characteristics and dynamic change pattern report, Fourier transform is used to perform spectrum analysis and consistency detection, and the continuity and consistency of the signal are evaluated to generate a signal continuity and consistency analysis report; Based on the signal continuity and consistency analysis report, a community detection algorithm and centrality analysis are used to analyze the network composed of facial data to generate a network structure abnormality pattern and attack indicator report; Based on the network structure abnormality pattern and attack indicator report, the results of the comprehensive inter-frame dynamic difference report, the adaptive resolution optimization data report, the graph structure pattern recognition report, the time-varying characteristics and dynamic change pattern report, and the signal continuity and consistency analysis report, the data fusion algorithm and multi-standard decision analysis method are used to perform the final diagnosis and processing and generate a comprehensive diagnosis decision plan.

2. The method for preventing live body attacks using facial video according to claim 1, characterized in that: The inter-frame dynamic difference report is specifically a quantitative data analysis report of motion blur and light changes, the adaptive resolution optimization data report is specifically a parameter and efficiency optimization data analysis report of resolution adjustment, the graph structure pattern recognition report is specifically an analysis report of the relationship between vertices in the graph and potential correlation patterns, the time-varying characteristics and dynamic change pattern report is specifically a report on the changing characteristics and dynamic patterns of facial data over time, the signal continuity and consistency analysis report is specifically a quantitative index report of signal continuity and consistency, the network structure abnormality pattern and attack index report is specifically a report on abnormal characteristics of the network structure and potential attack indexes, and the comprehensive diagnostic decision-making plan includes comprehensive analysis conclusions and adjustment strategies for identified problems.

3. The method for preventing live body attacks using facial video according to claim 1, characterized in that: Based on video frame data, the optical flow method and frame difference method are used to analyze the dynamic changes between consecutive frames, and the motion vector and change intensity of each frame are evaluated. The specific steps for generating the inter-frame dynamic difference report are as follows: Based on video frame data, the Lucas-Kanade optical flow method is used to analyze the motion vectors of pixels in consecutive frames and generate a pixel motion analysis report; Based on the pixel motion analysis report, an absolute difference method is used to perform inter-frame difference analysis to generate a preliminary inter-frame difference analysis report; Based on the preliminary analysis report of the inter-frame difference, a multi-scale image analysis method is used to evaluate the motion changes and background stability of the image and generate an inter-frame dynamic difference report.

4. The method for preventing liveness attacks using facial video according to claim 3, characterized in that: Based on the inter-frame dynamic difference report, the BRISQUE algorithm is used to automatically adjust the resolution of the image processing process and optimize the recognition efficiency. The steps for generating the adaptive resolution optimization data report are as follows: Based on the inter-frame dynamic difference report, the Laplacian operator is used to quantitatively evaluate the image clarity and generate an image clarity evaluation report; Based on the image clarity assessment report, the image pyramid downsampling technology is used to analyze the relationship between image resolution and recognition efficiency, and a resolution optimization strategy report is generated; Based on the resolution optimization strategy report, an adaptive bilinear interpolation algorithm is used to perform image sampling, and the image resolution is adjusted to generate an adaptive resolution optimization data report.

5. The method for preventing live body attacks using facial video according to claim 4, characterized in that: Based on the adaptive resolution optimization data report, the Node2Vec algorithm is used to convert the data into a graph structure and analyze the relationship between vertices to generate a graph structure pattern recognition report. The specific steps are as follows: Based on the adaptive resolution optimization data report, the facial recognition data is converted into a graph structure using the Node2Vec algorithm to generate a facial data graph structure report; Based on the facial data graph structure report, a graph centrality analysis method is used to analyze the relationship between vertices and generate a graph centrality correlation analysis report; Based on the graph centrality association analysis report, a subgraph matching algorithm is used to evaluate abnormal patterns in the graph structure and generate a graph structure pattern recognition report.

6. The method for preventing liveness attacks using facial video according to claim 5, characterized in that: Based on the graph structure pattern recognition report, time series analysis and wavelet transform are used to analyze the time-varying characteristics of facial data, and the steps of generating a time-varying characteristic and dynamic change pattern report are as follows: Based on the graph structure pattern recognition report, an autoregressive moving average model is used to analyze the time series of facial data to generate a time series analysis report; Based on the time series analysis report, a multi-scale time-frequency analysis is performed using discrete wavelet transform to generate a multi-scale time-varying characteristic report; Based on the multi-scale time-varying characteristic report and the statistical time series analysis method, the time-varying characteristics and dynamic change patterns of the facial data are comprehensively evaluated to generate a time-varying characteristic and dynamic change pattern report.

7. The method for preventing liveness attacks using facial video according to claim 6, characterized in that: Based on the time-varying characteristics and dynamic change pattern report, Fourier transform is used to perform spectrum analysis and consistency detection, and the continuity and consistency of the signal are evaluated. The steps for generating a signal continuity and consistency analysis report are as follows: Based on the time-varying characteristics and dynamic change pattern report, fast Fourier transform is used to perform spectrum analysis to generate a spectrum analysis report; Based on the spectrum analysis report, a phase consistency detection method is used to perform signal consistency evaluation and generate a signal stability analysis report; Based on the signal stability analysis report, a signal integrity verification method is used to verify the continuity and consistency of the signal and generate a signal continuity and consistency analysis report.

8. The method for preventing live body attacks using facial video according to claim 7, characterized in that: Based on the signal continuity and consistency analysis report, the community detection algorithm and centrality analysis are used to analyze the network composed of facial data. The specific steps for generating the network structure abnormality pattern and attack indicator report are as follows: Based on the signal continuity and consistency analysis report, a modular optimized community detection algorithm is used to analyze the community structure of the face data network and generate a community structure analysis report; Based on the community structure analysis report, a network centrality measurement method is used to analyze the importance of key nodes and connections in the network and generate a network key node analysis report; Based on the network key node analysis report, network pattern recognition technology is used to analyze the community structure and centrality, and generate a network structure abnormal pattern and attack indicator report.

9. The method for preventing liveness attacks using facial video according to claim 8, characterized in that: Based on the network structure anomaly pattern and attack indicator report, the results of the integrated inter-frame dynamic difference report, the adaptive resolution optimization data report, the graph structure pattern recognition report, the time-varying characteristics and dynamic change pattern report, and the signal continuity and consistency analysis report, the data fusion algorithm and multi-criteria decision analysis method are used to perform the final diagnosis and processing. The specific steps for generating a comprehensive diagnosis and decision plan are as follows: Based on the network structure anomaly pattern and attack indicator report, linear discriminant analysis is used to extract key information and indicators and generate a preliminary data fusion report; Based on the preliminary data fusion report, a high-dimensional data integration method is used to fuse the results of the inter-frame dynamic difference report, the adaptive resolution optimization data report, the graph structure pattern recognition report, the time-varying characteristics and dynamic change pattern report, and the signal continuity and consistency analysis report to generate an optimized data fusion report; Based on the optimized data fusion report, the hierarchical analysis method is used to consider the weights and importance of multiple indicators while performing final diagnosis and processing to generate a comprehensive diagnosis decision plan.

10. A system for preventing attacks using facial videos combined with liveness, characterized by: The method for preventing attacks using facial videos combined with live bodies according to any one of claims 1 to 9, wherein the system comprises a dynamic change analysis module, an image processing optimization module, a graph structure analysis module, a time-varying characteristic analysis module, a spectrum consistency detection module, a network structure analysis module, a data fusion decision module, and a comprehensive diagnosis decision module; The dynamic change analysis module uses optical flow and inter-frame difference methods based on video frame data to analyze the dynamic changes between consecutive frames, evaluate the motion vector and change intensity of each frame, and generate an inter-frame dynamic difference report; The image processing optimization module uses the BRISQUE algorithm to automatically adjust the resolution of the image processing process based on the inter-frame dynamic difference report, optimize the recognition efficiency, and generate an adaptive resolution optimization data report; The graph structure analysis module optimizes the data report based on adaptive resolution, uses the Node2Vec algorithm to transform the data into a graph structure, analyzes the relationship between vertices, and generates a graph structure pattern recognition report; The time-varying characteristics analysis module uses time series analysis and wavelet transform based on the graph structure pattern recognition report to analyze the time-varying characteristics of facial data and generate a time-varying characteristics and dynamic change pattern report; The spectrum consistency detection module uses Fourier transform to perform spectrum analysis and consistency detection based on time-varying characteristics and dynamic change pattern reports, evaluates signal continuity and consistency, and generates a signal continuity and consistency analysis report; The network structure analysis module analyzes the network composed of face data based on the signal continuity and consistency analysis report, using community detection algorithm and centrality analysis to generate network structure abnormality pattern and attack indicator report; The data fusion decision module, based on the network structure anomaly pattern and attack indicator report, adopts data fusion algorithm and multi-criteria decision analysis method to fuse the results of dynamic change analysis module, image processing optimization module, graph structure analysis module, time-varying characteristic analysis module, spectrum consistency detection module and network structure analysis module to generate an optimized data fusion report; The comprehensive diagnosis decision module is based on the optimized data fusion report and adopts the hierarchical analysis method to consider the weights and importance of multiple indicators while performing final diagnosis and processing to generate a comprehensive diagnosis decision plan.

Images