Portrait data information acquisition and analysis system based on big data

Abstract

The invention relates to the technical field of computer vision and big data processing, in particular to a portrait data information acquisition and analysis system based on big data. The method comprises the following steps: a data acquisition and storage unit acquires face images in real time by using a multi-thread data acquisition technology, and performs storage and index labeling on the face images; the data image processing unit is used for preprocessing and standardizing the portrait image acquired by the data acquisition and storage unit; the face feature analysis unit is used for extracting standardized face image features, analyzing the face image features by using a spectrogram theory in combination with a frequency domain analysis algorithm, and classifying and updating index labels of face images; and the security management control unit is used for being responsible for the overall security and compliance of the system and providing privacy protection, access control and anomaly detection functions. According to the portrait data information acquisition and analysis system based on the big data, through a feature fusion technology combining a spectrogram theory and frequency domain analysis, analysis, classification and labeling of portrait data are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and big data processing. Specifically, it relates to a big data-based portrait data information collection and analysis system. Background Art

[0002] The big data-based portrait data information collection and analysis system aims to efficiently manage and retrieve massive portrait data, accurately extract and analyze facial features, and control the balance between feature extraction accuracy and real-time performance through a feature fusion method that combines distributed index storage with spectral graph theory and frequency domain analysis, so as to achieve fast retrieval, precise identification, and dynamic update of index annotation of portrait information in a multi-source data environment.

[0003] Portrait data information collection and analysis systems usually have difficulty maintaining stable data management and analysis efficiency in high-concurrency and cross-regional data collection scenarios. Moreover, due to uneven distribution of data sources, uneven image quality, and high complexity of multi-channel feature fusion, it will lead to insufficient robustness of feature extraction, unstable classification results, and untimely update of index annotation, thus affecting the accuracy of portrait recognition and the efficiency of data retrieval. Therefore, a big data-based portrait data information collection and analysis system is designed. Summary of the Invention

[0004] The purpose of the present invention is to provide a big data-based portrait data information collection and analysis system to solve the problems mentioned in the above background art, that is, due to uneven distribution of data sources, uneven image quality, and high complexity of multi-channel feature fusion, it will lead to insufficient robustness of feature extraction, unstable classification results, and untimely update of index annotation, thus affecting the accuracy of portrait recognition and the efficiency of data retrieval.

[0005] To achieve the above purpose, the present invention aims to provide a big data-based portrait data information collection and analysis system, including:

[0006] A data collection and storage unit, which uses multi-threaded data collection technology to collect face images in real time, stores face images using an Elasticsearch distributed database, and performs index annotation on face images;

[0007] It further includes a data image processing unit, which is used to preprocess and standardize the portrait images collected by the data collection and storage unit;

[0008] It further includes a face feature analysis unit, which is used to extract the features of the standardized face images, analyze the face image features using spectral graph theory combined with a frequency domain analysis algorithm, classify and update the index annotation of the face images;

[0009] It also includes a security management control unit, which is responsible for the overall security and compliance of the system and provides functions such as privacy protection, access control, and anomaly detection.

[0010] As a further improvement of this technical solution, the data acquisition and storage unit includes a face image acquisition module and a storage annotation module;

[0011] Among them, the face image acquisition module uses the Python multi-threaded programming model to collect image data in real time from the data source, and uses the MTCNN face detection algorithm to identify and intercept the face area to obtain a face image; the data source includes cameras, mobile devices, and recorded videos;

[0012] The storage annotation module stores the face images in the Elasticsearch distributed database and generates index annotations for each face image. The index annotations include time, location, device ID, and person ID.

[0013] As a further improvement of this technical solution, the data image processing unit includes an image quality enhancement module and an image normalization module;

[0014] Among them, the image quality enhancement module is used to perform smoothing processing on the face image through Gaussian filtering;

[0015] The image normalization module is used to scale the face image to a predetermined standard size, apply a color normalization algorithm to unify the color distribution of the face image, and generate a final normalized face image.

[0016] As a further improvement of this technical solution, the face feature analysis unit includes a feature extraction module and a feature analysis index update module;

[0017] Among them, the feature extraction module uses the Canny edge detection algorithm to extract the facial contour features of the face image, uses the SIFT feature detection algorithm to extract the key point distribution features of the face image, and uses the local binary pattern method to extract the texture features of the face image; the face image features include facial contour features, key point distribution features, and texture features;

[0018] The feature analysis index update module uses spectral graph theory combined with a frequency domain analysis algorithm to analyze and classify the face image features and update the index annotations of the face image.

[0019] As a further improvement of this technical solution, the feature extraction module uses the Canny edge detection algorithm to extract the facial contour features of the face image, uses the SIFT feature detection algorithm to extract the key point distribution features of the face image, and uses the local binary pattern method to extract the texture features of the face image. The specific method steps are as follows:

[0020] S3.1.1. Extract the facial contour features of the face image using the Canny edge detection algorithm:

[0021] Based on the normalized face image , calculate the image gradient parallel non-maximum suppression and double-threshold hysteresis processing to obtain a binary edge map :

[0022] ;

[0023] Among them, is the pixel point coordinate on the face image ;

[0024] Extract the largest connected edge point set from the binary edge map , which is the facial contour of the face image; is the coordinate of the th edge point; is the total number of edge points;

[0025] Use elliptical fitting for the contour point set , and solve the parameters , that is, the facial contour features :

[0026] ;

[0027] Among them, is the abscissa of the center point of the ellipse; is the ordinate of the center point of the ellipse; is the horizontal radius of the ellipse; is the vertical radius of the ellipse; is the rotation angle of the ellipse relative to the horizontal axis;

[0028] S3.1.2. Extract the key point distribution features of the face image using the SIFT feature detection algorithm:

[0029] Based on the normalized face image , construct a Gaussian pyramid and a difference Gaussian pyramid:

[0030] ;

[0031] ;

[0032] Among them, is the standard Gaussian filter; is the Gaussian pyramid; is the difference Gaussian pyramid; is the constant of the scale interval; is the standard deviation of the Gaussian kernel at the current scale;

[0033] Detect the extreme points in the difference-of-Gaussians pyramid as the key points of the face image, and obtain the key point set :

[0034] ;

[0035] where, is the position coordinate of the key point ; is the scale of the key point ;

[0036] Calculate the gradient direction distribution of the key point in the local neighborhood and assign the main direction :

[0037] ;

[0038] ;

[0039] ;

[0040] where, is the point of the key point in the local neighborhood; is the pixel gradient direction; is the weighting function; is the indicator function; is the main direction of the key point ;

[0041] In the neighborhood of the key point , statistically compose the gradient direction and amplitude into a descriptor vector:

[0042] SIFT k =[ h 1 , h 2 ,…, h D s ] ;

[0043] where, is the local descriptor of the key point ; is the dimension of the SIFT descriptor; is the count value of the neighborhood gradient direction histogram of the key point ;

[0044] Obtain the key point distribution feature set:

[0045] ;

[0046] Among them, is the total number of key points; is the set of key point distribution characteristics;

[0047] S3.1.3. Extract the texture features of the face image using the local binary pattern method:

[0048] Let the central pixel value be ; the neighborhood pixel value be ;

[0049] Calculate the LBP code of pixel point as:

[0050] ;

[0051] Among them, is the LBP code of pixel point ; is the serial number of the neighborhood pixel; is the total number of neighborhood pixels; is the indicator function for calculating ;

[0052] Calculate the LBP codes of all pixel points in the face image:

[0053] ;

[0054] Among them, is the occurrence frequency of the pixel point with the LBP code value of in the entire face image; is the value of the LBP code, ; is the width of the face image; is the height of the face image; is the indicator function for calculating ;

[0055] Obtain the LBP histogram vector , which is the texture feature:

[0056] T=[ H LBP (0), H LBP (1),…, H LBP ( 2 N -1)] ;

[0057] Among them, is the texture feature.

[0058] As a further improvement of this technical solution, the feature analysis index update module analyzes and classifies the facial image features by using spectral graph theory combined with the frequency domain analysis algorithm, and updates the index annotation of the facial image. The specific method steps are as follows:

[0059] S3.2.1. Based on the facial contour features and the key point distribution feature set , construct a weighted undirected graph using spectral graph theory;

[0060] S3.2.2. Calculate the degree matrix and the adjacency matrix to construct the Laplacian matrix of the graph, and perform eigenvalue decomposition on the Laplacian matrix to extract the spectral feature vectors of the weighted undirected graph;

[0061] S3.2.3. Perform discrete Fourier transform on the texture features of the facial image to extract the frequency domain features, and perform dimensionality reduction and screening on the frequency domain features to obtain the texture frequency domain feature vectors;

[0062] S3.2.4. Perform feature-level fusion on the spectral feature vectors and the texture frequency domain feature vectors to obtain comprehensive feature vectors, and classify the comprehensive feature vectors using statistical classification methods, and update the index annotation of the facial image according to the classification results.

[0063] As a further improvement of this technical solution, in S3.2.1, based on the facial contour features and the key point distribution feature set , construct a weighted undirected graph using spectral graph theory. The specific method is as follows:

[0064] S3.2.1.1. Regard the key points of the facial image as the nodes of the graph, and define the node set:

[0065] ;

[0066] Among them, is the first node defined by the first key point, is the second node defined by the second key point, and so on, is the th node defined by the th key point; The set of nodes forms the node set ;

[0067] S3.2.1.2. Construct a weighted undirected graph according to the Euclidean distance between the key points and the similarity of the SIFT descriptor features:

[0068] Define the adjacency matrix :

[0069] ;

[0070] ;

[0071] Among them, is the Euclidean distance between node and node ; is the Euclidean distance between the SIFT descriptor of node and the SIFT descriptor of node ; is the control distance parameter; is the control feature similarity parameter; represents the correlation degree between node and node ;

[0072] Using the facial contour feature to obtain the global geometric reference of the human face in the image for key point position normalization and graph weight calculation; normalize the coordinates before calculating ;

[0073] In the aforementioned S3.2.2, calculate the degree matrix and the adjacency matrix to construct the Laplacian matrix of the graph, and perform eigenvalue decomposition on the Laplacian matrix to extract the spectral feature vectors of the weighted undirected graph. The specific method is as follows:

[0074] S3.2.2.1. Define the degree matrix as a diagonal matrix. The diagonal elements of the degree matrix are as follows:

[0075] ;

[0076] S3.2.2.2. Define the Laplacian matrix:

[0077] ;

[0078] Among them, is the Laplacian matrix;

[0079] S3.2.2.3. Perform eigenvalue decomposition on the Laplacian matrix :

[0080] ;

[0081] Among them, is the eigenvector of the Laplacian matrix ; is the eigenvalue of the Laplacian matrix ;

[0082] Sort the eigenvalues in ascending order, and take the first smaller non-zero eigenvalues and their corresponding eigenvectors to form a spectral eigenvector:

[0083] s=[ λ 1 , λ 2 ,…, λ M ] T ;

[0084] wherein, is the spectral eigenvector; is the transpose operation.

[0085] As a further improvement of this technical solution, in S3.2.3, the texture features of the face image are subjected to discrete Fourier transform to extract frequency domain features, and the frequency domain features are dimension-reduced and screened to obtain a texture frequency domain feature vector. The specific method is as follows:

[0086] S3.2.3.1. Perform discrete Fourier transform on the texture features of the face image to obtain a spectrum :

[0087] ;

[0088] F = [F(0), F(1), …, F( 2 N -1)] ;

[0089] wherein, is the frequency of the pixel with LBP code value appearing in the entire face image; is the index of the frequency domain component after discrete Fourier transform; is for after discrete Fourier transform, the complex spectrum value obtained at the frequency index ; is the imaginary unit; is the rotation factor of the discrete Fourier transform;

[0090] S3.2.3.2. Extract the amplitudes of the first frequency components as the texture frequency domain feature vector :

[0091] ;

[0092] wherein, is the spectrum At the frequency index the amplitude value; For the spectrum at the frequency index the amplitude value; .

[0093] As a further improvement of the technical solution, in S3.2.4, the spectral feature vector and the texture frequency domain feature vector are fused at the feature level into a comprehensive feature vector, and the comprehensive feature vector is classified using a statistical classification method, and the index annotation of the face image is updated according to the classification result. The specific method is as follows:

[0094] S3.2.4.1. Fuse the spectral feature vector and the texture frequency domain feature vector at the feature level into a comprehensive feature vector :

[0095] c = [s;t] ∈ R M + P ;

[0096] Among them, is dimensions;

[0097] S3.2.4.2. Classify the comprehensive feature vector using a support vector machine:

[0098] The decision function of the support vector machine is:

[0099] ;

[0100] Among them, is the number of support vectors; is the Lagrange multiplier; is the kernel function is the bias term; is the classification result;

[0101] The classification result includes the category, identity, and clustering label of the face image;

[0102] S3.2.4.3. According to the classification result , update the analysis result to the index annotation of the face image.

[0103] As a further improvement of the technical solution, the security management control unit includes a permission access control module, a privacy protection data encryption module, and an anomaly detection and auditing module;

[0104] Among them, the permission access control module is used to perform role-based access control on system users, processes, and applications, configure fine-grained permission policies, and implement multi-factor authentication measures;

[0105] The privacy protection data encryption module is used to encrypt and store and transmit portrait data;

[0106] The anomaly detection and auditing module is used to monitor the system operation status in real time, continuously audit and analyze access logs, operation behaviors, and data transmissions, use intrusion detection systems and security information and event management tools to identify potential security threats and anomaly events, and automatically trigger an early warning when an anomaly is detected.

[0107] Compared with the prior art, the beneficial effects of the present invention are:

[0108] 1. In this big data-based portrait data information collection and analysis system, based on the distributed index storage and high-concurrency data processing architecture, it can quickly achieve efficient data storage, flexible retrieval, and accurate matching when collecting a large amount of face image data simultaneously from multiple sources and multiple points.

[0109] 2. In this big data-based portrait data information collection and analysis system, through the feature fusion technology that combines spectrogram theory and frequency domain analysis, it realizes the fine classification and annotation update of face images. BRIEF DESCRIPTION OF THE DRAWINGS

[0110] Figure 1 is the overall flow block diagram of the present invention;

[0111] The meanings of each label in the figure are as follows:

[0112] 1. Data acquisition and storage unit; 2. Data image processing unit; 3. Face feature analysis unit; 4. Security management and control unit; 11. Face image acquisition module; 12. Storage and annotation module; 21. Image quality enhancement module; 22. Image normalization module; 31. Feature extraction module; 32. Feature analysis, index update module; 41. Permission access control module; 42. Privacy protection data encryption module; 43. Anomaly detection and auditing module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0113] 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 embodiments of 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.

[0114] Embodiment

[0115] Please refer to Figure 1 as shown, a big data-based portrait data information acquisition and analysis system is provided, including:

[0116] A data acquisition and storage unit 1, which uses multi-threaded data acquisition technology to collect face images in real time, stores face images using an Elasticsearch distributed database, and indexes and labels the face images;

[0117] In this embodiment, the data acquisition and storage unit 1 includes a face image acquisition module 11 and a storage and annotation module 12;

[0118] Among them, the face image acquisition module 11 uses the Python multi-threaded programming model to collect image data from data sources in real time, and uses the MTCNN face detection algorithm to identify and intercept the face area to obtain face images; the data sources include cameras, mobile devices, and recorded videos;

[0119] In this embodiment, the face image acquisition module 11 uses the Python multi-threaded programming model to collect face images from data sources in real time. The specific method steps are as follows:

[0120] Access various cameras and mobile devices through RTSP, HTTP or other protocols to establish a stable data stream connection; allocate independent acquisition threads for each data source to receive video streams or image frames in real time; extract continuous image frames from the video stream; use the Python thread-safe queue Queue to temporarily store the collected image frames;

[0121] In this embodiment, the MTCNN face detection algorithm is a face detection and alignment algorithm, which uses a cascaded multi-task convolutional neural network to perform face detection and key point localization simultaneously;

[0122] The storage and annotation module 12 stores the face images in the Elasticsearch distributed database and generates index annotations for each face image. The index annotations include time, location, device ID, and person ID.

[0123] In this embodiment, multiple Elasticsearch nodes are deployed to build a distributed database cluster, and a multi-field index based on time, location, and person ID is designed and created; the face images are stored in Base64 encoding, and the face image data and its metadata are batch inserted through the Elasticsearch API.

[0124] It further includes a data image processing unit 2, which is used to preprocess and standardize the portrait images collected by the data acquisition and storage unit 1;

[0125] In this embodiment, the data image processing unit 2 includes an image quality enhancement module 21 and an image normalization module 22;

[0126] Among them, the image quality enhancement module 21 is used to smooth the face image through Gaussian filtering;

[0127] The image normalization module 22 is used to scale the face image to a predetermined standard size, apply a color normalization algorithm to unify the color distribution of the face image, and generate a final normalized face image.

[0128] In this embodiment, the predetermined standard size is 224x224 pixels, and according to application requirements, the image background is removed or replaced to highlight the face main body and reduce the interference of the background on feature extraction.

[0129] It further includes a face feature analysis unit 3, and the face feature analysis unit 3 is used to extract the features of the normalized face image, analyze the face image features using the spectral graph theory combined with the frequency domain analysis algorithm, and classify and update the index annotation of the face image;

[0130] In this embodiment, the face feature analysis unit 3 includes a feature extraction module 31 and a feature analysis index update module 32;

[0131] Among them, the feature extraction module 31 uses the Canny edge detection algorithm to extract the facial contour features of the face image, uses the SIFT feature detection algorithm to extract the key point distribution features of the face image, and uses the local binary pattern method to extract the texture features of the face image; the face image features include facial contour features, key point distribution features, and texture features;

[0132] The feature analysis index update module 32 analyzes and classifies the face image features using the spectral graph theory combined with the frequency domain analysis algorithm, and updates the index annotation of the face image.

[0133] The feature extraction module 31 uses the Canny edge detection algorithm to extract the facial contour features of the face image, uses the SIFT feature detection algorithm to extract the key point distribution features of the face image, and uses the local binary pattern method to extract the texture features of the face image. The specific method steps are as follows:

[0134] S3.1.1. Use the Canny edge detection algorithm to extract the facial contour features of the face image:

[0135] Based on the normalized face image , calculate the image gradient parallel non-maximum suppression and double-threshold hysteresis processing to obtain a binary edge map :

[0136] ;

[0137] wherein, are the pixel coordinates on the face image ;

[0138] Extract the largest connected edge point set from the binary edge map , which is the facial contour of the face image; are the coordinates of the th edge point; are the coordinates of the edge point; is the total number of edge points;

[0139] Use ellipse fitting for the contour point set to solve the parameters , i.e., the facial contour features :

[0140] ;

[0141] wherein, is the abscissa of the center point of the ellipse; is the ordinate of the center point of the ellipse; is the horizontal radius of the ellipse; is the vertical radius of the ellipse; is the rotation angle of the ellipse relative to the horizontal axis;

[0142] S3.1.2. Use the SIFT feature detection algorithm to extract the key point distribution features of the face image:

[0143] Based on the standardized face image , construct the Gaussian pyramid and the difference Gaussian pyramid:

[0144] ;

[0145] ;

[0146] wherein, is the standard Gaussian filter; is the Gaussian pyramid; is the difference Gaussian pyramid; is the constant of the scale interval; is the standard deviation of the Gaussian kernel at the current scale;

[0147] Detect the extreme points in the difference Gaussian pyramid as the key points of the face image to obtain the key point set :

[0148] ;

[0149] wherein, are the position coordinates of the key point ; is the key point scale;

[0150] Calculate the key point gradient direction distribution in the local neighborhood and assign the main direction :

[0151] ;

[0152] ;

[0153] ;

[0154] wherein, is the key point point in the local neighborhood; is the pixel gradient direction; is the weighting function; is the indicator function; is the key point main direction;

[0155] In this embodiment, is the calculation indicator function, when the function returns 1; otherwise it returns 0, used to count the number of pixels with gradient direction in the image;

[0156] In the neighborhood of the key point statistical description sub-vector is formed for the gradient direction and amplitude:

[0157] SIFT k =[ h 1 , h 2 ,…, h D s ] ;

[0158] wherein, is the local descriptor of the key point ; is the dimension of the SIFT descriptor; is the key point count value of the neighborhood gradient direction histogram;

[0159] Obtain the key point distribution feature set:

[0160] ;

[0161] wherein, is the total number of key points; is the set of key point distribution characteristics;

[0162] S3.1.3. Extract the texture features of the face image using the local binary pattern method:

[0163] Let the central pixel value be ; the neighborhood pixel value be ;

[0164] Calculate the LBP code of pixel point as:

[0165] ;

[0166] where is the LBP code of pixel point ; is the serial number of the neighborhood pixel; is the total number of neighborhood pixels; is the indicator function for calculating ;

[0167] Calculate the LBP codes of all pixel points in the face image:

[0168] ;

[0169] where is the occurrence frequency of the pixel point with the LBP code value of in the entire face image; is the value of the LBP code, ; is the width of the face image; is the height of the face image; is the indicator function for calculating ;

[0170] Obtain the LBP histogram vector , which is the texture feature:

[0171] T=[ H LBP (0), H LBP (1),…, H LBP ( 2 N -1)] ;

[0172] where is the texture feature.

[0173] The feature analysis index update module 32 analyzes and classifies the facial image features by using spectral graph theory in combination with a frequency domain analysis algorithm, and updates the index annotation of the facial image. The specific method steps are as follows:

[0174] S3.2.1. Based on the facial contour features and the key point distribution feature set , construct a weighted undirected graph using spectral graph theory;

[0175] S3.2.2. Calculate the degree matrix and the adjacency matrix to construct the Laplacian matrix of the graph, and perform eigenvalue decomposition on the Laplacian matrix to extract the spectral feature vectors of the weighted undirected graph;

[0176] S3.2.3. Perform discrete Fourier transform on the texture features of the facial image to extract frequency domain features, and perform dimensionality reduction and screening on the frequency domain features to obtain texture frequency domain feature vectors;

[0177] S3.2.4. Perform feature-level fusion on the spectral feature vectors and the texture frequency domain feature vectors to obtain comprehensive feature vectors, and classify the comprehensive feature vectors using a statistical classification method. Update the index annotation of the facial image according to the classification results.

[0178] In this embodiment S3.2.1, based on the facial contour features and the key point distribution feature set , construct a weighted undirected graph using spectral graph theory. The specific method is as follows:

[0179] S3.2.1.1. Regard the key points of the facial image as the nodes of the graph, and define the node set:

[0180] ;

[0181] Among them, is the first node defined by the first key point, is the second node defined by the second key point, and so on, is the th node defined by the th key point; The node set constitutes the node set

[0182] S3.2.1.2. Construct a weighted undirected graph according to the Euclidean distance between the key points and the SIFT descriptor feature similarity:

[0183] Define the adjacency matrix :

[0184] ;

[0185] ;

[0186] wherein, is the Euclidean distance between node and node ; is the Euclidean distance between the SIFT descriptor of node and the SIFT descriptor of node ; is the control distance parameter; is the control feature similarity parameter; represents the correlation degree between node and node ;

[0187] Utilize the facial contour feature to obtain the global geometric reference of the human face in the image, which is used for key point position normalization and graph weight calculation; normalize the coordinates before calculating ;

[0188] In this embodiment S3.2.2, calculate the degree matrix and the adjacency matrix to construct the Laplacian matrix of the graph, and perform eigenvalue decomposition on the Laplacian matrix to extract the spectral feature vectors of the weighted undirected graph. The specific method is as follows:

[0189] S3.2.2.1. Define the degree matrix as a diagonal matrix. The diagonal elements of the degree matrix are as follows:

[0190] ;

[0191] S3.2.2.2. Define the Laplacian matrix:

[0192] ;

[0193] wherein, is the Laplacian matrix;

[0194] S3.2.2.3. Perform eigenvalue decomposition on the Laplacian matrix :

[0195] ;

[0196] wherein, is the eigenvector of the Laplacian matrix ; is the eigenvalue of the Laplacian matrix ;

[0197] Sort the eigenvalues in ascending order, and take the first A smaller non-zero eigenvalue and its corresponding eigenvector form a spectral eigenvector:

[0198] s=[ λ 1 , λ 2 ,…, λ M ] T ;

[0199] Wherein, is the spectral eigenvector; is the transpose operation.

[0200] In this embodiment S3.2.3, the texture feature of the face image is subjected to discrete Fourier transform to extract frequency-domain features, and the frequency-domain features are dimension-reduced and screened to obtain a texture frequency-domain feature vector. The specific method is as follows:

[0201] S3.2.3.1. Perform discrete Fourier transform on the texture feature of the face image to obtain a spectrum :

[0202] ;

[0203] F = [F(0), F(1), …, F( 2 N -1)] ;

[0204] Wherein, is the frequency at which the pixel with LBP code value appears in the entire face image; is the index of the frequency-domain component after discrete Fourier transform; is for after discrete Fourier transform, the complex spectrum value obtained at the frequency index ; is the imaginary unit; is the rotation factor of the discrete Fourier transform;

[0205] S3.2.3.2. Extract the amplitudes of the first frequency components as the texture frequency-domain feature vector :

[0206] ;

[0207] Wherein, is the amplitude value of the spectrum at the frequency index ; is the spectrum At the frequency index the amplitude value; .

[0208] In this embodiment S3.2.4, the spectral feature vector and the texture frequency domain feature vector are fused at the feature level into a comprehensive feature vector, and a statistical classification method is used to classify the comprehensive feature vector. According to the classification result, the index annotation of the face image is updated. The specific method is as follows:

[0209] S3.2.4.1. Fuse the spectral feature vector and the texture frequency domain feature vector at the feature level into a comprehensive feature vector :

[0210] c = [s;t] ∈ R M + P ;

[0211] Among them, is dimensions;

[0212] S3.2.4.2. Use a support vector machine to classify the comprehensive feature vector :

[0213] The decision function of the support vector machine is:

[0214] ;

[0215] Among them, is the number of support vectors; is the Lagrange multiplier; is the kernel function is the bias term; is the classification result;

[0216] The classification result includes the category, identity, and clustering label of the face image;

[0217] S3.2.4.3. According to the classification result , update the analysis result to the index annotation of the face image.

[0218] It also includes a security management control unit 4, and the security management control unit 4 is responsible for the overall security and compliance of the system, and provides privacy protection, access control, and anomaly detection functions;

[0219] In this embodiment, the security management control unit 4 includes a permission access control module 41, a privacy protection data encryption module 42, and an anomaly detection audit module 43;

[0220] Among them, the permission access control module 41 is used to perform role-based access control on system users, processes, and application programs, configure fine-grained permission policies, and implement multi-factor authentication measures;

[0221] In this embodiment, the fine-grained permission policy is designed as follows:

[0222] The system access is divided into multiple levels of roles and resource types, and the permission scope of each type of role for resources is clarified:

[0223] Role types:

[0224] Administrator: Has the highest-level permissions, can manage user accounts, assign roles, modify security policies, configure data sources and storage policies;

[0225] Data maintainer: Responsible for data cleaning, annotation, and preprocessing, has write and edit permissions for operations related to data processing, but has no right to modify security policies or access sensitive data in plaintext;

[0226] Analyst: Mainly performs read-only or restricted write operations such as data query, retrieval, feature analysis, and model training;

[0227] Read-only user: Can only perform queries and retrievals on specific data sets, cannot modify or download original image data, nor access sensitive metadata;

[0228] Resource types:

[0229] Data resources: Include original face images, feature data, and index annotation metadata;

[0230] System configuration resources: Include permission policy files, security configuration files, logs, and audit records;

[0231] Management operations: Include adding users, modifying permission policies, creating or deleting indexes;

[0232] The multi-factor authentication measures are designed as follows:

[0233] The multi-factor authentication requires users to provide at least two different types of identity factors when logging in, upgrading permissions, or performing sensitive operations:

[0234] Knowledge factors include passwords, PIN codes;

[0235] Possession factors include mobile phone one-time dynamic verification codes, USB tokens, or security keys;

[0236] Biometric factors include fingerprint, face recognition, or voiceprint verification, but privacy compliance needs to be observed;

[0237] Initial Login and Periodic Verification: When a user logs in for the first time and accesses sensitive data sets again at fixed intervals, after entering the username + password, a one-time verification code needs to be obtained through the mobile App or short message, or a dynamic password generated by a hardware Token is used for secondary verification before entering the system;

[0238] Secondary Verification during Permission Upgrade: When an ordinary analyst attempts to perform advanced analysis operations, the system requires them to additionally provide a one-time token or software and hardware certificate signature;

[0239] Geographical Location and Device Fingerprint: Optionally introduce geographical restrictions and device fingerprint technology; when a user attempts to access data at an abnormal location or on an unrecognized device, the system will trigger the MFA process and may require administrator approval;

[0240] Expiration and Re-authentication: To prevent sessions from remaining active for a long time, after a certain time limit, a secondary factor needs to be entered again to ensure operation only in active and trusted sessions;

[0241] The privacy protection data encryption module 42 is used for encrypting and storing and transmitting portrait data;

[0242] The anomaly detection and auditing module 43 is used to monitor the system operation status in real time, continuously audit and analyze access logs, operation behaviors, and data transmissions, use intrusion detection systems and security information and event management tools to identify potential security threats and abnormal events, and automatically trigger an early warning when an anomaly is detected.

[0243] In this embodiment, the intrusion detection system is a security tool that monitors network or system activities and issues alerts for events that violate security policies or exhibit suspicious behaviors; the security information and event management tool is a system that integrates, analyzes, and correlates security events and log data from multiple sources.

[0244] 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 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.

Claims

1. A portrait data information collection and analysis system based on big data, characterized in that: include: A data acquisition storage unit (1), wherein the data acquisition storage unit (1) uses a multi-threaded data acquisition technology to acquire face images in real time, uses an Elasticsearch distributed database to store the face images, and indexes and annotates the face images; A data image processing unit (2), the data image processing unit (2) being used to pre-process and standardize the portrait image collected by the data collection and storage unit (1); A facial feature analysis unit (3), the facial feature analysis unit (3) is used to extract standardized facial image features, analyze facial image features using spectrum theory combined with frequency domain analysis algorithm, and classify and update index annotations of facial images; A security management control unit (4), wherein the security management control unit (4) is responsible for the overall security and compliance of the system and provides privacy protection, access control and anomaly detection functions.

2. The portrait data information collection and analysis system based on big data according to claim 1 is characterized by: The data acquisition and storage unit (1) comprises a face image acquisition module (11) and a storage and annotation module (12); The face image acquisition module (11) uses a Python multi-threaded programming model to collect image data from a data source in real time, and uses an MTCNN face detection algorithm to identify and capture a face area to obtain a face image; the data source includes a camera, a mobile device, and a recorded video; The storage annotation module (12) stores the face images in the Elasticsearch distributed database and generates an index annotation for each face image, wherein the index annotation includes time, location, device ID and person ID.

3. The portrait data information collection and analysis system based on big data according to claim 1 is characterized by: The data image processing unit (2) comprises an image quality enhancement module (21) and an image normalization module (22); The image quality enhancement module (21) is used to perform smoothing processing on the face image by Gaussian filtering; The image normalization module (22) is used to scale the face image to a predetermined standard size, apply a color standardization algorithm to unify the color distribution of the face image, and generate a final standardized face image.

4. The portrait data information collection and analysis system based on big data according to claim 1 is characterized by: The face feature analysis unit (3) comprises a feature extraction module (31) and a feature analysis index update module (32); The feature extraction module (31) uses the Canny edge detection algorithm to extract facial contour features of the face image, uses the SIFT feature detection algorithm to extract key point distribution features of the face image, and uses the local binary pattern method to extract texture features of the face image; the face image features include facial contour features, key point distribution features and texture features; The feature analysis index updating module (32) uses spectrum graph theory combined with frequency domain analysis algorithm to analyze and classify the features of the face image, and updates the index annotation of the face image.

5. The portrait data information collection and analysis system based on big data according to claim 4 is characterized in that: The feature extraction module (31) uses the Canny edge detection algorithm to extract facial contour features of the face image, uses the SIFT feature detection algorithm to extract key point distribution features of the face image, and uses the local binary pattern method to extract texture features of the face image. The specific method steps are as follows: S3.1.1, Use Canny edge detection algorithm to extract facial contour features of face images: Based on the standardized face image , calculate the image gradient and perform non-maximum suppression and double threshold hysteresis processing to obtain a binary edge map : ； in, For face images The pixel coordinates on ; From the binary edge graph Extract the maximum connected edge point set from , which is the facial contour of the face image; For the The coordinates of the edge points; is the total number of edge points; Fitting a contour point set using an ellipse , solve for the parameters Facial contour features : ； in, is the horizontal coordinate of the center point of the ellipse; is the ordinate of the center point of the ellipse; is the horizontal radius of the ellipse; is the vertical radius of the ellipse; is the rotation angle of the ellipse relative to the horizontal axis; S3.1.2, use SIFT feature detection algorithm to extract key point distribution features of face images: Based on the standardized face image , construct Gaussian pyramid and difference Gaussian pyramid: ； ； in, is a standard Gaussian filter; is a Gaussian pyramid; is the difference Gaussian pyramid; is a constant of the scale interval; is the Gaussian kernel standard deviation of the current scale; Detect the extreme points in the differential Gaussian pyramid as the key points of the face image and obtain the key point set : ； in, For key points The location coordinates of For key points Scale of Calculate key points Distribution of gradient directions in a local neighborhood and assignment of main directions : ； ； ； in, For key points Points in a local neighborhood; is the pixel gradient direction; is the weighting function; is the indicator function; For key points The main direction of At the key point In the neighborhood, the gradient direction and amplitude statistics constitute a descriptor vector: SIFT k =[ h 1 , h 2 ,…, h D s ] ； in, For key points The local descriptor of is the dimension of SIFT descriptor; For key points The count value of the neighborhood gradient direction histogram; Get the key point distribution feature set: ； in, is the total number of key points; is a set of key point distribution features; S3.1.3, Use the local binary pattern method to extract the texture features of the face image: Let the center pixel value be ; The neighborhood pixel value is ; Counting pixels The LBP code is: ； in, Pixel LBP code; is the serial number of the neighborhood pixel; is the total number of neighborhood pixels; For calculation The indicator function of Calculate the LBP code of all pixels in the face image: ； in, The LBP code value is The frequency of occurrence of pixels in the entire face image; is the value of the LBP code, ; is the width of the face image; is the height of the face image; For calculation The indicator function of Get the LBP histogram vector , That is the texture feature: T=[ H LBP (0), H LBP (1),…, H LBP ( 2 N -1)] ； in, For texture features.

6. The portrait data information collection and analysis system based on big data according to claim 5 is characterized by: The feature analysis index update module (32) uses the spectrum theory combined with the frequency domain analysis algorithm to analyze and classify the features of the face image, and updates the index annotation of the face image. The specific method steps are as follows: S3.2.

1. Based on facial contour features And key point distribution feature set , construct weighted undirected graphs using spectral graph theory; S3.2.2, calculate the degree matrix and the adjacency matrix to construct the Laplacian matrix of the graph, and perform eigenvalue decomposition on the Laplacian matrix to extract the spectral eigenvector of the weighted undirected graph; S3.2.

3. Texture features of face images Perform discrete Fourier transform to extract frequency domain features, reduce the dimension of the frequency domain features and filter them to obtain the texture frequency domain feature vector; S3.2.

4. The spectral feature vector and the texture frequency domain feature vector are fused at the feature level into a comprehensive feature vector, and the comprehensive feature vector is classified using a statistical classification method, and the index annotation of the face image is updated according to the classification result.

7. The portrait data information collection and analysis system based on big data according to claim 6 is characterized by: In S3.2.1, based on facial contour features And key point distribution feature set , use spectral graph theory to construct a weighted undirected graph. The specific method is as follows: S3.2.1.

1. Consider the key points of the face image as nodes of the graph and define the node set : ； in, The first node defined for the first keypoint, The second node defined for the second key point, and so on, For the The key point defined by nodes; Node Set ; S3.2.1.

2. Construct a weighted undirected graph based on the Euclidean distance between key points and the SIFT descriptor feature similarity: Defining the adjacency matrix : ； ； in, For Node and nodes The Euclidean distance between For Node SIFT descriptors and nodes The Euclidean distance between SIFT descriptors; To control the distance parameter; To control the feature similarity parameter; Representation Node and nodes degree of relevance; Utilize facial features Obtain the global geometric reference of the face in the image for key point position normalization and graph weight calculation; Normalize the coordinates before . In S3.2.2, the degree matrix and the adjacency matrix are calculated to construct the Laplacian matrix of the graph, and the Laplacian matrix is ​​decomposed by eigenvalues ​​to extract the spectral eigenvector of the weighted undirected graph. The specific method is as follows: S3.2.2.

1. Defining the degree matrix is a diagonal matrix, the degree matrix The diagonal elements of are as follows: ； S3.2.2.2, define the Laplacian matrix: ； in, is the Laplace matrix; S3.2.2.

3. Laplace Matrix Perform eigenvalue decomposition: ； in, is the Laplace matrix The eigenvector of is the Laplace matrix The characteristic value of The eigenvalue Sort by smallest to largest and take the first The smaller non-zero eigenvalues ​​and their corresponding eigenvectors form the spectral eigenvectors: s=[ λ 1 , λ 2 ,…, λ M ] T ； in, is the spectral eigenvector; is the transpose operation.

8. The portrait data information collection and analysis system based on big data according to claim 7 is characterized by: In S3.2.3, the texture features of the face image are Discrete Fourier transform is performed to extract frequency domain features, and the frequency domain features are reduced in dimension and screened to obtain the texture frequency domain feature vector. The specific method is as follows: S3.2.3.1 Texture features of face images Perform discrete Fourier transform to get the spectrum : ； F=[F(0),F(1),…,F( 2 N -1)] ； in, The LBP code value is The frequency of the pixel in the entire face image; is the index of the frequency domain component after discrete Fourier transform; For After the discrete Fourier transform, the frequency index The complex spectrum value obtained at is an imaginary unit; is the rotation factor of discrete Fourier transform; S3.2.3.2 Before extraction The amplitude of the frequency component is taken as the texture frequency domain feature vector : t= |F(0)|,|F(1)|,…,|F(P-1)| T ； in, For spectrum In frequency index The amplitude value at ; and so on, For spectrum In frequency index The amplitude value at ; .

9. The portrait data information collection and analysis system based on big data according to claim 8, characterized in that: In S3.2.4, the spectral feature vector and the texture frequency domain feature vector are fused at the feature level into a comprehensive feature vector, and the comprehensive feature vector is classified using a statistical classification method, and the index annotation of the face image is updated according to the classification result. The specific method is as follows: S3.2.4.

1. Spectral feature vector and texture frequency domain feature vector Perform feature-level fusion into a comprehensive feature vector : c=[s;t]∈ R M+P ； in, for Dimensions; S3.2.4.

2. Use support vector machine to analyze the comprehensive feature vector To classify: The decision function of the support vector machine is: ； in, is the number of support vectors; is the Lagrange multiplier; is the kernel function; is the bias term; is the classification result; Classification results Includes category, identity, and cluster labels of face images; S3.2.4.

3. According to the classification results , update the analysis results to the index annotation of the face image.

10. The portrait data information collection and analysis system based on big data according to claim 1 is characterized by: The security management control unit (4) comprises an authority access control module (41), a privacy protection data encryption module (42) and an anomaly detection audit module (43); The permission access control module (41) is used to perform role-based access control on system users, processes and applications, configure fine-grained permission policies, and implement multi-factor authentication measures; The privacy protection data encryption module (42) is used to encrypt, store and transmit the portrait data; The anomaly detection audit module (43) is used to monitor the system operation status in real time, continuously audit and analyze access logs, operation behaviors and data transmission, use intrusion detection systems and security information and event management tools to identify potential security threats and abnormal events, and automatically trigger an early warning when an anomaly is detected.

Images