Methods and devices for protecting facial data privacy

Abstract

This application provides a method and apparatus for protecting facial data privacy. It constructs a trusted permission management mechanism through a distributed network and smart contracts, and innovatively designs a multi-dimensional feature extraction scheme that integrates geometric and texture features. The system employs data sharding and homomorphic encryption technology to securely process feature vectors, and distributes the encrypted data across a blockchain network, achieving decentralized management of feature data. Feature comparison and identity verification are performed within the encrypted domain, and the entire process is traceable by combining timestamps and operation records. This effectively solves the data security and privacy protection problems in traditional facial recognition systems, providing a secure and reliable technical solution for the field of biometric recognition.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a method and apparatus for protecting facial data privacy. Background Technology

[0002] Traditional facial recognition systems often employ centralized storage and management models, posing data security risks and privacy breaches. Existing technologies have significant shortcomings in data protection and access control, and lack a comprehensive lifecycle security protection mechanism for user biometric data. User privacy data is vulnerable to unauthorized access and malicious attacks.

[0003] Meanwhile, existing systems also have limitations in feature extraction and data encryption. Traditional methods often employ a single feature representation, failing to fully utilize the multidimensional feature information of facial images, and their encryption storage schemes are relatively simple, making it difficult to meet high security requirements. Furthermore, the system lacks an effective decentralized mechanism for data distribution and storage.

[0004] Furthermore, existing technologies need improvement in terms of identity verification and data traceability. The lack of reliable data access control policies and operational auditing mechanisms prevents full-process oversight of the use of signature data. Verification processes are typically conducted in plaintext, posing a risk of data leakage.

[0005] In terms of data sharing and interoperability, existing systems generally suffer from data silos and a lack of mutual trust mechanisms. The absence of unified data exchange standards and trusted consensus mechanisms makes secure data sharing between different organizations difficult. Solving these problems is crucial for building a secure and reliable facial recognition system. Summary of the Invention

[0006] To address the problems in the existing technology, this application provides a method and apparatus for protecting facial data privacy, which can solve the data security and privacy protection problems in traditional facial recognition systems and provide a safe and reliable technical solution for the field of biometric recognition.

[0007] To solve at least one of the above problems, this application provides the following technical solution:

[0008] Firstly, this application provides a method for protecting facial data privacy, including:

[0009] A distributed blockchain network is established, in which multiple nodes are configured and connected through a peer-to-peer communication protocol. A permission management module is built based on smart contracts, and the data access of the nodes is controlled through the permission management module.

[0010] Collect facial image data and extract key point coordinates, construct geometric feature vectors and texture feature vectors, fuse the feature vectors and reduce their dimensionality to obtain target feature vectors, segment and homomorphically encrypt the target feature vectors, package the encrypted feature segments and related operation information to generate blocks and store them in the blockchain distributed network.

[0011] Upon receiving a face verification request, the system performs feature extraction and encrypted fragmentation on the face image to be verified, retrieves the corresponding encrypted feature fragments from the blockchain distributed network, performs fragment reassembly and feature comparison within the ciphertext domain, and generates a new block by writing the verification conclusion and related information into the blockchain distributed network.

[0012] Furthermore, including:

[0013] A blockchain distributed network is established, multiple nodes are configured in the blockchain distributed network, the multiple nodes are connected through a peer-to-peer communication protocol, a unique identity and access key are generated in the multiple nodes, the unique identity and access key are written into a smart contract, a permission management module is built based on the smart contract, and the data access of the multiple nodes is controlled through the permission management module.

[0014] The process involves collecting facial image data, extracting key point coordinates from the facial image data, constructing a geometric feature vector based on the key point coordinates, calculating a texture feature vector from the facial image data, fusing the geometric feature vector and the texture feature vector to obtain a fused feature vector, performing a dimensionality reduction transformation on the fused feature vector to obtain a target feature vector, dividing the target feature vector into multiple data slices, performing homomorphic encryption on each of the multiple data slices to obtain encrypted feature slices, packaging the encrypted feature slices, operation timestamps, and operation records to generate blocks, distributing the blocks to multiple nodes in the blockchain distributed network, recording access operations to the encrypted feature slices based on the smart contract, and executing the access operation after verification by the permission management module.

[0015] The system receives a face verification request, obtains the face image to be verified, repeats the feature extraction and encryption sharding process to obtain the encrypted feature shard to be verified, retrieves the encrypted feature shard from the blockchain distributed network, performs sharding and reassembly and feature comparison on the encrypted feature shard to be verified and the encrypted feature shard in the ciphertext domain, generates a comparison result, determines the verification conclusion based on the comparison result, generates a new block by combining the verification conclusion, verification timestamp and verification operation record, and writes the new block into the blockchain distributed network.

[0016] Secondly, this application provides a facial data privacy protection device, comprising:

[0017] A network construction module is used to establish a blockchain distributed network, in which multiple nodes are configured and connected through a peer-to-peer communication protocol. A permission management module is built based on smart contracts to control the data access of nodes.

[0018] The image processing module is used to acquire face image data and extract key point coordinates, construct geometric feature vectors and texture feature vectors, fuse the feature vectors and reduce their dimensionality to obtain target feature vectors, segment and homomorphically encrypt the target feature vectors, package the encrypted feature segments and related operation information to generate blocks and store them in the blockchain distributed network.

[0019] The privacy protection module is used to receive face verification requests, perform feature extraction and encrypted fragmentation on the face image to be verified, retrieve the corresponding encrypted feature fragments from the blockchain distributed network, perform fragment reorganization and feature comparison in the ciphertext domain, and generate a new block by writing the verification conclusion and related information into the blockchain distributed network.

[0020] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the facial data privacy protection method.

[0021] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned face data privacy protection method.

[0022] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the aforementioned face data privacy protection method.

[0023] As can be seen from the above technical solution, this application provides a method and apparatus for protecting facial data privacy. It constructs a trusted permission management mechanism through a distributed network and smart contracts, and innovatively designs a multi-dimensional feature extraction scheme that integrates geometric and texture features. The system employs data sharding and homomorphic encryption technology to securely process feature vectors, and distributes the encrypted data in a distributed manner within a blockchain network, achieving decentralized management of feature data. Feature comparison and identity verification are performed within the encrypted domain, and the entire process is traceable by combining timestamps and operation records. This effectively solves the data security and privacy protection problems in traditional facial recognition systems, providing a secure and reliable technical solution for the field of biometric recognition. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is one of the flowcharts illustrating the face data privacy protection method in the embodiments of this application;

[0026] Figure 2 This is the second flowchart illustrating the face data privacy protection method in this application embodiment;

[0027] Figure 3 This is the third flowchart illustrating the face data privacy protection method in this application embodiment;

[0028] Figure 4 This is the fourth flowchart illustrating the face data privacy protection method in the embodiments of this application;

[0029] Figure 5 This is the fifth flowchart illustrating the face data privacy protection method in the embodiments of this application;

[0030] Figure 6 This is the sixth flowchart illustrating the face data privacy protection method in the embodiments of this application;

[0031] Figure 7 This is the seventh flowchart illustrating the face data privacy protection method in the embodiments of this application;

[0032] Figure 8 This is a structural diagram of the face data privacy protection device in the embodiments of this application;

[0033] Figure 9 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0034] Figure label:

[0035] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0038] To address the problems existing in current technologies, this application provides a method and apparatus for protecting facial data privacy. It constructs a trusted access control mechanism through a distributed network and smart contracts, and innovatively designs a multi-dimensional feature extraction scheme that integrates geometric and texture features. The system employs data sharding and homomorphic encryption techniques to securely process feature vectors, and distributes the encrypted data across a blockchain network, achieving decentralized management of feature data. Feature comparison and identity verification are performed within the encrypted domain, and the entire process is traceable by combining timestamps and operation records. This effectively solves the data security and privacy protection problems in traditional facial recognition systems, providing a secure and reliable technical solution for the field of biometric recognition.

[0039] To address the data security and privacy protection issues in traditional facial recognition systems and provide a secure and reliable technical solution for the field of biometric identification, this application provides an embodiment of a facial data privacy protection method, see [link to embodiment]. Figure 1 The facial data privacy protection method specifically includes the following:

[0040] Step S101: Establish a blockchain distributed network, configure multiple nodes in the network and connect them through a peer-to-peer communication protocol, build a permission management module based on smart contracts, and control the data access of the nodes through the permission management module;

[0041] Specifically, this includes: establishing a blockchain distributed network, configuring multiple nodes in the blockchain distributed network, connecting the multiple nodes through a peer-to-peer communication protocol, generating unique identifiers and access keys in the multiple nodes, writing the unique identifiers and access keys into a smart contract, constructing a permission management module based on the smart contract, and controlling data access of the multiple nodes through the permission management module;

[0042] Optionally, this embodiment employs a consortium blockchain architecture to construct a distributed network. The network topology adopts a fully connected structure, with each node establishing direct point-to-point connections with other nodes. Node communication is based on the libp2p protocol framework, implementing multiplexing, flow control, and congestion control mechanisms. To optimize network performance, a persistent connection pool is established between nodes, maintaining a stable communication channel through long TCP connections. During node discovery, new nodes obtain network topology information by bootstrapping nodes, then establish a neighbor node list based on the ping-pong mechanism, and periodically update node status and routing tables.

[0043] This embodiment employs a multi-layered cryptographic mechanism for node identity generation. First, a hardware fingerprinting module collects characteristic information such as the device's CPU ID, MAC address, and hard drive serial number. This information is then used to calculate a unique device identifier using the SHA3-256 algorithm. Next, an asymmetric key pair is generated based on the secp256k1 elliptic curve algorithm. The private key is used for signature operations, and the public key is used to generate a node address using the Keccak-256 hash algorithm. Finally, the device identifier and node address are combined, and an access key is generated using the PBKDF2 key derivation function to achieve identity authentication and data encryption.

[0044] This embodiment implements a complete permission management system at the smart contract level. The contracts adopt a layered design: the bottom layer is responsible for identity registration and key management; the middle layer implements permission rules and policy configuration; and the top layer provides permission verification and access control interfaces. Contract states use a mapping data structure to store node information, including attributes such as node type, permission level, and validity period. Permission rules are described using a state machine model, supporting complex combinations of conditions and state transitions.

[0045] This embodiment designs a permission management architecture based on the RBAC (Role-Based Access Control) model. First, a role hierarchy is defined, including super administrators, zone administrators, ordinary nodes, and read-only nodes. Each role is associated with a set of permissions, with permission granularity refined to the specific data operation level. Role assignment employs a multi-signature mechanism, requiring the joint signatures of administrator nodes reaching a preset threshold for it to take effect. Permission inheritance relationships are represented by a directed acyclic graph, supporting flexible permission transfer and delegation.

[0046] This embodiment implements dynamic access control policies in permission control. The policy rules encompass multiple dimensions: a time dimension controlling access periods, a spatial dimension restricting access to geographical locations, and a frequency dimension constraining the number of operations. The rule engine employs a forward inference mechanism to evaluate in real-time whether access requests meet policy requirements. Policy configurations support dynamic updates; management nodes can modify rule parameters through a smart contract interface, and changes take effect immediately.

[0047] This embodiment employs a layered encryption scheme for inter-node communication. The network layer uses the TLS 1.3 protocol to establish a secure channel, supporting perfect forward secrecy. The transport layer uses the ChaCha20-Poly1305 authentication encryption algorithm to provide high-performance encryption and authentication services. The application layer encrypts messages based on the shared access key between nodes and uses ECDH key negotiation to generate session keys. Messages include digital signatures and incrementing nonce values ​​to prevent replay attacks.

[0048] This embodiment constructs a distributed auditing system. Each node maintains a local audit log, recording detailed operation information, including the requester, operation type, timestamp, accessed resources, and execution result. Log entries are organized using a MerklePatricia Trie to ensure data immutability. Audit data is periodically synchronized to the blockchain network, forming a globally consistent audit history. The contract provides multi-dimensional query interfaces to support complex audit analysis.

[0049] This embodiment implements a complete fault tolerance mechanism. Node status monitoring uses the Gossip protocol for propagation and detects node liveness through heartbeat messages. When a node failure is detected, an automatic failover process is triggered: first, the faulty node is isolated and the network topology is updated; then, the backup node is activated and the connection relationship is rebuilt; finally, permission data is synchronized and the service status is restored. The entire process ensures network consistency through a consensus mechanism.

[0050] This embodiment implements fine-grained control over permission delegation. Delegation operations are executed through smart contracts, supporting partial delegation of permissions and time-limited permissions. The delegation chain is maintained through a tree structure to control delegation depth and prevent permission proliferation. Delegation status is updated to the blockchain in real time, ensuring that all nodes receive the latest permission information. This solution achieves secure and controllable permission management, providing a trusted data access environment for facial recognition systems.

[0051] Step S102: Collect face image data and extract key point coordinates, construct geometric feature vectors and texture feature vectors, fuse the feature vectors and reduce their dimensionality to obtain target feature vectors, perform fragmentation and homomorphic encryption on the target feature vectors, package the encrypted feature fragments and related operation information to generate blocks and store them in the blockchain distributed network.

[0052] Specifically, the process includes: collecting facial image data, extracting key point coordinates from the facial image data, constructing a geometric feature vector based on the key point coordinates, calculating a texture feature vector from the facial image data, fusing the geometric feature vector and the texture feature vector to obtain a fused feature vector, performing a dimensionality reduction transformation on the fused feature vector to obtain a target feature vector, dividing the target feature vector into multiple data slices, performing homomorphic encryption on each of the multiple data slices to obtain encrypted feature slices, packaging the encrypted feature slices, operation timestamps, and operation records to generate blocks, distributing the blocks to multiple nodes in the blockchain distributed network, recording access operations to the encrypted feature slices based on the smart contract, and executing the access operation after verification by the permission management module.

[0053] Optionally, this embodiment employs a multispectral image acquisition scheme, configuring a dual-camera array for visible light and near-infrared light to achieve all-weather face image acquisition. The camera array achieves spatial alignment through precise calibration, and real-time exposure control and white balance adjustment are performed during the acquisition process. The image preprocessing module first performs illumination compensation, uses an adaptive histogram equalization algorithm to enhance image contrast, and then performs multi-scale decomposition through Gaussian pyramids to construct an image pyramid model that supports feature extraction at different resolutions.

[0054] This embodiment employs an improved cascaded shape regression algorithm for keypoint localization. First, the face region is located using a Haar feature detector. Then, a multi-layer cascaded regressor is used to progressively optimize the keypoint positions. The regressor uses a gradient boosting tree structure, with each layer containing multiple weak classifiers, which are combined to improve localization accuracy. The model training uses a constrained objective function to ensure that the keypoint distribution conforms to facial anatomical features. The final output is the coordinates of 68 standard facial keypoints.

[0055] This embodiment constructs multi-dimensional geometric features based on keypoint coordinates. First, the Euclidean distance matrix between keypoints is calculated to capture the relative positional relationships of facial organs. Then, triangle features are extracted, and keypoints are selected to form Delaunay triangulations. The area ratio and angle ratio of the triangles are calculated, and these features are highly robust to changes in facial pose. Simultaneously, the curvature features of the keypoints are calculated to describe the local geometric properties of the facial contour.

[0056] This embodiment employs a multi-scale Gabor filter bank for texture feature extraction. The filter parameters cover 8 directions and 5 scales, forming 40 filter kernels. After convolution operations on the image, amplitude and phase information are extracted to construct a feature map. Local Binary Pattern (LBP) features are calculated on the feature map to capture local variation patterns of the texture. Then, histogram features are generated through block-based statistics, and finally, principal component analysis is used to reduce the dimensionality and obtain the texture feature vector.

[0057] This embodiment designs an adaptive feature fusion algorithm. First, geometric and texture features are standardized to eliminate dimensional differences. Then, the feature discriminative power is calculated based on the Fisher criterion, and feature weight coefficients are designed. A weighted concatenation method is used to combine the two types of features into a high-dimensional feature vector. To improve the expressive power of the features, a kernel function is introduced to map the features to a high-dimensional space, enhancing the nonlinear expressive power of the features.

[0058] This embodiment achieves feature dimensionality reduction through local-preserving projection. First, a nearest-neighbor graph is constructed to describe the local structure between samples, and edge weights are calculated based on a hot kernel function. Then, the generalized eigenvalue problem is solved to obtain the optimal projection matrix. The projection process preserves the local geometric structure of the samples while simultaneously reducing dimensionality. Finally, a fixed-dimensional target feature vector is obtained, facilitating subsequent piecewise processing and encrypted storage.

[0059] This embodiment employs a threshold secret sharing scheme for feature sharding. First, the target feature vector is divided into multiple shards of fixed length, and each shard is encrypted independently. The Shamir threshold scheme distributes the secret among multiple holders; a sufficient number of shards are needed to reconstruct the original data. Random perturbations are introduced during the sharding process to increase data security.

[0060] This embodiment employs an improved homomorphic encryption algorithm in feature encryption. A homomorphic encryption system is constructed based on the BGV scheme, supporting addition and multiplication operations on the ciphertext field. Key generation utilizes a ring learning problem to ensure encryption strength. Noise terms are introduced into the encryption process to prevent statistical analysis attacks. Each feature fragment is encrypted independently, generating corresponding ciphertext data.

[0061] This embodiment designs a complete block packaging mechanism. Cryptographic feature fragments, timestamps, and operation records are organized into transaction data, and the Merkle tree root hash of the data is calculated as the block identifier. The block header also includes fields such as version number and difficulty target. The block body adopts a compact encoding format to optimize storage space. Finally, a consensus algorithm synchronizes the block to multiple nodes in the distributed network.

[0062] This embodiment implements access control through smart contracts. The contract defines the access rules for feature fragments, including read permissions, usage limits, and timeliness restrictions. Access requests require permission verification, and the contract records access logs and updates usage counts. This solution achieves secure storage and controllable access to facial feature data, providing effective protection for privacy.

[0063] Step S103: Receive a face verification request, perform feature extraction and encrypted fragmentation on the face image to be verified, retrieve the corresponding encrypted feature fragment from the blockchain distributed network, perform fragment reorganization and feature comparison in the ciphertext domain, and generate a new block by generating the verification conclusion and related information and write it into the blockchain distributed network.

[0064] Specifically, the process includes: receiving a face verification request, obtaining a face image to be verified, repeatedly executing the feature extraction and encryption sharding process to obtain an encrypted feature shard to be verified, retrieving the encrypted feature shard from the blockchain distributed network, performing sharding and feature comparison on the encrypted feature shard to be verified and the encrypted feature shard in the ciphertext domain, generating a comparison result, determining a verification conclusion based on the comparison result, generating a new block by combining the verification conclusion, verification timestamp, and verification operation record, and writing the new block into the blockchain distributed network.

[0065] Optionally, this embodiment employs a multi-queue priority scheduling mechanism in the verification request processing. The request queues are divided into high-priority and normal-priority queues based on urgency, and a token bucket algorithm is used to control the request processing rate. The verification request includes a timestamp, location information, and access credentials, and the requester's identity is verified using zero-knowledge proofs. The request parsing module extracts verification parameters, including the target identity identifier and verification scenario type, to ensure request integrity.

[0066] This embodiment designs an adaptive image acquisition controller. Camera parameters, including exposure time, gain, and aperture, are dynamically adjusted based on ambient light intensity. A motion detection algorithm is used to identify the stable state of the face region, triggering image acquisition at the optimal time. The acquired image sequence is quality-assessed, and the image with the best sharpness and pose is selected as the verification sample. The image enhancement module improves image quality through adaptive gamma correction and noise reduction.

[0067] This embodiment parallelizes the computational tasks during feature extraction. First, the face image is divided into multiple overlapping local regions, and feature extraction is performed independently in each region. Geometric feature extraction and texture feature extraction are executed in parallel across different processing units, with processing efficiency optimized through a data pipeline. The feature fusion stage employs an attention mechanism, dynamically adjusting feature weights based on the discriminative power of different regions.

[0068] This embodiment employs an improved data sharding strategy. Based on correlation analysis of feature vectors, highly correlated feature elements are assigned to the same shards, reducing information redundancy between shards. A checksum is added to each shard to support data integrity verification. Random permutation is introduced during the sharding process to increase the uniformity of data distribution. Finally, each shard is independently homomorphically encrypted to generate encrypted feature shards to be verified.

[0069] This embodiment employs a distributed caching mechanism for blockchain data retrieval. A multi-level caching structure is constructed, including in-memory caching and persistent storage. The caching strategy is based on access frequency and temporal locality, managing cached content through the LRU algorithm. Retrieval requests first search the local cache; if no match is found, data is then retrieved from the blockchain network. Network requests are made asynchronously to avoid blocking the verification process.

[0070] This embodiment implements a ciphertext domain feature comparison algorithm. First, based on the properties of homomorphic encryption, the Euclidean distance between encrypted feature fragments is calculated. The comparison results of the distance values ​​are obtained through partial homomorphic decryption, without needing to completely decrypt the original data. The feature reconstruction process employs a secure multi-party computation protocol to ensure that intermediate results do not leak original information. Finally, the similarity score is recovered through threshold decryption.

[0071] This embodiment designs an adaptive threshold decision mechanism. A base threshold is set based on the security level of the verification scenario, and a dynamic threshold is updated using an exponential moving average. The decision process considers multiple indicators, including feature similarity, confidence level, and quality score. A fuzzy inference system is used to fuse multiple indicators to generate the final verification conclusion. The system supports multi-level verification results, including confirmed match, suspected match, and no match.

[0072] This embodiment achieves efficient data organization during block generation. Verification conclusions are packaged together with detailed verification process information, including feature comparison scores, threshold settings, and decision criteria. Compression encoding is used to reduce block size while retaining key audit information. The block header includes a timestamp, parent block hash, and Merkle root to ensure data temporality and integrity.

[0073] This embodiment employs an improved consensus mechanism for block writing. A consensus protocol is built based on a practical Byzantine fault-tolerant algorithm, supporting consistency in asynchronous network environments. The consensus process consists of three phases: pre-preparation, preparation, and confirmation, ensuring block validity through multiple rounds of voting. After block confirmation, it is synchronized to all network nodes via a broadcast mechanism, guaranteeing data consistency. This scheme provides end-to-end security protection for face verification and supports large-scale distributed deployment scenarios.

[0074] As described above, the face data privacy protection method provided in this application can construct a trusted permission management mechanism through a distributed network and smart contracts, and innovatively designs a multi-dimensional feature extraction scheme that integrates geometric and texture features. The system employs data sharding and homomorphic encryption technology to securely process feature vectors, and distributes the encrypted data across a blockchain network, achieving decentralized management of feature data. Feature comparison and identity verification are performed within the encrypted domain, and the entire process is traceable by combining timestamps and operation records, effectively solving the data security and privacy protection problems in traditional face recognition systems, and providing a secure and reliable technical solution for the field of biometric recognition.

[0075] In one embodiment of the face data privacy protection method of this application, see [link to relevant documentation]. Figure 2 It can also specifically include the following:

[0076] Step S201: Use Gabor wavelet transform to decompose the face image data at multiple scales, extract texture information of the image at different directions and frequencies, map the texture information to the feature space to construct a texture feature matrix, perform dimensionality reduction processing on the texture feature matrix through principal component analysis to obtain texture feature vectors, use a deep convolutional neural network to locate the coordinates of facial key points in the face image data, and calculate the distance and angle values ​​between facial organs based on the coordinates of the facial key points to construct geometric feature vectors;

[0077] Step S202: Normalize the texture feature vector and the geometric feature vector, use a weighted fusion algorithm to combine the normalized texture feature vector and geometric feature vector into a high-dimensional feature matrix, and use an adaptive feature selection algorithm to select feature components with strong discriminative power from the high-dimensional feature matrix, and construct the selected feature components into a fused feature vector.

[0078] Optionally, this embodiment employs an adaptive parameter configuration strategy in the Gabor wavelet transform. The filter bank comprises 5 scales and 8 directions. The scale parameters are dynamically adjusted based on the image resolution, while the direction parameters uniformly cover the range from 0 to π. The center frequency and bandwidth of each filter kernel are optimized through grid search to maximize the discriminative power of texture features. Convolution operations are accelerated using Fast Fourier Transform, and image boundaries are processed using an overlap-addition method to ensure the continuity of feature extraction.

[0079] This embodiment designs a multi-level texture feature extraction framework. First, local statistical features, including mean, variance, skewness, and kurtosis, are calculated on the Gabor response map. Then, a spatial pyramid structure is constructed, and gradient direction histograms are extracted at different resolutions. Feature mapping uses kernel functions to project nonlinear features into a high-dimensional feature space, enhancing the expressive power of the features. Finally, features from different levels are concatenated to form a texture feature matrix.

[0080] This embodiment implements an incremental learning mechanism in principal component analysis. The feature matrix is ​​first zero-mean and whitened to improve the statistical properties of the data. The principal eigenvectors are calculated using a modified power iteration method, and the eigenvalue energy threshold is adaptively determined. The dimensionality reduction process retains principal components whose cumulative contribution rate reaches a preset threshold, ensuring controllable information loss. The dimensionality-reduced eigenvectors preserve the geometric structure of the feature space through orthogonal transformations.

[0081] This embodiment constructs a hierarchical deep convolutional neural network. The network backbone adopts a residual structure, and skip connections are used to alleviate the gradient vanishing problem. The feature extraction layer uses dilated convolutions to expand the receptive field, and a feature pyramid network is used for multi-scale feature fusion. The keypoint detection head is designed with a multi-task learning framework, simultaneously predicting keypoint heatmaps and displacement vectors. The network training adopts a soft and hard example mining strategy, dynamically adjusting the sample weights.

[0082] This embodiment designs a geometric feature extraction module based on anatomical features. First, the centroid positions of facial organs are calculated, constructing a standardized coordinate system. Then, multiple sets of geometric measurements are extracted, including absolute distances such as interocular distance, nose length, and mouth width, as well as angular features such as eye angle, nasal bridge angle, and jaw angle. Feature calculation considers changes in facial pose, achieving pose normalization through projection transformation. Simultaneously, contour curvature features are extracted to describe the geometric characteristics of the facial contour.

[0083] This embodiment employs a robust standardization method for feature normalization. For texture features, an improved Z-score standardization is used, replacing the mean and standard deviation with the median and interquartile range to reduce the impact of outliers. Geometric features are normalized using min-max normalization, with interval boundaries determined by kernel density estimation. The standardization parameters are updated via a sliding window to adapt to dynamic changes in the feature distribution.

[0084] This embodiment implements an adaptive feature weighted fusion algorithm. First, the discriminative power of each feature component is evaluated using the Fisher discriminant criterion, and inter-class dispersion and intra-class clustering are calculated. The weight coefficients are optimized using the Lagrange multiplier method to maximize the discriminative power of the features. The fusion process employs a weighted concatenation method to generate a high-dimensional feature matrix. The matrix structure contains local and global feature information, supporting multi-scale feature representation.

[0085] This embodiment designs a dynamic feature selection mechanism. A feature importance score is constructed based on the mutual information criterion, and a feature subset is iteratively selected using a sequential forward selection strategy. The selection process considers the redundancy between features, balancing the discriminativeness and independence of features through the minimum redundancy and maximum correlation criterion. The feature selection results are dynamically updated according to the sample distribution, maintaining the optimality of the feature components.

[0086] This embodiment generates a fused feature vector through feature component reconstruction. The reconstruction process preserves the spatiotemporal structure of the features and adjusts the contributions of different feature domains through an attention mechanism. The vector dimension is determined through cross-validation, achieving a balance between feature expressive power and computational efficiency. This scheme achieves efficient extraction and optimized representation of facial features, providing a reliable foundation for subsequent feature encryption and comparison.

[0087] In one embodiment of the face data privacy protection method of this application, see [link to relevant documentation]. Figure 3 It can also specifically include the following:

[0088] Step S301: The fused feature vector is nonlinearly reduced in dimensionality using a local linear embedding algorithm to construct a local linear relationship matrix between feature points. The optimal projection direction is calculated based on the local linear relationship matrix. The fused feature vector is projected onto a low-dimensional space to obtain the target feature vector. The target feature vector is segmented according to a preset segment length to generate a fixed-length data segment sequence. A unique identifier is assigned to each data segment and the segment position information is recorded.

[0089] Step S302: Generate a public key pair and a private key pair based on the homomorphic encryption algorithm, perform encryption operation on each data fragment in the data fragment sequence using the public key, combine the encrypted data fragment with the corresponding fragment identifier and location information to form an encrypted feature fragment, construct a fragment index table to record the association between the encrypted feature fragments, and write the fragment index table into the smart contract.

[0090] Optionally, this embodiment employs an adaptive neighborhood construction strategy in the local linear embedding algorithm. First, a fast nearest neighbor search is used to determine the local neighborhood of each feature point, with the neighborhood size dynamically adjusted based on the local data density. Neighborhood relationships are described using a hot kernel weighting function, where the weight coefficients decay exponentially with distance. Threshold pruning is used when constructing the sparse weight matrix to preserve significant local connectivity relationships, improving computational efficiency and robustness.

[0091] This embodiment implements improved local linear relationship modeling. For each feature point, the reconstructed weights are obtained by solving a constrained least squares problem. The constraints include a weight sum of 1 and preservation of local geometry. The optimization process employs the augmented Lagrangian method, obtaining the optimal weights through iterative convergence. The reconstruction error is used as a regularization term to balance local preservation and global structure. Sparsity of the weight matrix is ​​achieved through a soft thresholding strategy to improve storage efficiency.

[0092] This embodiment designs an efficient feature projection optimization framework. A global covariance structure is constructed based on the reconstructed weight matrix, and the optimal projection direction is solved through generalized eigenvalue decomposition. The eigenvalue decomposition employs the Lanzos iterative method, significantly improving the efficiency of large-scale data processing. The projection dimension is determined by accumulating the reconstruction error, ensuring that the features retain sufficient discriminative information after dimensionality reduction. Orthogonalization of the projection matrix ensures the orthogonality of the feature space.

[0093] This embodiment employs a hierarchical data sharding strategy. First, the target feature vector is normalized to ensure uniform numerical distribution. The shard length is determined through feature correlation analysis, with highly correlated feature elements assigned to the same shard. Redundant checksums are added to each shard to support data integrity verification. Random permutation is introduced during the sharding process to increase the uniformity of data distribution.

[0094] This embodiment implements a traceable fragment identification mechanism. The identifier is generated by combining a timestamp and a random number, ensuring global uniqueness. Location information includes the fragment's starting position and length in the original vector, and compression encoding is used to reduce storage overhead. Identification information is linked through hash chains to construct an association structure, supporting fragment tracing and integrity verification.

[0095] This embodiment employs an improved BGV scheme in homomorphic encryption. Key generation is based on the ring learning problem, constructing public-private key pairs through random multinomials. The public key contains multiple components, supporting batch encryption. The private key is stored in shards and distributed to multiple authorized nodes using a threshold scheme. Key updates are implemented through a blinding factor, with periodic rotation enhancing security.

[0096] This embodiment employs a fragment-level parallel encryption mechanism. Each data fragment is encrypted independently, supporting multi-threaded parallel execution. Random noise is introduced into the encryption process to prevent statistical analysis attacks. Ciphertext compression is achieved through analog-to-digital switching, optimizing storage space. The encrypted feature fragment contains ciphertext data, fragment identifiers, and location information, using a unified serialization format.

[0097] This embodiment constructs a highly efficient sharded index structure. The index table is organized using skip lists, supporting fast range queries and insertion operations. Index entries contain shard identifiers, location information, and relationships, and are stored using a prefix tree for compressed storage. Index updates employ optimistic concurrency control to improve concurrent access efficiency.

[0098] This embodiment implements index management through smart contracts. The contract defines the interfaces for adding, deleting, modifying, and querying the index, and access permissions are controlled by roles. Index operations generate event logs, recording the operation type and execution result. The contract state is synchronized to the blockchain network to ensure the consistency of the index data. This scheme achieves secure sharded storage of feature data and supports efficient data retrieval and access control.

[0099] The technical solution in this embodiment preserves the essential structure of features through nonlinear dimensionality reduction, ensures data security through fragmented storage and homomorphic encryption, and enables trusted index management through smart contracts. The overall design meets the efficiency and security requirements of facial recognition systems for feature processing.

[0100] In one embodiment of the face data privacy protection method of this application, see [link to relevant documentation]. Figure 4 It can also specifically include the following:

[0101] Step S401: Combine the encrypted feature fragments with the operation timestamp to form a data block, calculate the hash value of the data block as the block header, write the operation record information into the block body, construct the association structure between the block header and the block body based on the Merkle tree algorithm, use a consensus algorithm to verify the block among multiple nodes, generate blockchain connection information, write the verified block into the blockchain distributed network, and distribute the block to different nodes according to a preset distribution strategy;

[0102] Step S402: Set access control rules in the smart contract, construct an access operation log structure to record data access request information, verify the legality of the access request based on the zero-knowledge proof algorithm, match the access request information with the authorization information of the corresponding node, generate access credentials and set the access permission validity period, and send the access credentials to the node that initiated the access request.

[0103] Optionally, this embodiment employs a hierarchical data block organization structure. The data block header includes a version number, timestamp, previous block hash, and difficulty target. The timestamp is accurate to the millisecond and synchronized via a network time protocol. The data blocks use a compact serialization format to optimize storage efficiency. The block body contains indexes and operation records for multiple cryptographic feature fragments, recording a detailed data processing history.

[0104] This embodiment implements an improved Merkle tree algorithm in block construction. First, the block data is hashed hierarchically. Leaf nodes store the hash value of the original data, while intermediate nodes are calculated using the hashes of their child nodes. The hash function used is SHA3-256, providing stronger collision resistance. The tree structure employs a balanced optimization, maintaining tree height balance through rotation operations. The root node hash serves as the unique identifier for the block, supporting rapid data verification.

[0105] This embodiment designs an efficient consensus mechanism. A consensus framework is built based on a practical Byzantine fault-tolerant algorithm, supporting asynchronous network environments. The consensus process consists of three phases: pre-preparation, preparation, and commit. Nodes confirm the validity of blocks through voting, requiring a consensus of more than two-thirds of the nodes. Byzantine fault tolerance ensures that the system can still operate normally even if some nodes fail or act maliciously.

[0106] This embodiment employs a dynamic block distribution strategy. Blocks are mapped to a node ring based on a consistent hashing algorithm, ensuring even data distribution. Each block has a replication factor set according to its importance, determining the number of backups. Node selection considers geographical location and network latency to optimize access performance. Data synchronization uses the gossip protocol to ensure eventual network consistency.

[0107] This embodiment implements fine-grained access control within a smart contract. The contract defines a multi-layered permission model, including administrator permissions, operation permissions, and read-only permissions. Permission rules support time limits and operation count limits. Access policies are described using a state machine, supporting complex combinations of conditions. Contract state changes are communicated to relevant nodes via an event mechanism.

[0108] This embodiment constructs a complete access log system. The log structure includes request time, requester identity, operation type, and accessed resource. Log entries are organized in a chain structure to ensure the immutability of records. Log queries support multi-dimensional filtering for easy audit analysis. Logs are synchronized to a blockchain network, forming a distributed audit trail.

[0109] This embodiment employs an improved Schnorr protocol in zero-knowledge proofs. First, the prover's identity commitment is constructed, including the public key and a random challenge. The verification process is completed through interactive proofs, without disclosing the original credentials. The proof structure uses a compressed format to reduce communication overhead. The zero-knowledge nature ensures that sensitive information is not disclosed during the verification process.

[0110] This embodiment employs a secure authorization mechanism. Authorization information includes the node's public key, permission level, and validity period. The authorization process utilizes a multi-signature scheme, requiring a sufficient number of authorized nodes to sign for it to take effect. Authorization status is managed through smart contracts, supporting the revocation and updating of authorizations. Status changes trigger event notifications to ensure timely synchronization of authorization information.

[0111] This embodiment implements efficient access credential management. Credentials are in JSON Web Token format and include permission statements and signature information. A random factor is incorporated into the credential generation process to prevent replay attacks. The validity period uses a sliding window mechanism to support automatic renewal. Credential revocation is implemented through a blacklist to ensure security.

[0112] This embodiment achieves secure data storage and access control through a distributed network. Blockchain technology ensures data immutability, smart contracts enable flexible permission management, and zero-knowledge proofs ensure secure identity verification. This solution provides a reliable data protection mechanism for facial recognition systems.

[0113] In one embodiment of the face data privacy protection method of this application, see [link to relevant documentation]. Figure 5It can also specifically include the following:

[0114] Step S501: Receive the face verification request sent by the receiving node, parse the target identity information and access credentials in the verification request, verify the validity of the access credentials, call the image acquisition module to obtain the face image to be verified, perform illumination compensation and pose correction on the face image to be verified, and standardize the corrected face image according to a preset resolution.

[0115] Step S502: Perform feature extraction on the standardized face image to be verified. Use the same algorithm configuration as the original feature extraction to extract texture feature vectors and geometric feature vectors. After fusing the feature vectors, perform dimensionality reduction transformation to obtain the target feature vector. Segment the target feature vector based on a preset segmentation strategy. Use the public key to perform encryption operation on the segmented data segments to obtain the encrypted feature segments to be verified.

[0116] Optionally, this embodiment employs a layered authentication mechanism in the verification request processing. First, a token parser extracts the identity identifier and access credentials from the verification request. The credential format adopts the JWT standard, including a permission statement and a digital signature. The authentication process is divided into two stages: identity authentication and permission verification. Asymmetric encryption is used to verify the signature validity, ensuring the request source is trustworthy. Permission verification is based on the RBAC model, combined with timestamps to check the timeliness of access permissions.

[0117] This embodiment features an adaptive image acquisition controller. The acquisition module is equipped with a multispectral camera array, including visible light and near-infrared sensors, supporting image acquisition under various lighting conditions. Camera parameters, including exposure time, gain, and aperture, are adjusted in real time based on feedback. The image quality evaluation module selects the optimal image frames based on metrics such as sharpness, contrast, and facial pose.

[0118] This embodiment implements a multi-stage illumination compensation algorithm. First, adaptive histogram equalization enhances image contrast, with weighting coefficients dynamically adjusted based on local brightness distribution. Then, an improved Retinex algorithm is used to separate illumination and reflection components, and illumination estimation is optimized through multi-scale decomposition. Finally, gamma correction adjusts the image dynamic range to ensure clear visibility of facial details.

[0119] This embodiment employs a 3D model-assisted method for pose correction. First, a deep learning model estimates the 3D pose parameters of the face, including Euler angles and translation vectors. Then, a facial mesh in a standard pose is generated based on a 3D deformation model, and perspective projection establishes the correspondence between the 2D image and the 3D model. Finally, thin-plate spline interpolation is used to reproject the image, generating a frontal view of the face.

[0120] This embodiment designs an adaptive image normalization processing flow. Image scaling employs a bicubic interpolation algorithm to maintain a smooth transition of edge details. Resolution normalization considers the needs of subsequent feature extraction, ensuring consistency in feature description. Image alignment is based on eye corner calibration points, achieving geometric normalization through affine transformation. The preprocessed image meets the input requirements of the feature extraction module.

[0121] This embodiment implements a parallel computing framework for feature extraction. The extraction processes for texture and geometric features are executed in parallel on independent processing units. Texture features are extracted using a Gabor filter bank to extract multi-scale features, with filter parameters consistent with those of the original feature extraction. Geometric features are calculated based on facial key points, including distance ratios and angle ratios, ensuring pose invariance of the features.

[0122] This embodiment employs a dynamic feature fusion strategy. The feature vectors are first standardized to eliminate scale differences between different feature domains. The fusion weights are dynamically adjusted based on feature discriminative capabilities, taking into account the characteristics of the current verification scenario. Feature combination uses a weighted concatenation method to maintain the structural integrity of the features. The fused feature vectors undergo local preserving projection dimensionality reduction to maintain the local geometric structure of the samples.

[0123] This embodiment implements a consistent feature sharding mechanism. The sharding strategy remains consistent with the original feature processing, ensuring the comparability of encrypted features. The sharding process considers feature correlation, assigning highly correlated feature elements to the same shard. Verification information is added to each shard to support integrity verification. Random perturbations are introduced into the sharding process to enhance security.

[0124] This embodiment employs optimized homomorphic encryption in feature encryption. The encryption process uses the system public key, supporting feature comparison across the ciphertext field. The encryption operation utilizes a parallel processing architecture to improve processing efficiency. Ciphertext compression is achieved through analog-to-digital switching, optimizing storage overhead. The format of the encrypted feature fragments remains consistent with the original encrypted feature, facilitating subsequent feature comparison.

[0125] This embodiment achieves feature preparation for face verification through a standardized processing flow. The entire process, from image acquisition to feature encryption, maintains consistency with the registration phase, ensuring feature comparability. This scheme supports efficient face recognition verification while protecting the security of feature data.

[0126] In one embodiment of the face data privacy protection method of this application, see [link to relevant documentation]. Figure 6 It can also specifically include the following:

[0127] Step S601: Retrieve encrypted feature fragments from the blockchain distributed network according to the fragment index information recorded in the smart contract, verify the integrity and validity of the encrypted feature fragments, sort the retrieved encrypted feature fragments based on the fragment identifier and location information, reassemble the sorted encrypted feature fragments according to preset rules, and calculate the Euclidean distance between the encrypted feature fragment to be verified and the reassembled encrypted feature fragment in the ciphertext domain using a homomorphic encryption algorithm;

[0128] Step S602: Set a feature matching threshold in the ciphertext domain, compare the calculated Euclidean distance value with the feature matching threshold, calculate the similarity score between feature vectors based on the addition and multiplication operations of the homomorphic encryption algorithm, perform normalization processing on the similarity score, generate the feature comparison result after interval mapping, and combine the feature comparison result with the corresponding confidence value to form the verification result data.

[0129] Optionally, this embodiment employs a distributed caching mechanism in shard retrieval. First, a smart contract queries the shard index table to obtain the shard distribution information of the target feature. The index query uses a Bloom filter for pre-screening to reduce invalid queries. Retrieval requests are routed to storage nodes via the DHT network, supporting parallel retrieval of multiple shards. The caching strategy is based on access frequency and temporal locality, managing cached content using the LRU algorithm.

[0130] This embodiment implements multi-layered integrity verification. Shard verification first checks the validity of the digital signature and timestamp to ensure that the data has not been tampered with. Then, it verifies the existence of the shard in the blockchain using Merkel proof, the proof process being based on a hash chain. Integrity verification employs erasure coding technology, supporting data recovery from partially corrupted shards. Shards that fail verification are reacquired through backup nodes.

[0131] This embodiment designs an adaptive fragment reassembly algorithm. First, an ordered sequence is constructed based on fragment identifiers, and the relative order of the fragments is determined by positional information. The reassembly process employs a sliding window mechanism to dynamically adjust the batch size. Data alignment is achieved through synchronization markers to ensure the accuracy of fragment splicing. The reassembly results are verified through checksums to guarantee data consistency.

[0132] This embodiment employs optimized homomorphic operations in the ciphertext domain computation. The Euclidean distance calculation first converts the feature vectors into inner product form, then utilizes the multiplicative homomorphic property of homomorphic encryption to calculate the vector inner product. The computation process uses the Montgomery algorithm to accelerate modular exponentiation, improving computational efficiency. Noise management is achieved through modulus switching to control ciphertext growth.

[0133] This embodiment implements a dynamic threshold management mechanism. Threshold settings consider the security requirements of the verification scenario, and a baseline value is determined through historical data analysis. Threshold updates employ an exponential moving average to adapt to dynamic changes in feature distribution. Comparison operations are performed in the ciphertext domain using a homomorphically encrypted comparison circuit. Threshold adjustments support multi-level security policies to meet the needs of different application scenarios.

[0134] This embodiment employs a multi-feature fusion strategy in similarity calculation. First, similarity scores for different feature components are calculated, with score weights dynamically adjusted based on feature importance. The fusion process considers feature reliability, using confidence-based weighting. The calculation is performed entirely in the encrypted domain, preserving the original feature information.

[0135] This embodiment employs an adaptive normalization mechanism. First, the statistical characteristics of the similarity distribution are estimated, and a probability model is constructed using kernel density estimation. Normalization uses a piecewise linear mapping to ensure a uniform score distribution. The mapping parameters are optimized through cross-validation to balance the recognition rate and the false recognition rate. The normalization result is limited to a preset interval to facilitate subsequent decision-making.

[0136] This embodiment implements a complete confidence assessment framework. Confidence calculation considers multiple factors, including feature quality, matching degree, and environmental conditions. The assessment model employs a fuzzy inference system, using a rule base to describe the relationships between factors. The inference process supports uncertainty handling and outputs interpretable confidence values.

[0137] This embodiment establishes a standardized verification result format. The result data includes comparison scores, confidence levels, and timestamp information. The data is organized in a hierarchical structure to support result traceability and auditing. The format design considers subsequent processing needs, facilitating result storage and analysis.

[0138] This embodiment implements face verification through a secure feature comparison mechanism. The entire process, from slice retrieval to result generation, is completed in the encrypted domain, ensuring the security of feature data. Multi-layered verification mechanisms and dynamic threshold management provide reliable verification results. This scheme achieves efficient face feature comparison while protecting privacy.

[0139] In one embodiment of the face data privacy protection method of this application, see [link to relevant documentation]. Figure 7 It can also specifically include the following:

[0140] Step S701: Based on the preset scoring rules, analyze the feature comparison results, compare the similarity score of the feature comparison results with the confidence threshold, generate a verification conclusion including verification status, verification timestamp and scoring details, digitally sign the verification conclusion, combine the signed verification conclusion with the verification operation record to form transaction data, and calculate the hash value of the transaction data to construct the block header;

[0141] Step S702: Combine the block header with the transaction data to generate a new block, broadcast the block in the blockchain network using the proof-of-work mechanism, verify the new block using a distributed consensus algorithm, generate blockchain connection information, append the verified block to the end of the existing blockchain structure, update the verification status information in the smart contract, and synchronize the block to other nodes in the blockchain distributed network.

[0142] Optionally, this embodiment employs a hierarchical scoring mechanism in result parsing. First, a scoring rule tree is constructed, containing multiple dimensions of evaluation metrics, such as similarity score, feature quality, and environmental factors. Rule parsing uses a recursive approach, calculating scores at each level from bottom to top. Scoring weights are determined through expert knowledge and data analysis, supporting dynamic adjustment. The rule engine employs a forward inference strategy to ensure the interpretability of the scoring process.

[0143] This embodiment implements an adaptive confidence comparison strategy. The confidence threshold is dynamically set based on the security level of the verification scenario, with higher security levels using stricter thresholds. The comparison process considers confidence intervals and handles uncertainties using fuzzy logic. Threshold updates employ a feedback mechanism, automatically adjusting based on verification accuracy. Comparison results are correlated with historical data to identify abnormal patterns.

[0144] This embodiment designs a standardized verification conclusion format. The conclusion data includes a verification status code, timestamp, scoring details, and environmental parameters. The status code defines multiple levels to provide a detailed description of the verification results. The timestamp uses UTC format, supporting cross-timezone synchronization. The scoring details record the scores and weights of each dimension for easy subsequent analysis. The environmental parameters include hardware configuration and algorithm version information.

[0145] This embodiment employs a multi-signature mechanism in digital signatures. The signing process first calculates a digest of the verification result using a hash function, and then generates a signature using the node's private key. The signature algorithm uses EdDSA, providing high-speed verification. Multi-signature requires the participation of multiple authorized nodes, enhancing the credibility of the result. Signature verification supports batch processing mode, improving processing efficiency.

[0146] This embodiment constructs a complete transaction data structure. The transaction data includes verification conclusions, operation records, and signature information. The operation records detail the key steps of the verification process, supporting post-audit. Data serialization uses a compact format to optimize storage space. A random factor is incorporated into the transaction construction process to prevent replay attacks.

[0147] This embodiment implements optimized hash calculations in block construction. The block header includes a version number, timestamp, previous block hash, and difficulty target. The hash algorithm used is SHA3, providing stronger security. The difficulty target maintains a stable block time through a dynamic adjustment mechanism. The block structure adopts a layered design, supporting rapid verification.

[0148] This embodiment designs an efficient block broadcasting mechanism. The broadcasting adopts a layered network architecture, optimizing the propagation path through relay nodes. Node discovery is based on the Kademlia protocol, supporting dynamic node management. Compressed transmission is used during the broadcasting process to reduce network load. New block verification employs pipelined processing to improve concurrency performance.

[0149] This embodiment employs an improved proof-of-work mechanism in its consensus algorithm. First, the computational difficulty target is adjusted using random numbers to balance the distribution of network computing power. The verification process supports parallel computation, improving verification efficiency. Reaching consensus requires meeting a minimum number of confirmations to ensure transaction security. Fork handling follows the longest chain principle to maintain network consistency.

[0150] This embodiment implements a reliable blockchain connection mechanism. The link information includes block hashes and sequence numbers, and block relationships are maintained through a doubly linked list. Block appending uses atomic operations to ensure state consistency. The chain structure supports fast traversal and retrieval, optimizing query performance. Fork detection is implemented using a hash tree to promptly identify anomalies.

[0151] This embodiment manages and verifies state through smart contracts. State updates employ a transaction processing mechanism to ensure data consistency. The contract defines state transition rules, supporting complex business logic. State synchronization uses an incremental update strategy to improve synchronization efficiency. Event logs are generated during contract execution for easy monitoring and analysis.

[0152] The technical solution in this embodiment achieves reliable storage and synchronization of verification results. The entire process, from result evaluation to blockchain synchronization, ensures data integrity and traceability. The distributed architecture and consensus mechanism ensure the system's reliability and security. This solution provides a complete verification result management solution for facial recognition systems.

[0153] To address the data security and privacy protection issues in traditional facial recognition systems and provide a secure and reliable technical solution for the field of biometric identification, this application provides an embodiment of a facial data privacy protection device for implementing all or part of the aforementioned facial data privacy protection method. See [link to embodiment]. Figure 8 The facial data privacy protection device specifically includes the following components:

[0154] Network construction module 10 is used to establish a blockchain distributed network, configure multiple nodes in the network and connect them through a peer-to-peer communication protocol, and build a permission management module based on smart contracts to control the data access of nodes.

[0155] Image processing module 20 is used to acquire face image data and extract key point coordinates, construct geometric feature vectors and texture feature vectors, fuse the feature vectors and reduce their dimensionality to obtain target feature vectors, segment and homomorphically encrypt the target feature vectors, package the encrypted feature segments and related operation information to generate blocks and store them in the blockchain distributed network.

[0156] The privacy protection module 30 is used to receive face verification requests, perform feature extraction and encrypted fragmentation on the face image to be verified, retrieve the corresponding encrypted feature fragments from the blockchain distributed network, perform fragment reorganization and feature comparison in the ciphertext domain, and generate a new block by writing the verification conclusion and related information into the blockchain distributed network.

[0157] As described above, the facial data privacy protection device provided in this application embodiment can construct a trusted permission management mechanism through a distributed network and smart contracts, and innovatively designs a multi-dimensional feature extraction scheme that integrates geometric and texture features. The system employs data sharding and homomorphic encryption technology to securely process feature vectors, and distributes the encrypted data across a blockchain network, achieving decentralized management of feature data. Feature comparison and identity verification are performed within the encrypted domain, and the entire process is traceable by combining timestamps and operation records, effectively solving the data security and privacy protection problems in traditional facial recognition systems, and providing a secure and reliable technical solution for the field of biometric recognition.

[0158] From a hardware perspective, in order to solve the data security and privacy protection issues in traditional facial recognition systems and provide a secure and reliable technical solution for the field of biometric recognition, this application provides an embodiment of an electronic device for implementing all or part of the facial data privacy protection method. The electronic device specifically includes the following components:

[0159] The system comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the face data privacy protection device and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the face data privacy protection method and the face data privacy protection device in the embodiments, the content of which is incorporated herein, and repeated details will not be described again.

[0160] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0161] In practical applications, some aspects of the facial data privacy protection method can be executed on the electronic device side as described above, or all operations can be completed in the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0162] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0163] Figure 9 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 9 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 9 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0164] In one embodiment, the facial data privacy protection method function can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following controls:

[0165] Step S101: Establish a blockchain distributed network, configure multiple nodes in the network and connect them through a peer-to-peer communication protocol, build a permission management module based on smart contracts, and control the data access of the nodes through the permission management module;

[0166] Step S102: Collect face image data and extract key point coordinates, construct geometric feature vectors and texture feature vectors, fuse the feature vectors and reduce their dimensionality to obtain target feature vectors, perform fragmentation and homomorphic encryption on the target feature vectors, package the encrypted feature fragments and related operation information to generate blocks and store them in the blockchain distributed network.

[0167] Step S103: Receive a face verification request, perform feature extraction and encrypted fragmentation on the face image to be verified, retrieve the corresponding encrypted feature fragment from the blockchain distributed network, perform fragment reorganization and feature comparison in the ciphertext domain, and generate a new block by generating the verification conclusion and related information and write it into the blockchain distributed network.

[0168] As described above, the electronic device provided in this application embodiment constructs a trusted permission management mechanism through a distributed network and smart contracts, and innovatively designs a multi-dimensional feature extraction scheme that integrates geometric and texture features. The system employs data sharding and homomorphic encryption technology to securely process feature vectors, and distributes the encrypted data across a blockchain network, achieving decentralized management of feature data. Feature comparison and identity verification are performed within the encrypted domain, and the entire process is traceable by combining timestamps and operation records, effectively solving the data security and privacy protection problems in traditional face recognition systems, and providing a secure and reliable technical solution for the field of biometric recognition.

[0169] In another embodiment, the face data privacy protection device can be configured separately from the central processing unit 9100. For example, the face data privacy protection device can be configured as a chip connected to the central processing unit 9100, and the face data privacy protection method function can be implemented through the control of the central processing unit.

[0170] like Figure 9 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 9 All components shown; in addition, the electronic device 9600 may also include Figure 9 For components not shown, please refer to existing technologies.

[0171] like Figure 9 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0172] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0173] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0174] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0175] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device for communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0176] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0177] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 9130 is coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored audio via the speaker 9131.

[0178] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the face data privacy protection method with a server or client as the execution subject in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the face data privacy protection method with a server or client as the execution subject in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0179] Step S101: Establish a blockchain distributed network, configure multiple nodes in the network and connect them through a peer-to-peer communication protocol, build a permission management module based on smart contracts, and control the data access of the nodes through the permission management module;

[0180] Step S102: Collect face image data and extract key point coordinates, construct geometric feature vectors and texture feature vectors, fuse the feature vectors and reduce their dimensionality to obtain target feature vectors, perform fragmentation and homomorphic encryption on the target feature vectors, package the encrypted feature fragments and related operation information to generate blocks and store them in the blockchain distributed network.

[0181] Step S103: Receive a face verification request, perform feature extraction and encrypted fragmentation on the face image to be verified, retrieve the corresponding encrypted feature fragment from the blockchain distributed network, perform fragment reorganization and feature comparison in the ciphertext domain, and generate a new block by generating the verification conclusion and related information and write it into the blockchain distributed network.

[0182] As described above, the computer-readable storage medium provided in this application embodiment constructs a trusted permission management mechanism through a distributed network and smart contracts, and innovatively designs a multi-dimensional feature extraction scheme that integrates geometric and texture features. The system employs data sharding and homomorphic encryption technology to securely process feature vectors, and distributes the encrypted data across a blockchain network, achieving decentralized management of feature data. Feature comparison and identity verification are performed within the encrypted domain, and the entire process is traceable by combining timestamps and operation records, effectively solving the data security and privacy protection problems in traditional face recognition systems, and providing a secure and reliable technical solution for the field of biometric recognition.

[0183] Embodiments of this application also provide a computer program product capable of implementing all steps of the face data privacy protection method with the execution subject being a server or client in the above embodiments. When executed by a processor, this computer program / instruction implements the steps of the face data privacy protection method. For example, the computer program / instruction implements the following steps:

[0184] Step S101: Establish a blockchain distributed network, configure multiple nodes in the network and connect them through a peer-to-peer communication protocol, build a permission management module based on smart contracts, and control the data access of the nodes through the permission management module;

[0185] Step S102: Collect face image data and extract key point coordinates, construct geometric feature vectors and texture feature vectors, fuse the feature vectors and reduce their dimensionality to obtain target feature vectors, perform fragmentation and homomorphic encryption on the target feature vectors, package the encrypted feature fragments and related operation information to generate blocks and store them in the blockchain distributed network.

[0186] Step S103: Receive a face verification request, perform feature extraction and encrypted fragmentation on the face image to be verified, retrieve the corresponding encrypted feature fragment from the blockchain distributed network, perform fragment reorganization and feature comparison in the ciphertext domain, and generate a new block by generating the verification conclusion and related information and write it into the blockchain distributed network.

[0187] As described above, the computer program product provided in this application constructs a trusted permission management mechanism through a distributed network and smart contracts, and innovatively designs a multi-dimensional feature extraction scheme that integrates geometric and texture features. The system employs data sharding and homomorphic encryption technology to securely process feature vectors, and distributes the encrypted data across a blockchain network, achieving decentralized management of feature data. Feature comparison and identity verification are performed within the encrypted domain, and the entire process is traceable by combining timestamps and operation records, effectively solving the data security and privacy protection problems in traditional face recognition systems, and providing a secure and reliable technical solution for the field of biometric recognition.

[0188] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0189] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0190] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0191] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0192] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for protecting facial data privacy, characterized in that, The method includes: A distributed blockchain network is established, in which multiple nodes are configured and connected through a peer-to-peer communication protocol. A permission management module is built based on smart contracts, and the data access of the nodes is controlled through the permission management module. A multispectral image acquisition scheme is adopted, configuring a dual-camera array of visible light and near-infrared to achieve all-weather face image acquisition. An improved cascaded shape regression algorithm is used, employing a multi-layer cascaded regressor to progressively optimize key point positions. The regressor uses a gradient boosting tree structure to extract the coordinates of key points in the face image data. Based on the key point coordinates, multi-dimensional geometric features are constructed, the Euclidean distance matrix between key points is calculated, triangle features are extracted, and key points are selected to form a Delaunay triangulation. Geometric feature vectors are constructed based on the key point coordinates. A multi-scale Gabor filter bank is used for texture feature extraction, with filter parameters covering 8 directions and 5 scales. The texture feature vector of the face image data is calculated, and an adaptive feature fusion algorithm is designed. Based on the Fisher criterion, the feature discrimination capability is calculated, and feature weight coefficients are designed. The geometric feature vector and the texture feature vector are fused to obtain the fused feature. The vector is used to achieve feature dimensionality reduction through local preservation projection, and a nearest neighbor graph is constructed to describe the local structure between samples. The fused feature vector is subjected to dimensionality reduction transformation to obtain the target feature vector. A threshold secret sharing scheme is used for feature fragmentation, and the Shamir threshold scheme is used to distribute the secret to multiple holders. The target feature vector is divided into multiple data fragments. A homomorphic encryption system is constructed based on the BGV scheme, which supports addition and multiplication operations on the ciphertext field. An improved homomorphic encryption algorithm is used, and the key generation adopts the ring learning problem to ensure encryption strength. The multiple data fragments are homomorphically encrypted to obtain encrypted feature fragments. The encrypted feature fragments, operation timestamps, and operation records are packaged to generate blocks, and the blocks are distributed and stored on multiple nodes in the blockchain distributed network. The access operation of the encrypted feature fragment is recorded based on the smart contract, and the access operation is executed after the permission management module verifies it. The verification request processing adopts a multi-queue priority scheduling mechanism, verifies the identity of the requester through zero-knowledge proof, receives face verification requests, performs feature extraction and encrypted fragmentation on the face image to be verified, retrieves the corresponding encrypted feature fragments from the blockchain distributed network, implements the ciphertext domain feature comparison algorithm, calculates the Euclidean distance between encrypted feature fragments based on the properties of homomorphic encryption, performs fragment reorganization and feature comparison within the ciphertext domain, adopts an improved consensus mechanism to realize block writing, constructs a consensus protocol based on a practical Byzantine fault-tolerant algorithm, and generates a new block by writing the verification conclusion and related information into the blockchain distributed network.

2. A facial data privacy protection device, characterized in that, The device includes: A network construction module is used to establish a blockchain distributed network, in which multiple nodes are configured and connected through a peer-to-peer communication protocol. A permission management module is built based on smart contracts to control the data access of nodes. The image processing module employs a multispectral image acquisition scheme, configuring a dual-camera array of visible light and near-infrared light to achieve all-weather face image acquisition. It utilizes an improved cascaded shape regression algorithm, employing a multi-layer cascaded regressor to progressively optimize key point positions. The regressor uses a gradient boosting tree structure to extract key point coordinates from the face image data. Based on these coordinates, it constructs multi-dimensional geometric features, calculates the Euclidean distance matrix between key points, extracts triangle features, selects key points to form a Delaunay triangulation, and constructs a geometric feature vector based on the key point coordinates. A multi-scale Gabor filter bank is used for texture feature extraction, with filter parameters covering 8 directions and 5 scales. The module calculates the texture feature vector from the face image data, designs an adaptive feature fusion algorithm, calculates feature discrimination capability based on the Fisher criterion, designs feature weight coefficients, and fuses the geometric feature vector and the texture feature vector. The process involves fusing feature vectors, performing feature dimensionality reduction through local preservation projection, constructing a nearest neighbor graph to describe the local structure between samples, performing dimensionality reduction transformation on the fused feature vectors to obtain the target feature vector, employing a threshold secret sharing scheme for feature fragmentation, distributing the secret to multiple holders using the Shamir threshold scheme, dividing the target feature vector into multiple data fragments, constructing a homomorphic encryption system based on the BGV scheme to support addition and multiplication operations on the ciphertext domain, using an improved homomorphic encryption algorithm, and employing a ring learning problem for key generation to ensure encryption strength, performing homomorphic encryption on each of the multiple data fragments to obtain encrypted feature fragments; packaging the encrypted feature fragments, operation timestamps, and operation records into blocks, distributing the blocks to multiple nodes in the blockchain distributed network; recording access operations to the encrypted feature fragments based on the smart contract, and executing the access operation after verification by the permission management module; The privacy protection module employs a multi-queue priority scheduling mechanism in verification request processing. It verifies the identity of the requester through zero-knowledge proofs, receives face verification requests, performs feature extraction and encrypted fragmentation on the face image to be verified, retrieves the corresponding encrypted feature fragments from the blockchain distributed network, implements a ciphertext domain feature comparison algorithm, calculates the Euclidean distance between encrypted feature fragments based on the properties of homomorphic encryption, performs fragment reassembly and feature comparison within the ciphertext domain, implements block writing using an improved consensus mechanism, constructs a consensus protocol based on a practical Byzantine fault-tolerant algorithm, and generates a new block by writing the verification conclusion and related information into the blockchain distributed network.

3. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the face data privacy protection method according to claim 1.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the face data privacy protection method of claim 1.

Images