Real-name did and biometric fusion rapid verification system

Abstract

The application discloses a real-name DID and biological feature fusion rapid verification system, comprising the following modules: a biological feature processing module for generating a nested biological feature vector; a hash generation module for performing hash operation and constructing a Merkle tree; a real-name verification module for generating real-name identity data; a DID construction module for generating a DID document and adding a BioBinding extension field; a proof generation module for generating a verification record; a blockchain writing module for forming a chain real-name DID record; a verification module for matching with the chain real-name DID record; a confidence control module for comparing the matching result with a preset verification threshold, if the verification is passed, authorized access is completed; if the verification is not passed, a multi-factor re-verification mechanism is triggered for secondary verification, a DID confidence value is calculated, if the secondary verification is still not passed, identity freezing is executed. The application fuses multi-modal biological recognition and real-name DID generation mechanism, and realizes chain trusted verification.

Description

Technical Field

[0001] This invention relates to the field of digital identity recognition and data privacy technology, and in particular to a rapid verification system that integrates real-name DID and biometric features. Background Technology

[0002] With the widespread development of information technology, digital identity authentication has become a core component of cyberspace security. Especially in high-security scenarios such as financial services, government travel, healthcare, and social media platforms, identity recognition systems often need to strengthen user privacy protection and data trustworthiness while ensuring recognition efficiency. However, traditional username-password authentication methods, due to their vulnerabilities such as easy credential theft, high risk of replay attacks, and irreversibility, have been gradually replaced by more secure biometric authentication systems.

[0003] Currently mature biometric technologies include facial recognition, fingerprint recognition, voiceprint recognition, and iris recognition. While these single-modal recognition methods can improve verification security, they still have several problems in practical applications. For example, single-modal recognition is limited by the accuracy of the acquisition equipment and environmental interference, and its recognition accuracy is easily affected by factors such as image blurring, noise pollution, or spoofing attacks. To address this, researchers have recently begun to explore the introduction of multimodal fusion recognition technology, which integrates information from multiple biometric channels to improve recognition robustness. However, existing multimodal fusion schemes mostly remain at the level of shallow feature splicing or simple weighting, lacking effective methods for deep nested structure modeling, making it difficult to fully explore the potential structural coupling relationships between multimodalities, resulting in limited fusion effects.

[0004] On the other hand, with the rise of blockchain technology, decentralized identity (DID) is gradually becoming a key path for next-generation digital identity management. DIDs are generated and controlled by users, possessing characteristics such as autonomy, strong privacy protection, and high verifiability, and have been recognized as a core standard by the W3C. However, most current DID implementations still use static public keys or user-declared attributes for identity binding, lacking a deep integration mechanism with the user's real physical characteristics (especially biometrics in high-security scenarios). Since DIDs themselves do not possess real-name attributes, without a trusted binding mechanism, even if the DID record on the blockchain is complete, it is difficult to effectively trace the user's authenticity, failing to meet the stringent requirements of trusted on-chain identities in government, finance, and judicial scenarios.

[0005] Some academic research and industry practices have attempted to improve the credibility of the DID system by writing biometric hashes onto the blockchain and establishing a weak binding relationship with the DID. However, due to the limited accuracy of biometric hash algorithms and the high cost of on-chain storage, most implementations only use compressed single-modal hash values ​​on the blockchain, lacking modal fusion expression capabilities and dynamic update and identity status backtracking mechanisms. This results in the identity chain being unable to be adjusted or revoked once established. Furthermore, there is currently a lack of technical paths to extend multimodal fusion biometrics with DID structural specifications (such as the W3C DID document structure) with high consistency, leading to a chaotic on-chain identity record structure that is detrimental to system parsing and verification.

[0006] In terms of security verification, most existing on-chain identity verification schemes rely on plaintext comparison, fingerprint deduplication, or signature comparison, lacking a verifiable logical expression of the binding relationship between identity and biometric hashes. While some schemes based on multi-party secure computation (MPC) or homomorphic encryption offer privacy protection, their high computational overhead and complex on-chain deployment make them unsuitable for identity verification applications with high real-time requirements. Zero-knowledge proofs (ZKP), as a significant direction in privacy computing in recent years, are well-suited for private verification in the identity binding process due to their ability to verify facts without exposing the original data. However, existing DID systems have not yet widely implemented ZKP integration mechanisms with biometric binding, lacking complete protocol implementation and structural definitions.

[0007] Furthermore, in actual identity verification, user states are often not static. Device changes, environmental shifts, and changes in behavioral characteristics can all affect the credibility of an identity verification system's assessment of the current identity. However, traditional identity authentication systems often lack the ability to model dynamic user behavior states and cannot automatically adjust verification strength based on risk levels. For example, a user may switch from a trusted device to an unfamiliar terminal at a certain point in time. If the system cannot detect and intervene in time, it may allow a forged identity to pass authentication. Conversely, if multi-factor reverberation can be triggered when confidence fluctuates abnormally, and the identity credibility score can be dynamically adjusted based on a sliding window mechanism, the risk can be effectively reduced. However, current DID systems generally do not integrate this kind of dynamic confidence adjustment capability.

[0008] Therefore, how to provide a rapid verification system that integrates real-name DID and biometrics is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0009] One objective of this invention is to propose a rapid verification system that integrates real-name DID and biometric features. This invention integrates multimodal biometrics and real-name DID generation mechanisms, constructs a nested feature encoding and BioBinding extension structure, and combines zero-knowledge proofs and on-chain matching to achieve a highly reliable, low-leakage, and highly efficient identity verification process. It has the advantages of strong privacy protection, high verification accuracy, fast response speed, and scalable structure, and is suitable for high-security scenarios such as government affairs, finance, and immigration.

[0010] The rapid verification system for fusing real-name DID and biometric features according to embodiments of the present invention includes the following modules:

[0011] The biometric processing module is used to collect users' multimodal biometric data and generate nested biometric vectors through a variational nested encoder.

[0012] The hash generation module is used to perform hash operations on the nested biological feature vectors and construct a Merkle tree to generate biological hash root values.

[0013] The real-name verification module is used to verify the authenticity of users' real-name information and generate real-name identity data;

[0014] The DID building module is used to generate DID documents that conform to the W3C standard format and add the BioBinding extension field to record the on-chain reference address of the biohash root value;

[0015] The proof generation module is used to construct zero-knowledge proofs based on the real-name identity data and the biometric hash root value, and generate verification records;

[0016] The blockchain writing module is used to write the DID document, BioBinding extended fields and verification records into the blockchain to form an on-chain real-name DID record;

[0017] The verification module is used to collect lightweight biometric data of a user's single modality during identity verification, generate a summary vector, and match it with the real-name DID record on the blockchain.

[0018] The confidence control module compares the matching result with the preset verification threshold. If the verification passes, the authorized access is completed. If the verification fails, the multi-factor re-verification mechanism is triggered for a second verification, and the DID confidence value is calculated. If the second verification still fails, the identity is frozen.

[0019] The rapid verification method for fusing real-name DID and biometric features according to embodiments of the present invention includes the following steps:

[0020] S1. Collect the user's multimodal biometric data and input it into the variational nested encoder model, then fuse and compress the mapping to generate a nested biometric vector;

[0021] S2. Perform a hash operation on the nested biological feature vectors and construct a Merkle tree to generate a biological hash root value;

[0022] S3. Obtain the real-name information provided by the user, verify the authenticity of the real-name information, and generate real-name identity data;

[0023] S4. Generate a DID document conforming to the W3C standard format based on the real-name identity data, and add a BioBinding extended field to record the on-chain reference address of the bio-hash root value;

[0024] S5. Construct a zero-knowledge proof protocol input, use the zero-knowledge proof protocol to verify the binding relationship between the real-name identity data and the biometric hash root value, and generate a verification record;

[0025] S6. Write the DID document, BioBinding extended fields and verification records into the blockchain to generate an immutable on-chain real-name DID record.

[0026] S7. When performing identity verification, collect the user's single-modality lightweight biometric data, process the single-modality lightweight biometric data locally to generate a summary vector, and submit it to the on-chain comparison engine to match it with the on-chain real-name DID record.

[0027] S8. Compare the matching result with the preset verification threshold. If the verification passes, the authorized access is completed. If the verification fails, the multi-factor re-verification mechanism is triggered for a second verification. The DID confidence value is calculated. If the second verification still fails, the identity is frozen.

[0028] Optionally, the multimodal biometric data includes facial images, fingerprint images, and voiceprint signals.

[0029] Optionally, S1 specifically includes:

[0030] S11. The variational nested encoder model includes a modal feature extraction layer, a cross-modal fusion layer, a nested variational encoder layer, and a compression mapping output layer;

[0031] S12. Input the multimodal biometric data into the modality feature extraction layer, and perform convolutional encoding, normalization and activation processing on each modality to extract preliminary modality features;

[0032] S13. Input the preliminary modal features into the cross-modal fusion layer. The cross-modal fusion layer includes a multi-head attention mechanism and a gating mechanism to dynamically adjust the weight relationship between each modality and generate a modal fusion representation.

[0033] S14. Input the modal fusion representation into the nested variational encoder layer. The nested variational encoder layer consists of a master encoder and three sub-encoders. The master encoder uses a multilayer perceptron structure to model the modal fusion representation and is used to extract the global latent mean and variance vector shared by all modalities.

[0034] S15. The three sub-encoders correspond to the facial image, fingerprint image and voiceprint signal respectively. They adopt independent variational channels in the shared architecture to model the local variation information of each modality and output modality-specific latent vectors.

[0035] S16. The global latent mean and variance vector output by the main encoder are fused with the modal local latent vectors output by each sub-encoder to generate nested latent feature vectors.

[0036] S17. Input the nested latent feature vector into the compressed mapping output layer, and obtain the structure-preserving nested biological feature vector through nonlinear dimensionality reduction mapping.

[0037] Optionally, S2 specifically includes:

[0038] S21. The nested biometric vectors are structurally divided into several sub-vector segments according to the modality source and feature level. Each sub-vector segment corresponds to a set of features in a biometric modality or coding level, which is used for distributed hash calculation.

[0039] S22. The SHA-3 hash algorithm is used to perform three rounds of hashing operations on each sub-vector segment. The first round is to perform hashing after vector standardization, the second round is to perform hashing after vector dimension rotation, and the third round is to perform scrambling hashing after introducing mask parameters, generating multiple hash sub-values ​​as leaf node values ​​of the Merkle tree.

[0040] S23. Construct the underlying structure of the Merkle tree. Use sequential pairing to concatenate two adjacent hash values ​​and then execute the SHA-3 hash algorithm again to generate the corresponding parent node hash value. Iterate upwards layer by layer to generate intermediate nodes until the root node is generated.

[0041] S24. During the construction of each level of the Merkle tree, a modality identifier label is attached to each intermediate node. The modality identifier label represents the modality source combination contained in the intermediate node through an encoding mapping method, which is used to trace the modality composition relationship.

[0042] S25. A local update mechanism is introduced during the construction process. The input hash combination relationship of each intermediate node is stored in a local buffer and an index table is built. When a certain sub-vector segment changes, only the corresponding branch and the upper-level node are updated to avoid repeated construction operations of the entire Merkle tree.

[0043] S26. After completing the calculation of all intermediate nodes, the final output is the biological hash root value, which serves as the unique identifier hash root value of the nested biological feature vector.

[0044] Optionally, S4 specifically includes:

[0045] S41. Construct a basic identity information object based on real-name identity data. The basic identity information object includes a user identifier, an issuing authority identifier, an issuance timestamp, an expiration timestamp, and an identity recognition code, and is organized in a key-value pair manner.

[0046] S42. Generate a DID document according to the W3C decentralized identity technical specification. The DID document includes a context field, an id field, an authentication field, and a verificationMethod field, wherein the context field declares the specification version, and the id field corresponds to the user's DID identifier.

[0047] S43. Set the public key identifier data based on elliptic curves in the verificationMethod field, and set the signature method identifier used for authentication in the authentication field;

[0048] S44. Add an extended field BioBinding to the DID document. The BioBinding field includes four sub-fields: bioMerkleRoot, bindingTime, dataType, and storageReference, which correspond to the biohash root value, binding time, feature data type, and on-chain storage address information of the nested biofeature vector, respectively.

[0049] S45. Encapsulate the DID document containing the BioBinding extended field into JSON-LD format.

[0050] Optionally, S5 specifically includes:

[0051] S51. Construct the input data set for the zero-knowledge proof protocol, and represent real-name identity data and biometric hash root values ​​as private inputs of the relationship to be proved;

[0052] S52. Introduce a zero-knowledge proof protocol for verification. Define constraints based on the relationship to be proved and represent the constraints in circuit form. The circuit form contains the relationship constraint logic between real-name identity data and biometric hash root value.

[0053] S53. Generate a zero-knowledge proof public-private key pair based on the constraints of the circuit form, wherein the private key is used to generate the proof and the public key is used for on-chain verification.

[0054] S54. Using the private key, perform zero-knowledge proof to generate proof data for the binding relationship between real-name identity data and biometric hash root value, and generate proof data including proof subject, random challenge number and response factor;

[0055] S55. The proof data is structured and digitally signed.

[0056] S56. Combine the digitally signed proof data with the corresponding public key to form a verification record, which serves as the verification record for the binding relationship between the real-name identity data and the biometric hash root value. The verification record includes signature digest and timestamp information.

[0057] Optionally, when writing the DID document, BioBinding extended fields, and verification records to the blockchain, the pre-deployed smart contract registration interface is called to use the on-chain real-name DID record as the transaction payload and record it in the block event log.

[0058] Optionally, S7 specifically includes:

[0059] S71. When performing identity verification, the user's single-modality lightweight biometric data is collected in real time through the terminal device, and the single-modality lightweight biometric data is subjected to scale normalization, noise filtering and standardization to obtain a feature preprocessing vector.

[0060] S72. A fixed random projection matrix and bias vector are stored locally on the terminal device, wherein the random projection matrix is ​​a predefined random numerical matrix and the bias vector is a predefined random constant vector. A summary vector is generated by mapping the feature preprocessing vector with the nonlinear activation function ReLU.

[0061] S73. For each element of the digest vector, the sign function of the Local Aware Hash algorithm is used to perform binary quantization dimension by dimension, converting each element into a corresponding binary sign bit, thus forming a 64-bit Local Aware Hash string:

[0062]

[0063] Among them, h i s represents the value of the i-th bit in the locally perceptual hash string.i This represents the i-th element in the summary vector;

[0064] S74. Establish a communication connection with the blockchain network through the terminal device, encapsulate the local perception hash string into an on-chain transaction payload, and submit it to the on-chain comparison engine.

[0065] S75. The on-chain comparison engine reads the on-chain real-name DID record, parses the DID document to verify the source validity, extracts the bio-hash root value from the BioBinding extended field, and verifies the signature digest and timestamp information in the record, and receives the local perceptual hash string and the bio-hash root value for matching.

[0066] S76. The on-chain comparison engine uses the Hamming distance function to calculate the similarity d between the locally perceptual hash string and the bio-hash root value. H :

[0067]

[0068] Among them, H bio,i This represents the value of the i-th bit of the bio hash root value, d H Output as the matching result.

[0069] d H Compare with the preset verification threshold θ, when d H When d ≤ θ, it means the verification passed; when d ≤ θ, it means the verification passed. H >θ indicates that the verification failed.

[0070] Optionally, the multi-factor re-verification mechanism specifically includes calculating the DID confidence value based on the user's device type, historical login geographic location, access frequency, and verification behavior trajectory, and calculating the DID confidence change rate based on a sliding time window. When the confidence change rate exceeds the set confidence threshold, it is determined that the second verification has still failed, and the identity is frozen.

[0071] The beneficial effects of this invention are:

[0072] First, this invention introduces a variational nested encoder model to perform deep, structured fusion of multimodal biometric features such as facial images, fingerprint images, and voiceprint signals. It then extracts global latent semantics and modal local variation information through a master-sub nested modeling approach, significantly improving the expressive integrity and discriminative power of the fused biometric features. Compared to traditional shallow splicing or simple weighted fusion methods, this invention achieves deep collaboration and unified representation among modalities, providing high-quality, low-redundancy feature inputs for subsequent identity binding.

[0073] Secondly, this invention constructs an improved multimodal hashing mechanism based on Merkle trees. Through modal structure identification and a locally updatable strategy, it achieves an efficient and traceable nested hash calculation process and generates a unique biometric hash root value for identity binding. This hashing mechanism not only improves the accuracy of fuzzy matching but also significantly enhances the structural expressiveness and dynamic adaptability of biometric data during on-chain recording, solving the problems of non-scalability and non-reconstructability of hashing in existing systems.

[0074] Furthermore, this invention proposes a standardized extension method for the DID document structure. Based on the W3C specification, it introduces the BioBinding extension field and records the biohash root value, binding time, bio data type, and on-chain storage address through a clear field structure. This achieves a strong binding between real-name identity data and multimodal biohash, and has the technical capabilities of on-chain structure parsing and field verification, filling the gap in the existing DID system's lack of native support for biobinding information.

[0075] Furthermore, this invention integrates a zero-knowledge proof protocol, achieving verifiable modeling of the binding relationship between real-name identity data and biometric hash root values ​​without exposing original identity data and biometric features. This mechanism expresses the binding logic through circuit constraints, generates a verifiable proof structure, and binds it to the DID document to form a verification record. This meets the requirements for high privacy, high trust, and high compliance in real-name verification, and is superior to traditional MPC or hash signature schemes in terms of technical approach, exhibiting stronger practicality and scalability.

[0076] Finally, in the verification phase, this invention designs a complete process supporting locally perceptual hashing, on-chain fuzzy comparison, and dynamic confidence level adjustment. During user identity verification, the system collects lightweight biometric features of a single modality, generates a digest hash vector through random projection and symbolic mapping, and calculates the Hamming distance between this vector and the biometric hash root value in the on-chain real-name DID record, ensuring a balance between matching efficiency and error tolerance. Simultaneously, this invention constructs an identity situation map based on multi-source behavioral features such as user device, location, and frequency, and introduces a sliding window mechanism to calculate the DID confidence change rate in real time. This enables the identity system to automatically trigger multi-factor re-verification or execute identity freezing when encountering abnormal behavior, effectively improving the overall verification system's risk response capability and dynamic reliability. Attached Figure Description

[0077] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0078] Figure 1 This is a schematic diagram of the rapid verification system that fuses real-name DID with biometric features proposed in this invention.

[0079] Figure 2 This is an overall flowchart of the rapid verification method that fuses real-name DID with biometric features proposed in this invention. Detailed Implementation

[0080] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0081] refer to Figure 1 A rapid verification system that integrates real-name DID with biometrics includes the following modules:

[0082] The biometric processing module is used to collect users' multimodal biometric data and generate nested biometric vectors through a variational nested encoder.

[0083] The hash generation module is used to perform hash operations on the nested biological feature vectors and construct a Merkle tree to generate biological hash root values.

[0084] The real-name verification module is used to verify the authenticity of users' real-name information and generate real-name identity data;

[0085] The DID building module is used to generate DID documents that conform to the W3C standard format and add the BioBinding extension field to record the on-chain reference address of the biohash root value;

[0086] The proof generation module is used to construct zero-knowledge proofs based on the real-name identity data and the biometric hash root value, and generate verification records;

[0087] The blockchain writing module is used to write the DID document, BioBinding extended fields and verification records into the blockchain to form an on-chain real-name DID record;

[0088] The verification module is used to collect lightweight biometric data of a user's single modality during identity verification, generate a summary vector, and match it with the real-name DID record on the blockchain.

[0089] The confidence control module compares the matching result with the preset verification threshold. If the verification passes, the authorized access is completed. If the verification fails, the multi-factor re-verification mechanism is triggered for a second verification, and the DID confidence value is calculated. If the second verification still fails, the identity is frozen.

[0090] This invention constructs a rapid real-name DID verification system that integrates modules such as biometric processing, hash generation, real-name verification, DID construction, proof generation, on-chain writing, verification processing, and confidence control, thus fully establishing a trusted fusion path between user biometrics and decentralized on-chain identity. The system design features a clear structure and well-defined module boundaries, enabling a closed-loop process from data collection, fusion encoding, identity generation, privacy verification to on-chain comparison and confidence control. Compared to traditional DID systems that rely solely on static public keys or declared data for binding, this system enhances the real-person credibility and on-chain verification response capabilities of DIDs through structured multimodal biometric input and dynamic security control mechanisms. It boasts higher security, flexibility, and scalability, making it suitable for high-security scenarios such as financial authentication, government real-name registration, and immigration control.

[0091] refer to Figure 2 A rapid verification method that integrates real-name DID with biometrics includes the following steps:

[0092] S1. Collect the user's multimodal biometric data and input it into the variational nested encoder model, then fuse and compress the mapping to generate a nested biometric vector;

[0093] S2. Perform a hash operation on the nested biological feature vectors and construct a Merkle tree to generate a biological hash root value;

[0094] S3. Obtain the real-name information provided by the user, verify the authenticity of the real-name information, and generate real-name identity data;

[0095] S4. Generate a DID document conforming to the W3C standard format based on the real-name identity data, and add a BioBinding extended field to record the on-chain reference address of the bio-hash root value;

[0096] S5. Construct a zero-knowledge proof protocol input, use the zero-knowledge proof protocol to verify the binding relationship between the real-name identity data and the biometric hash root value, and generate a verification record;

[0097] S6. Write the DID document, BioBinding extended fields and verification records into the blockchain to generate an immutable on-chain real-name DID record.

[0098] S7. When performing identity verification, collect the user's single-modality lightweight biometric data, process the single-modality lightweight biometric data locally to generate a summary vector, and submit it to the on-chain comparison engine to match it with the on-chain real-name DID record.

[0099] S8. Compare the matching result with the preset verification threshold. If the verification passes, the authorized access is completed. If the verification fails, the multi-factor re-verification mechanism is triggered for a second verification. The DID confidence value is calculated. If the second verification still fails, the identity is frozen.

[0100] This method defines the interaction methods and order between modules in the system, clarifying the operational flow of the entire real-name DID generation and verification process. Through sequential processing from S1 to S8, the system first integrates multimodal biometric features to generate a unique identifier hash value, then binds it to real-name data using zero-knowledge binding and writes it to the blockchain. This hash value is then rapidly compared and dynamically controlled during subsequent verification stages. This process design not only ensures consistency between data structures but also achieves a closed-loop chain through upstream and downstream data dependencies, effectively improving verification efficiency and identity credibility. This process avoids problems such as fragmented modules, inconsistent data sources, and unverifiable verification links found in existing technologies, enhancing the system's operability and consistency in real-world application scenarios.

[0101] In this embodiment, the multimodal biometric data includes facial images, fingerprint images, and voiceprint signals. This modal combination can provide complementary identification information in different environments. For example, in scenarios with limited image acquisition, it can switch to fingerprint or voiceprint recognition, improving the stability and accuracy of verification. Simultaneously, multimodal input can effectively reduce the risk of spoofing attacks and prevent security vulnerabilities caused by the forgery of a single modality. Compared with single-modal recognition schemes in existing systems, this approach has the advantages of strong anti-interference capability, wide feature coverage, and high authentication robustness, making it suitable for secure identity verification in complex scenarios such as high-risk, cross-device, and multi-terminal environments.

[0102] In this embodiment, S1 specifically includes:

[0103] S11. The variational nested encoder model includes a modal feature extraction layer, a cross-modal fusion layer, a nested variational encoder layer, and a compression mapping output layer;

[0104] S12. Input the multimodal biometric data into the modality feature extraction layer, and perform convolutional encoding, normalization and activation processing on each modality to extract preliminary modality features;

[0105] S13. Input the preliminary modal features into the cross-modal fusion layer. The cross-modal fusion layer includes a multi-head attention mechanism and a gating mechanism to dynamically adjust the weight relationship between each modality and generate a modal fusion representation.

[0106] S14. Input the modal fusion representation into the nested variational encoder layer. The nested variational encoder layer consists of a master encoder and three sub-encoders. The master encoder uses a multilayer perceptron structure to model the modal fusion representation and is used to extract the global latent mean and variance vector shared by all modalities.

[0107] S15. The three sub-encoders correspond to the facial image, fingerprint image and voiceprint signal respectively. They adopt independent variational channels in the shared architecture to model the local variation information of each modality and output modality-specific latent vectors.

[0108] S16. The global latent mean and variance vector output by the main encoder are fused with the modal local latent vectors output by each sub-encoder to generate nested latent feature vectors.

[0109] S17. Input the nested latent feature vector into the compressed mapping output layer, and obtain the structure-preserving nested biological feature vector through nonlinear dimensionality reduction mapping.

[0110] Step S1 defines the hierarchical structure of the variational nested encoder, including a modal feature extraction layer, a cross-modal fusion layer, a nested variational encoder layer, and a compressed mapping output layer. This structural design ensures the hierarchical, interactive, and compact representation of multimodal features. This structure can model the feature distribution within a modality and the collaborative relationships between modalities, and extracts global and local semantics through a master-child nesting mechanism, ultimately outputting a unique and compressible nested biometric vector. Compared to traditional feature concatenation encoding methods, this approach has stronger discriminative power and generation consistency, significantly improving the accuracy and efficiency of subsequent hash matching and identity binding. It represents an effective innovation in the structural representation of biometric technology in decentralized identity systems.

[0111] In this embodiment, S2 specifically includes:

[0112] S21. The nested biometric vectors are structurally divided into several sub-vector segments according to the modality source and feature level. Each sub-vector segment corresponds to a set of features in a biometric modality or coding level, which is used for distributed hash calculation.

[0113] S22. The SHA-3 hash algorithm is used to perform three rounds of hashing operations on each sub-vector segment. The first round is to perform hashing after vector standardization, the second round is to perform hashing after vector dimension rotation, and the third round is to perform scrambling hashing after introducing mask parameters, generating multiple hash sub-values ​​as leaf node values ​​of the Merkle tree.

[0114] S23. Construct the underlying structure of the Merkle tree. Use sequential pairing to concatenate two adjacent hash values ​​and then execute the SHA-3 hash algorithm again to generate the corresponding parent node hash value. Iterate upwards layer by layer to generate intermediate nodes until the root node is generated.

[0115] S24. During the construction of each level of the Merkle tree, a modality identifier label is attached to each intermediate node. The modality identifier label represents the modality source combination contained in the intermediate node through an encoding mapping method, which is used to trace the modality composition relationship.

[0116] S25. A local update mechanism is introduced during the construction process. The input hash combination relationship of each intermediate node is stored in a local buffer and an index table is built. When a certain sub-vector segment changes, only the corresponding branch and the upper-level node are updated to avoid repeated construction operations of the entire Merkle tree.

[0117] S26. After completing the calculation of all intermediate nodes, the final output is the biological hash root value, which serves as the unique identifier hash root value of the nested biological feature vector.

[0118] Step S2 proposes an improved process for structural partitioning, multi-round scrambling hashing, and Merkle tree construction of nested biometric feature vectors, forming a nested hash structure with modality traceability and an update mechanism. By using modality and hierarchy partitioning mechanisms, introducing mask perturbations, and using node labels in the Merkle tree, not only is the collision resistance and security of biometric hashing improved, but the collaborative expression of feature modalities and structure is also achieved during hash chain construction. In particular, the addition of a local update mechanism during construction allows only the corresponding node to be updated when a modality changes, avoiding full tree reconstruction and significantly improving system real-time performance and computational efficiency. This scheme enhances the uniqueness and adaptability of on-chain bound data, making it suitable for scenarios with frequent changes in multimodal features.

[0119] In this embodiment, S4 specifically includes:

[0120] S41. Construct a basic identity information object based on real-name identity data. The basic identity information object includes a user identifier, an issuing authority identifier, an issuance timestamp, an expiration timestamp, and an identity recognition code, and is organized in a key-value pair manner.

[0121] S42. Generate a DID document according to the W3C decentralized identity technical specification. The DID document includes a context field, an id field, an authentication field, and a verificationMethod field, wherein the context field declares the specification version, and the id field corresponds to the user's DID identifier.

[0122] S43. Set the public key identifier data based on elliptic curves in the verificationMethod field, and set the signature method identifier used for authentication in the authentication field;

[0123] S44. Add an extended field BioBinding to the DID document. The BioBinding field includes four sub-fields: bioMerkleRoot, bindingTime, dataType, and storageReference, which correspond to the biohash root value, binding time, feature data type, and on-chain storage address information of the nested biofeature vector, respectively.

[0124] S45. Encapsulate the DID document containing the BioBinding extended field into JSON-LD format.

[0125] Step S4 implements a standardized extension to the DID document structure. Without compromising the integrity of the W3C standard structure, a new BioBinding extension field is added to record the biometric hash root value and its related on-chain reference information. Through the refined design of fields such as bioMerkleRoot, bindingTime, dataType, and storageReference, modular expansion and field-level verifiability of the on-chain identity structure are achieved. This extension method not only has good system compatibility but also improves the efficiency of DID document parsing and the readability of the on-chain structure, effectively solving the problem that traditional DID systems cannot carry structured biometric information, and providing the identity system with a more complete and auditable on-chain representation capability.

[0126] In this embodiment, S5 specifically includes:

[0127] S51. Construct the input data set for the zero-knowledge proof protocol, and represent real-name identity data and biometric hash root values ​​as private inputs of the relationship to be proved;

[0128] S52. Introduce a zero-knowledge proof protocol for verification. Define constraints based on the relationship to be proved and represent the constraints in circuit form. The circuit form contains the relationship constraint logic between real-name identity data and biometric hash root value.

[0129] S53. Generate a zero-knowledge proof public-private key pair based on the constraints of the circuit form, wherein the private key is used to generate the proof and the public key is used for on-chain verification.

[0130] S54. Using the private key, perform zero-knowledge proof to generate proof data for the binding relationship between real-name identity data and biometric hash root value, and generate proof data including proof subject, random challenge number and response factor;

[0131] S55. The proof data is structured and digitally signed.

[0132] S56. Combine the digitally signed proof data with the corresponding public key to form a verification record, which serves as the verification record for the binding relationship between the real-name identity data and the biometric hash root value. The verification record includes signature digest and timestamp information.

[0133] Step S5 introduces a zero-knowledge proof protocol to achieve privacy verification of the binding relationship between real-name identity data and biometric hash root values, constructing an identity association mechanism that balances authenticity and privacy. Through circuit-level constraint modeling, key generation, proof construction, and signature encapsulation, a verifiable proof structure that can verify the binding relationship without disclosing the original data is completed. This scheme not only reduces the risk of sensitive information being exposed on-chain but also strengthens the auditability and non-repudiation capability of the binding relationship. Compared with existing plaintext signatures or static binding methods, it has a higher level of security and data protection capabilities, making it suitable for identity authentication scenarios with high privacy and high compliance requirements.

[0134] In this embodiment, when writing the DID document, BioBinding extended fields, and verification records into the blockchain, the pre-deployed smart contract registration interface is called to use the on-chain real-name DID record as the transaction payload and record it in the block event log.

[0135] By uniformly writing DID documents, BioBinding extended fields, and verification records as transaction payloads into the blockchain and recording them in the block event log, on-chain trusted evidence storage of real-name identity structures is achieved. This mechanism not only ensures the immutability and end-to-end traceability of data but also standardizes the data writing process through contract interfaces, improving system consistency and compatibility. Compared to traditional off-chain records or centralized database storage methods, this solution enhances the long-term trustworthiness and cross-entity sharing capabilities of identity data, representing a crucial technological path for achieving decentralized governance of real-name identity.

[0136] In this embodiment, S7 specifically includes:

[0137] S71. When performing identity verification, the user's single-modality lightweight biometric data is collected in real time through the terminal device, and the single-modality lightweight biometric data is subjected to scale normalization, noise filtering and standardization to obtain a feature preprocessing vector.

[0138] S72. A fixed random projection matrix and bias vector are stored locally on the terminal device, wherein the random projection matrix is ​​a predefined random numerical matrix and the bias vector is a predefined random constant vector. A summary vector is generated by mapping the feature preprocessing vector with the nonlinear activation function ReLU.

[0139] S73. For each element of the digest vector, the sign function of the Local Aware Hash algorithm is used to perform binary quantization dimension by dimension, converting each element into a corresponding binary sign bit, thus forming a 64-bit Local Aware Hash string:

[0140]

[0141] Among them, h i s represents the value of the i-th bit in the locally perceptual hash string. i This represents the i-th element in the summary vector;

[0142] S74. Establish a communication connection with the blockchain network through the terminal device, encapsulate the local perception hash string into an on-chain transaction payload, and submit it to the on-chain comparison engine.

[0143] S75. The on-chain comparison engine reads the on-chain real-name DID record, parses the DID document to verify the source validity, extracts the bio-hash root value from the BioBinding extended field, and verifies the signature digest and timestamp information in the record, and receives the local perceptual hash string and the bio-hash root value for matching.

[0144] S76. The on-chain comparison engine uses the Hamming distance function to calculate the similarity d between the locally perceptual hash string and the bio-hash root value. H :

[0145]

[0146] Among them, H bio,i This represents the value of the i-th bit of the bio hash root value, d H Output as the matching result.

[0147] d H Compare with the preset verification threshold θ, when d H When d ≤ θ, it means the verification passed; when d ≤ θ, it means the verification passed. H >θ indicates that the verification failed.

[0148] Step S7 proposes a fast verification mechanism based on lightweight feature extraction and fuzzy hash matching, significantly improving the system's response speed and fault tolerance. By mapping feature vectors to 64-bit hash strings using a local perceptual hash function and calculating Hamming distance with on-chain hash values ​​for similarity evaluation, a lightweight comparison process without uploading the original image is achieved. This method supports approximate matching and fault tolerance while ensuring privacy and security, effectively solving the problem of recognition failure due to biometric data collection errors in existing systems. It is suitable for applications with high real-time requirements, such as airport security checks, access control, and remote account opening.

[0149] In this embodiment, the multi-factor re-verification mechanism specifically includes calculating the DID confidence value based on user device type, historical login geographic location, access frequency, and verification behavior trajectory, and calculating the DID confidence change rate based on a sliding time window. When the confidence change rate exceeds a set confidence threshold, it is determined that the second verification has failed, and identity freezing is executed. By introducing a dynamic calculation mechanism for DID confidence value and change rate, risk perception and behavioral trend tracking of user identity status are achieved. By combining multi-dimensional behavioral data such as user device type, location trajectory, and access frequency to construct an identity situation map, and judging the confidence fluctuation trend based on a sliding time window, the system can trigger multi-factor re-verification when the risk is high, and automatically freeze the identity after verification failure. This mechanism greatly improves the system's response capability to abnormal behavior, has the advantages of dynamic controllability, self-learning, and state recovery, and is a key component for realizing the intelligent and defensive upgrade of the trusted identity management system.

[0150] Example 1:

[0151] To verify the feasibility of this invention in practice, it was applied to the intelligent identity verification system of a large-scale urban integrated government service center. This service center serves over 3,000 users daily, covering multiple high-security service windows including identity registration, household registration changes, social security processing, and real estate registration. The original system primarily relied on users presenting physical ID cards, comparing their identities with those of the ID card reader and a facial recognition module. However, in actual operation, problems frequently arose such as lost documents, low device recognition rates, and significant light interference. Especially under conditions of multiple concurrent users, high environmental noise, or users wearing masks, the system's false recognition rate was high, requiring frequent manual intervention and resulting in a poor user experience. The center urgently needed a more efficient, secure, and privacy-friendly identity authentication solution that could improve access efficiency and system stability while ensuring compliance with real-name registration requirements.

[0152] This invention deploys a rapid verification system that fuses real-name DID with multimodal biometrics at the center's verification portal and business front end. Upon initial system access, on-site personnel assist users in inputting their facial image, fingerprint image, and voiceprint data via a self-service terminal, and simultaneously verify their identity using real-name information accessed through the public security portal. After data collection, the system fuses and encodes the three modalities using the variational nested encoder described in this invention, constructs a Merkle hash tree to generate a biometric hash root value, and then generates a DID document containing the BioBinding extended field. A zero-knowledge proof protocol is used to verify the relationship between the user's real-name information and the biometric hash, and the complete DID structure is written to the blockchain. The entire process is modeled and compressed off-chain, with only the structure hash value and verification record uploaded to the blockchain, ensuring that privacy data remains on-chain and not left on the blockchain.

[0153] When a user revisits the center or performs online authentication, they only need to collect lightweight biometric information of either modality via a facial recognition camera or fingerprint scanning module. The system will generate a compressed digest vector locally and map it to a 64-bit hash string using Local Aware Hash (LAH) and submit it to the on-chain comparison engine. The system extracts the biometric hash root value from the BioBinding field in the user's bound DID document and calculates the Hamming distance to determine if the verification standard is met. If the user's behavior environment is stable, the device is consistent, and the DID confidence level is normal, the system will quickly return the authorization result, and the business window will directly allow access. If there are signs of access from a different location, frequent device switching, or other risk signals, the system will initiate a multi-factor re-verification process, prompting the user to supplement with other modalities or perform secondary verification, and automatically adjust the DID confidence level based on the behavioral graph.

[0154] Based on operational data collected over six months after deployment, the system demonstrated significant advantages in both verification accuracy and user access efficiency. Compared to the original system, the biometric false rejection rate decreased from 2.7% to 0.6%, and the multimodal verification pass rate increased to 99.2%. Regarding identity verification time, the initial DID entry took an average of approximately 14.8 seconds, while subsequent verifications took only 2.6 seconds. More notably, the on-chain identity structure exhibited strong traceability and version control capabilities. On average, each user performed 1.7 identity updates within six months, yet no conflicts or failed identity reconstruction records were observed in the overall on-chain structure. In handling suspected abnormal access incidents, 52 high-risk logins were identified through DID confidence sliding analysis, successfully blocking 37 risky paths. Four cases involving identity theft were investigated in assistance to the police, all of which could be fully traced back through system logs.

[0155] This embodiment demonstrates that the present invention can achieve the goals of data privacy protection, improved verification accuracy, and optimized system performance in high-concurrency, high-risk identity verification scenarios, and has strong practicality, good scalability, and significant commercial value.

[0156] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-name DID and biometric fusion rapid verification system, characterized in that, Includes the following modules: The biometric processing module is used to collect users' multimodal biometric data and generate nested biometric vectors through a variational nested encoder. The variational nested encoder model includes a modal feature extraction layer, a cross-modal fusion layer, a nested variational encoder layer, and a compression mapping output layer; The multimodal biometric data is input into the modality feature extraction layer, and convolutional encoding, normalization, and activation processing are performed on each modality to extract preliminary modality features. The preliminary modal features are input into the cross-modal fusion layer, which includes a multi-head attention mechanism and a gating mechanism to dynamically adjust the weight relationship between each modality and generate a modal fusion representation. The modal fusion representation is input into the nested variational encoder layer, which consists of a master encoder and three sub-encoders. The master encoder uses a multilayer perceptron structure to model the modal fusion representation and is used to extract the global latent mean and variance vector shared by all modalities. The three sub-encoders correspond to facial images, fingerprint images, and voiceprint signals, respectively. They adopt independent variational channels in the shared architecture to model the local variation information of each modality and output modality-specific latent vectors. The global latent mean and variance vector output by the main encoder are fused with the modal local latent vectors output by each sub-encoder to generate nested latent feature vectors. The nested latent feature vectors are input into the compressed mapping output layer, and the structure-preserving nested biological feature vectors are obtained through nonlinear dimensionality reduction mapping. The hash generation module is used to perform hash operations on the nested biological feature vectors and construct a Merkle tree to generate biological hash root values. The real-name verification module is used to verify the authenticity of users' real-name information and generate real-name identity data; The DID construction module is used to generate DID documents that conform to the W3C standard format and add the BioBinding extension field to record the on-chain reference address of the biohash root value. The BioBinding field includes four sub-fields: bioMerkleRoot, bindingTime, dataType, and storageReference, which correspond to the biohash root value, binding time, feature data type, and on-chain storage address information of the nested biofeature vector, respectively. The proof generation module is used to construct zero-knowledge proofs based on the real-name identity data and the biometric hash root value, and generate verification records; The blockchain writing module is used to write the DID document, BioBinding extended fields and verification records into the blockchain to form an on-chain real-name DID record; The verification module is used to collect lightweight biometric data of a user's single modality during identity verification, generate a summary vector, and match it with the real-name DID record on the blockchain. The confidence control module is used to compare the matching results with the preset verification threshold. If the verification is successful, the authorized access is completed. If the verification fails, a multi-factor re-verification mechanism is triggered for a second verification, and the DID confidence value is calculated. If the second verification still fails, the identity is frozen.

2. The real-name DID and biometric fusion quick verification system according to claim 1, characterized in that, The modules are connected in the following way: S1. Collect the user's multimodal biometric data and input it into the variational nested encoder model, then fuse and compress the mapping to generate a nested biometric vector; S2. Perform a hash operation on the nested biological feature vectors and construct a Merkle tree to generate a biological hash root value; S3. Obtain the real-name information provided by the user, verify the authenticity of the real-name information, and generate real-name identity data; S4. Generate a DID document conforming to the W3C standard format based on the real-name identity data, and add a BioBinding extended field to record the on-chain reference address of the bio-hash root value; S5. Construct a zero-knowledge proof protocol input, use the zero-knowledge proof protocol to verify the binding relationship between the real-name identity data and the biometric hash root value, and generate a verification record; S6. Write the DID document, BioBinding extended fields and verification records into the blockchain to generate an immutable on-chain real-name DID record. S7. When performing identity verification, collect the user's single-modality lightweight biometric data, process the single-modality lightweight biometric data locally to generate a summary vector, and submit it to the on-chain comparison engine to match it with the on-chain real-name DID record. S8. Compare the matching results with the preset verification threshold. If the verification passes, the authorized access is completed. If the verification fails, a multi-factor re-verification mechanism is triggered for a second verification, and the DID confidence value is calculated. If the second verification still fails, the identity is frozen.

3. The real-name DID and biometric fusion quick verification system according to claim 2, characterized in that, The multimodal biometric data includes facial images, fingerprint images, and voiceprint signals.

4. The real-name DID and biometric fusion quick verification system according to claim 2, characterized in that, S2 specifically includes: S21. The nested biometric vectors are structurally divided into several sub-vector segments according to the modality source and feature level. Each sub-vector segment corresponds to a set of features in a biometric modality or coding level, which is used for distributed hash calculation. S22. The SHA-3 hash algorithm is used to perform three rounds of hashing operations on each sub-vector segment. The first round is to perform hashing after vector standardization, the second round is to perform hashing after vector dimension rotation, and the third round is to perform scrambling hashing after introducing mask parameters, generating multiple hash sub-values ​​as leaf node values ​​of the Merkle tree. S23. Construct the underlying structure of the Merkle tree. Use sequential pairing to concatenate two adjacent hash values ​​and then execute the SHA-3 hash algorithm again to generate the corresponding parent node hash value. Iterate upwards layer by layer to generate intermediate nodes until the root node is generated. S24. During the construction of each level of the Merkle tree, a modality identifier label is attached to each intermediate node. The modality identifier label represents the modality source combination contained in the intermediate node through an encoding mapping method, which is used to trace the modality composition relationship. S25. A local update mechanism is introduced during the construction process. The input hash combination relationship of each intermediate node is stored in a local buffer and an index table is built. When a certain sub-vector segment changes, only the corresponding branch and the upper-level node are updated to avoid repeated construction operations of the entire Merkle tree. S26. After completing the calculation of all intermediate nodes, the final output is the biological hash root value, which serves as the unique identifier hash root value of the nested biological feature vector.

5. The rapid verification system for fusing real-name DID and biometrics according to claim 2, characterized in that, S4 specifically includes: S41. Construct a basic identity information object based on real-name identity data. The basic identity information object includes a user identifier, an issuing authority identifier, an issuance timestamp, an expiration timestamp, and an identity recognition code, and is organized in a key-value pair manner. S42. Generate a DID document according to the W3C decentralized identity technical specification. The DID document includes a context field, an id field, an authentication field, and a verificationMethod field, wherein the context field declares the specification version, and the id field corresponds to the user's DID identifier. S43. Set the public key identifier data based on elliptic curves in the verificationMethod field, and set the signature method identifier used for authentication in the authentication field; S44. Add the extended field BioBinding to the DID document; S45. Encapsulate the DID document containing the BioBinding extended field into JSON-LD format.

6. The real-name DID and biometric fusion quick verification system according to claim 2, characterized in that, S5 specifically includes: S51. Construct the input data set for the zero-knowledge proof protocol, and represent real-name identity data and biometric hash root values ​​as private inputs of the relationship to be proved; S52. Introduce a zero-knowledge proof protocol for verification. Define constraints based on the relationship to be proved and represent the constraints in circuit form. The circuit form contains the relationship constraint logic between real-name identity data and biometric hash root value. S53. Generate a zero-knowledge proof public-private key pair based on the constraints of the circuit form, wherein the private key is used to generate the proof and the public key is used for on-chain verification. S54. Using the private key, perform zero-knowledge proof to generate proof data for the binding relationship between real-name identity data and biometric hash root value, and generate proof data including proof subject, random challenge number and response factor; S55. The proof data is structured and digitally signed. S56. Combine the digitally signed proof data with the corresponding public key to form a verification record, which serves as the verification record for the binding relationship between the real-name identity data and the biometric hash root value. The verification record includes signature digest and timestamp information.

7. The real-name DID and biometric fusion quick verification system according to claim 2, characterized in that, When writing the DID document, BioBinding extended fields, and verification records to the blockchain, the pre-deployed smart contract registration interface is called to use the on-chain real-name DID record as the transaction payload and record it in the block event log.

8. The real-name DID and biometric fusion quick verification system according to claim 2, characterized in that, Specifically, S7 includes: S71. When performing identity verification, the user's single-modality lightweight biometric data is collected in real time through the terminal device, and the single-modality lightweight biometric data is subjected to scale normalization, noise filtering and standardization to obtain a feature preprocessing vector. S72. A fixed random projection matrix and bias vector are stored locally on the terminal device, wherein the random projection matrix is ​​a predefined random numerical matrix and the bias vector is a predefined random constant vector. A summary vector is generated by mapping the feature preprocessing vector with the nonlinear activation function ReLU. S73. For each element of the digest vector, the sign function of the Local Aware Hash algorithm is used to perform binary quantization dimension by dimension, converting each element into a corresponding binary sign bit, thus forming a 64-bit Local Aware Hash string: ； wherein, denotes the value of the bit at position in the local perceptual hashing string, denotes the element at position in the digest vector; S74. Establish a communication connection with the blockchain network through the terminal device, encapsulate the local perception hash string into an on-chain transaction payload, and submit it to the on-chain comparison engine. S75. The on-chain comparison engine reads the on-chain real-name DID record, parses the DID document to verify the source validity, extracts the bio-hash root value from the BioBinding extended field, and verifies the signature digest and timestamp information in the record, and receives the local perceptual hash string and the bio-hash root value for matching. S76. The on-chain comparison engine uses the Hamming distance function to calculate the similarity between the locally perceptual hash string and the bio-hash root value. : ； in, The first value representing the biological hash root value The value of each bit will Output as the matching result.

9. The rapid verification system for fusing real-name DID and biometrics according to claim 2, characterized in that, The multi-factor re-verification mechanism specifically includes calculating the DID confidence value based on user device type, historical login geographic location, access frequency and verification behavior trajectory, and calculating the DID confidence change rate based on a sliding time window. When the confidence change rate exceeds the set confidence threshold, it is determined that the second verification has failed and the identity is frozen.

Images