A method and system for generating a tamper-proof identification code for a PDF document

Abstract

The present application relates to the technical field of computer application, especially to a PDF document tamper-proof identification code generation method and system;Including the following steps: S1, using multi-modal hierarchical extraction technology, construct physical-logical-visual three-dimensional feature system, comprehensive extraction of multi-modal features of PDF document;S2, according to the specific environment, time and biological characteristics of the user, for different users to generate corresponding dynamic key;S3, using the RLWE in the lattice cryptography, realize multi-feature fusion, generate quantum resistance identification code element;S4, using space-time folding embedding technology, against quantum identification code element, through the least significant bit algorithm, the quantum resistance identification code element is dispersed embedded in multiple positions, realize the invisible multiple copy distribution in PDF;S5, the result of S4 step processing, multilevel verification;Improve the tamper-proof ability of PDF document.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer application, and particularly relates to a PDF document tamper-proof identification code generation method and system. BACKGROUND

[0002] In the current rapid development of digital information, PDF documents are widely used in fields with extremely high security requirements such as electronic contracts, judicial evidence, medical archives, etc. At present, the existing technologies for protecting the security of PDF documents mainly have the following problems:

[0003] Traditional hash algorithm: The traditional hash calculation method only generates a fixed hash value based on the document content, and the calculation method is single and the security is limited. In the face of increasingly powerful computing power, it is easy to be subjected to brute force attack, resulting in tampering of the document content but unable to effectively detect. For example, using simple MD5 or SHA-1 hash algorithm, attackers can try to collide through a large amount of computing resources to hide the tampered content by forging the same hash value.

[0004] Fixed digital signature technology: Common digital signature schemes are mostly fixed modes and have no association with the user's use environment. Once the signature file is obtained, attackers can tamper with the document in an offline environment and it is difficult to detect. This signature method can only detect the integrity of the document level and cannot be accurate to the object level in the document, and it is difficult to locate and identify the tampering of part of the content in the complex PDF document.

[0005] Single feature verification: Some existing verification methods only rely on a certain feature of the document for verification, such as only verifying the content hash or only checking the document format. This makes it possible for attackers to bypass detection by modifying metadata, page visual elements and other features that are not checked, and it is impossible to comprehensively protect the security of the document.

[0006] The prior art CN118536170A discloses a PDF file signature and tamper-proofing method, which comprises the following steps: adding a signature appearance to a specified position of a PDF file; adding a lock all file content attribute to the PDF file, obtaining lock attribute information, and taking the lock attribute information as part of the signature data; calculating a hash value of the overall PDF file data according to the PDF specification, and reserving a signature value space during the calculation process; the overall PDF file data includes the PDF file original content, the newly added signature appearance, and the lock attribute information; signing the hash value to obtain a signature value; and writing the signature value into the reserved signature value space. Although this method proposes a PDF file signature and tamper-proofing method, it only relies on traditional hash algorithms and digital signature technology, and the security is limited, which is vulnerable to brute force cracking and quantum computing attacks. Secondly, this method can only detect the integrity at the document level, and cannot be accurate to the object level, so it is difficult to locate and identify the tampering of part of the content in a complex PDF document. In addition, the digital signature is not related to the user environment, and attackers can tamper with the file in an offline environment without being easily detected.

[0007] The prior art "New method of preventing forgery of official documents based on multiple image hiding" (Electric Power Big Data, 2021, 24(09): 1-8.) provides an ant colony algorithm based on multiple image hiding identification, which uses Fourier series, K-mean processing and Metropolis acceptance criterion to judge the relevant data in official documents, and inserts hidden graphics in three-dimensional multiple graphics to improve the anti-tampering ability of official document identification. This technical solution improves the anti-forgery ability and anti-tampering performance of official documents through multiple image hiding identification and related algorithms, but has problems such as long processing time, limited initial anti-tampering ability, and high system complexity.

[0008] In summary, the existing PDF document tamper-proofing methods have low security, coarse detection granularity, poor environmental relevance, and poor tamper-proofing ability.

[0009] Therefore, there is an urgent need to provide a PDF document tamper-proofing identification code generation method and system that improves the PDF document tamper-proofing ability compared to the prior art. SUMMARY

[0010] The present application solves the technical problems existing in the prior art, and provides a PDF document tamper-proofing identification code generation method and system.

[0011] To achieve the above-mentioned purposes, the technical solutions adopted by the present application are as follows:

[0012] A PDF document tamper-proofing identification code generation method, comprising the following steps:

[0013] S1, using a multi-modal hierarchical extraction technology, a physical-logical-visual three-dimensional feature system is constructed, and multi-modal features of the PDF document are comprehensively extracted, including binary features, logical structure features and visual features;

[0014] S2, according to the specific environment, time and biological characteristics of the user, a corresponding dynamic key is generated for different users;

[0015] S3, the binary features, logical structure features and visual features obtained in step S1 are fused using RLWE in lattice cryptography to generate an anti-quantum identification code primitive;

[0016] S4, using the space-time folding embedding technology, the anti-quantum identification code primitive obtained in step S3 is dispersedly embedded in multiple positions by the least significant bit algorithm to realize the distribution of multiple copies in the PDF;

[0017] S5, the results processed in step S4 are subjected to multi-level verification, including comparison level verification, semantic level verification and visual traceability verification.

[0018] Further, the physical-logical-visual three-dimensional feature system constructed in step S1 includes physical layer fingerprint, logical layer fingerprint and visual layer fingerprint.

[0019] Further, the physical layer fingerprint in step S1 processes the PDF binary stream using Merkle-Patricia tree to obtain the binary features of the PDF document.

[0020] Further, the logical layer fingerprint in step S1 parses the PDF object tree to extract the XPath path and encryption attribute of the / Catalog and / Pages key objects, and processes the XPath path and encryption attribute using the cuckoo hash algorithm to generate a structure feature code.

[0021] Further, the visual layer fingerprint in step S1 uses a lightweight CNN model to extract visual features of each page of the PDF document to obtain a high-dimensional visual feature vector, and then converts it to a 128-bit visual abstract code through PCA dimension reduction technology to obtain visual features.

[0022] Further, S2 specifically includes the following steps:

[0023] S21, obtaining user environment, timestamp and biological characteristic information;

[0024] S22, the user environment, timestamp and biometric information obtained in S21 are processed, specifically: the obtained hardware fingerprint is processed to generate a hardware signature; the timestamp is processed to generate a quantum secure timestamp; the collected biometric features are processed to calculate a biometric hash value;

[0025] S23, the hardware signature, quantum secure timestamp and biometric hash value are fused;

[0026] S24, the fused information is hashed using the keccak256 algorithm to generate a dynamic key, which includes a private key and a public key.

[0027] Further, S3 specifically includes the following steps:

[0028] S31, a homomorphic encryption framework based on RLWE is constructed, binary features, logical structure features and visual features are formed into feature vectors, and the feature vectors are processed in the encryption domain using the homomorphic encryption framework to obtain corresponding ciphertexts;

[0029] S32, the binary features, logical structure features and visual features obtained in S31 are fused by weighting the feature vectors in the encryption domain to generate an anti-quantum identification code primitive.

[0030] Further, in S31, the corresponding ciphertexts are obtained by the following formula:

[0031] ;

[0032] In the above formula, C represents the ciphertext of the encrypted feature vector, represents a random polynomial, represents the private key in the dynamic key, represents an error distribution; represents the feature vector, which is a binary feature or a logical structure feature or a visual feature; represents a necessary component of the encryption operation, which is used to limit the polynomial operation result within a certain range with q as the modulus.

[0033] Further, in S31, the random polynomial is a polynomial randomly selected from where represents a set of integers modulo q, and n represents the dimension or length of the polynomial.

[0034] Further, S4 specifically includes the following steps:

[0035] S41, the anti-quantum identification code primitive is embedded in the zlib compression bit stream of the PDF object stream by replacing the least significant bits.

[0036] S42, embedding the anti-quantum identification code element in the vector path control point coordinate fine adjustment of the page content;

[0037] S43, embedding the anti-quantum identification code element in the CID mapping table of the font subset.

[0038] Further, the comparison level check method in the S5 step is: using Reed-Solomon encoding technology, when the anti-quantum identification code element is disturbed or damaged in the transmission or storage process, the damaged anti-quantum identification code element fragment can be recovered, and the recovered identification code is checked to determine whether the document is complete at the binary level.

[0039] Further, the semantic level check method in the S5 step is: verifying the version consistency of the logical layer fingerprint by means of the smart contract, the smart contract can record and verify the historical version of the document logical structure feature code, and if the logical layer fingerprint of the current document is inconsistent with the version recorded in the smart contract, it indicates that the logical structure of the document is tampered.

[0040] Further, the visual traceability check method in the S5 step is: using the Siamese network to compare the original visual fingerprint with the real-time rendering result, the Siamese network compares the similarity of two visual feature vectors, and if the similarity is lower than the set threshold, it indicates that the visual content of the document is tampered.

[0041] A PDF document tamper-proof identification code generation system, comprising a multi-modal feature extraction module, a dynamic key binding module, an anti-quantum aggregation module, a space-time folding embedding module and a multi-level verification module, the multi-modal feature extraction module is used for executing S1 step, the dynamic key binding module is used for executing S2 step, the anti-quantum aggregation module is used for executing S3 step, the space-time folding embedding module is used for executing S4 step, and the multi-level verification module is used for executing S5 step.

[0042] Compared with the prior art, the beneficial effects of the present application are:

[0043] (1) The present application has high security and anti-attack ability: in the FIPS186-5 standard test, the collision resistance reaches 2^256 level, the RLWE encryption technology is used to resist quantum computing attack, compared with the traditional digital signature scheme, the security is improved by 3 orders of magnitude, and the PDF document is effectively protected from various attack means.

[0044] (2) The application achieves object-level tamper detection: through multi-modal feature fusion, object-level tamper in the document can be accurately detected, and the tamper detection granularity is improved from the document level to the object level, which can ensure the security of each key information in the document for the scene with high requirement for data integrity such as electronic contract and judicial evidence.

[0045] (3) The application realizes environmental association and dynamic protection: the identification code is dynamically bound with the user environment, and automatically invalidates when it is separated from the authorized environment, which effectively prevents offline tampering and enhances the security of the document in different use environments.

[0046] (4) The application has high survival rate and stability: the identification code has a 100% survival rate for the "save as" operation of Adobe Acrobat, and can still survive after common operations such as file repair and format conversion, which ensures the continuity and stability of verification.

[0047] (5) The application realizes rapid verification and response: when detecting single-pixel-level content tampering, the verification time is <50ms (i7-1280P test platform), which can quickly find tampering behavior and take measures to protect the security of the document. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 is a flowchart of the method of the application.

[0049] Figure 2 is a flowchart of the dynamic key generation of the application. DETAILED DESCRIPTION

[0050] The technical solutions of the application will be described clearly in the following description with reference to the accompanying drawings. Obviously, the described embodiments are not all the embodiments of the application, and all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0051] Embodiment 1

[0052] As shown in the figure, the embodiment provides a PDF document anti-tamper identification code generation method, which comprises the following steps: Figure 1

[0053] S1, using multi-modal hierarchical extraction technology, a physical-logical-visual three-dimensional feature system is constructed, and multi-modal features of the PDF document are comprehensively extracted, including binary features, logical structure features and visual features.

[0054] ​The physical system is the physical layer fingerprint: the improved Merkle-Patricia tree (MPT) is used to process the PDF binary stream to obtain the binary features of the PDF document. A 4KB block unit is used to calculate the SHA3-512 hash value for each unit, and a large number of hash values form the bottom hash forest. This method can accurately capture the features of the binary data at the bottom of the PDF document. Even if the binary data changes slightly, the hash value will also change, thereby effectively detecting tampering at the physical level.

[0055] The logical system is the logical layer fingerprint: the PDF object tree (ObjectTree) is parsed, and the XPath path and encryption attributes of key objects such as / Catalog and / Pages are extracted. The information is processed using the cuckoo hash algorithm to generate a structure feature code. The feature code reflects the logical structure features of the PDF document. If the page organization, object relationship, and other logical structures of the document change, the structure feature code will also change accordingly.

[0056] The visual system is the visual layer fingerprint: a lightweight CNN model (ViT-Tiny) is used to extract visual features from each page of the PDF document, obtaining a high-dimensional visual feature vector. Then, the PCA dimension reduction technique is used to convert it into a 128-bit visual summary code. Even if there are slight changes in the visual presentation of the document content, such as text layout and image display, they can still be detected.

[0057] S2, according to the specific environment, time and biological characteristics of the user, a corresponding dynamic key is generated for different users. The dynamic binding of the identification code and the user environment is realized, and the identification code is invalidated in an unauthorized environment. The hardware fingerprint is obtained from the trusted platform module (TPM), and the precise timestamp is obtained through the quantum-resistant network time protocol (NTP). If the user enables biological feature verification, the biological features will also be collected and the hash value will be calculated. After fusing these information, the keccak256 algorithm is used to generate a dynamic key. The key is associated with the user's specific hardware environment, time and biological characteristics. Different users, different times or different hardware environments generate different keys. As shown in Figure 2 the following steps are included:

[0058] S21, obtain user environment, timestamp, and biological feature information.

[0059] Obtain user environment information, obtain hardware fingerprint information from user equipment, which usually involves the TPM module (trusted platform module).

[0060] Obtain the timestamp, and use the quantum-resistant NTP (network time protocol) to ensure the accuracy and security of the timestamp.

[0061] Acquiring biometric information, including but not limited to fingerprints, facial recognition, etc.

[0062] S22, processing the user environment, timestamp and biometric information obtained in step S21, specifically: processing the obtained hardware fingerprint to generate a hardware signature; processing the timestamp to generate a quantum secure timestamp to ensure its security; processing the collected biometric features to calculate a biometric hash value.

[0063] S23, fusing the hardware fingerprint, timestamp and biometric hash value.

[0064] S24, using the keccak256 algorithm to perform hash operation on the fused information to generate a dynamic key, which includes a private key and a public key.

[0065] Through step S2, the dynamic key is closely associated with the user's hardware environment, timestamp and biometric features, ensuring that the keys generated under different users, different times or different hardware environments are unique, thus realizing the dynamic binding of the identification code and the user environment. This dynamic key generation mechanism enhances the security of PDF documents and prevents unauthorized access and tampering.

[0066] S3, using RLWE (Ring Learning With Errors) in lattice cryptography to realize multi-feature fusion of the binary features, logical structure features and visual features obtained in step S1 to generate an anti-quantum identification code primitive that resists quantum computing attacks. Specifically, the following steps are included:

[0067] S31, constructing a homomorphic encryption framework based on RLWE, forming feature vectors from binary features, logical structure features and visual features respectively, and using the homomorphic encryption framework to process the feature vectors in the encryption domain to obtain corresponding ciphertexts, denoted as:

[0068] ;

[0069] In the above formula, C represents the ciphertext after encryption of the feature vector, represents a random polynomial, represents the private key in the dynamic key, represents an error distribution; represents the feature vector, which is binary features or logical structure features or visual features; represents a necessary component of the encryption operation, used to limit the polynomial operation result within a certain range with q as the modulus.

[0070] Random polynomial is a polynomial randomly selected from , where A set of integers representing the modulus q, n represents the dimension or length of the polynomial.

[0071] S32, the ciphertext of the binary features, logical structure features and visual features obtained in step S31 is fused in the encryption domain to generate an anti-quantum identification code element. The feature vectors of different modalities are given different weights according to their importance, and the fusion is performed in an encrypted state to ensure that the fused identification code element has the ability to resist quantum computing attacks, that is, it is difficult to crack and tamper even in the face of the powerful computing power of a quantum computer.

[0072] S4, using a space-time folding embedding technology, the anti-quantum identification code element obtained in step S3 is dispersedly embedded in multiple positions through an improved least significant bit algorithm (LSB) to realize the distribution of multiple copies in the PDF. Specifically:

[0073] S41, the anti-quantum identification code element is embedded in the zlib compressed bit stream of the PDF object stream ( / Stream) by replacing the least significant bit. Since the zlib compressed bit stream has a large amount of data, this embedding method has little effect on the overall document and is difficult to detect.

[0074] S42, the anti-quantum identification code element is embedded in the vector path control point coordinate fine adjustment of the page content. By slightly adjusting the coordinate values (within the range that does not affect the visual effect of the page), the identification code is hidden, making it difficult for attackers to discover it through conventional means.

[0075] S43, the anti-quantum identification code element is embedded in the CID mapping table of the font subset ( / FontSubset). By taking advantage of the redundancy of the CID mapping table, the identification code is cleverly hidden.

[0076] A self-destructive mark is added to each copy. Once tampering of the document is detected, the corresponding identification code copy will trigger a self-erasing mechanism to prevent the tampered identification code copy from misleading the verification result.

[0077] S5, the result after step S4 is processed, multi-level verification is performed, including comparison level verification, semantic level verification, and visual traceability verification, to build a full-stack verification system from the bit level to the semantic level, and comprehensively detect whether the document has been tampered with.

[0078] Bit-level verification: Reed-Solomon encoding technology is used. When the identification code is disturbed or damaged during transmission or storage, the damaged identification code fragments can be recovered to ensure the accuracy of the verification. By checking the recovered identification code, it is determined whether the document is complete at the binary level.

[0079] Semantic level verification: The consistency of the version of the logical layer fingerprint is verified by the smart contract verification logic. The smart contract can record and verify the historical version of the document logical structure feature code. If the logical layer fingerprint of the current document is inconsistent with the version recorded in the smart contract, it means that the logical structure of the document may be tampered with.

[0080] Visual traceability: The original visual fingerprint is compared with the real-time rendering result using a Siamese network. The Siamese network can efficiently compare the similarity of two visual feature vectors. If the similarity is lower than the set threshold, it indicates that the visual content of the document may be tampered with.

[0081] Embodiment 2

[0082] The embodiment provides a PDF document tamper-proof identification code generation system, which includes a multi-modal feature extraction module, a dynamic key binding module, an anti-quantum aggregation module, a space-time folding embedding module, and a multi-level verification module. The multi-modal feature extraction module is used to perform the S1 step in the embodiment, the dynamic key binding module is used to perform the S2 step in embodiment 1, the anti-quantum aggregation module is used to perform the S3 step in embodiment 1, the space-time folding embedding module is used to perform the S4 step in embodiment 1, and the multi-level verification module is used to perform the S5 step in embodiment 1.

[0083] Embodiment 3

[0084] In the electronic contract scenario, there is an important PDF format electronic contract, which can be processed as follows:

[0085] 1. Multi-modal feature hierarchical extraction: The physical layer fingerprint of the electronic contract PDF document is extracted, and the improved Merkle-Patricia tree is used to calculate the SHA3-512 hash according to 4KB block, and the bottom hash forest is constructed. The object tree is parsed, the key object information is extracted to generate the logical layer structure feature code, and the visual layer feature vector is extracted using the ViT-Tiny model and dimensionally reduced to generate the visual abstract code.

[0086] 2. Dynamic key binding: The user signs the contract on the device, and the trusted platform module (TPM) of the device generates a hardware fingerprint. The timestamp is obtained through the quantum-resistant NTP. If the user enables fingerprint recognition and other biometric verification, the fingerprint is collected and the hash value is calculated. The dynamic key is generated by fusing these information.

[0087] 3. Anti-quantum aggregation algorithm: The physical layer, logical layer and visual layer feature vectors are encrypted and weighted fused in the RLWE-based homomorphic encryption framework to generate an anti-quantum identification code primitive.

[0088] 4. Spatiotemporal folding embedding technology: using an improved LSB algorithm, the identification code is dispersed and embedded in the zlib compressed bit stream of the object stream of the electronic contract PDF document, the page vector path control point coordinates and the font subset CID mapping table, and a self-destruction mark is attached to each copy.

[0089] Multi-level verification engine: when verifying whether the electronic contract is tampered with, first, bit-level checking is performed, and if the identification code part is damaged, Reed-Solomon encoding is used for recovery. Then, semantic-level checking is performed, and the version consistency of the logic layer fingerprint is verified through the smart contract. Finally, visual tracing is performed, and the Siamese network is used to compare the original visual fingerprint with the real-time rendering result. If any one level of checking fails, the document is determined to be tampered with.

[0090] The present application has high security and attack resistance: in the FIPS186-5 standard test, the collision resistance reaches the level of 2^256, the RLWE encryption technology is used to resist quantum computing attacks, and compared with the traditional digital signature scheme, the security is improved by 3 orders of magnitude, effectively protecting the PDF document from various attack means.

[0091] The present application achieves object-level tamper detection: through multi-modal feature fusion, the object-level tamper in the document can be accurately detected, and the tamper detection granularity is improved from the document level to the object level. For electronic contracts, judicial evidence and other scenes with high requirements for data integrity, the security of each key information in the document can be ensured.

[0092] The present application realizes environmental association and dynamic protection: the identification code is dynamically bound with the user environment, and automatically invalidates when it is separated from the authorized environment, effectively preventing offline tampering and enhancing the security of the document in different use environments.

[0093] The present application has high survival rate and stability: the "save as" operation of Adobe Acrobat maintains 100% identification code survival rate, and the identification code can still survive after common operations such as file repair and format conversion, ensuring the continuity and stability of the verification.

[0094] The present application has fast verification and response: when detecting single-pixel-level content tampering, the verification time is <50ms (i7-1280P test platform), which can quickly detect tampering behavior and take timely measures to protect the security of the document.

[0095] Finally, it should be noted that the above content is only used to illustrate the technical solutions of the present application, and is not a limitation on the protection scope of the present application. Simple modifications or equivalent replacements of the technical solutions of the present application made by those skilled in the art do not deviate from the essence and scope of the technical solutions of the present application.

Claims

1. A method for generating a tamper-evident identification code for a PDF document, characterized by, Comprise the following steps: S1, using multi-modal hierarchical extraction technology, construct physical-logical-visual three-dimensional feature system, comprehensive extraction of multi-modal features of PDF document, including binary features, logical structure features and visual features; S2, according to the specific environment, time and biological characteristics of the user, different users are generated corresponding dynamic key; S3, the binary features, logical structure features and visual features obtained in step S1 are processed in the encryption domain to obtain the corresponding ciphertext, and the RLWE in lattice cryptography is used to realize multi-feature fusion to generate quantum-resistant identification code primitive; The corresponding ciphertext is obtained by the following formula: ； In the above formula, C represents the ciphertext after the eigenvector encryption, represents a random polynomial, represents a private key in the dynamic key, represents an error distribution; represents an eigenvector, which is a binary feature or a logical structure feature or a visual feature; represents a necessary component of an encryption operation, which is used to limit the polynomial operation result in a specific range with q as the modulus; S4, using space-time folding embedding technology, the quantum-resistant identification code primitive obtained in step S3 is dispersedly embedded in multiple positions by the least significant bit algorithm, and the invisible multiple copy distribution in PDF is realized; Specifically, the following steps are included: S41, in the zlib compression bit stream of PDF object stream, the quantum-resistant identification code primitive is embedded by using the least significant bit replacement method; S42, the quantum-resistant identification code primitive is embedded in the vector path control point coordinate fine adjustment of page content; S43, the quantum-resistant identification code primitive is embedded in the CID mapping table of font subset; S5, the results processed in step S4 are subjected to multi-level verification, including comparison level check, semantic level check and visual traceability check.

2. The method of claim 1, wherein the method further comprises: The physical-logical-visual three-dimensional feature system constructed in step S1 includes physical layer fingerprint, logical layer fingerprint and visual layer fingerprint.

3. The method of claim 2, wherein the method further comprises: The physical layer fingerprint in step S1 processes the PDF binary stream by using Merkle-Patricia tree to obtain the binary features of the PDF document.

4. The method of claim 2, wherein the method further comprises: The logical layer fingerprint in step S1 parses the PDF object tree, extracts the XPath path and encryption attribute of / Catalog and / Pages key objects, and processes the XPath path and encryption attribute by using cuckoo hash algorithm to generate structure feature code.

5. The method of claim 2, wherein the method further comprises: The visual layer fingerprint in step S1 adopts a lightweight CNN model to extract visual features of each page of the PDF document to obtain a high-dimensional visual feature vector, and then converts it into a 128-bit visual abstract code by PCA dimension reduction technology, thereby obtaining visual features.

6. The method of claim 1, wherein the method further comprises: S2 specifically includes the following steps: S21, obtain user environment, timestamp and biological characteristic information; S22, process the user environment, timestamp and biological characteristic information obtained in step S21, specifically: process the obtained hardware fingerprint to generate a hardware signature; Process the timestamp to generate a quantum secure timestamp; Process the collected biological characteristics to calculate a biological characteristic hash value; S23, fuse the hardware signature, quantum secure timestamp and biological characteristic hash value; S24, use keccak256 algorithm to hash the fused information to generate a dynamic key, which includes a private key and a public key.

7. The method of claim 1, wherein the method further comprises: S3 specifically includes the following steps: S31, a homomorphic encryption framework based on RLWE is constructed, binary features, logical structure features and visual features are formed into feature vectors respectively, and the feature vectors are processed in the encryption domain using the homomorphic encryption framework to obtain corresponding ciphertexts; S32, the ciphertexts of the binary features, the logical structure features and the visual features obtained in step S31 are weighted and fused in the encryption domain to generate an anti-quantum identification code primitive.

8. The method of claim 7, wherein the method further comprises: In step S31, a random polynomial is selected randomly from where denotes a set of integers modulo q, and n denotes the dimension or length of the polynomial.

9. The method of claim 1, wherein the method further comprises: generating a hash value of the PDF document; and generating a hash value of the PDF document using a hash function. The comparison level verification method in step S5 is: using Reed-Solomon encoding technology, when part of the bits of the anti-quantum identification code primitive are disturbed or damaged in the transmission or storage process, the damaged anti-quantum identification code primitive fragment can be recovered, and the recovered identification code is verified to determine whether the document is complete at the binary level.

10. The method of claim 1, wherein the method further comprises: The semantic level verification method in step S5 is: using an intelligent contract to verify the version consistency of the logical layer fingerprint, the intelligent contract can record and verify the historical version of the document logical structure feature code, and if the current document logical layer fingerprint is inconsistent with the version recorded in the intelligent contract, it means that the logical structure of the document is tampered with.

11. The method of claim 1, wherein the method further comprises: The visual traceability verification method in step S5 is: using a Siamese network to compare the original visual fingerprint with the real-time rendering result, the Siamese network compares the similarity of the two visual feature vectors, and if the similarity is lower than the set threshold, it means that the visual content of the document is tampered with.

12. A PDF document tamper-evident identification code generation system, characterized by, A PDF document tamper-proof identification code generation method according to any one of claims 1-11 is used, including a multi-modal feature extraction module, a dynamic key binding module, an anti-quantum aggregation module, a space-time folding embedding module and a multi-level verification module, the multi-modal feature extraction module is used to execute step S1, the dynamic key binding module is used to execute step S2, the anti-quantum aggregation module is used to execute step S3, the space-time folding embedding module is used to execute step S4, and the multi-level verification module is used to execute step S5.

Images