Blockchain-based performance lossless watermarking and trusted provenance method and system

Abstract

The application discloses a kind of performance lossless watermarking credible traceability method and system based on blockchain, method includes: in generative model latent space, embedding of extended watermark information is carried out based on distribution consistent sampling to obtain embedded watermark generative content;Watermark metadata is converted into hash value and stored in blockchain and node trust management is carried out and shared data storage and consensus mechanism;Embedded watermark generative content is extracted by inversion to obtain watermark information and calculate hash value, compare with the hash value of watermark information stored in blockchain, consistent with comparison then watermark verification passes, inconsistent then watermark verification fails;When watermark verification fails, node signature verification is carried out, if node signature verification fails or data is abnormal, then start on-chain traceability process, according to watermark metadata, track data generation path, find the node or data transmission point where abnormal occurs.The application can ensure high quality of generative content, while realizing lossless embedding of watermark information and credible circulation and reliable traceability.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of deep learning information security and blockchain technology, and particularly relates to a performance lossless watermark trusted traceability method and system based on a blockchain. BACKGROUND

[0002] With the wide development of blockchain technology and generative models in the field of digital economy and content generation, copyright protection and trusted traceability of generated content have gradually become problems to be solved. Blockchain, with its distributed storage, tamper resistance and multi-party consensus mechanism, has shown great application potential in the fields of finance, government affairs and supply chain. However, with the popularity of image generation technology similar to diffusion model, its convenient generation capability and diversified application scenarios have also brought a series of problems including infringement, false information dissemination, etc. In recent years, generative models have been misused to generate false news, fake digital art or images of fictitious events, which not only disrupts the market order, but also may cause public panic, posing a serious challenge to the social trust system and economic stability.

[0003] In order to solve these problems, higher requirements are put forward for the traceability, copyright protection and content authenticity of generated content. Digital watermarking, as an important means of copyright protection and content tracing, realizes the traceability and copyright authentication of generated content by embedding invisible identification information into the content. However, existing watermarking techniques face many limitations when applied to deep generative models.

[0004] Firstly, traditional watermark embedding methods are prone to cause performance degradation of content generation, because the embedding of watermark often needs to modify the original content to some extent, thereby affecting the quality of generated content and user experience. Secondly, most methods need to modify the parameters of the generative model or retrain the model, which not only increases the computational complexity of the system, but also may cause further degradation of the quality of generated content. In practical applications, such performance degradation and quality loss will directly affect user experience and the overall efficiency of the system. In addition, existing watermarking techniques lack effective trust management and traceability capabilities when facing malicious nodes in distributed systems or complex data flow scenarios. In a distributed system, data flows between multiple nodes, and malicious nodes may tamper with watermark information, making content tracing and verification more complex. At the same time, watermark information may be maliciously tampered with by external attacks such as cropping, compression and noise during cross-platform transmission and storage, further exacerbating the difficulty of tracing and verification. More seriously, the robustness of existing watermarking methods in dealing with various adversarial attacks or backdoor attacks is not strong enough. In some cases, the watermarking method may even become a means of backdoor implantation, not only causing loss or tampering of watermark information, but also affecting the stability and usability of the system, and leading to instability and error rate increase in the verification link.

[0005] In summary, the existing watermark embedding and management methods have significant limitations in meeting the performance preservation after content embedding and the complex traceability requirements in multi-platform environments. In particular, in terms of how to achieve an effective balance between system performance and generated content quality during the watermark embedding process, traditional technical solutions usually rely on adjustments to system parameters or additional training of models, which not only significantly increases the computational and time costs, but also can lead to a decline in the quality of generated content, and the lack of robustness will also bring serious security risks. Therefore, in practical applications, how to maintain system performance while achieving effective copyright protection and trusted traceability has become a problem to be solved. SUMMARY

[0006] In view of the above, the purpose of the present application is to provide a blockchain-based performance lossless watermark trusted traceability method and system, which combines watermark embedding technology with the distributed ledger system of blockchain, ensuring high-quality generated content while achieving lossless embedding of watermark information and trusted circulation and reliable traceability.

[0007] To achieve the above-mentioned purpose of the application, the technical solutions provided by the present application are as follows:

[0008] In a first aspect, the present application provides a blockchain-based performance lossless watermark trusted traceability method, comprising the following steps:

[0009] The watermark information to be embedded is encoded, encrypted and redundantly expanded to form extended watermark information, and the embedding of the extended watermark information is performed in the latent space of the generative model based on uniform distribution sampling, and finally the generated content with embedded watermark is obtained;

[0010] Based on the embedding process of the extended watermark information, watermark metadata including watermark ID, timestamp and node signature are generated and converted into a hash value and stored in the blockchain, and node trust management, shared data storage and consensus mechanism are performed on the nodes in the blockchain;

[0011] The generated content with embedded watermark is inverted to extract the watermark information embedded in the latent space and calculate the hash value, and the hash value is compared with the hash value of the watermark information stored in the blockchain. If the comparison is consistent, the watermark verification is passed, and if the comparison is inconsistent, the watermark verification fails;

[0012] When the watermark verification fails, the node signature verification is performed, and if the node signature verification fails or the data is abnormal, the on-chain traceability process is started, the data generation path is tracked according to the watermark ID, timestamp and node signature stored in the blockchain, and the abnormal node or data transmission point is found for abnormal handling.

[0013] Preferably, the encoding, encryption and redundancy expansion of the watermark information to be embedded form an expanded watermark information, comprising:

[0014] The watermark information to be embedded is binary coded, the coded watermark information is encrypted using the ChaCha20 encryption algorithm to generate a pseudo-random binary stream, and the binary stream m is copied multiple times in the latent space using redundancy design in space and channel to obtain the expanded watermark information.

[0015] Preferably, the embedding of the expanded watermark information in the latent space of the generative model based on distribution consistent sampling comprises:

[0016] The original latent representation z generated by the generative model in the latent space T ~N(0, I) is a high-dimensional vector subject to a standard normal distribution, I is a standard deviation, and the expanded watermark information s d is embedded into the original latent representation z T in multiple dimensions, denoted as:

[0017]

[0018]

[0019] wherein, is the latent representation after watermark embedding in the i-th dimension, z T [i] is the original latent representation in the i-th dimension, s d [i] is the embedded expanded watermark information in the i-th dimension, and λ is a weight coefficient controlling the embedding strength, if(·) is a conditional function.

[0020] In the watermark embedding process, the latent representation after watermark embedding z is consistent with the distribution of the original latent representation z T ~N(0, I) through distribution consistent sampling, and the sampling process is represented as:

[0021]

[0022] wherein, PPF(·) is the inverse function of the normal distribution, i is the current position of the embedded watermark, i.e., the dimension of the latent representation, U(0, 1) is uniformly distributed random noise, and 2 l is the l-th power of 2.

[0023] Preferably, after completing the embedding of the expanded watermark information, the generated latent representation after watermark embedding z is verified, and the consistency of the latent representation after watermark embedding z with the original latent representation z T is verified, denoted as:

[0024]

[0025] Wherein, Mean(·) is the mean operation, Var(·) is the variance operation, and the verification passes to the blockchain storage link.

[0026] Preferably, the embedding process based on extended watermark information generates watermark metadata including watermark ID, timestamp and node signature and is converted into a hash value stored in the blockchain, including:

[0027] The embedding process based on extended watermark information generates watermark metadata including watermark ID, timestamp and node signature, and stores the watermark metadata in the blockchain in the form of a Merkle tree structure. Each leaf node of the Merkle tree stores the hash value H of a piece of data, expressed as:

[0028] H=Hash(WatermarkID∥Time∥Sign)

[0029] Wherein, Hash(·) is the hash value calculation, WatermarkID is the watermark ID, Time is the timestamp of the watermark embedding, and Sign is the node signature generated by the blockchain;

[0030] When forming the Merkle tree structure, the adjacent two hash values are merged and continue to be calculated until the root hash value H is formed root , the root hash value H root is stored in the blockchain.

[0031] Preferably, the node trust management and shared data storage and consensus mechanism for the nodes in the blockchain include:

[0032] The node trust management is performed on the nodes in the blockchain, the data processing behavior of the nodes is monitored in real time, and the trust score of the nodes is calculated according to the compliance, consensus participation rate and abnormal detection result, the malicious nodes are identified according to the trust score and the permission is limited, and the trust score is expressed as:

[0033] T n =α·P n +β·C n +γ·A n

[0034] Wherein, T n is the trust score of the nth node, P n , C n and A n are the compliance score, consensus participation rate and abnormal behavior detection rate of the nth node, respectively, and α, β and γ are weight coefficients.

[0035] The alliance chain architecture is used to realize shared data storage and consensus mechanism among multiple nodes. Each node saves a complete copy of the ledger to synchronize and maintain consistency of watermark information across the entire network. The PBFT consensus algorithm is used to ensure the reliability of on-chain data and keep the system running normally in the event of failure or malicious behavior of some nodes.

[0036] Preferably, when watermark verification failure is detected, the node signature data is tracked and the node signature is verified, which is expressed as:

[0037] Verify(Sign k ,PublicKey k ,H k )=True

[0038] Among them, Verify(·) is the verification operation, Sign k 、PublicKey k and H k They are the node signature, public key, and hash value of the kth node respectively. If the verification is True, it means the verification is successful. Otherwise, if the verification fails, the current node is marked as suspicious and the exception handling mechanism is triggered.

[0039] Preferably, the real-time monitoring is performed to determine whether the node submits data on time and frequently, and whether the data submitted by the node is consistent with the data of other nodes. If the data is submitted frequently or inconsistent, it is marked as an abnormal node and the exception handling mechanism is triggered.

[0040] In a second aspect, to achieve the above-mentioned purpose of the invention, an embodiment of the present invention further provides a blockchain-based performance lossless watermark trusted traceability system, which is implemented using the above-mentioned blockchain-based performance lossless watermark trusted traceability method, including: a watermark embedding module, a blockchain trusted storage module, an on-chain and off-chain watermark verification module, and an anomaly detection and traceability module;

[0041] The watermark embedding module is used to encode, encrypt and redundantly expand the watermark information to be embedded to form extended watermark information, embed the extended watermark information based on distribution consistent sampling in the latent space of the generative model, and finally obtain the generated content embedded with the watermark;

[0042] The blockchain trusted storage module is used to generate watermark metadata including watermark ID, timestamp and node signature based on the embedding process of extended watermark information and convert it into a hash value and store it in the blockchain, and perform node trust management and shared data storage and consensus mechanism on the nodes in the blockchain;

[0043] The on-chain off-chain watermark verification module is used for inverse extraction of the watermark information embedded in the generated content in a potential space and calculation of a hash value, comparison of the hash value with a hash value of the watermark information stored on the blockchain, and watermark verification passing when the comparison is consistent and watermark verification failure when the comparison is inconsistent.

[0044] The anomaly detection and tracing module is used for node signature verification when watermark verification fails, starting of an on-chain tracing process when node signature verification fails or data is abnormal, generation of a path according to the watermark ID, timestamp and node signature of the tracked data stored on the blockchain, and finding of an abnormal node or data transmission point for anomaly processing.

[0045] In a third aspect, to achieve the above-mentioned object, an electronic device is provided, comprising a memory and a processor, the memory is used for storing a computer program, and the processor is used for implementing the above-mentioned blockchain-based performance lossless watermark credible tracing method when executing the computer program.

[0046] Compared with the prior art, the present application has at least the following beneficial effects:

[0047] (1) The present application combines watermark embedding technology with the distributed ledger system of the blockchain, ensures the high quality of the generated content, and realizes lossless embedding and verification of the watermark information. By embedding watermark information in the potential space of the generated content and using the distributed consistent sampling technology, the precision indicators of the content after embedding the watermark are ensured to be consistent with the original content. The tamper-proof nature of the blockchain technology ensures the security of the watermark metadata during transmission and storage, and improves the transparency and credibility of the traceability management. An efficient traceability system covering the whole life cycle of the generated content is constructed through the organic combination of the blockchain and the watermark embedding technology.

[0048] (2) The present application performs node trust management, shared data storage and consensus mechanism on the nodes in the blockchain, evaluates and manages the behavior of the nodes in real time, and ensures that all nodes in the system can participate in data transmission and processing according to the established rules. After the watermark embedding of the generated content is completed, the watermark information is automatically uploaded to the blockchain for archiving, forming a chain-on-chain off-link trust management system. In the cross-platform and multi-system scenario, the interoperability of the blockchain provides security guarantee for the safe flow of data and the credible management of the watermark.

[0049] (3) The present application forms an extended watermark information for watermark embedding, has high robustness, can maintain the integrity and effectiveness of the watermark information when facing strong noise, cropping and compression and other malicious attacks, and ensures the watermark detectability in some adversarial environments, so as to realize stable tracing of the generated content. BRIEF DESCRIPTION OF DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings required by the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0051] Figure 1 is a flowchart of the performance lossless watermark trusted traceability method based on blockchain provided by the embodiments of the present application;

[0052] Figure 2 is a framework diagram of the performance lossless watermark trusted traceability method based on blockchain provided by the embodiments of the present application;

[0053] Figure 3 is a structural diagram of the performance lossless watermark trusted traceability system based on blockchain provided by the embodiments of the present application. DETAILED DESCRIPTION

[0054] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the protection scope of the present application.

[0055] The inventive concept of the present application is that: in view of the problem that the existing watermark embedding and management method has significant limitations in meeting the performance preservation after content embedding and the complex traceability requirements in the multi-platform environment, the embodiments of the present application provide a performance lossless watermark trusted traceability method and system based on blockchain, which embeds watermark information into the potential space of generated content without affecting the performance of content generation, and combines with the distributed consistent sampling technology to ensure the natural fusion of the embedding process and the quality preservation of the generated content. At the same time, relying on the tamper-proof characteristics and multi-party consensus mechanism of the blockchain, transparent storage and dynamic management of watermark information on and off the chain are realized, ensuring the reliable traceability of watermark information and the reliable traceability of data in the multi-node and multi-platform environment, and enhancing the robustness of the system, even in complex application scenarios such as noise interference, data cropping or compression operations, etc. The integrity and verifiability of the watermark can still be ensured.

[0056] Figure 1 is a flowchart of the performance lossless watermark trusted traceability method based on blockchain provided by the embodiments of the present application, Figure 2 is a framework diagram of the performance lossless watermark trusted traceability method based on blockchain provided by the embodiments of the present application. As Figure 1 and Figure 2As shown, the embodiment provides a blockchain-based performance lossless watermark trusted traceability method, including the following steps:

[0057] S1, the watermark information to be embedded is encoded, encrypted and redundantly expanded to form extended watermark information, and the embedding of the extended watermark information is performed in the latent space of the generative model based on distribution consistent sampling, and finally the generative content with embedded watermark is obtained.

[0058] S1.1, input model and data. Input the generative model and the watermark information to be embedded, and in the embodiment, the generative model adopts a diffusion model and is used to generate images. The latent space of the diffusion model is used to carry the watermark information, so as to ensure that the embedding process does not affect the quality of the generative content. The latent representation z generated by the diffusion model T ~N(0, I) is a high-dimensional vector subject to standard normal distribution, and I is the standard deviation. Before watermark embedding, the original sample is input into the diffusion model to generate an image, and the generated image is segmented based on the Gaussian Shading method, and the segmented main region is taken as the target region for watermark embedding, so as to improve the efficiency and concealment of watermark embedding.

[0059] S1.2, encoding and encryption of watermark data. The watermark information s to be embedded is binary encoded to form a watermark bit stream. The ChaCha20 encryption algorithm is used to encrypt the encoded watermark information to generate a pseudo-random binary stream m, which is represented as:

[0060] m=ChaCha20(s,K)

[0061] Where ChaCha20(·) is the ChaCha20 encryption algorithm, K is the key, and the encrypted m is a uniformly distributed binary stream, which is used to avoid easy guessing of the watermark information by malicious attackers.

[0062] S1.3, diffusion and redundancy design. In order to enhance the robustness of the watermark, spatial and channel redundancy design is adopted, and the binary stream m is copied multiple times in the latent space to obtain the extended watermark information s d , which is represented as:

[0063]

[0064] Where Repeat(·) is the copy operation, f c is the channel redundancy factor, is the width-height redundancy factor. Through multiple iterations of the copy operation in the diffusion process, it is ensured that even under cropping, compression and noise attacks, part of the watermark information can still be completely extracted.

[0065] S1.2, watermark embedding and fusion based on distribution consistent sampling. The extended watermark information sd Embedding into the original latent representation z T The watermark is embedded bit by bit through the following formula:

[0066]

[0067]

[0068] Wherein, is the latent representation after watermark embedding in the i-th dimension, z T is the original latent representation in the i-th dimension, s d is the embedded extended watermark information in the i-th dimension, if(·) is a conditional function, λ is a weight coefficient controlling the embedding strength, and by adjusting λ, it is ensured that the embedding process does not affect the distribution of the latent representation and the quality of the generated content.

[0069] In the watermark embedding process, by uniformly distributed sampling, the latent representation after watermark embedding is consistent with the distribution of the original latent representation z T ~N(0,I), avoiding the influence of the embedding process on the performance of the generation model, and the sampling process is represented as:

[0070]

[0071] Wherein, PPF(·) (Percent Point Function) is the inverse function of the normal distribution, i is the current position of the embedded watermark, i.e. the dimension of the latent representation, U(0,1) is uniformly distributed random noise, and 2 l is the l-th power of 2. Through this sampling method, it can be ensured that the embedded latent representation is statistically consistent with the representation without embedding the watermark.

[0072] S1.3, after completing the embedding of the extended watermark information, the generated watermark-embedded latent representation is verified, and the watermark-embedded latent representation is consistent with the distribution of the original latent representation z T , which is represented as:

[0073]

[0074] Wherein, Mean(·) is the mean operation, and Var(·) is the variance operation. After verification, it can proceed to the next step of blockchain storage.

[0075] S2, generating watermark metadata including watermark ID, timestamp and node signature based on embedding process of extended watermark information and transforming into hash value stored in blockchain, performing node trust management on nodes in blockchain and sharing data storage and consensus mechanism.

[0076] S2.1, generating watermark metadata. Based on the embedding process of extended watermark information, generate watermark metadata including watermark ID, timestamp and node signature to ensure the uniqueness and integrity of data. Watermark ID is used to track the source of generated content, and timestamp ensures the timing accuracy of data generation.

[0077] S2.2, blockchain storage. Store watermark metadata in Merkle tree structure in blockchain, each leaf node of Merkle tree stores the hash value H of one piece of data, denoted as:

[0078] H = Hash(WatermarkID || Time || Sign)

[0079] Where Hash(·) is hash value calculation, WatermarkID is watermark ID, Time is watermark embedding timestamp, and Sign is node signature generated by blockchain. In this way, the integrity of the data can be quickly verified when needed, and the metadata cannot be tampered with.

[0080] When storing in Merkle tree structure, calculate the hash value of each piece of metadata (including watermark ID, timestamp and node signature):

[0081] H k = Hash(WatermarkID k ∥Time∥Sign k )

[0082] Where subscript k is the index of hash value, watermark ID and node signature.

[0083] By merging the adjacent two hash values and continuing to calculate until the root hash value H root is formed:

[0084] H root = Iteration(Hash(H j || H j+1 ))

[0085] Where Iteration(·) is iteration, and subscripts j and j+1 are the indices of the adjacent two hash values.

[0086] Store the root hash value H root in the blockchain to ensure the integrity of the data when verifying later.

[0087] S2.3, node trust management, node behavior monitoring and trust score calculation. Perform node trust management on the nodes in the blockchain, monitor the data processing behavior of the nodes in real time, and calculate the trust score according to the compliance, consensus participation rate and abnormal detection result, identify malicious nodes according to the trust score and limit their permissions, and the trust score is represented as:

[0088] T n = a P n + b C n + g A n

[0089] Where T n is the trust score of the nth node, P n , C n and A n are the compliance score, consensus participation rate and abnormal behavior detection rate of the nth node, respectively, and a, b and g are weight coefficients.

[0090] S2.4, cross-node consensus and synchronization of on-chain data. Use the consortium chain (Consortium Blockchain) architecture to realize multi-node shared data storage and consensus mechanism, each node saves a complete copy of the ledger to make the watermark information synchronized and consistent in the whole network, and through the PBFT (Practical Byzantine Fault Tolerance, Practical Byzantine Fault Tolerance) consensus algorithm to ensure the reliability of the on-chain data, and keep the system running normally in the case of partial node failure or malicious behavior.

[0091] S3, reverse the embedded watermark of the generated content to extract the watermark information embedded in the latent space and calculate the hash value, and compare it with the hash value of the watermark stored on the blockchain. If the comparison is consistent, the watermark verification is passed, and if the comparison is inconsistent, the watermark verification fails.

[0092] S3.1, off-chain watermark information extraction. Extract the watermark information in the latent space from the generated content through the DDIM inversion (Denoising Diffusion Implicit Models Inversion) algorithm. The inversion process is as follows:

[0093] z0= DDIM_Inverse(X s )

[0094] Where DDIM_Inverse(·) is the DDIM inversion calculation, X s is the generated image embedded with the watermark, and z0 is the latent representation obtained by inversion calculation. Extract the embedded watermark information contained in z0, and calculate the hash value H of the embedded watermark information.extracted , H extracted It will be used to compare with the data stored on the chain during the watermark verification phase.

[0095] S3.2, on-chain and off-chain watermark verification. Use the watermark ID as an index to quickly locate the corresponding watermark metadata on the chain. By comparing the metadata stored on the chain with the watermark information extracted off-chain, ensure that the content has not been tampered with. The specific verification is based on the watermark ID of the watermark information extracted off-chain. s , find the corresponding hash value H on the blockchain s , H s With H extracted Make a comparison.

[0096] S4, when watermark verification fails, node signature verification is performed. If node signature verification fails or data is abnormal, the chain traceability process is started. The data is tracked according to the watermark ID, timestamp and node signature stored on the blockchain to generate a path, find the node or data transmission point where the abnormality occurred, and handle the abnormality.

[0097] S4.1, Anomaly Detection Mechanism. The core of the anomaly detection mechanism is to monitor the behavior of all nodes in the blockchain, such as data tampering and malicious data release. This mechanism monitors in real time whether nodes submit data on time, frequently, and whether the data submitted by a node is consistent with that of other nodes. If data is submitted frequently or inconsistently, the node is marked as an anomaly and the anomaly handling mechanism is triggered.

[0098] S4.2, traceability mechanism. When watermark verification fails, the node signature data will be tracked to analyze whether there are malicious nodes involved in data tampering. The verification formula of the node signature is expressed as:

[0099] Verify(Sign k ,PublicKey k ,H k )=True

[0100] Among them, Verify(·) is the verification operation, Sign k 、PublicKey k and H k They are the node signature, public key, and hash value of the kth node respectively. If the verification is True, it means the verification is successful. Otherwise, if the verification fails, the current node is marked as suspicious and the exception handling mechanism is triggered.

[0101] In the case of node signature verification failure or data anomaly, the system will automatically start the on-chain traceability process to track the data generation path according to the timestamp and node signature on the blockchain. The system checks the signature and data transmission link of each node along the history record of the watermark ID until the abnormal node or data transmission point is found for abnormal processing.

[0102] In summary, the performance lossless watermark trusted traceability method based on the blockchain provided by the embodiment of the present application can ensure the high quality of the generated content while realizing the lossless embedding and verification of the watermark information by introducing the Gaussian Shading watermark embedding technology and combining it with the distributed ledger system of the blockchain. The security of the watermark metadata in the transmission and storage process is realized, the transparency and credibility of the traceability management are improved, and the high robustness of the watermark embedding is ensured. Through the construction of the efficient traceability system covering the whole life cycle of the generated content, the integrity and effectiveness of the watermark information can be maintained when facing strong noise, cropping and compression and other malicious attacks, so as to realize the stable traceability of the generated content.

[0103] Based on the same inventive concept, as shown in Figure 3 The embodiment of the present application also provides a performance lossless watermark trusted traceability system 300 based on the blockchain, which comprises a watermark embedding module 310, a blockchain trusted storage module 320, an on-chain and off-chain watermark verification module 330 and an abnormality detection and traceability module 340.

[0104] The watermark embedding module 310 is used for encoding, encrypting and redundancy expanding the watermark information to be embedded to form extended watermark information, embedding the extended watermark information in the latent space of the generative model based on uniform distribution sampling, and finally obtaining the generated content with embedded watermark;

[0105] The blockchain trusted storage module 320 is used for generating watermark metadata including watermark ID, timestamp and node signature based on the embedding process of the extended watermark information and converting the watermark metadata into a hash value for storage in the blockchain, performing node trust management on the nodes in the blockchain, and sharing data storage and consensus mechanism;

[0106] The on-chain and off-chain watermark verification module 330 is used for inverse extraction of the watermark information embedded in the latent space from the generated content with embedded watermark and calculation of the hash value, comparison of the hash value with the hash value of the watermark information stored on the blockchain, and comparison of the hash value with the hash value of the watermark information stored on the blockchain.

[0107] The anomaly detection traceability module 340 is configured to perform node signature verification when watermark verification fails, and to start an on-chain traceability process if the node signature verification fails or data is abnormal, to generate a path according to the watermark ID, the timestamp and the node signature traceability data stored on the blockchain, and to find the node or data transmission point where the anomaly occurs for anomaly processing.

[0108] Based on the same inventive concept, the embodiments of the present application also provide an electronic device comprising a memory and a processor, the memory being configured to store a computer program, and the processor being configured to implement the above-mentioned blockchain-based performance lossless watermark trusted traceability method when executing the computer program.

[0109] It should be noted that the blockchain-based performance lossless watermark trusted traceability system and the electronic device provided by the above-mentioned embodiments belong to the same inventive concept as the blockchain-based performance lossless watermark trusted traceability method, and the specific implementation process is described in detail in the blockchain-based performance lossless watermark trusted traceability method embodiments, which will not be repeated here.

[0110] The specific embodiments described above have described the technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only the most preferred embodiment of the present application and is not intended to limit the present application. Any modifications, supplements and equivalent replacements made within the principle range of the present application shall be included in the protection scope of the present application.

Claims

1. A blockchain-based performance lossless watermark trusted provenance method, characterized in that, The method comprises the following steps: The embedded watermark information is encoded, encrypted and redundantly expanded to form extended watermark information, the embedding of the extended watermark information is performed based on distribution consistent sampling in the latent space of the generative model, and finally the generative content embedded with the watermark is obtained; wherein the embedding of the extended watermark information based on the distribution consistent sampling in the latent space of the generative model comprises: generating the original latent representation z T ~N(0, I) is a high-dimensional vector subject to a standard normal distribution, I is a standard deviation, and the extended watermark information s d is embedded into the original latent representation z T in multiple dimensions, and is represented as: wherein, is the latent representation after watermark embedding for the i-th dimension, z T is the original latent representation for the i-th dimension, s d is the embedded extended watermark information for the i-th dimension, λ is the weight coefficient to control the embedding strength, if(·) is the conditional function; in the watermark embedding process, the latent representation after watermark embedding is consistent with the original latent representation z T ~N(0, I) distribution, the sampling process is represented as: where PPF(·) is the inverse function of normal distribution, i is the current position of embedding watermark, i.e., the dimension of latent representation, U(0, 1) is the random noise of uniform distribution, 2 l is a power of 2; The embedding process based on the extended watermark information generates watermark metadata including a watermark ID, a timestamp, and a node signature and is converted into a hash value stored in a blockchain, node trust management is performed on the nodes in the blockchain, and a shared data storage and consensus mechanism are implemented; The generated content embedded with the watermark is reversed to extract the watermark information embedded in the latent space and calculate a hash value, which is compared with the hash value of the watermark information stored in the blockchain, and if the comparison is consistent, the watermark verification is passed, and if the comparison is inconsistent, the watermark verification fails; When the watermark verification fails, node signature verification is performed, and if the node signature verification fails or the data is abnormal, an on-chain tracing process is started, the data generation path is tracked according to the watermark ID, timestamp, and node signature stored in the blockchain, and the abnormal node or data transmission point is found for abnormal processing.

2. The blockchain-based performance lossless watermark trusted provenance method of claim 1, wherein, The encoding, encryption, and redundancy expansion of the watermark information to be embedded form the extended watermark information, which comprises: The watermark information to be embedded is binary coded, the ChaCha20 encryption algorithm is used to encrypt the coded watermark information to generate a pseudo-random binary stream, and the binary stream m is copied multiple times in the latent space to obtain the extended watermark information by using redundancy design in space and channel. 3.The blockchain-based performance lossless watermark trusted provenance method of claim 1, wherein, After the embedding of the extended watermark information is completed, the generated watermark-embedded latent representation is verified, verifying the watermark-embedded latent representation is consistent with the distribution of the original latent representation z T is represented as: Wherein, Mean(·) is the mean operation, Var(·) is the variance operation, and after verification, it enters the blockchain storage link. 4.The blockchain-based performance lossless watermark trusted provenance method of claim 1, wherein, The embedding process based on the extended watermark information generates watermark metadata including a watermark ID, a timestamp, and a node signature and is converted into a hash value stored in a blockchain, which comprises: The embedding process based on the extended watermark information generates watermark metadata including a watermark ID, a timestamp, and a node signature, and the watermark metadata is stored in the blockchain in a Merkle tree structure, and each leaf node of the Merkle tree stores the hash value H of one piece of data, which is represented as: H = Hash(WatermarkID || Time || Sign) Wherein, Hash(·) is a hash value calculation, WatermarkID is a watermark ID, Time is a timestamp of watermark embedding, and Sign is a node signature generated by a blockchain; In forming the Merkle tree structure, by merging two adjacent hash values and continuing to calculate until the root hash value H root is formed root is stored in the blockchain.

5. The blockchain-based performance lossless watermark trusted provenance method of claim 1, wherein, The node trust management is performed on the nodes in the blockchain, the trust score of each node is calculated by monitoring the data processing behavior of the node in real time and according to compliance, consensus participation rate, and abnormal detection result, the malicious node is identified according to the trust score, and the permission of the malicious node is limited, and the trust score is represented as: The alliance chain architecture is adopted to realize the shared data storage and consensus mechanism of multiple nodes, each node saves a complete copy of the ledger, so that the watermark information is synchronized and consistent in the whole network, and the PBFT consensus algorithm is used to ensure the reliability of the data on the chain, so that the system can operate normally in the case of failure or malicious behavior of part of the nodes. T n = a · P n + β · C n + γ · A n Among them, T n is the trust score of the nth node, P n 、C n and A n are the compliance score, consensus participation rate, and abnormal behavior detection rate of the nth node, respectively. α, β, and γ are the weight coefficients, respectively. When the watermark verification fails, the node signature data is tracked, and the node signature is verified, which is represented as:

6. The blockchain-based performance lossless watermark trusted provenance method of claim 1, wherein, ​ Verify(Sign k , PublicKey k , H k ) = True where Verify(·) is a verification operation, Sign k , PublicKey k and H k are the node signature, public key and hash value of the kth node, respectively. If the verification is True, it indicates that the verification is successful, otherwise, the current node is marked as suspicious and an exception handling mechanism is triggered.

7. The blockchain-based performance lossless watermark trusted provenance method of claim 1, wherein, Real-time monitoring whether the node submits data on time, whether it submits frequently, and determining whether the data submitted by the node is consistent with the data of other nodes, if the data is submitted frequently or the data is inconsistent, the node is marked as an abnormal node, and an abnormal processing mechanism is triggered.

8. A blockchain-based performance lossless watermark trusted provenance system, implemented by using the blockchain-based performance lossless watermark trusted provenance method of any one of claims 1-7, characterized in that, It comprises: a watermark embedding module, a blockchain trusted storage module, an on-chain and off-chain watermark verification module, and an abnormality detection and tracing module; The watermark embedding module is configured to encode, encrypt and redundantly expand the watermark information to be embedded to form expanded watermark information, embed the expanded watermark information in the latent space of the generative model based on distribution consistent sampling, and finally obtain the generative content with the embedded watermark; wherein the embedding of the expanded watermark information in the latent space of the generative model based on distribution consistent sampling comprises: generating an original latent representation z by the generative model in the latent space T ~N(0, I) is a high-dimensional vector subject to a standard normal distribution, I is a standard deviation, and the expanded watermark information s d is embedded into the original latent representation z T in multiple dimensions, and is represented as: where, is the latent representation after watermark embedding for the i-th dimension, z T is the original latent representation for the i-th dimension, s d is the embedded extended watermark information for the i-th dimension, λ is the weight coefficient to control the embedding strength, if(·) is the conditional function; during the watermark embedding process, the latent representation after watermark embedding is consistent with the original latent representation z T ~ N(0, I) is consistent, and the sampling process is represented as: where PPF(·) is the inverse function of the normal distribution, i is the current position of embedding the watermark, i.e., the dimension of the latent representation, U(0, 1) is the random noise of uniform distribution, 2 l is a power of 2; The blockchain trusted storage module is used to generate watermark metadata including watermark ID, timestamp, and node signature based on the embedding process of extended watermark information and convert it into a hash value for storage in the blockchain, perform node trust management on the nodes in the blockchain, and store and consensus mechanism for shared data; The on-chain and off-chain watermark verification module is used to extract the watermark information embedded in the potential space from the generated content embedded with the watermark and calculate the hash value, and compare it with the hash value of the watermark information stored on the blockchain, if the comparison is consistent, the watermark verification is passed, if the comparison is inconsistent, the watermark verification fails; The abnormality detection and tracing module is used to perform node signature verification when the watermark verification fails, and if the node signature verification fails or the data is abnormal, start the on-chain tracing process, trace the data generation path according to the watermark ID, timestamp, and node signature stored on the blockchain, find the node or data transmission point where the abnormality occurs, and perform abnormality processing.

9. An electronic device comprising a memory and a processor, the memory for storing a computer program, characterized in that, The processor is used to implement the blockchain-based performance lossless watermark trusted tracing method of any one of claims 1-7 when executing the computer program.

Images