An AI-generated content multi-modal semantic watermark embedding and witness certificate issuing method and system, and a storage medium

Abstract

The application discloses an AI-generated content multi-modal semantic watermark embedding and witness certificate issuing method and system and a storage medium, and belongs to the fields of data security and artificial intelligence. The modality of AI-generated content is identified; semantic watermarks (text zero-width characters, image DCT domain, audio / video frequency domain) are embedded into the AI-generated content according to the modality; witness certificates (recording content fingerprints and generation information) are issued for the AI-generated content; and the semantic watermarks and the witness certificates are associated and mutually verified to support the authenticity identification and generation traceability of the AI-generated content. AI generation is a data-related behavior, and the witness certificate is a behavior witness certificate. The method is located at the generation side and forms an AI-generated content credible loop with an authenticity identification method located at the identification side.

Description

Technical Field

[0001] This invention relates to the fields of data security and artificial intelligence, and in particular to a method, system, and storage medium for embedding semantic watermarks into multimodal AI-generated content and issuing witness certificates. Background Technology

[0002] The authenticity and traceability of AI-generated content (AIGC) need to be marked and documented at the generation stage: watermarks should be embedded in the content, and a verification certificate should be issued for the content so that it can be authenticated later. Different watermarking methods should be used for different modalities of content.

[0003] Current practices often only add watermarks to a single modality, or only store evidence without watermarking, or the watermark is not associated with the certificate, resulting in insufficient evidence for authenticity verification. Therefore, how to embed semantic watermarks according to modality, issue witness certificates, and associate and mutually verify watermarks with certificates is a key issue for the trusted marking on the AIGC generation side. Summary of the Invention

[0004] (a) Technical problems to be solved To address the issues of existing AIGC-generated watermarks being singular and not associated with the credentials, this invention provides a method, system, and storage medium for embedding multimodal semantic watermarks of AI-generated content and issuing witness credentials.

[0005] (II) Technical Solution This invention identifies content modalities, embeds semantic watermarks (zero-width characters for text / DCT for images / audio / video frequency domains) according to the modality, issues witness certificates for the content (recording content fingerprints and generation information), and associates and mutually verifies the watermarks and certificates. The witness certificates are used for authentication and retrieval of cross-domain authenticity verification methods for AIGC-generated content in a public witness library; AI generation is the behavior, and the certificate is the witness certificate for data-related behavior proof methods; this invention, on the generation side, and the authentication side of the AIGC-generated content cross-domain authenticity verification method in the public witness library form a trusted closed loop.

[0006] (III) Beneficial Effects First, semantic watermarks are embedded according to modality, covering text / images / audio / video; second, the watermarks are linked and mutually verified with the witness certificate, providing dual support for authenticity verification; third, the AIGC trusted closed loop is formed between the generation side and the cross-domain authenticity verification method of the publicly available witness library AIGC-generated content (complete engineering design has been completed). Attached Figure Description

[0007] Figure 1 A schematic diagram of the overall process for embedding multimodal semantic watermarks into AI-generated content and issuing witness certificates (CSOP).

[0008] Figure 2 This is a schematic diagram of multimodal semantic watermark embedding.

[0009] Figure 3 This is a diagram illustrating the issuance of CSOP witness certificates.

[0010] Figure 4 This is a flowchart of the watermark embedding and certificate issuance process.

[0011] Figure 5 This diagram illustrates the connection between the cross-domain authenticity verification method and the data-related behavior proof method for content generated by the publicly available witness library AIGC. Detailed Implementation

[0012] The present invention will be further described in detail below with reference to the accompanying drawings, taking the embedding of watermarks into AIGC content and the issuance of certificates as an example.

[0013] 1. Overall Mechanism See Figure 1 Semantic watermarks are embedded in multimodal AIGC content and CSOP witness certificates are issued. The association between the watermark and the certificate supports the verification of authenticity.

[0014] 2. Multimodal watermark embedding See Figure 2 Watermarks can be embedded in modalities: zero-width characters for text, DCT domain for images, and frequency domain for audio and video.

[0015] 3. Issuance of CSOP Witness Certificate See Figure 3 The system issues CSOPs (records content fingerprints and generation information) for content, which are linked to watermarks. The watermark and credentials provide dual support for authenticity verification.

[0016] 4. Embedding and Issuance Process See Figure 4 It identifies modalities, embeds watermarks according to modalities, issues CSOPs, and associates watermarks with vouchers to support authenticity verification.

[0017] 5. Integration with methods for cross-domain authenticity verification of AIGC-generated content in publicly available witness libraries and methods for proving data-related behaviors. See Figure 5 This invention (on the generation side) embeds a watermarked signature CSOP, which, together with the cross-domain authenticity verification method for content generated by the publicly available AIGC witness library, forms a trusted closed loop for AIGC; CSOP serves as a witness credential for data-related behavior proof methods. The above embodiments are for illustrative purposes only and not for limitation; all equivalent substitutions and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for embedding multimodal semantic watermarks into AI-generated content and issuing witness certificates, characterized in that, include: Identify the modality of AI-generated content; embed semantic watermarks into the AI-generated content according to the modality; issue witness certificates for the AI-generated content; The semantic watermark is associated with the witness certificate to support the authentication of the AI-generated content.

2. The method according to claim 1, characterized in that, The modal watermarking includes: zero-width character watermarks for text, DCT domain watermarks for images, and frequency domain watermarks for audio and video.

3. The method according to claim 1, characterized in that, The witness certificate records the content fingerprint and generation information of the AI-generated content.

4. The method according to claim 1, characterized in that, The semantic watermark is embedded in the content itself, and the witness certificate is stored in the witness database. The two are linked and mutually verified.

5. The method according to claim 1, characterized in that, The witness certificate is used for authenticity verification and retrieval based on a public witness database.

6. The method according to claim 1, characterized in that, The AI ​​generates data-related behaviors, and the witness certificate is a behavior witness certificate.

7. The method according to claim 1, characterized in that, The watermark and the witness certificate provide dual support for authenticity verification and traceability.

8. The method according to claim 1, characterized in that, The method described above is located on the generation side of AI-generated content, forming a reliable closed loop with the authenticity verification method located on the authentication side.

9. The method according to claim 1, characterized in that, The content fingerprint is a cryptographic digest.

10. The method according to claim 1, characterized in that, The embedding and issuance process will be recorded.

11. The method according to claim 1, characterized in that, The witness certificate uses the SM2 algorithm for signature and the SM3 algorithm for content fingerprint.

12. A system for embedding multimodal semantic watermarks into AI-generated content and issuing witness certificates, characterized in that, include: The recognition module is used to identify content modalities; The embedding module is used to embed semantic watermarks according to modality; the issuance module is used to issue witness certificates. The association module is used to associate watermarks with vouchers; The system is used to perform the method according to any one of claims 1 to 11.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 11.

Images