Watermark information embedding method, device, storage medium and program product

Abstract

The present application provides a watermark information embedding method, device, storage medium and program product, wherein the method includes obtaining the target watermark information to be embedded and the original carrier information in response to a watermark embedding instruction; mapping the target watermark information to a latent variable space to obtain a watermark latent vector, and mapping the original carrier information to the latent variable space to obtain a carrier latent vector; fusing the watermark latent vector and the carrier latent vector to generate a noise latent vector with a watermark; denoising the noise latent vector to generate a target carrier information with the target watermark information. The present application realizes that when the target carrier information is subjected to various complex editing processes, the watermark information can still be effectively extracted, thereby improving the robustness of the watermark and providing a reliable data basis for subsequent digital information ownership protection.

Description

Technical Field

[0001] The present application relates to the field of information processing technology, and in particular to a watermark information embedding method, device, storage medium and program product. Background Art

[0002] In recent years, the large-scale model technology of artificial intelligence generated content (AIGC) has flourished at an unprecedented rate, greatly promoting innovations in multiple fields such as content creation, media communication, education and entertainment. These models can automatically generate text, images, audio and even video content, with amazing realism and creativity, bringing unlimited possibilities to the digital world. However, with the increasing application of AIGC, a series of security and compliance issues have become increasingly prominent and have become challenges that cannot be ignored: the first is copyright and originality issues. For example, applications such as image reconstruction and style transfer based on AIGC technology have made content creation extremely convenient, but at the same time have also triggered disputes about the originality of works and copyright infringement. Therefore, how to protect copyright ownership and ensure the authenticity of content in the context of the booming development of AIGC has become a key issue that needs to be solved urgently.

[0003] Traditional watermarking technology achieves copyright protection by embedding visible or invisible noise in the carrier information. Taking images as an example, the robustness of watermarks can be improved by frequency domain transformation to resist simple image processing operation attacks. However, with the rapid progress of AIGC technology, especially the rise of deep learning-driven image reconstruction and editing methods, these noise-based watermarking schemes seem to be unable to cope with the situation. AIGC can easily modify the pixel structure and feature distribution of an image, resulting in the watermark information embedded by traditional methods being easily lost or damaged after the image is processed by advanced editing, making it difficult to effectively verify the originality of the image and track copyright information. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to provide a watermark information embedding method, device, storage medium and program product, so that the watermark information can still be effectively extracted when the generated target carrier information is faced with various complex editing processes, thereby improving the robustness of the watermark and providing a reliable data basis for subsequent digital information ownership protection.

[0005] In a first aspect, an embodiment of the present application provides a watermark information embedding method, comprising: in response to a watermark embedding instruction, obtaining target watermark information to be embedded and original carrier information; mapping the target watermark information to a latent variable space to obtain a watermark latent vector, and mapping the original carrier information to the latent variable space to obtain a carrier latent vector; fusing the watermark latent vector and the carrier latent vector to generate a noisy latent vector with a watermark; denoising the noise latent vector to generate a target carrier information with the target watermark information.

[0006] In one embodiment, the obtaining of the target watermark information to be embedded includes: obtaining original watermark information to be embedded; and performing checksum coding and / or error correction coding on the original watermark information to generate the target watermark information.

[0007] In one embodiment, mapping the target watermark information to a latent variable space to obtain a watermark latent vector includes: mapping the target watermark information to a latent variable space to obtain an original watermark latent vector; adding preset noise that conforms to a preset distribution feature to the original watermark latent vector to generate a final watermark latent vector, wherein the preset distribution feature is a distribution feature of an initial latent vector of a preset generation model.

[0008] In one embodiment, mapping the target watermark information to the latent variable space to obtain the watermark latent vector includes: mapping the target watermark information to the latent variable space through a pre-trained watermark mapping model to obtain the watermark latent vector, wherein the training process of the watermark mapping model constrains the reconstruction loss of the sample watermark and the divergence loss between the distribution characteristics of the sample latent vector and the preset distribution characteristics as training objectives, and the preset distribution characteristics are the distribution characteristics of the initial latent vector of the preset generation model.

[0009] In one embodiment, the preset distribution feature is a standard normal distribution; mapping the target watermark information to a latent variable space to obtain a watermark latent vector includes: mapping the target watermark information to a latent variable space to obtain an original watermark latent vector; adding preset noise that conforms to a standard normal distribution to the original watermark latent vector to generate a final watermark latent vector.

[0010] In one embodiment, mapping the original carrier information to the latent variable space to obtain the carrier latent vector includes: inputting the original carrier information into a pre-trained carrier mapping model so that the carrier mapping model maps the original carrier information to the latent variable space and outputs the carrier latent vector, wherein the carrier mapping model learns the potential representation of the sample carrier information under a preset distribution feature during the training process, and the preset distribution feature is the distribution feature of the initial latent vector of the preset generation model.

[0011] In one embodiment, the fusing the watermark latent vector and the carrier latent vector to generate a noise latent vector with a watermark includes: determining weight information between the watermark latent vector and the carrier latent vector; and in the latent variable space, fusing the watermark latent vector and the carrier latent vector according to the weight information to generate a noise latent vector with a watermark.

[0012] In one embodiment, it also includes: in response to a watermark detection instruction, obtaining the carrier information to be processed; mapping the carrier information to be processed to the latent variable space to obtain the target latent vector of the carrier information to be processed; and extracting the watermark information to be processed carried by the carrier information to be processed according to the target latent vector.

[0013] In one embodiment, the method further includes: decoding the watermark information to be processed to obtain a check code and / or an error correction code carried by the watermark information to be processed; and checking the watermark information to be processed according to the check code and / or the error correction code.

[0014] In one embodiment, it also includes: inputting the watermark latent vector as initial noise into a preset generation model, so that the preset generation model performs denoising processing on the watermark latent vector according to the prompt word to generate target carrier information with the target watermark information.

[0015] In a second aspect, an embodiment of the present application provides a method for embedding watermark information in a product image, comprising: in response to a watermark embedding instruction, obtaining target watermark information to be embedded and an original product image, wherein the product image comprises product information; mapping the target watermark information to a latent variable space to obtain a watermark latent vector, and mapping the original product image to the latent variable space to obtain an image latent vector; fusing the watermark latent vector and the image latent vector to generate a noise latent vector with a watermark; and denoising the noise latent vector to generate a target product image with the target watermark information.

[0016] In a third aspect, an embodiment of the present application provides a watermark information embedding device, including:

[0017] An acquisition module, used for acquiring target watermark information to be embedded and original carrier information in response to a watermark embedding instruction;

[0018] A mapping module, used to map the target watermark information to a latent variable space to obtain a watermark latent vector, and to map the original carrier information to the latent variable space to obtain a carrier latent vector;

[0019] A fusion module, used for fusing the watermark latent vector and the carrier latent vector to generate a noise latent vector with a watermark;

[0020] The generating module is used to perform denoising on the noise latent vector to generate target carrier information with the target watermark information.

[0021] In one embodiment, the acquisition module is used to acquire original watermark information to be embedded; perform checksum coding and / or error correction coding on the original watermark information to generate the target watermark information.

[0022] In one embodiment, the mapping module is used to map the target watermark information to the latent variable space to obtain the original watermark latent vector; add preset noise that meets the preset distribution characteristics to the original watermark latent vector to generate the final watermark latent vector, wherein the preset distribution characteristics are the distribution characteristics of the initial latent vector of the preset generation model.

[0023] In one embodiment, a mapping module is used to map the target watermark information to the latent variable space through a pre-trained watermark mapping model to obtain the watermark latent vector, wherein the training process of the watermark mapping model takes the reconstruction loss of the constrained sample watermark and the divergence loss between the distribution characteristics of the sample latent vector and the preset distribution characteristics as training objectives, and the preset distribution characteristics are the distribution characteristics of the initial latent vector of the preset generation model.

[0024] In one embodiment, the preset distribution feature is a standard normal distribution; a mapping module is used to map the target watermark information to a latent variable space to obtain an original watermark latent vector; and preset noise that conforms to the standard normal distribution is added to the original watermark latent vector to generate a final watermark latent vector.

[0025] In one embodiment, a mapping module is used to input the original carrier information into a pre-trained carrier mapping model so that the carrier mapping model maps the original carrier information to the latent variable space and outputs the carrier latent vector, wherein the carrier mapping model learns the potential representation of the sample carrier information under a preset distribution feature during training, and the preset distribution feature is the distribution feature of the initial latent vector of the preset generation model.

[0026] In one embodiment, the fusion module is used to determine the weight information between the watermark latent vector and the carrier latent vector; in the latent variable space, the watermark latent vector and the carrier latent vector are fused according to the weight information to generate a noise latent vector with a watermark.

[0027] In one embodiment, it also includes: a detection module, which is used to respond to a watermark detection instruction to obtain the carrier information to be processed; map the carrier information to be processed to the latent variable space to obtain the target latent vector of the carrier information to be processed; and extract the watermark information to be processed carried by the carrier information to be processed according to the target latent vector.

[0028] In one embodiment, it also includes: a verification module, which is used to decode the watermark information to be processed to obtain a verification code and / or an error correction code carried by the watermark information to be processed; and verify the watermark information to be processed according to the verification code and / or the error correction code.

[0029] In one embodiment, the generation module is further used to input the watermark latent vector as initial noise into a preset generation model, so that the preset generation model denoises the watermark latent vector according to the prompt word to generate target carrier information with the target watermark information.

[0030] In a fourth aspect, an embodiment of the present application provides an electronic device, including:

[0031] at least one processor; and

[0032] a memory communicatively coupled to the at least one processor;

[0033] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the electronic device to execute the method described in any one of the above aspects.

[0034] In a fifth aspect, an embodiment of the present application provides a cloud device, including:

[0035] at least one processor; and

[0036] a memory communicatively coupled to the at least one processor;

[0037] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the cloud device to execute the method described in any one of the above aspects.

[0038] In a sixth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When a processor executes the computer-executable instructions, the method described in any one of the above aspects is implemented.

[0039] In a seventh aspect, an embodiment of the present application provides a computer program product, including a computer program, which implements the method described in any of the above aspects when executed by a processor.

[0040] The watermark information embedding method, device, storage medium and program product provided in the embodiment of the present application respectively maps the target watermark information to be embedded and the original carrier information to the latent variable space to obtain the representation in the latent variable space, and then fuses the watermark latent vector and the carrier latent vector in the latent variable space to generate a noise latent vector with the watermark, and generates the target carrier information with the target watermark information by denoising the noise latent vector. Since the watermark information is embedded in the latent variable space, the generated target carrier information can still be effectively extracted when facing various complex editing processes, thereby improving the robustness of the watermark. It provides a reliable data basis for the subsequent protection of digital information ownership. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. Obviously, the drawings described below are some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor.

[0042] Figure 1 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0043] Figure 2 A schematic diagram of an application scenario of a watermark information embedding system provided in an embodiment of the present application;

[0044] Figure 3 A schematic diagram of a process of embedding watermark information according to an embodiment of the present application;

[0045] Figure 4 A schematic diagram of a process for encoding original watermark information provided in an embodiment of the present application;

[0046] Figure 5 A schematic diagram of the structural principle of a watermark mapping model provided in this embodiment;

[0047] Figure 6 A schematic diagram of a watermark detection process provided in an embodiment of the present application;

[0048] Figure 7 A schematic diagram of an application scenario of a watermark information embedding solution provided in an embodiment of the present application;

[0049] Figure 8 A schematic diagram of an application scenario of a watermark embedding solution provided in an embodiment of the present application;

[0050] Fig. 9A schematic diagram of a process for embedding watermark information in a commodity image according to an embodiment of the present application;

[0051] Fig.10 A schematic diagram of the structure of a watermark information embedding device provided in an embodiment of the present application;

[0052] Fig.11 A schematic diagram of the structure of a cloud device provided in an embodiment of the present application.

[0053] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0054] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0055] The term "and / or" in this article is used to describe the association relationship of associated objects, specifically indicating that there may be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0056] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0057] In order to clearly describe the technical solution of the embodiment of the present application, the terms involved in the present application are first defined:

[0058] AI: Artificial Intelligence.

[0059] AIGC: Artificial Intelligence Generated Content, including but not limited to text, images, audio and video.

[0060] AI Editing: Use artificial intelligence technology to edit content, such as image style transfer, filter application, object recognition and removal, background replacement, etc.

[0061] SD: Stable Diffusion, a deep learning model for generating images.

[0062] Diffusion Model: Diffusion Model is a model for generating data, such as the SD model. The basic idea is to start with random noise and gradually "denoise" it to generate target data. A key feature of diffusion models is that they can generate high-quality and diverse samples, which makes them widely used in image generation, text generation and other fields.

[0063] VAE: Variational Autoencoder, variational autoencoder.

[0064] LDM: Latent Diffusion Model, latent diffusion model, is a generative model that obtains intermediate representation results by denoising step by step in a latent representation space, and then decodes it into a complete generated image. LDM first converts high-dimensional raw data (such as images) into low-dimensional latent vectors through an encoder (such as variational autoencoder VAE), and then applies diffusion and inverse diffusion processes in the latent space. The advantage of this is that the compactness of the latent space and the semantic characteristics of the latent variables can be utilized, making the model more efficient and controllable when generating high-resolution images, videos or other complex data. Compared with the diffusion model based on pixel space, the latent representation space significantly reduces the amount of computation. It combines the advantages of the diffusion model and the variational autoencoder (VAE) to generate high-quality images from text or other input patterns.

[0065] DDIM: Denoising Diffusion Implicit Models, denoising diffusion implicit model, is an implicit denoising process that closely matches the probability distribution of the diffusion process by optimizing the sampling step while maintaining the high quality and diversity of the generated samples. It can be applied to the denoising step of the diffusion model or the potential diffusion model.

[0066] DDIM Inversion: The process of inversely mapping a specific image to its representation in the latent space. It refers to the process of mapping a given image or sample back to the latent vector space using the back diffusion process in the DDIM framework. The main goal of this process is to find the latent representation that matches the image so that further sampling, editing, or generation of new images can be performed.

[0067] Prompt: The prompt word of the large model is an input used to guide the model to generate a specific type of output. It tells the model what type of content should be generated, or in what style. In image generation, the prompt may be a text description of the required image. In AI editing, the prompt may be a description of the image editing process.

[0068] Digital watermark: Add some digital information that is not easily perceived by the human eye to digital media information (such as images, sounds, videos, etc.) in order to protect the copyright of digital media, prove the authenticity of products, track piracy or add additional information to products. The watermark information is embedded in the carrier file and does not affect the visibility and integrity of the original file.

[0069] Image watermark: embedding specific information or patterns in digital images, usually invisible or imperceptible, to identify the source of the image, copyright information, or track the use and distribution of the image. This technology is mainly used for copyright protection, content authentication, and preventing unauthorized image copying or tampering.

[0070] FFT: Fast Fourier Transform, fast Fourier transform.

[0071] U-Net: U-Net: Convolutional Networks for Biomedical Image Segmentation, a convolutional neural network architecture for image segmentation.

[0072] MSE: Mean Squared Error, is an indicator commonly used in statistics and signal processing to measure the difference between the predicted value and the true value.

[0073] KL Kullback-Leibler Divergence Loss, divergence loss. The watermark information embedding method of the embodiment of the present application can be applied to any field involving watermark processing.

[0074] Artificial Intelligence (AI) is the study of the technology of using computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), and tasks that usually require human intelligence to complete are performed by computer systems or machines. Artificial Intelligence technology is widely used in various fields.

[0075] In recent years, the large-scale model technology of artificial intelligence generated content (AIGC) has flourished at an unprecedented rate, greatly promoting innovations in multiple fields such as content creation, media communication, education and entertainment. These models can automatically generate text, images, audio and even video content, with amazing realism and creativity, bringing unlimited possibilities to the digital world. However, with the increasing application of AIGC, a series of security and compliance issues have become increasingly prominent and have become challenges that cannot be ignored: the first is the issue of copyright and originality. Applications such as image reconstruction and style transfer based on AIGC technology have made content creation extremely convenient, but at the same time have also triggered disputes about the originality of works and copyright infringement. Therefore, how to protect copyright ownership and ensure the authenticity of content in the context of the booming development of AIGC has become a key issue that needs to be solved urgently.

[0076] Traditional watermarking technology achieves copyright protection by embedding visible or invisible noise in the image, and uses frequency domain transformation to improve robustness to resist simple image processing operations. However, with the rapid progress of AIGC technology, especially the rise of deep learning-driven image reconstruction and editing methods, these noise-based watermarking schemes seem to be unable to cope with the situation. AIGC can easily modify the pixel structure and feature distribution of an image, resulting in the watermark information embedded by traditional methods being easily lost or damaged after the image is processed by advanced editing, making it difficult to effectively verify the originality of the image and track copyright information. Therefore, watermarking technology for videos and images that are resistant to AIGC reconstruction is particularly important.

[0077] Taking the e-commerce platform scenario as an example, merchants on the e-commerce platform can use AI big models to generate different product images to update the products. If others see that the product is selling well, they are likely to misappropriate the product image without authorization, and then use AIGC technology and other methods to process the product image in an attempt to escape infringement. For example, AI editing is used to remove the watermark information of the product image. Traditional watermark technology cannot resist the attack of this AIGC technology. The product image will soon lose its ownership information, seriously infringing the user's information security.

[0078] In a related embodiment, an artificially designed annular watermark can be embedded in the FFT domain of the initial noise vector sampled by the diffusion model as the initial noise of the diffusion model, and then the image generated by the diffusion model is watermarked by the denoising process. This scheme is based on the special properties of Fourier transform for scaling, cropping, rotation, etc. to ensure the robustness of the method for operations such as cropping, scaling, flipping and rotation.

[0079] However, this scheme uses artificially designed watermarks to embed watermarks in the Fourier domain of the initial noise, which will destroy the standard normal distribution of the initial noise, thereby affecting the quality of the final generated image. In addition, this scheme can only be used for endogenous watermarks in the diffusion model and cannot embed watermarks in existing images.

[0080] In order to solve at least one of the above problems, the embodiment of the present application provides a watermark information embedding solution, which maps the target watermark information to be embedded and the original carrier information to the latent variable space respectively, obtains the representation in the latent variable space, and then fuses the watermark latent vector and the carrier latent vector in the latent variable space to generate a noise latent vector with a watermark, and generates the target carrier information with the target watermark information by denoising the noise latent vector. Since the watermark information is embedded in the latent variable space, the generated target carrier information can still be effectively extracted when facing various complex editing processes, thereby improving the robustness of the watermark. It provides a reliable data foundation for the subsequent protection of digital information ownership.

[0081] Some embodiments of the present application are described in detail below in conjunction with the accompanying drawings. In the case where there is no conflict between the embodiments, the following embodiments and the features in the embodiments can be combined with each other. In addition, the step sequence in the following method embodiments is only an example and not a strict limitation.

[0082] like Figure 1 As shown, this embodiment provides an electronic device 1, including: at least one processor 11 and a memory 12, Figure 1 A processor is taken as an example. The processor 11 and the memory 12 are connected via a bus 10. The memory 12 stores instructions that can be executed by the processor 11, and the instructions are executed by the processor 11 so that the electronic device 1 can execute all or part of the process of the method in the following embodiment, so as to achieve that the generated target carrier information can still be effectively extracted when facing various complex editing processes, thereby improving the robustness of the watermark. Provide a reliable data basis for the subsequent protection of digital information ownership.

[0083] In one embodiment, the electronic device 1 may be a mobile phone, a tablet computer, a laptop computer, a desktop computer, or a large computing system composed of multiple computers.

[0084] Figure 2 A schematic diagram of an application scenario 200 of a watermark information embedding system provided in an embodiment of the present application. Figure 2 As shown, the system includes: a server 210 and a terminal 220, wherein:

[0085] Server 210 may be a data platform that provides watermark information embedding services, such as a digital information security service platform for e-commerce. In actual scenarios, an e-commerce platform may have multiple servers 210. Figure 2 In the figure, one server 210 is taken as an example.

[0086] The terminal 220 may be a computer, mobile phone, tablet or other device used by the user to log in to the digital information security service platform. There may be multiple terminals 220. Figure 2 Two terminals 220 are taken as an example for illustration.

[0087] The terminal 220 and the server 210 can transmit information via the Internet, so that the terminal 220 can access the data on the server 210. The terminal 220 and / or the server 210 can be implemented by the electronic device 1.

[0088] The watermark information embedding solution of the embodiment of the present application can be deployed on the server 210, can also be deployed on the terminal 220, or can be partially deployed on the server 210 and partially deployed on the terminal 220. In actual scenarios, it can be selected based on actual needs, and this embodiment does not limit it.

[0089] When the watermark information embedding solution is fully or partially deployed on the server 210 , a calling interface may be opened to the terminal 220 to provide algorithm support to the terminal 220 .

[0090] The method provided in the embodiment of the present application can be implemented by executing the corresponding software code by the electronic device 1, and can be implemented by exchanging data with the server. The electronic device 1 can be a local terminal device. When the method is run on the server, the method can be implemented and executed based on the cloud interaction system, wherein the cloud interaction system includes a server and a client device.

[0091] In a possible implementation, the method provided in the embodiment of the present application provides a graphical user interface through a terminal device, wherein the terminal device can be the local terminal device mentioned above, or can be the client device in the cloud interaction system mentioned above.

[0092] Please see Figure 3 , which is a watermark information embedding method according to an embodiment of the present application, the method can be Figure 1 The electronic device 1 shown is used to perform and can be applied to Figure 2 The watermark information shown in is embedded in the application scenario, so that the generated target carrier information can still be effectively extracted when facing various complex editing processes, thereby improving the robustness of the watermark. It provides a reliable data basis for the subsequent digital information ownership protection. In this embodiment, the terminal 220 is used as an execution end as an example, and the method includes the following steps:

[0093] Step 301: In response to a watermark embedding instruction, obtain target watermark information to be embedded and original carrier information.

[0094] In this step, the original carrier information refers to the multimedia information that needs to be embedded with the watermark information, including but not limited to images, videos, audios, texts and other information. The target watermark information refers to the watermark information that needs to be embedded in the original carrier information. The watermark embedding instruction can be actively triggered by the user. For example, if the user wants to add watermark information to a specified image, the watermark embedding instruction can be triggered for the image. The watermark embedding instruction can also be automatically triggered by the system, such as automatically triggering the watermark embedding instruction for the image generated by the system. The original carrier information can be uploaded by the user, or obtained by the system from local storage or remote storage, or generated by the system in real time.

[0095] In one embodiment, step 301 may include: obtaining original watermark information to be embedded, performing checksum coding and / or error correction coding on the original watermark information, and generating target watermark information.

[0096] In this embodiment, the original watermark information can be customized watermark information in various forms. In order to ensure the reliability of the enhanced watermark, the original watermark information can be encoded. For example, by performing checksum coding and / or error correction coding on the original watermark information to generate the target watermark information, the checksum coding can be used to verify the integrity of the watermark information to ensure that when the watermark is extracted, it can be detected whether the information has been tampered with or damaged. This is particularly important for application scenarios that require information security and integrity. Through error correction coding, the system can use less redundant information to achieve higher error correction capabilities when embedding watermarks, thereby optimizing embedding efficiency. Efficient error detection and correction can still be achieved without significantly increasing the embedding burden. Therefore, the checksum and / or error correction code can effectively detect and correct errors that may occur during watermark embedding, transmission or extraction. This encoding mechanism improves the reliability of the watermark information, ensuring that the watermark information can still be correctly identified and restored even after being disturbed or damaged to a certain extent. It is particularly suitable for scenarios where multimedia content is frequently edited and transmitted.

[0097] Optionally, there may be multiple check codes and error correction codes that can be used, and the check codes include but are not limited to parity check codes, CRC and other check codes. Error correction codes include but are not limited to Hamming codes, BCH codes, RS codes, convolutional codes, turbo codes, LDPC codes and other error correction codes.

[0098] like Figure 4As shown, it is a schematic diagram of a process of encoding original watermark information provided in an embodiment of the present application. Assuming that the original watermark information M is binary watermark information (0101...), after performing checksum coding and error correction coding on the original watermark information M, the target watermark information W (010110001...) containing the checksum code and the error correction code can be generated.

[0099] Step 302: Map the target watermark information to the latent variable space to obtain the watermark latent vector, and map the original carrier information to the latent variable space to obtain the carrier latent vector.

[0100] In this step, the target watermark information and the original carrier information are mapped to the latent variable space respectively, and the representation of the target watermark information and the original carrier information in the latent variable space is obtained, namely, the watermark latent vector and the carrier latent vector, so that the watermark can be embedded in the latent variable space later to improve the reliability of the watermark. In actual scenarios, the mapping process can be implemented by an encoder or a specific mapping function.

[0101] In one embodiment, mapping the target watermark information to the latent variable space to obtain the watermark latent vector includes: mapping the target watermark information to the latent variable space to obtain the original watermark latent vector. Adding a preset noise that meets the preset distribution characteristics to the original watermark latent vector to generate a final watermark latent vector, wherein the preset distribution characteristics are the distribution characteristics of the initial latent vector of the preset generation model.

[0102] In this embodiment, the preset generation model refers to the preset generation model in step 304, and the preset generation model is used to generate the target carrier information with the target watermark information. The preset distribution feature is the distribution feature of the initial latent vector of the preset generation model. By adding the preset noise that conforms to the distribution feature of the initial latent vector to the original watermark latent vector, the robustness of watermark transformation and restoration can be enhanced.

[0103] In one embodiment, the preset distribution feature is a standard normal distribution. Step 302 may specifically include: mapping the target watermark information to a latent variable space to obtain an initial latent vector. Adding preset noise that conforms to a standard normal distribution to the initial latent vector to generate a watermark latent vector.

[0104] In this embodiment, assuming that the initial latent variables of the preset generation model in the reasoning process are close to the standard normal distribution, the final watermark latent vector can be generated by adding preset noise that conforms to the standard normal distribution to the initial watermark latent vector in the latent variable space. For example, the preset noise X∈N(0,1) is noise that conforms to the standard normal distribution, which is used to enhance the robustness of watermark transformation and restoration.

[0105] In one embodiment, the target watermark information is mapped to the latent variable space to obtain the watermark latent vector, including: mapping the target watermark information to the latent variable space through a pre-trained watermark mapping model to obtain the watermark latent vector, wherein the training process of the watermark mapping model constrains the reconstruction loss of the sample watermark and the divergence loss between the distribution characteristics of the sample latent vector and the preset distribution characteristics as training objectives, and the preset distribution characteristics are the distribution characteristics of the initial latent vector of the preset generation model.

[0106] In this embodiment, during the reasoning process of the preset generation model, if the initial noise input into the preset generation model does not conform to the preset distribution characteristics, the target carrier information finally generated will deviate from the original carrier information, resulting in distortion. Taking the image generation model as an example, the original carrier information is the original image. Assuming that the preset distribution characteristics of the initial latent vector of the image generation model are standard normal distribution, in order to ensure that the target image finally generated is consistent with the original image and not distorted, a pre-trained watermark mapping model can be used. During the training process, the model takes the reconstruction loss and the distribution feature divergence loss as the optimization targets, thereby ensuring the accuracy and concealment of the watermark information during the mapping process. The reconstruction loss constrains the information loss of the sample watermark during the mapping and reconstruction process, so that the watermark information can be accurately reconstructed. The distribution feature divergence loss ensures that the generated watermark latent vector matches the distribution characteristics of the initial latent vector of the preset generation model, so that the watermark latent vector is better integrated into the working mechanism of the generation model, thereby reducing the distortion of the target carrier information with the final watermark, improving the generation quality of the model, and reducing the instability that may occur in the generation model during the processing process. This mapping method not only improves the security of watermark information, but also provides higher flexibility and reliability for subsequent watermark information processing and application. The explanation of the preset distribution characteristics can refer to the description of the above embodiment.

[0107] like Figure 5 As shown in FIG. 1 , a schematic diagram of the structural principle of a watermark mapping model provided in this embodiment is shown. Taking the preset generation model as a model based on LDM as an example, it is assumed that the preset distribution feature is a standard normal distribution. Taking the aforementioned watermark scene in the image as an example, the watermark mapping model may include a watermark encoder and a watermark decoder based on VAE. During the training process, the encoder is responsible for mapping the sample watermark information W1 to the sample watermark latent vector Z W1 , the decoder is responsible for converting the sample watermark hidden vector Z W1 The watermark information W1' is obtained by decoding, where X∈N(0, 1) is a preset noise conforming to the standard normal distribution, which is used to enhance the robustness of watermark transformation and restoration.

[0108] During the training process, the training loss function Loss can be defined as:

[0109] Loss=Lre c +λLdiv

[0110] Lrec=MSE(W1,W1′)

[0111] LdivKL(P||Q)

[0112] Among them, W1 is the sample watermark, W1' is the reconstructed sample watermark, Lrec represents the reconstruction loss of the sample watermark W1, and Ldiv represents the sample latent vector Z W1 The divergence loss between the distribution characteristics and the standard normal distribution. λ represents the scale factor, which can be selected according to actual needs.

[0113] P is the hidden vector Z W1 distribution, Q∈N(0,1). KL represents the divergence loss between P and Q. Since the initial latent vector in the LDM derivation process is close to the standard normal distribution, L can be used div To ensure that the watermark latent vector obtained by watermark encoding is also close to the standard normal distribution, to ensure the quality of the final generated target image, and to reduce the distortion of the image before and after the watermark is embedded.

[0114] After training is completed, you can Figure 5 The encoder shown maps the target watermark information W to the latent variable space and obtains the watermark latent vector Z W The decoder can be responsible for converting the watermark hidden vector Z W Decode and obtain the watermark information W'.

[0115] This embodiment transforms the watermark into Gaussian noise that conforms to the standard normal distribution by training a watermark mapping model of an encoder-decoder structure, so that the latent vector after adding the watermark still conforms to the standard normal distribution, thereby avoiding the image distortion that may be caused by the non-Gaussian distribution characteristics of the latent vector.

[0116] Optionally, Figure 5 The embodiment shown uses a watermark mapping model based on a VAE encoder-decoder structure to map the watermark to a latent vector that conforms to the standard normal distribution. In actual scenarios, the watermark mapping model can also be implemented in other ways, such as based on deep learning networks such as UNet, or other methods.

[0117] In one embodiment, the original carrier information is mapped to the latent variable space to obtain the carrier latent vector, including: inputting the original carrier information into a pre-trained carrier mapping model so that the carrier mapping model maps the original carrier information to the latent variable space and outputs the carrier latent vector, wherein the carrier mapping model learns the potential representation of the sample carrier information under the preset distribution characteristics during the training process, and the preset distribution characteristics are the distribution characteristics of the initial latent vector of the preset generation model.

[0118] In this embodiment, by using a pre-trained carrier mapping model, the model learns the potential representation of sample carrier information under preset distribution features during the training process, ensuring the integrity and effectiveness of the carrier information during the mapping process. Through this mapping method, the carrier information is converted into a latent vector in the latent variable space, making the representation of the information more compact and efficient while retaining the key features of the carrier information. The use of preset distribution features ensures that the generated carrier latent vector is consistent with the initial latent vector distribution features of the generation model, thereby enhancing the stability and consistency of information processing, ensuring the consistency between the final generated target carrier information with watermark and the original carrier information, and reducing information distortion.

[0119] Taking the scenario of embedding watermark information in an image as an example, the preset generation model can be a DDIM model, and the preset distribution feature is a standard normal distribution. The carrier mapping model can be implemented using a DDIM Inversion model based on zero text prompts. DDIM Inversion is the inverse process of DDIM in the following step 304. It compresses and encodes the rich visual information of the original carrier image into a low-dimensional latent vector (i.e., the carrier latent vector). This latent vector can be regarded as a compact and abstract representation of the image content, which is convenient for subsequent watermarking in the latent vector space. After this step, the original carrier image is transformed into the latent space to obtain the carrier latent vector Z I , and the vector hidden vector Z I Close to a standard normal distribution.

[0120] Step 303: The watermark latent vector and the carrier latent vector are merged to generate a noise latent vector with a watermark.

[0121] In this step, the watermark latent vector and the carrier latent vector are fused in the latent variable space to generate a noise latent vector with watermark. The fusion process utilizes the characteristics of the latent variable space so that the watermark information can be embedded in the carrier information in a hidden way to form a unified representation. The generated noise latent vector with watermark not only retains the main features of the original carrier information, but also effectively embeds the watermark information, ensuring the concealment and security of the watermark. This method reduces the information loss and distortion in the information fusion process by operating in the latent variable space, and improves the anti-attack ability and robustness of the watermark information.

[0122] In one embodiment, step 303 may specifically include: determining weight information between the watermark latent vector and the carrier latent vector. In the latent variable space, the watermark latent vector and the carrier latent vector are fused according to the weight information to generate a noise latent vector with a watermark.

[0123] In this embodiment, by determining the weight information between the watermark latent vector and the carrier latent vector, the contribution ratio of the watermark information and the carrier information in the fusion process can be flexibly adjusted. The weighted fusion method ensures that the generated noise latent vector can not only effectively embed the watermark information, but also maintain the integrity and characteristic expression of the carrier information. The introduction of weight information provides controllability of watermark strength and carrier information fidelity, enhances the flexibility and adaptability of watermark embedding, and not only improves the effectiveness of information protection, but also provides higher practical value for diversified application scenarios.

[0124] Optionally, taking the scenario of embedding watermark information in an image as an example, assuming that the preset generation model can be a DDIM model, the preset distribution feature is a standard normal distribution, and the watermark adding operation is performed in the latent vector space (the initial noise space of the DDIM model). After the operation in step 302, the target watermark information and the original carrier image are respectively mapped to the latent vector space close to the standard normal distribution, and the watermark latent vectors Z and Z are respectively obtained. W and the vector hidden vector Z I , watermarking can be performed in the latent vector space based on the following formula:

[0125]

[0126] Among them, Z T represents the fused noise latent vector, λ1∈(0,1) is Z w and Z I The weight factor of the fusion, due to Z w , Z I are independent and close to the standard normal distribution, so Z T It is also close to the standard normal distribution. λ1 is used to determine the strength of the watermark. The larger λ1 is, the stronger Z T The stronger the watermark in the image, the stronger the watermark in the final generated image. In actual scenarios, if λ1 is too large, the watermark will be too strong, which will affect the quality of the original carrier image. If λ1 is too small, the final watermark extraction effect will be poor. Therefore, according to actual needs, the appropriate λ1 can be selected based on the principle of minimizing the impact of the watermark on the image quality and maximizing the watermark extraction effect. After this process, the target watermark information is added to the noise space of the original carrier image in the form of a latent vector.

[0127] Step 304: De-noise the noise latent vector to generate target carrier information with target watermark information.

[0128] In this step, the noise latent vector obtained in step 303 can be used as the initial noise input of the preset generation model to the preset generation model. Here, the preset generation model can be associated with the aforementioned carrier mapping model, such as sharing the same model parameters. Therefore, by denoising the noise latent vector through the preset generation model, the target carrier information containing the watermark information can be effectively extracted and reconstructed. This method not only eliminates the redundant and interfering information in the noise latent vector, but also ensures the complete embedding of the target watermark information and the high-quality reconstruction of the target carrier information. Since the watermark information is embedded in the latent variable space, the generated target carrier information can still be effectively extracted when facing various complex editing processes, so that the watermark can remain stable and effective in various environments. It not only improves the effect of information protection and copyright management, but also provides strong support for the secure transmission and storage of digital content.

[0129] Taking the scenario of embedding watermark information in an image as an example, assuming that the preset generation model can be a DDIM model, the preset generation model DDIM here shares the same model parameters with the aforementioned DDIM Inversion, so the standard DDIM denoising process can be used to transform the noise latent vector Z containing the target watermark information into T Denoising After denoising, the potential expression Z0 of the target carrier image can be obtained, and then Z0 is decoded by VAE to obtain the target carrier image containing the watermark.

[0130] The above scheme adopts a watermark embedding mechanism of latent vector fusion. First, through the encoder-decoder architecture, the original watermark data is processed by deep learning at the encoder layer, and the watermark is mapped to a highly flexible noise domain that conforms to the Gaussian distribution. Then, through the fusion mechanism based on weighted superposition, the watermark signal can be naturally integrated into the host data, avoiding the image distortion that may be caused by the non-Gaussian distribution characteristics of the latent variables of the diffusion model.

[0131] In one embodiment, after step 304, the method may further include a watermark detection process, such as Figure 6 FIG. 1 is a flow chart of a watermark detection process provided by an embodiment of the present application, comprising the following steps:

[0132] Step 601: In response to a watermark detection instruction, obtain carrier information to be processed.

[0133] Step 602: Map the carrier information to be processed to the latent variable space to obtain the target latent vector of the carrier information to be processed.

[0134] Step 603: extracting the watermark information to be processed carried by the carrier information to be processed according to the target latent vector.

[0135] In this embodiment, when a watermark detection instruction is received, the carrier information to be processed is first mapped to the latent variable space to generate a target latent vector, making full use of the characteristics of the latent variable space to convert the complex carrier information into a form that is easier to analyze and process. Based on the target latent vector, the watermark information embedded in the carrier information to be processed can be accurately extracted, ensuring the integrity and accuracy of the watermark information. The use of the latent variable space not only improves the accuracy of watermark information extraction, but also enhances the processing capability of complex carrier information. The watermark detection instruction can be actively triggered by the user or automatically triggered by the system.

[0136] Assuming that the carrier information to be processed is the image to be processed, the DDIM Inversion process with zero text prompt can be used to transform the image to be processed into the latent vector Z T ', then based on the above Figure 5 The decoder trained in the embodiment in the example above is used to train the latent vector Z T 'Perform decoding processing to obtain the potential watermark information to be processed in the image to be processed.

[0137] In one embodiment, the method may further include step 604: decoding the watermark information to be processed to obtain a check code and / or an error correction code carried by the watermark information to be processed. Step 605: verifying the watermark information to be processed according to the check code and / or the error correction code. If the check passes, it is determined that the watermark extraction is successful; otherwise, if the check fails, it is determined that the watermark extraction fails.

[0138] In this embodiment, the error correction code is decoded for the watermark information to be processed, and the check code is checked for the information string obtained after decoding. If the check is passed, the watermark extraction is successful, and if the check is not passed, the watermark extraction fails. The decoding process can extract the check code and error correction code from the watermark information to be processed, and these codes are used to detect and correct potential errors or tampering in the watermark information. Through the use of the check code, the integrity and authenticity of the watermark information can be quickly verified to ensure that it has not been tampered with during transmission and storage, and through the check code, the extracted potential watermark can be self-checked without the need for original information comparison. The introduction of the error correction code further enhances the robustness of the system, compensates for the partial distortion phenomenon caused by the watermark method based on latent variable embedding in the DDIM inverse process, so that the watermark information can still be restored and verified in the presence of noise or data damage. This dual verification mechanism improves the reliability and security of the watermark information.

[0139] like Figure 7 As shown in FIG. 1 , a schematic diagram of an application scenario of a watermark information embedding solution provided in an embodiment of the present application is shown. Taking the scenario of embedding a watermark in an original image as an example, assuming that the preset generation model is a DDIM model, the solution may include a watermark embedding process and a watermark extraction process. The specific process steps are as follows: Figure 8As shown, it may include:

[0140] Watermark embedding process: First, through the DDIM Inversion process, the original image is mapped to the latent variable space to obtain the image latent vector Z I , after the original watermark information M is verified and coded, the target watermark information is generated, and the target watermark information is obtained by a pre-trained encoder (such as Figure 5 As shown in the figure), the target watermark information should be set in the latent variable space to obtain the watermark latent vector Z W Then in the latent variable space, the image latent vector Z I and watermark hidden vector Z W Fusion is performed to obtain the noise latent vector Z with watermark after fusion T , through the DDIM model to the noise latent vector Z T Perform T steps (where T represents the number of denoising times and T is a positive integer) of denoising to obtain the denoised information Z0, and then pass it through the decoder (such as Figure 5 As shown in the figure, the information Z0 is decoded to generate a target image containing a watermark.

[0141] Watermark extraction process: In actual scenarios, the target image may be subjected to different attacks, such as AI editing, cropping, compression, etc. For the attacked image, the embodiment of the present application can effectively extract the watermark information therein through the watermark extraction process. Specifically, it may include: mapping the attacked image to the latent variable space through the DDIM Inversion process, and obtaining the latent vector Z of the attacked image T ', and then by Figure 5 The decoder shown in Figure 2 is used to represent the latent vector Z T 'Decode to obtain the latent watermark, perform error correction code decoding on the latent watermark, and perform checksum verification on the information string obtained after decoding. If the verification passes, the original watermark information M is obtained, and the watermark extraction is successful. If the verification fails, the watermark extraction fails.

[0142] The above method first maps the original image to the latent vector space through DDIM Inversion, superimposes the watermark latent vector, and then obtains the watermarked image through DDIM denoising, thereby realizing the end-to-end watermark adding process. In view of the problem that the watermark information is detected by hypothesis testing in the related art and it is necessary to know that the watermark has been embedded when detecting the watermark, the embodiment of the present application verifies and error-corrects the original watermark, and realizes the self-verification of the extracted potential watermark through verification coding, without the need to know that the watermark has been embedded, thereby realizing blind extraction of the watermark and broadening the scope of application.

[0143] In one embodiment, the method may further include: inputting the watermark latent vector as initial noise into a preset generation model, so that the preset generation model performs denoising processing on the watermark latent vector according to the prompt word to generate target carrier information with target watermark information.

[0144] In this embodiment, by taking the watermark latent vector as the initial input, the preset generation model can identify and retain the key features of the watermark information in the process of denoising according to the prompt word, and seamlessly embed it into the generated target carrier information. In this way, the watermark embedding scheme can not only realize the addition of watermarks on the basis of existing original carrier information, but also be applied to the generation process of target carrier information, directly generating target carrier information to be watermarked based on the set target watermark information, thereby expanding the scope of application.

[0145] Taking the preset generation model as a diffusion model as an example, assuming that the target carrier information is an image, this example solution can also be used as an endogenous watermark solution for diffusion models such as LDM. By mapping the watermark to the latent variable space through the encoder in step 302, a watermark latent vector is obtained, and the watermark latent vector is used as the initial noise of LDM, so that the image generated by the model can be natively equipped with the target watermark information. This makes the watermark embedding technology no longer limited to specific processing scenarios.

[0146] like Figure 8 As shown in FIG. 1 , a schematic diagram of an application scenario of a watermark embedding solution provided in an embodiment of the present application is shown. Taking the embedding process of the endogenous watermark in the LDM model as an example, firstly, the set original watermark information M is subjected to verification coding and error correction coding to generate the target watermark information, and then the target watermark information is generated by Figure 5 The encoder shown maps the target watermark information to the latent variable space and obtains the watermark latent vector Z W , and then use the watermark hidden vector Z W is the initial noise, combined with the LDM prompt word: "smiling girl", the initial noise Z is modeled by the DDIM model. W Perform T-step denoising processing, and use a decoder to decode the denoised information to generate a target image containing a watermark.

[0147] It can be seen that this embodiment can not only add a watermark to the original image, but also be integrated into the diffusion model so that the image generated by the diffusion model has its own watermark.

[0148] The above method is based on DDIM Inversion. It embeds the watermark information in the latent variable space of the image by going deep into the semantic level of the image, and cleverly combines it with the high-level semantic features of the image content to resist attacks such as AI reconstruction and editing. Since this scheme is embedded in the latent vector space, it can be used as an end-to-end (input image, output watermarked image) image watermarking scheme, and can also be integrated into diffusion models such as LDM as a large model endogenous watermarking scheme, so that the image generated by the diffusion model has its own watermark.

[0149] The watermark embedding scheme of the embodiment of the present application does not need to add a watermark directly to the final generated image, but embeds customized watermark information in the latent variable space of the diffusion model: first, the original image is restored to the noise space (hidden variable) by the inverse process of DDIM, and then the watermark is added to the latent variable, and then the watermarked image is obtained through the denoising process of the diffusion model. This process ensures the deep integration of the watermark and the content, and the effectiveness of the watermark can be maintained even after multiple reconstructions and editing by AI, which greatly enhances the protection of AIGC content and provides strong technical support for maintaining the clarity and security of the network environment.

[0150] For details of each step of the above method, please refer to the relevant description of the above embodiment, which will not be repeated here.

[0151] Please see Fig. 9 , which is a method for embedding watermark information about a commodity image in an embodiment of the present application, the method can be Figure 1 The electronic device 1 shown is used to perform and can be applied to Figure 2 The watermark information shown in is embedded in the application scenario to generate a product image with a watermark. In this embodiment, the terminal 220 is used as the execution end, and the product image scenario in the e-commerce scenario is used as an example. The method includes the following steps:

[0152] Step 901: In response to a watermark embedding instruction, target watermark information to be embedded and an original product image are acquired, where the product image includes product information.

[0153] In this step, product information includes but is not limited to product-related knowledge information, appearance information, product promotion information, product-related order information, product-related service information, etc. AI-generated content about products includes but is not limited to product images, product introduction text, product introduction audio and video, etc., such as the AI ​​main picture of the product, the AI ​​details page of the product, etc.

[0154] Step 902: Map the target watermark information to the latent variable space to obtain the watermark latent vector, and map the original product image to the latent variable space to obtain the image latent vector.

[0155] Step 903: In the latent variable space, the watermark latent vector and the image latent vector are fused to generate a noise latent vector with a watermark.

[0156] Step 904: input the noise latent vector into the preset generation model so that the preset generation model performs denoising processing on the noise latent vector to generate a target product image with target watermark information.

[0157] The above-mentioned method for embedding watermark information of commodity images maps the target watermark information to be embedded and the original commodity image to the latent variable space respectively to obtain the representation in the latent variable space, and then fuses the watermark latent vector and the image latent vector in the latent variable space to generate a noise latent vector with a watermark, and denoises the noise latent vector through a preset generation model to generate a target commodity image with the target watermark information. Since the watermark information is embedded in the latent variable space, the generated target commodity image can still effectively extract the watermark information when facing various complex editing processes, thereby improving the robustness of the watermark. It provides a reliable data basis for the subsequent protection of the ownership of commodity images.

[0158] For details of each step of the above method, please refer to the relevant description of the above embodiment, which will not be repeated here.

[0159] Please see Fig.10 , which is a watermark information embedding device 1000 of an embodiment of the present application, which can be applied to Figure 1 The electronic device 1 shown can be applied to Figure 2 The watermark information shown in is embedded in the application scenario to achieve that the generated target carrier information can still be effectively extracted when facing various complex editing processes, thereby improving the robustness of the watermark. It provides a reliable data basis for the subsequent protection of digital information ownership. The device includes: an acquisition module, a mapping module, a fusion module and a generation module, and the functional principles of each module are as follows:

[0160] The acquisition module is used to acquire the target watermark information to be embedded and the original carrier information in response to the watermark embedding instruction.

[0161] The mapping module is used to map the target watermark information to the latent variable space to obtain the watermark latent vector, and to map the original carrier information to the latent variable space to obtain the carrier latent vector.

[0162] The fusion module is used to fuse the watermark latent vector and the carrier latent vector to generate a noise latent vector with a watermark.

[0163] The generation module is used to denoise the noise latent vector and generate target carrier information with target watermark information.

[0164] In one embodiment, the acquisition module is used to acquire the original watermark information to be embedded, and perform checksum coding and / or error correction coding on the original watermark information to generate target watermark information.

[0165] In one embodiment, the mapping module is used to map the target watermark information to the latent variable space to obtain the original watermark latent vector. A preset noise that meets the preset distribution characteristics is added to the original watermark latent vector to generate the final watermark latent vector, wherein the preset distribution characteristics are the distribution characteristics of the initial latent vector of the preset generation model.

[0166] In one embodiment, a mapping module is used to map the target watermark information to the latent variable space through a pre-trained watermark mapping model to obtain a watermark latent vector, wherein the training process of the watermark mapping model constrains the reconstruction loss of the sample watermark and the divergence loss between the distribution characteristics of the sample latent vector and the preset distribution characteristics as training objectives, and the preset distribution characteristics are the distribution characteristics of the initial latent vector of the preset generation model.

[0167] In one embodiment, the preset distribution feature is a standard normal distribution. A mapping module is used to map the target watermark information to a latent variable space to obtain an original watermark latent vector. A preset noise that conforms to the standard normal distribution is added to the original watermark latent vector to generate a final watermark latent vector.

[0168] In one embodiment, a mapping module is used to input the original carrier information into a pre-trained carrier mapping model so that the carrier mapping model maps the original carrier information to a latent variable space and outputs a carrier latent vector, wherein the carrier mapping model learns the potential representation of the sample carrier information under a preset distribution feature during the training process, and the preset distribution feature is the distribution feature of the initial latent vector of the preset generation model.

[0169] In one embodiment, the fusion module is used to determine the weight information between the watermark latent vector and the carrier latent vector. In the latent variable space, the watermark latent vector and the carrier latent vector are fused according to the weight information to generate a noise latent vector with a watermark.

[0170] In one embodiment, the method further includes: a detection module, which is used to obtain the carrier information to be processed in response to the watermark detection instruction, map the carrier information to be processed to the latent variable space, and obtain the target latent vector of the carrier information to be processed, and extract the watermark information to be processed carried by the carrier information to be processed according to the target latent vector.

[0171] In one embodiment, the system further includes: a verification module, which is used to decode the watermark information to be processed, obtain the verification code and / or error correction code carried by the watermark information to be processed, and verify the watermark information to be processed according to the verification code and / or error correction code.

[0172] In one embodiment, the generation module is further used to input the watermark latent vector as initial noise into a preset generation model, so that the preset generation model performs denoising processing on the watermark latent vector according to the prompt word to generate target carrier information with target watermark information.

[0173] For a detailed description of the watermark information embedding device 1000, please refer to the description of the relevant method steps in the above embodiment. The implementation principle and technical effect are similar and will not be repeated in this embodiment.

[0174] Fig.11 The following is a schematic diagram of a cloud device 110 provided in an exemplary embodiment of the present application. The cloud device 110 can be used to run the method provided in any of the above embodiments. Fig.11 As shown, the cloud device 110 may include: a memory 1104 and at least one processor 1105, Fig.11 A processor is taken as an example.

[0175] The memory 1104 is used to store computer programs and can be configured to store various other data to support operations on the cloud device 110. The memory 1104 can be an object storage service (OSS).

[0176] Memory 1104 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0177] The processor 1105 is coupled to the memory 1104 and is used to execute the computer program in the memory 1104 to implement the solution provided by any of the above method embodiments. The specific functions and technical effects that can be achieved are not repeated here.

[0178] Furthermore, if Fig.11 The cloud device also includes: a firewall 1101, a load balancer 1102, a communication component 1106, a power supply component 1103 and other components. Fig.11 Only some components are shown schematically, which does not mean that the cloud device only includes Fig.11 Components shown.

[0179] In one embodiment, the above Fig.11The communication component 1106 in is configured to facilitate wired or wireless communication between the device where the communication component 1106 is located and other devices. The device where the communication component 1106 is located can access a wireless network based on a communication standard, such as WiFi, 2G, 3G, 4G, LTE (Long Term Evolution, Long Term Evolution, referred to as LTE), 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component 1106 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 1106 also includes a near field communication (Near Field Communication, referred to as NFC) module to facilitate short-range communication. For example, the NFC module can be based on Radio Frequency Identification (Radio Frequency Identification, referred to as RFID) technology, Infrared Data Association (Infrared Data Association, referred to as IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (bluetooth, referred to as BT) technology and other technologies to achieve.

[0180] In one embodiment, the above Fig.11 The power supply component 1103 provides power to various components of the device where the power supply component 1103 is located. The power supply component 1103 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device where the power supply component is located.

[0181] An embodiment of the present application further provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the method of any of the aforementioned embodiments is implemented.

[0182] An embodiment of the present application also provides a computer program product, including a computer program, which implements the method of any of the aforementioned embodiments when executed by a processor.

[0183] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0184] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the methods of various embodiments of the present application.

[0185] It should be understood that the above-mentioned processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The memory may include a high-speed RAM (Random Access Memory) memory, and may also include non-volatile storage NVM (Nonvolatile memory, NVM for short), such as at least one disk storage, and can also be a USB flash drive, a mobile hard disk, a read-only memory, a disk or an optical disk, etc.

[0186] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special computer.

[0187] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0188] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, clothing or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, clothing or device. In the absence of more restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, clothing or device including the element.

[0189] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0190] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods of each embodiment of the present application.

[0191] In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of user data and other information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0192] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A watermark information embedding method, characterized in that: include: In response to the watermark embedding instruction, obtaining target watermark information to be embedded and original carrier information; Mapping the target watermark information to the latent variable space to obtain a watermark latent vector, and mapping the original carrier information to the latent variable space to obtain a carrier latent vector; The watermark latent vector and the carrier latent vector are merged to generate a noise latent vector with a watermark; The noise latent vector is denoised to generate target carrier information with the target watermark information.

2. The method according to claim 1, characterized in that The step of obtaining target watermark information to be embedded includes: Obtaining the original watermark information to be embedded; The original watermark information is subjected to checksum coding and / or error correction coding to generate the target watermark information.

3. The method according to claim 1, characterized in that Mapping the target watermark information to a latent variable space to obtain a watermark latent vector includes: Mapping the target watermark information to the latent variable space to obtain the original watermark latent vector; A preset noise conforming to a preset distribution feature is added to the original watermark latent vector to generate a final watermark latent vector, wherein the preset distribution feature is a distribution feature of an initial latent vector of a preset generation model.

4. The method according to claim 1 or 3, characterized in that: Mapping the target watermark information to a latent variable space to obtain a watermark latent vector includes: The target watermark information is mapped to the latent variable space through a pre-trained watermark mapping model to obtain the watermark latent vector, wherein the training process of the watermark mapping model takes the reconstruction loss of the constrained sample watermark and the divergence loss between the distribution characteristics of the sample latent vector and the preset distribution characteristics as training objectives, and the preset distribution characteristics are the distribution characteristics of the initial latent vector of the preset generation model.

5. The method according to claim 3, characterized in that: The preset distribution feature is a standard normal distribution; and mapping the target watermark information to a latent variable space to obtain a watermark latent vector includes: Mapping the target watermark information to the latent variable space to obtain the original watermark latent vector; A preset noise conforming to a standard normal distribution is added to the original watermark latent vector to generate a final watermark latent vector.

6. The method according to claim 1, characterized in that The mapping the original carrier information to the latent variable space to obtain the carrier latent vector includes: The original carrier information is input into a pre-trained carrier mapping model so that the carrier mapping model maps the original carrier information to the latent variable space and outputs the carrier latent vector, wherein the carrier mapping model learns the potential representation of the sample carrier information under a preset distribution feature during the training process, and the preset distribution feature is the distribution feature of the initial latent vector of the preset generation model.

7. The method according to claim 1, characterized in that The step of fusing the watermark latent vector with the carrier latent vector to generate a noise latent vector with a watermark includes: Determining weight information between the watermark latent vector and the carrier latent vector; In the latent variable space, the watermark latent vector and the carrier latent vector are fused according to the weight information to generate a noise latent vector with a watermark.

8. The method according to claim 1, characterized in that Also includes: In response to the watermark detection instruction, obtaining the carrier information to be processed; Mapping the carrier information to be processed to the latent variable space to obtain a target latent vector of the carrier information to be processed; The watermark information to be processed carried by the carrier information to be processed is extracted according to the target latent vector.

9. The method according to claim 8, characterized in that Also includes: Decoding the watermark information to be processed to obtain a check code and / or an error correction code carried by the watermark information to be processed; The watermark information to be processed is verified according to the verification code and / or the error correction code.

10. The method according to claim 1, characterized in that Also includes: The watermark latent vector is input as initial noise into a preset generation model, so that the preset generation model performs denoising processing on the watermark latent vector according to the prompt word to generate target carrier information with the target watermark information.

11. A method for embedding watermark information into a commodity image, characterized in that: include: In response to a watermark embedding instruction, acquiring target watermark information to be embedded and an original commodity image, wherein the commodity image includes commodity information; Mapping the target watermark information to a latent variable space to obtain a watermark latent vector, and mapping the original product image to the latent variable space to obtain an image latent vector; The watermark latent vector and the image latent vector are merged to generate a noise latent vector with a watermark; The noise latent vector is denoised to generate a target product image with the target watermark information.

12. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the electronic device to perform the method described in any one of claims 1-11.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the processor executes the computer-executable instructions, the method according to any one of claims 1 to 11 is implemented.

14. A computer program product, characterized in that The method comprises a computer program, which implements the method according to any one of claims 1 to 11 when being executed by a processor.

Images