Black box watermark model-based copyright protection method, apparatus and device, and storage medium

Through the copyright protection method based on the black box watermark model, the target watermark image is generated and the suspicious images are extracted and compared, which solves the problem of image copyright source identification in the prior art, and effectively protects copyright files and tracks of infringement activities. .

CN119006257APending Publication Date: 2024-11-22MACAO POLYTECHNIC INST

Patent Information

Application Number
CN202411046648.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing digital watermarking technology is difficult to accurately identify the copyright source of images and cannot effectively prevent information subversion and copyright infringement activities.

Method used

Using the copyright protection method based on the black box watermark model, the target watermark image is generated by inputting the image text information into the trained black box watermark model, and watermark extraction and comparison detection are carried out on suspicious images to determine the source of the image, and the copyright file is realized. effective detection and protection.

Benefits of technology

It realizes accurate judgment of the source of image copyright, effectively protects copyright files, and enhances the ability to identify the source of suspicious images and track copyright infringement activities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006257A_ABST
    Figure CN119006257A_ABST
Patent Text Reader

Abstract

The invention discloses a copyright protection method, device and equipment based on a black box watermark model and a storage medium, and belongs to the technical field of information protection. The method comprises the following steps: inputting image text information into a trained black box watermark model for image generation preparation; performing image generation based on the prepared black box watermark model, and outputting a target watermark image; obtaining a target suspicious image and carrying out watermark extraction on the target suspicious image to obtain suspicious image watermark information; and performing comparison detection based on the suspicious image watermark information and the target watermark image to obtain an image comparison detection result. According to the method, the corresponding target watermark image is generated based on the text information, and then the suspicious image watermark information extracted from the suspicious image is subjected to consistency comparison with the watermark information of the target watermark image, so that the source of the suspicious image is determined, and effective detection and protection of the copyright file are realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of information protection technology, and in particular to a copyright protection method, device, equipment and storage medium based on a black box watermark model. Background Art

[0002] In today's rapidly developing information age, data is growing at an unprecedented rate. In this context, the maintenance of intellectual property rights and the prevention of information subversion have become increasingly important. Digital watermarking technology plays a key role in regulating artificial intelligence generated content (AIGC) and maintaining copyright protection. It can be embedded in LDM and its generated output as a digital fingerprint, which is essential for identifying AIGC and tracking copyright infringement activities.

[0003] However, although the current digital watermarking methods have many potentials and application prospects, they still have some problems and functional deficiencies, such as the inability to accurately identify the copyright source of suspicious images.

[0004] In view of this, there is an urgent need for a black box watermark model copyright protection method to accurately determine the source of image copyright.

[0005] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0006] The main purpose of the present invention is to provide a copyright protection method, device, equipment and storage medium based on a black box watermark model, aiming to solve the technical problem of how to determine the copyright source of an image based on the black box watermark technology in the prior art.

[0007] To achieve the above object, the present invention provides a copyright protection method based on a black box watermark model, the method comprising the following steps:

[0008] Input the image text information into the trained black box watermark model to prepare for image generation;

[0009] Generate an image based on the prepared black box watermark model and output the target watermark image;

[0010] Acquire a target suspicious image and extract a watermark from the target suspicious image to obtain watermark information of the suspicious image;

[0011] A comparison detection is performed based on the suspicious image watermark information and the target watermark image to obtain an image comparison detection result.

[0012] Optionally, image generation is performed based on the prepared black box watermark model to output a target watermark image, specifically including:

[0013] Generate a process image based on the prepared black box watermark model and the image text information to obtain a target process image;

[0014] The target process image is watermarked to obtain a target watermark image.

[0015] Optionally, embedding a watermark into the target process image to obtain a target watermark image specifically includes:

[0016] Generate a watermark based on the trained black box watermark model to obtain a first watermark to be embedded;

[0017] Performing embedding preprocessing on the first watermark to be embedded to obtain a second watermark to be embedded;

[0018] The target process image is watermarked according to the second watermark to be embedded to obtain a target watermark image.

[0019] Optionally, performing embedding preprocessing on the first watermark to be embedded to obtain a second watermark to be embedded specifically includes:

[0020] Generate a feature matrix based on the first watermark to be embedded to obtain a first feature matrix;

[0021] Acquire watermark embedding parameters based on the trained black box watermark model;

[0022] Watermark embedding matching is performed on the first feature matrix according to the watermark embedding parameters to obtain the second watermark to be embedded.

[0023] Optionally, acquiring a target suspicious image and performing watermark extraction on the target suspicious image to obtain watermark information of the suspicious image specifically includes:

[0024] Acquire a suspicious target image;

[0025] Performing dimension reshaping processing based on the suspicious image to obtain a processed suspicious image;

[0026] The processed suspicious image is input into a watermark extraction network to extract the watermark, and the watermark information of the suspicious image is obtained.

[0027] Optionally, it is characterized in that, based on the watermark information of the suspicious image and the target watermark image, a comparison detection is performed to obtain an image comparison detection result, which specifically includes:

[0028] Analyze the target watermark image to obtain target watermark information;

[0029] The target watermark information is compared with the suspicious image watermark information for consistency to obtain the image comparison detection result.

[0030] Optionally, the watermark extraction process of the trained black box watermark model during the model training phase specifically includes:

[0031] Get training set image data;

[0032] Inputting the training set image data into an image attack network to perform image attack simulation, thereby obtaining the training set image data after attack simulation;

[0033] Watermark extraction is performed based on the training set image data after the attack simulation to obtain training set image watermark information.

[0034] In addition, to achieve the above purpose, the present invention also proposes a copyright protection device based on a black box watermark model, the copyright protection device based on the black box watermark model comprises:

[0035] Information input module: inputs image text information into the trained black box watermark model to prepare for image generation;

[0036] Image generation module: generates images based on the prepared black box watermark model and outputs the target watermark image;

[0037] Watermark extraction module: obtains a target suspicious image and extracts watermarks from the target suspicious image to obtain watermark information of the suspicious image;

[0038] Image comparison module: performs image comparison based on the watermark information of the suspicious image and the target watermark image to obtain an image comparison detection result.

[0039] In addition, to achieve the above-mentioned purpose, the present invention also proposes a copyright protection device based on a black box watermark model, and the copyright protection device based on the black box watermark model includes: a memory, a processor, and a copyright protection program based on the black box watermark model stored in the memory and executable on the processor, and the copyright protection program based on the black box watermark model is configured to implement the steps of the copyright protection method based on the black box watermark model as described above.

[0040] In addition, to achieve the above-mentioned purpose, the present invention also proposes a computer-readable storage medium storing a computer program, wherein the storage medium stores a copyright protection program based on a black box watermark model, and when the copyright protection program based on the black box watermark model is executed by a processor, the steps of the copyright protection method based on the black box watermark model as described above are implemented.

[0041] The present invention inputs the image text information into the trained black box watermark model to prepare for image generation; generates the image based on the prepared black box watermark model and outputs the target watermark image; obtains the target suspicious image and extracts the watermark of the target suspicious image to obtain the watermark information of the suspicious image; performs a comparison test based on the watermark information of the suspicious image and the target watermark image to obtain the image comparison test result. The present invention first generates the corresponding target watermark image based on the text information, and then compares the watermark information of the suspicious image extracted from the suspicious image with the watermark information of the target watermark image for consistency, thereby determining the source of the suspicious image and realizing effective detection and protection of copyright files. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a structural diagram of a copyright protection device based on a black box watermark model in a hardware operating environment involved in an embodiment of the present invention;

[0043] Figure 2 It is a flowchart of the first embodiment of the copyright protection method based on the black box watermark model of the present invention;

[0044] Figure 3 It is a flowchart of a second embodiment of a copyright protection method based on a black box watermark model of the present invention;

[0045] Figure 4 It is a flowchart of a third embodiment of a copyright protection method based on a black box watermark model of the present invention;

[0046] Figure 5 It is a flowchart of a fourth embodiment of a copyright protection method based on a black box watermark model of the present invention;

[0047] Figure 6 It is a structural block diagram of the first embodiment of the copyright protection device based on the black box watermark model of the present invention.

[0048] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0049] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0050] Reference Figure 1 , Figure 1 It is a schematic diagram of the structure of a copyright protection device based on a black box watermark model in the hardware operating environment involved in the embodiment of the present invention.

[0051] like Figure 1As shown, the copyright protection device based on the black box watermark model may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (Wireless-Fidelity, Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM), or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk storage. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0052] Those skilled in the art will understand that Figure 1 The structure shown in the figure does not constitute a limitation on the copyright protection device based on the black box watermark model, and may include more or less components than those shown in the figure, or combine certain components, or arrange the components differently.

[0053] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a copyright protection program based on a black box watermark model.

[0054] exist Figure 1 In the copyright protection device based on the black box watermark model shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the copyright protection device based on the black box watermark model of the present invention can be set in the copyright protection device based on the black box watermark model, and the copyright protection device based on the black box watermark model calls the copyright protection program based on the black box watermark model stored in the memory 1005 through the processor 1001, and executes the copyright protection method based on the black box watermark model provided by the embodiment of the present invention.

[0055] The embodiment of the present invention provides a copyright protection method based on a black box watermark model. Figure 2 , Figure 2 The figure is a flowchart of a first embodiment of a copyright protection method based on a black box watermark model of the present invention.

[0056] In this embodiment, the copyright protection method based on the black box watermark model includes the following steps:

[0057] Step S10: inputting the image text information into the trained black box watermark model to prepare for image generation;

[0058] It should be noted that, in the specific implementation, the user can edit the image text information according to his own needs, and then input the image text information into the trained black box watermark model to prepare for image generation, that is, the final generated image can meet the user's own needs.

[0059] It should also be noted that, in the specific implementation, the trained black box watermark model includes a latent diffusion model (LDM, Latent Diffusion Model), which can be used to generate images based on image text information, that is, to realize the model function of artificial intelligence generated content (AIGC, Artificial Intelligence Generated Content).

[0060] Step S20: generating an image based on the prepared black box watermark model and outputting a target watermark image;

[0061] It should be noted that, in the specific implementation, the watermark embedding process involved in the target watermark image is implemented in the process of generating the target watermark image. Specifically, it means that in the process of generating the target watermark image by the latent diffusion model, the watermark embedding network will embed the watermark into the latent representation layer of the image.

[0062] Step S30: acquiring a target suspicious image and extracting a watermark from the target suspicious image to obtain watermark information of the suspicious image;

[0063] It should be noted that, in a specific implementation, suspicious images include pictures that are circulated on various social media platforms and are similar to the target watermark image, and the suspicious image watermark information extracted from them can be used for image source detection.

[0064] Step S40: performing a comparison test based on the suspicious image watermark information and the target watermark image to obtain an image comparison test result.

[0065] It can be understood that, in a specific implementation, the purpose of performing a comparison test based on the watermark information of the suspicious image and the target watermark image is to determine whether the suspicious image is legally derived from the target watermark image.

[0066] This embodiment inputs the image text information into the trained black box watermark model to prepare for image generation; generates an image based on the prepared black box watermark model and outputs a target watermark image; obtains a target suspicious image and extracts watermarks from the target suspicious image to obtain suspicious image watermark information; performs a comparison test based on the suspicious image watermark information and the target watermark image to obtain an image comparison test result. This embodiment first generates a corresponding target watermark image based on the text information, and then compares the suspicious image watermark information extracted from the suspicious image with the watermark information of the target watermark image for consistency, thereby determining the source of the suspicious image and achieving effective detection and protection of copyright files.

[0067] refer to Figure 3 , Figure 3 It is a flowchart diagram of a second embodiment of a copyright protection method based on a black box watermark model of the present invention.

[0068] Based on the above first embodiment, in this embodiment, step S20 specifically includes:

[0069] Step S21: Generate a process image based on the prepared black box watermark model and image text information to obtain a target process image;

[0070] It should be noted that, in the specific implementation, the image text information specifically refers to the image description information input by the user into the trained black box watermark model according to his own needs, which specifically includes description information related to the image content, image color information, and image format information.

[0071] Step S22: embedding a watermark into the target process image to obtain a target watermark image.

[0072] It can be understood that, in the specific implementation, the watermark embedding of the target process image is specifically achieved through the watermark embedding network. Specifically, in the process of LDM generating the image, the watermark is embedded into the potential representation of the image through the watermark embedding network. This embedding method can deeply integrate the watermark information into the potential features of the image, so that the watermark becomes a part of the image generation process, which increases the concealment and robustness of the watermark. Even after the image is subjected to various attacks (such as compression, noise, brightness adjustment, etc.), the watermark can still be maintained and extracted.

[0073] This embodiment obtains a target process image based on the prepared black box watermark model and image text information, and then performs watermark embedding based on the target process image to obtain a target watermark image, thereby realizing watermark encryption of image data.

[0074] refer to Figure 4 , Figure 4It is a flowchart diagram of a third embodiment of a copyright protection method based on a black box watermark model of the present invention.

[0075] Based on the above second embodiment, in this embodiment, step S22 specifically includes:

[0076] Step S221: generating a watermark based on the trained black box watermark model to obtain a first watermark to be embedded;

[0077] It should be noted that, in the specific implementation, the first watermark to be embedded is generated based on the watermark embedding network in the trained black-box watermark model, and its specific generation process is to generate a feature matrix adapted to the StableDiffusion model standard through the convolution layer of the watermark embedding network, that is, the first watermark to be embedded. For example, if the potential representation dimension of the StableDiffusion model standard is 4×96×96, and the potential representation dimension of the selected model block is 128×768×768, the watermark image generated based on the trained black-box watermark model can be processed by two convolution layers to generate a 128×256×256 feature matrix, that is, the first watermark to be embedded.

[0078] Step S222: performing embedding preprocessing on the first watermark to be embedded to obtain a second watermark to be embedded;

[0079] It can be understood that, in a specific implementation, the embedding preprocessing is essentially to expand the dimension of the first watermark to be embedded for matching the potential representation, thereby achieving a deep fusion of the watermark feature and the potential representation, so that the network can naturally combine the watermark feature and the potential representation. For example, if the potential representation dimension of the Stable Diffusion model standard is 4×96×96, the potential representation dimension of the selected model block is 128×768×768, and the corresponding feature matrix generated is 128×256×256, nine such feature matrices can be repeatedly stacked to form a 128×768×768 matrix, i.e., the second watermark to be embedded.

[0080] Step S223: embedding a watermark into the target process image according to the second watermark to be embedded to obtain a target watermark image.

[0081] It should be noted that, in the specific implementation, the watermark embedding process is implemented based on the watermark embedding network, and in order to further enhance the effectiveness of watermark information embedding, a zero convolution layer is also included at the end of the watermark embedding network. In the early stage of model training, the output result of the watermark embedding network will produce noise. The zero convolution layer can effectively suppress the irrelevant noise input in the early stage of training, thereby ensuring the high-quality generation of the watermark image.

[0082] This embodiment generates a first watermark to be embedded based on the trained black box watermark model, and then performs embedding preprocessing based on the first watermark to be embedded to obtain a second watermark to be embedded, so as to finally obtain the corresponding target watermark image through watermark embedding.

[0083] Furthermore, embedding preprocessing is performed on the first watermark to be embedded to obtain a second watermark to be embedded, specifically including: generating a feature matrix based on the first watermark to be embedded to obtain a first feature matrix; acquiring watermark embedding parameters based on the trained black box watermark model; and performing watermark embedding matching on the first feature matrix according to the watermark embedding parameters to obtain a second watermark to be embedded.

[0084] It should be noted that, in the specific implementation, the watermark embedding parameters essentially refer to the potential representation dimension data based on the Stable Diffusion model standard. For example, if the potential representation dimension of the Stable Diffusion model standard is 4×96×96, the potential representation dimension of the selected model block is 128×768×768, and the corresponding feature matrix generated is 128×256×256, nine such feature matrices can be repeatedly stacked to form a 128×768×768 matrix, i.e., the second watermark to be embedded.

[0085] refer to Figure 5 , Figure 5 It is a flowchart diagram of a fourth embodiment of a copyright protection method based on a black box watermark model of the present invention.

[0086] Based on the above first embodiment, in this embodiment, step S30 specifically includes:

[0087] Step S31: Acquire a target suspicious image;

[0088] It can be understood that, in a specific implementation, the suspicious images include pictures that are circulated on various social media platforms and are similar to the target watermark image.

[0089] Step S32: performing dimension reshaping processing based on the suspicious image to obtain a processed suspicious image;

[0090] It should be noted that, in a specific implementation, the dimension reshaping process is to adaptively adjust the format dimension of the suspicious image so that it can meet the image format requirements of subsequent watermark extraction. For example, the suspicious image can be reshaped from the dimension of 3×768×768 to a lower dimensional feature space of (3×3×3)×256×256.

[0091] Step S33: input the processed suspicious image into the watermark extraction network to extract the watermark, and obtain the suspicious image watermark information.

[0092] It should also be noted that, in the specific implementation, the watermark extraction process specifically uses the U-Net network (watermark extraction network) to extract watermark information, where the encoder of the U-Net network is based on the EfficientNet network, and the decoder includes the traditional Conv-GN-SiLU module combination.

[0093] This embodiment performs dimensional reshaping processing on the target suspicious image after acquiring it, thereby obtaining a processed suspicious image, and then performs watermark extraction to obtain corresponding suspicious image watermark information, providing a basis for subsequent image source comparison.

[0094] Furthermore, a comparison test is performed based on the suspicious image watermark information and the target watermark image to obtain an image comparison test result, which specifically includes: parsing based on the target watermark image to obtain the target watermark information; and comparing the target watermark information with the suspicious image watermark information for consistency to obtain the image comparison test result.

[0095] It is understandable that in a specific implementation, the target watermark information is compared with the suspicious image watermark information for consistency, specifically, the initial source of the suspicious image is determined based on the suspicious image watermark information, that is, whether the formation process of the suspicious image is consistent with that of the target watermark image is determined.

[0096] Furthermore, the watermark extraction process of the trained black box watermark model during the model training stage specifically includes: obtaining training set image data; inputting the training set image data into the image attack network to perform image attack simulation to obtain the training set image data after the attack simulation; performing watermark extraction based on the training set image data after the attack simulation to obtain the training set image watermark information.

[0097] It can be understood that in the specific implementation, the trained black box watermark model can be provided with an image attack simulation layer during the model training stage in order to simulate various attack situations that may be encountered during the actual use and dissemination of the image. The image attack simulation layer can simulate eight common types of attacks, namely, random brightness adjustment, random contrast adjustment, random saturation adjustment, horizontal flip, vertical flip, median blur, Gaussian noise and JPEG compression.

[0098] In addition, an embodiment of the present invention also proposes a computer-readable storage medium storing a computer program, wherein the storage medium stores a copyright protection program based on a black box watermark model, and when the copyright protection program based on the black box watermark model is executed by a processor, the steps of the copyright protection method based on the black box watermark model as described above are implemented.

[0099] Since the storage medium adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought by the technical solutions of the above embodiments, which will not be described one by one here.

[0100] Reference Figure 6 , Figure 6 It is a structural block diagram of the first embodiment of the copyright protection device based on the black box watermark model of the present invention.

[0101] like Figure 6 As shown, the copyright protection device based on the black box watermark model proposed in the embodiment of the present invention includes:

[0102] Information input module 10: inputs image text information into the trained black box watermark model to prepare for image generation;

[0103] Image generation module 20: generates an image based on the prepared black box watermark model and outputs a target watermark image;

[0104] Watermark extraction module 30: obtains a target suspicious image and extracts watermarks from the target suspicious image to obtain watermark information of the suspicious image;

[0105] Image comparison module 40: performs image comparison based on the watermark information of the suspicious image and the target watermark image to obtain an image comparison detection result.

[0106] This embodiment inputs the image text information into the trained black box watermark model to prepare for image generation; generates an image based on the prepared black box watermark model and outputs a target watermark image; obtains a target suspicious image and extracts watermarks from the target suspicious image to obtain suspicious image watermark information; performs a comparison test based on the suspicious image watermark information and the target watermark image to obtain an image comparison test result. This embodiment first generates a corresponding target watermark image based on the text information, and then compares the suspicious image watermark information extracted from the suspicious image with the watermark information of the target watermark image for consistency, thereby determining the source of the suspicious image and achieving effective detection and protection of copyright files.

[0107] It should be understood that the above is only an example and does not constitute any limitation on the technical solution of the present invention. In specific applications, technicians in this field can make settings as needed, and the present invention does not limit this.

[0108] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present invention. In practical applications, technicians in this field can select part or all of them according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0109] In addition, for technical details not described in detail in this embodiment, reference can be made to the copyright protection method based on the black box watermark model provided in any embodiment of the present invention, and will not be repeated here.

[0110] In addition, 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, article or system 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, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0111] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0112] 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 invention, 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 a read-only memory (ROM) / RAM, a magnetic 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, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0113] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A copyright protection method based on black box watermark model, It is characterized in that include: Input the image text information into the trained black box watermark model to prepare for image generation; Generate an image based on the prepared black box watermark model and output the target watermark image; Acquire a target suspicious image and extract a watermark from the target suspicious image to obtain watermark information of the suspicious image; A comparison detection is performed based on the suspicious image watermark information and the target watermark image to obtain an image comparison detection result.

2. The copyright protection method based on the black box watermark model according to claim 1, It is characterized in that Image generation is performed based on the prepared black box watermark model, and the target watermark image is output, which includes: Generate a process image based on the prepared black box watermark model and the image text information to obtain a target process image; The target process image is watermarked to obtain a target watermark image.

3. The copyright protection method based on the black box watermark model according to claim 2, It is characterized in that Embedding a watermark into the target process image to obtain a target watermark image specifically includes: Generate a watermark based on the trained black box watermark model to obtain a first watermark to be embedded; Performing embedding preprocessing on the first watermark to be embedded to obtain a second watermark to be embedded; The target process image is watermarked according to the second watermark to be embedded to obtain a target watermark image.

4. The copyright protection method based on the black box watermark model according to claim 3, It is characterized in that Performing embedding preprocessing on the first watermark to be embedded to obtain a second watermark to be embedded specifically includes: Generate a feature matrix based on the first watermark to be embedded to obtain a first feature matrix; Acquire watermark embedding parameters based on the trained black box watermark model; Watermark embedding matching is performed on the first feature matrix according to the watermark embedding parameters to obtain the second watermark to be embedded.

5. The copyright protection method based on the black box watermark model according to claim 1, It is characterized in that Acquiring a target suspicious image and extracting a watermark from the target suspicious image to obtain watermark information of the suspicious image specifically includes: Acquire a suspicious target image; Performing dimension reshaping processing based on the suspicious image to obtain a processed suspicious image; The processed suspicious image is input into a watermark extraction network to extract the watermark, and the watermark information of the suspicious image is obtained.

6. The copyright protection method based on the black box watermark model according to any one of claims 1 to 5, It is characterized in that Based on the watermark information of the suspicious image and the target watermark image, a comparison detection is performed to obtain an image comparison detection result, which specifically includes: Analyze the target watermark image to obtain target watermark information; The target watermark information is compared with the suspicious image watermark information for consistency to obtain the image comparison detection result.

7. The copyright protection method based on the black box watermark model according to claim 6, It is characterized in that The watermark extraction process of the trained black box watermark model during the model training phase specifically includes: Get training set image data; Inputting the training set image data into an image attack network to perform image attack simulation, thereby obtaining the training set image data after attack simulation; Watermark extraction is performed based on the training set image data after the attack simulation to obtain training set image watermark information.

8. A copyright protection device based on a black box watermark model, It is characterized in that The copyright protection device based on the black box watermark model includes: Information input module: inputs image text information into the trained black box watermark model to prepare for image generation; Image generation module: generates images based on the prepared black box watermark model and outputs the target watermark image; Watermark extraction module: obtains a target suspicious image and extracts watermarks from the target suspicious image to obtain watermark information of the suspicious image; Image comparison module: performs image comparison based on the watermark information of the suspicious image and the target watermark image to obtain an image comparison detection result.

9. A copyright protection device based on a black box watermark model, It is characterized in that The copyright protection device based on the black box watermark model includes: a memory, a processor, and a copyright protection program based on the black box watermark model stored in the memory and executable on the processor, wherein the copyright protection program based on the black box watermark model is configured to implement the copyright protection method based on the black box watermark model according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, It is characterized in that When the computer program is executed by a processor, the steps in the copyright protection method based on the black box watermark model described in any one of claims 1 to 7 can be implemented.

Citation Information

Patent Citations

  • Image processing model and property right protection method of image processing algorithm

    CN111311472A

  • Deep learning model protection method based on robust watermark

    CN114329365A

  • Watermark detection method and device, model training method and device and electronic equipment

    CN114596188A

  • Traceable deep learning model black box watermark scheme

    CN115482139A

  • Poison image generation method and black box classification model fingerprint watermarking method

    CN117172999A

Cited By

  • Image generation model training method, image generation method, and image detection method

    CN119863352A