Digital content inspection methods, equipment, storage media and program products

Abstract

This application provides a digital content detection method, device, storage medium, and program product. The method includes, in response to a detection task, acquiring target digital content and reference digital content to be processed; inputting the target digital content into a preset feature extraction model, causing the feature extraction model to output a first data feature of the target digital content, wherein the feature extraction model is trained on a first sample dataset, the first sample dataset including at least one pair of sample digital content, and sample digital content pairs with the same sample prompt words and the same generation model are labeled as having the same source tag; and determining whether the target digital content and the reference digital content are from the same source based on the first data feature and the reference digital content. This application effectively verifies whether the target digital content and the reference digital content are from the same source, improves the accuracy of digital content tracing and matching detection, and provides reliable data support for subsequent digital content ownership protection.

Description

Technical Field

[0001] This application relates to the field of information processing technology, and in particular to a digital content detection method, device, storage medium, and program product. Background Technology

[0002] With the continuous advancement of Artificial Intelligence (AI) technology, AI-generated content (AIGC) has become a revolutionary force in digital content creation. In content question answering, painting, photography, video production, and other media fields, AIGC technology has demonstrated its powerful automation and innovation capabilities, providing creators with unprecedented tools and possibilities. For example, in the field of painting, through AI models, users only need to input text descriptions to create images with aesthetic appeal and creativity. Similarly, on commercial platforms, users can more directly transform text into visual works, driving the development of digital art.

[0003] However, with the widespread application of AIGC technology, especially in the arts, the issues of originality and intellectual property ownership of AI-generated content have gradually surfaced, sparking significant controversies regarding creative property rights and copyright. While AI can generate compelling visual works, the generation process often lacks transparency, and the ownership of these works remains unclear. Therefore, protecting copyright ownership and ensuring the authenticity and credibility of content amidst the booming development of AIGC has become a critical issue that urgently needs to be addressed.

[0004] Currently, some progress has been made in detecting AI-generated content, and many solutions can detect whether digital content was generated by AI. However, most of these solutions only focus on detection and lack the ability to trace the source of digital content generation, making it difficult to effectively verify the originality and copyright information of digital content. Summary of the Invention

[0005] The main objective of this application is to provide a digital content detection method, device, storage medium, and program product, which effectively verifies whether target digital content and reference digital content are from the same source, improves the accuracy of digital content tracing and matching detection, and provides reliable data support for subsequent digital content ownership protection.

[0006] In a first aspect, embodiments of this application provide a digital content detection method, comprising: in response to a detection task, acquiring target digital content to be processed and reference digital content; inputting the target digital content into a preset feature extraction model, so that the feature extraction model outputs a first data feature of the target digital content, wherein the feature extraction model is trained on a first sample dataset, the first sample dataset including at least one sample digital content pair, and sample digital content pairs having the same sample prompt words and the same generation model are labeled as homologous; determining whether the target digital content and the reference digital content are homologous based on the first data feature and the reference digital content.

[0007] In one embodiment, before inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content, the method further includes: determining whether there is a target prompt word used to generate the target digital content in the input information of the detection task; if there is no target prompt word in the input information, performing the step of inputting the target digital content into the preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content.

[0008] In one embodiment, the method further includes: if the target prompt word exists in the input information, determining whether the target prompt word is the same as the reference prompt word corresponding to the reference digital content; if the target prompt word is the same as the reference prompt word, inputting the target digital content and the reference digital content into a preset classification model respectively, so that the preset classification model outputs first source information of the target digital content and second source information of the reference digital content; and determining whether the target digital content and the reference digital content are from the same source based on the first source information and the second source information.

[0009] In one embodiment, before inputting the target digital content and the reference digital content into a preset classification model respectively, so that the preset classification model outputs the first source information of the target digital content and the second source information of the reference digital content, the method further includes: obtaining a second sample dataset, the second sample dataset including multiple sample digital contents generated by different generation models, each sample digital content being labeled with information of the corresponding generation model; and training a preset neural network using a cross-entropy loss function based on the second sample dataset to obtain the preset classification model.

[0010] In one embodiment, the reference digital content is a comparison object specified by the user; determining whether the target digital content and the reference digital content are from the same source based on the first data feature and the reference digital content includes: obtaining a second data feature of the reference digital content; calculating the feature similarity between the first data feature and the second data feature; if the feature similarity is greater than a first preset threshold, determining that the target digital content and the reference digital content are from the same source; if the feature similarity is less than or equal to the first preset threshold, determining that the target digital content and the reference digital content are from different sources.

[0011] In one embodiment, there are multiple reference digital contents, and the second data feature and corresponding metadata of each reference digital content are stored in a preset database; determining whether the target digital content and the reference digital content are from the same source based on the first data feature and the reference digital content includes: calculating the feature similarity between the first data feature and the second data feature of each reference digital content in the preset database; and determining whether the target digital content and the reference digital content are from the same source based on the feature similarity and the metadata corresponding to each reference content.

[0012] In one embodiment, determining whether the target digital content and the reference digital content are from the same source based on the feature similarity and the metadata corresponding to each reference content includes: determining a set of similar digital content from a plurality of reference digital content whose feature similarity to the target digital content is greater than a second preset threshold based on the feature similarity; comparing the metadata of the target digital content with the metadata of each digital content in the set of similar digital content; if there is a specific digital content in the set of similar digital content that has the same metadata as the target digital content, determining that the target digital content and the specific digital content are from the same source; and / or, outputting the specific digital content; if there is no specific digital content in the set of similar digital content that has the same metadata as the target digital content, determining that the target digital content and the reference digital content are not from the same source.

[0013] In one embodiment, before inputting the target digital content into a preset feature extraction model to output the first data feature of the target digital content, the method further includes: acquiring multiple different sample prompt words; acquiring multiple different content generation models; inputting each of the multiple different sample prompt words into each of the content generation models, so that each of the content generation models generates corresponding sample digital content based on the multiple different sample prompt words, thereby obtaining an original sample dataset; pairing the sample digital content in the original sample dataset in pairs to form multiple sample digital content pairs, and labeling sample digital content pairs with the same sample prompt words and the same generation model as source tags, thereby forming the final first sample dataset.

[0014] In one embodiment, before inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content, the method further includes: fine-tuning and training a preset feature extractor using a contrastive loss function based on the first sample dataset to obtain the feature extraction model.

[0015] In one embodiment, the first sample dataset is a sample image set; before inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content, the method further includes: fine-tuning and training a preset feature extractor based on the sample image set, wherein the feature extractor includes an image feature extractor and a text feature extractor; during the training process, the image feature extractor and the text feature extractor are fine-tuned and trained using an inter-class contrast loss function and an image-text contrast loss function, respectively, to obtain the feature extraction model.

[0016] Secondly, embodiments of this application provide a method for detecting a product image, comprising: in response to a detection task, acquiring a target product image to be processed and a reference image, wherein the target product image includes product information; inputting the target product image into a preset feature extraction model, so that the feature extraction model outputs a first image feature of the target product image, wherein the feature extraction model is trained through a first sample image set, the first sample image set including at least one sample image pair, and sample image pairs having the same sample prompt words and the same generation model are labeled as homologous; determining whether the target product image and the reference image are homologous based on the first image feature and the reference image.

[0017] Thirdly, embodiments of this application provide a digital content detection device, comprising:

[0018] The acquisition module is used to acquire the target digital content to be processed and the reference digital content in response to the detection task;

[0019] An extraction module is used to input the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content. The feature extraction model is trained through a first sample dataset, which includes at least one sample digital content pair. Sample digital content pairs with the same sample prompt words and the same generation model are labeled as homologous tags.

[0020] The determining module is used to determine whether the target digital content and the reference digital content are from the same source based on the first data feature and the reference digital content.

[0021] In one embodiment, the system further includes: a first judgment module, configured to determine whether a target prompt word used to generate the target digital content exists in the input information of the detection task before the target digital content is input into a preset feature extraction model so that the feature extraction model outputs a first data feature of the target digital content; if the target prompt word does not exist in the input information, the system executes the step of inputting the target digital content into the preset feature extraction model so that the feature extraction model outputs a first data feature of the target digital content.

[0022] In one embodiment, the system further includes: a second judgment module, configured to determine whether the target prompt word is the same as the reference prompt word corresponding to the reference digital content if the target prompt word exists in the input information; a classification module, configured to input the target digital content and the reference digital content into a preset classification model respectively if the target prompt word is the same as the reference prompt word, so that the preset classification model outputs first source information of the target digital content and second source information of the reference digital content; the classification module is further configured to determine whether the target digital content and the reference digital content are from the same source based on the first source information and the second source information.

[0023] In one embodiment, the system further includes: a first training module, configured to: obtain a second sample dataset before the target digital content and the reference digital content are respectively input into a preset classification model so that the preset classification model outputs first source information of the target digital content and second source information of the reference digital content; the second sample dataset includes multiple sample digital contents generated by different generation models, each sample digital content being labeled with information of the corresponding generation model; and train a preset neural network using a cross-entropy loss function based on the second sample dataset to obtain the preset classification model.

[0024] In one embodiment, the reference digital content is a comparison object specified by the user; the determining module is used to obtain a second data feature of the reference digital content; calculate the feature similarity between the first data feature and the second data feature; if the feature similarity is greater than a first preset threshold, determine that the target digital content and the reference digital content are from the same source; if the feature similarity is less than or equal to the first preset threshold, determine that the target digital content and the reference digital content are from different sources.

[0025] In one embodiment, there are multiple reference digital contents, and the second data feature and corresponding metadata of each reference digital content are stored in a preset database; the determining module is used to calculate the feature similarity between the first data feature and the second data feature of each reference digital content in the preset database; and determine whether the target digital content and the reference digital content are from the same source based on the feature similarity and the metadata corresponding to each reference content.

[0026] In one embodiment, the determining module is specifically configured to: determine a set of similar digital content from a plurality of reference digital content based on the feature similarity, wherein the feature similarity to the target digital content is greater than a second preset threshold; compare the metadata of the target digital content with the metadata of each digital content in the set of similar digital content; if there is a specific digital content in the set of similar digital content that has the same metadata as the target digital content, determine that the target digital content and the specific digital content are from the same source; and / or, output the specific digital content; if there is no specific digital content in the set of similar digital content that has the same metadata as the target digital content, determine that the target digital content and the reference digital content are not from the same source.

[0027] In one embodiment, the system further includes: a data construction module, configured to: acquire multiple different sample prompt words; acquire multiple different content generation models; input the multiple different sample prompt words one by one into each of the content generation models, so that each of the content generation models generates corresponding sample digital content based on the multiple different sample prompt words, thereby obtaining an original sample dataset; pair the sample digital content in the original sample dataset in pairs to form multiple sample digital content pairs, and label the sample digital content pairs with the same sample prompt words and the same generation model as source tags, thereby forming the final first sample dataset.

[0028] In one embodiment, the system further includes a second training module, configured to, before inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs first data features of the target digital content, fine-tune and train a preset feature extractor using a contrastive loss function to obtain the feature extraction model.

[0029] In one embodiment, the first sample dataset is a sample image set; it further includes: a third training module, used to fine-tune and train a preset feature extractor based on the sample image set before the target digital content is input into a preset feature extraction model so that the feature extraction model outputs the first data features of the target digital content, wherein the feature extractor includes an image feature extractor and a text feature extractor; during the training process, the image feature extractor and the text feature extractor are fine-tuned and trained using an inter-class contrast loss function and an image-text contrast loss function, respectively, to obtain the feature extraction model.

[0030] Fourthly, embodiments of this application provide an electronic device, including:

[0031] At least one processor; and

[0032] A memory that is communicatively connected to the at least one processor;

[0033] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, cause the electronic device to perform the method described in any of the above aspects.

[0034] Fifthly, embodiments of this application provide a cloud device, including:

[0035] At least one processor; and

[0036] A memory that is communicatively connected to the at least one processor;

[0037] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, cause the cloud device to perform the method described in any of the above aspects.

[0038] Sixthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method described in any of the above aspects.

[0039] In a seventh aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the methods described in any of the above aspects.

[0040] The digital content detection method, device, storage medium, and program product provided in this application automatically extract representative first data features from target digital content using a pre-trained feature extraction model. These first data features characterize the source information of the target digital content. Since the feature extraction model is trained on a first sample dataset, where sample digital content pairs with the same prompt words and generation models are labeled as source-same, and both the prompt words and generation models characterize the source information of the digital content, the trained feature extraction model can effectively capture the first data features that accurately characterize the source information in the target digital content. This effectively verifies whether the target digital content and reference digital content share the same source, improving the accuracy of digital content tracing and matching detection, and providing reliable data support for subsequent digital content ownership protection. Attached Figure Description

[0041] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are some embodiments of the invention, and that those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0042] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0043] Figure 2 This is a schematic diagram illustrating an application scenario of a digital content detection system provided in an embodiment of this application;

[0044] Figure 3 This is a flowchart illustrating the digital content detection method according to an embodiment of this application;

[0045] Figure 4 A schematic diagram illustrating a dataset construction process provided in an embodiment of this application;

[0046] Figure 5 This is a schematic diagram of the training process of a feature extraction model provided in an embodiment of this application;

[0047] Figure 6 This is a schematic diagram of the training process of a feature extraction model provided in an embodiment of this application;

[0048] Figure 7 A schematic diagram illustrating the training process of a preset classification model provided in an embodiment of this application;

[0049] Figure 8 A flowchart illustrating a digital content detection method provided in an embodiment of this application;

[0050] Figure 9 A flowchart illustrating a digital content detection method provided in an embodiment of this application;

[0051] Figure 10 A schematic flowchart illustrating a product image detection method provided in an embodiment of this application;

[0052] Figure 11 This is a schematic diagram of the structure of a digital content detection device provided in an embodiment of this application;

[0053] Figure 12 This is a schematic diagram of the structure of a cloud device provided in an embodiment of this application.

[0054] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0055] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0056] In this article, the term "and / or" is used to describe the relationship between related objects. Specifically, it means that there can be three kinds of relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, or B exists alone.

[0057] 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, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0058] To clearly describe the technical solutions of the embodiments of this application, the terms involved in this application are first defined as follows:

[0059] AI: Artificial Intelligence.

[0060] AIGC: Artificial Intelligence Generated Content. This includes, but is not limited to, text, images, audio, and video.

[0061] AI editing: Using artificial intelligence technology to edit content, such as style transfer of images, application of filters, object recognition and removal, background replacement, etc.

[0062] SD: Stable Diffusion, is a deep learning model used to generate images.

[0063] GAN: Generative Adversarial Networks.

[0064] Prompt: A cue word for a large model, it's an input used to guide the model in generating 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 might be a textual description of the desired image.

[0065] Diffusion Model: A diffusion model is a type of data generation model, such as the SD model. Based on the principle of random diffusion, it gradually eliminates noise through optimization techniques like gradient descent to generate highly realistic images. This model is particularly suitable for creating new visual content based on given text descriptions and has been applied in various fields such as artistic creation and game design.

[0066] Artificial intelligence-generated images (AI-Generated Images) refer to images automatically generated using artificial intelligence techniques such as diffusion models and generative adversarial networks (GANs). These images can be landscapes, objects, people, etc., and they usually have a high degree of realism, making it difficult for the average person to distinguish them from fakes with the naked eye.

[0067] AI-Generated Image Provenance Detection aims to trace the origin of AI-generated images by analyzing and comparing their features to determine which specific AI generation model or platform was used. This is crucial for applications such as copyright protection, liability tracking, and algorithm performance evaluation, and is essential for ensuring the transparency and credibility of AI-generated content.

[0068] CLIP (Contrastive Language–Image Pretraining) is an innovative method proposed by OpenAI that uses contrastive learning to pretrain from large-scale image-text pairs. CLIP uses two independent encoders, one for images and one for text, mapping them to the same embedding space. It maximizes the similarity between correctly paired images and text using a contrastive loss function while minimizing the similarity between unrelated pairs, thereby training a visual model with strong transfer capabilities.

[0069] ViT: Vision Transformer, is a model that applies the Transformer architecture to computer vision tasks.

[0070] ConvNext: refers to a new type of convolutional neural network architecture.

[0071] BLIP (Bootstrapping Language-Image Pre-training) is a multimodal model designed to improve model performance on visual and language tasks by pre-training with a combination of language and image data. BLIP-2 is an improvement and extension of the original BLIP model, aiming to further enhance the performance of multimodal learning. Similar to its predecessor, BLIP-2 enhances model performance on visual and language tasks by combining language and image data for pre-training.

[0072] ResNet: Residual Network.

[0073] DenseNet: Densely Connected Convolutional Networks.

[0074] MobileNet: A lightweight convolutional neural network architecture.

[0075] EfficientNet: A highly efficient neural network.

[0076] HRNet: High-Resolution Network.

[0077] Swin Transformer: Shifted Window Transformer is a hierarchical visual Transformer model that achieves a balance between computational efficiency and performance by computing self-attention within local windows and moving between windows.

[0078] MobileViT: Mobile Vision Transformer, is a high-efficiency vision model specifically designed for mobile devices.

[0079] UGC: User-Generated Content.

[0080] The digital content detection method of this application embodiment can be applied to any field involving watermarking.

[0081] With the continuous advancement of Artificial Intelligence (AI) technology, AI-generated content (AIGC) has become a revolutionary force in digital content creation. In content question answering, painting, photography, video production, and other media fields, AIGC technology has demonstrated its powerful automation and innovation capabilities, providing creators with unprecedented tools and possibilities. For example, in the field of painting, through AI models, users only need to input text descriptions to create images with aesthetic appeal and creativity. Similarly, on commercial platforms, users can more directly transform text into visual works, driving the development of digital art.

[0082] However, with the widespread application of AIGC technology, especially in the arts, the issues of originality and intellectual property ownership of AI-generated content have gradually surfaced, sparking significant controversies regarding creative property rights and copyright. While AI can generate appealing visual works, the generation process often lacks transparency, and the ownership of these works remains unclear. Furthermore, to generate high-quality images, AI models require vast amounts of data for training. If the training data includes copyrighted works, the generated images may infringe on the original authors' copyrights, leading to various disputes.

[0083] Meanwhile, with the rise of the open-source community for AI generative models, an increasing number of practitioners are releasing models derived from the basic diffusion model series. These diverse versions of diffusion models bring high generalization requirements to the retrieval and matching of generated images in real-world scenarios. Since generated images in real-world scenarios are often transmitted through distorted channels such as social media platforms, they may undergo post-processing such as scaling, cropping, and secondary compression during transmission. Therefore, in real-world detection scenarios, detectors must be able to resist these common editing processes.

[0084] Taking e-commerce platforms as an example, merchants can use AI models to generate different product images for new product listings. If others see that a product is selling well, they may manipulate the product image by scaling, cropping, or recompressing it to try to evade copyright infringement, seriously harming the information security of the original user.

[0085] While some progress has been made in AI-generated content detection, enabling the identification of whether digital content was generated by AI, most of these solutions remain at the detection level, lacking the ability to trace the methods used to generate the digital content. This makes it difficult to effectively verify the originality and copyright information of the digital content.

[0086] To address at least one of the aforementioned problems, embodiments of this application provide a digital content detection scheme. By using a pre-trained feature extraction model, a representative first data feature is automatically extracted from the target digital content. This first data feature characterizes the source information of the target digital content. Since the feature extraction model is trained on a first sample dataset, where sample digital content pairs with the same prompt words and the same generation model are labeled as having the same source, the trained feature extraction model can effectively capture the first data feature characterizing the source information in the target digital content. This effectively verifies whether the target digital content and reference digital content share the same source, improving the accuracy of digital content tracing and matching detection, and providing reliable data support for subsequent digital content ownership protection.

[0087] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Where there is no conflict between the embodiments, the following embodiments and features can be combined with each other. Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0088] like Figure 1 As shown, this embodiment provides an electronic device 1, including: at least one processor 11 and a memory 12. Figure 1 Taking a processor 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. The instructions are executed by the processor 11 to enable the electronic device 1 to perform all or part of the process of the method in the following embodiments, so as to effectively verify whether the target digital content and the reference digital content are from the same source, improve the accuracy of digital content tracing and matching detection, and provide reliable data support for subsequent digital content ownership protection.

[0089] In one embodiment, the electronic device 1 may be a mobile phone, tablet computer, laptop computer, desktop computer, or a large computing system composed of multiple computers.

[0090] Figure 2 This is a schematic diagram illustrating an application scenario 200 of a digital content detection system provided in an embodiment of this application. For example... Figure 2 As shown, the system includes: a server 210 and a terminal 220, wherein:

[0091] Server 210 could be a data platform providing digital content inspection services, such as a digital information security service platform for e-commerce. In a real-world scenario, an e-commerce platform might have multiple servers 210. Figure 2 Taking a single server (210) as an example.

[0092] Terminal 220 can be a computer, mobile phone, tablet, or other device used by the user to log in to the digital information security service platform. There can also be multiple terminals 220. Figure 2 The following example uses two terminals, 220, for illustration.

[0093] Terminal 220 and server 210 can transmit information via the Internet, enabling terminal 220 to access data on server 210. Both terminal 220 and / or server 210 can be implemented by electronic device 1.

[0094] The digital content detection scheme of this application embodiment can be deployed on server 210, on terminal 220, or partially on server 210 and partially on terminal 220. The appropriate deployment method can be chosen based on actual needs in a real-world scenario, and this embodiment does not impose any limitations.

[0095] When the digital content detection solution is deployed entirely or partially on server 210, an interface can be opened to terminal 220 to provide algorithm support to terminal 220.

[0096] The method provided in this application embodiment can be implemented by electronic device 1 executing corresponding software code, and is achieved through data interaction with a server. Electronic device 1 can be a local terminal device. When the method runs on a server, it can be implemented and executed based on a cloud interaction system, which includes a server and client devices.

[0097] In one possible implementation, the method provided in this application provides a graphical user interface through a terminal device, wherein the terminal device may be the aforementioned local terminal device or a client device in the aforementioned cloud interaction system.

[0098] Please refer to Figure 3 This is a digital content detection method according to an embodiment of this application. The method can be performed by... Figure 1 The electronic device 1 shown is used to perform this action and can be applied to... Figure 2In the digital content detection application scenario shown, this method effectively verifies whether target digital content and reference digital content share the same origin, improving the accuracy of digital content tracing and matching detection, and providing reliable data support for subsequent digital content ownership protection. This embodiment uses terminal 220 as the execution end as an example, and the method includes the following steps:

[0099] Step 301: In response to the detection task, obtain the target digital content to be processed and the reference digital content.

[0100] In this step, the detection task can be triggered by the user or automatically by the system. The detection task can specify the target digital content to be processed. Target digital content refers to the digital content that needs to be traced and matched for detection; this digital content includes, but is not limited to, images, text, audio, and video. Reference digital content refers to standard digital content used to assist in detection; reference digital content includes, but is not limited to, images, text, audio, and video generated based on AI generation models. Reference digital content can be specified by the user. For example, if a user believes their image has been infringed upon, they can use a user-specified potentially infringed original image as reference digital content, and determine the potentially infringing image provided by the user as the target digital content to be tested. Optionally, reference digital content can also be content specified by the system, such as certified original images pre-stored in the system database.

[0101] Step 302: Input the target digital content into the preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content. The feature extraction model is trained through a first sample dataset. The first sample dataset includes at least one sample digital content pair. Sample digital content pairs with the same sample prompt words and the same generation model are labeled as homologous tags.

[0102] In this embodiment, the feature extraction model is trained based on a first sample dataset. In this first sample dataset, sample digital content pairs with the same sample prompt words and the same generation model are labeled as source-related tags. Both the prompt words and the generation model can represent the source information of the digital content. Therefore, the feature extraction model trained on this first sample dataset can learn the feature representations related to the prompt words and generation models in the digital content, so that the feature extraction model can effectively capture the first data features representing the source information in the target digital content during the inference stage, and the first data features can accurately represent the source information of the target digital content.

[0103] In one embodiment, prior to step 302, a dataset construction process is further included, comprising: acquiring multiple different sample prompt words; acquiring multiple different content generation models; inputting each of the multiple different sample prompt words into each content generation model, so that each content generation model generates corresponding sample digital content based on the multiple different sample prompt words, thereby obtaining an original sample dataset; pairing the sample digital content in the original sample dataset into pairs to form multiple sample digital content pairs, and labeling sample digital content pairs with the same sample prompt words and the same generation model as source-related labels, thereby forming the final first sample dataset.

[0104] In this embodiment, diverse sample digital content is generated by acquiring multiple different sample prompts and content generation models. This diversity helps train the feature extraction model, enabling it to better adapt to different types and styles of original digital content and improving the model's generalization ability. By pairing the generated original sample digital content in pairs and labeling sample digital content pairs with the same prompts and generation models as homologous labels, a structured sample dataset is constructed, which can provide more meaningful contextual information for subsequent feature extraction. By labeling homologous labels, the model can more effectively identify and distinguish the data features of homologous and heterologous digital content, improving the accuracy of feature extraction.

[0105] like Figure 4 The diagram shown illustrates a process for constructing a dataset according to this application. Taking image data as an example, the process may include the following steps:

[0106] Step 1: Collect a massive amount of text prompts (e.g., more than 100,000), and use a filtering algorithm to remove duplicate text prompts, ensuring that each sample text prompt is unique and non-repeating. Figure 4 The sample prompt in the example is "A knight in a spacesuit rides a Ferghana horse and strolls on the moon."

[0107] Step 2: Build an automated AI-generated image system that includes multiple open-source AI generation models (e.g., SDv1.4, SDv2.1, extended versions SDXL and SD3, and FLUX (a generation model based on a diffusion model)). Assume that the multiple content generation models are numbered, including generation model 1, generation model 2, ..., generation model n, where n is an integer.

[0108] Step 3: Input the sample prompts from Step 1 one by one into the automated AI image generation system in Step 2 to generate batches of AI-generated image pairs. Each input sample prompt (assuming all other input conditions remain the same) should generate at least two AI-generated images in each AI generation model. For example... Figure 4In the process, the sample prompt "A knight in a spacesuit riding a Ferghana horse strolls on the moon" is input into generation model 1, generation model 2, ..., generation model n, respectively. Each generation model generates two images. Figure 4 The example shows image pairs generated by generative model 1 and generative model 2. In this way, a raw sample dataset composed of multiple AI-generated images can be obtained.

[0109] Optionally, the first sample dataset is used to train the feature extraction model. Before training the model, training data labels are first defined. Image pairs generated by the same AI generation model using the same text prompt words (and other input conditions are also the same) in the original dataset are considered to be of the same category and labeled with the same source label. Otherwise, they are considered to be of different categories (e.g., two images generated by the same AI generation model using different text prompt words or by different AI generation models using the same text prompt words are considered to be of different categories). Therefore, assuming that the sample dataset is constructed using i different text prompt words and n AI generation models, the entire first sample dataset can contain c = ixn different category labels, and each category can contain at least 2 images, where i is a positive integer and n is a positive integer.

[0110] Optionally, to adapt to different training tasks, the original sample dataset can be processed accordingly. For example, to enhance the diversity of the sample dataset, random data augmentation schemes can be used to enhance the original sample dataset. Taking the sample image set as an example, specific augmentation schemes include, but are not limited to, random cropping, maximum long side size, compression quality, horizontal flipping, Gaussian noise, Gaussian blur, random 90-degree rotation, edge completion, completion to a fixed size, brightness and contrast transformation, grid dropout, grayscale conversion, translation and scale rotation, color jittering, random cropping and scaling, center cropping, random brightness adjustment, color jittering, color space conversion, random occlusion, salt and pepper noise, affine transformation, etc.

[0111] In one embodiment, before step 302, the method may further include: fine-tuning a preset feature extractor using a contrastive loss function based on the first sample dataset to obtain a feature extraction model.

[0112] In this embodiment, before inputting the target digital content into the preset feature extraction model, the preset feature extractor can be fine-tuned and trained using a contrastive loss function based on a first sample dataset. This effectively improves the accuracy and robustness of the feature extraction model. The contrastive loss function enhances the model's ability to distinguish between different categories of features, enabling the feature extraction model to extract representative first data features more accurately when processing target digital content. This not only improves the generalization ability of the feature extraction model but also reduces the risk of overfitting, thus better adapting to diverse digital content inputs in practical applications and improving the overall system performance and reliability.

[0113] Optionally, the preset feature extractor can be a feature extractor based on the CLIP pre-trained model. CLIP uses two independent encoders, one for images and one for text, mapping them to the same embedding space. It maximizes the similarity between correctly paired images and text through a contrastive loss function, while minimizing the similarity of unrelated pairs, thereby training a visual model with strong transfer capabilities. In real-world scenarios, the original pre-trained CLIP image feature extractor can be used to extract feature vectors for AI-generated images. However, this feature vector can only distinguish whether two AI-generated images use the same generation prompts, but it cannot distinguish between two AI-generated images generated using the same generation prompts in two different generation models. Therefore, it cannot accurately trace the origin of AI-generated images. Therefore, in this embodiment, in the first sample dataset, sample digital content pairs with the same sample prompt words and the same generation model are labeled as homologous. Based on the first sample dataset as training data, the CLIP image feature extractor is fine-tuned using a contrastive loss function. Then, the fine-tuned CLIP image feature extractor is used as a feature extraction model, so that the feature extraction model can learn the feature representations related to the prompt words and generation model in the digital content. Thus, the image feature vector extracted by the feature extraction model can not only distinguish whether the two AI-generated images use the same generation prompt words, but also distinguish two AI-generated images generated in two different generation models using the same generation prompt words, thereby completing the accurate source tracing of AI-generated images.

[0114] Optionally, the backbone network of the CLIP pre-trained model involved in this application embodiment can be a ViT structure pre-trained model or a ConvNext structure pre-trained model. Alternatively, image feature extractors and text feature extractors from BLIP and BLIP2 (Graphic-to-Text Model, the opposite task of CLIP) can also be used.

[0115] like Figure 5 The diagram shown illustrates the training process of a feature extraction model provided in this application embodiment. Taking a sample image dataset as an example, assume that the first sample dataset includes the following content with category labels:

[0116] Category 1: Image pairs generated by cue word 1 in model 1.

[0117] Category 2: Image pairs generated by cue word 1 in model 2.

[0118] Category 3: Image pairs generated by cue word 2 in model 1.

[0119] ...

[0120] Category c: Image pairs generated by model n for cue word i.

[0121] The CLIP image feature extractor was fine-tuned and trained using the first sample dataset mentioned above. During the training process, the CLIP image feature extractor extracted the feature vector for each sample image, for example:

[0122] The feature vectors of the sample image pairs in category 1 are f1_1 and f1_2, respectively.

[0123] The feature vectors of the sample image pairs in category 2 are f2_1 and f2_2, respectively.

[0124] The feature vectors of the sample image pairs in category 3 are f3_1 and f3_2, respectively.

[0125] ...

[0126] The feature vectors of the sample image pairs of category c are fc_1 and fc_2, respectively.

[0127] The CLIP image feature extractor is fine-tuned using a contrastive loss function to obtain the final feature extraction model.

[0128] In optional embodiments, the classification loss function used for training the feature extraction model may also be: cross-entropy loss and Focal loss (a loss function to solve class imbalance), etc. The embodiments of this application do not limit the type of loss function.

[0129] In one embodiment, the first sample dataset is a set of sample images. Before step 3602, the method may further include: fine-tuning and training a preset feature extractor based on the sample image set, wherein the feature extractor includes an image feature extractor and a text feature extractor. During training, the image feature extractor and the text feature extractor are fine-tuned and trained using an inter-class contrast loss function and an image-text contrast loss function, respectively, to obtain a feature extraction model.

[0130] In this embodiment, multimodal contrastive learning combining image and text data provides a more effective training method for tracing and matching the source of AI-generated images. The feature extractor consists of an image feature extractor and a text feature extractor, which are fine-tuned during training using inter-class contrastive loss and image-text contrastive loss functions, respectively. This allows the image feature extractor to better distinguish image features from different sources, improving the accuracy of image feature extraction. Simultaneously, the image-text contrastive loss function enhances the correlation between the text and image feature extractors, enabling the model to more effectively match and correlate visual and textual information. This dual-contrast loss training strategy not only improves the model's multimodal understanding ability but also enhances its generalization ability across different tasks, thus providing more reliable and accurate feature extraction results in practical applications. By fine-tuning the feature extractor using a sample image set, the performance and adaptability of the feature extraction model are significantly improved.

[0131] Alternatively, the feature extractor can be a CLIP image feature extractor and a CLIP text feature extractor.

[0132] like Figure 6 The diagram shown illustrates the training process of a feature extraction model provided in this application. Taking a sample image dataset as an example, the definition of sample labels can be referred to the description in the foregoing embodiments. Assume the first sample dataset includes the following content with category labels:

[0133] Category 1: Image pairs generated by cue word 1 in model 1.

[0134] Category 2: Image pairs generated by cue word 1 in model 2.

[0135] Category 3: Image pairs generated by cue word 2 in model 1.

[0136] ...

[0137] Category c: Image pairs generated by model n for cue word i.

[0138] The images in the above sample image set are applied to the CLIP image feature extractor, and the CLIP image feature extractor is used to extract the image features f of the sample images. The text of the sample prompt words corresponding to the above sample images is applied to the CLIP text feature extractor, and the CLIP text feature extractor is used to extract the text features v of the sample prompt words, resulting in the following feature vectors:

[0139] In the sample image pair of category 1, the feature vectors of the first sample image are f1_1 and the feature vector of the corresponding sample prompt word is v1_1, and the feature vectors of the second sample image are f1_2 and the feature vector of the corresponding sample prompt word is v1_2.

[0140] In the sample image pair of category 2, the feature vectors of the first sample image are f2_1 and the feature vector of the corresponding sample prompt word is v2_1, and the feature vectors of the second sample image are f2_2 and the feature vector of the corresponding sample prompt word is v2_2.

[0141] In the sample image pair of category 3, the feature vectors of the first sample image are f3_1 and the feature vector of the corresponding sample prompt word is v3_1, and the feature vectors of the second sample image are f3_2 and the feature vector of the corresponding sample prompt word is v3_2.

[0142] ...

[0143] In the sample image pair of category c, the feature vectors of the first sample image are fc_1 and the feature vector of the corresponding sample prompt word is vc_1, and the feature vectors of the second sample image are fc_2 and the feature vector of the corresponding sample prompt word is vc_2.

[0144] The CLIP image feature extractor and text feature extractor are fine-tuned using inter-class contrast loss function and image-text contrast loss function, respectively. The image-text contrast loss function ensures that the image features extracted by the fine-tuned image feature extractor from the target product image are aligned with the text features describing the content of the target product image, thereby achieving multimodal training and improving the performance of the final feature extraction model.

[0145] Alternatively, the inter-class contrastive loss function used for training the feature extraction model may also include: contrastive loss, triplet loss, circle loss (a loss function that learns through unified classification and sample pairs), InfoNCE loss, N-pair loss (information maximization unconditional estimation loss), and soft-nearest neighbor loss, etc., which are not limited in this application.

[0146] Step 303: Based on the first data feature and the reference digital content, determine whether the target digital content and the reference digital content are from the same source.

[0147] In this embodiment, the first data feature extracted in step 302 can accurately characterize the source information of the target digital content. Therefore, it can be used to effectively verify whether the target digital content and the reference digital content are from the same source, thereby improving the accuracy of digital content tracing and matching detection and providing reliable data support for the subsequent protection of digital content ownership.

[0148] In one embodiment, before step 302, the method further includes: determining whether there is a target prompt word in the input information of the detection task used to generate the target digital content. If there is no target prompt word in the input information, the method performs a step of inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content.

[0149] In this embodiment, the user can specify input information for the detection task, such as uploading a target product image and a reference image. Alternatively, the user can provide text prompts for the large model to generate the target product image or for the reference image. The text prompts for the target product image are the target prompts, and the text prompts for the reference image are the reference prompts. By determining whether target prompts exist in the input information before inputting the target digital content into the preset feature extraction model, if no target prompts are found, no prompt processing is needed, and the feature extraction step 302 can be directly executed. This avoids unnecessary prompt processing and simplifies the data processing flow. It ensures that prompt processing is only performed when necessary, reducing computational resource waste and accelerating feature extraction. Furthermore, this method reduces the risk of feature extraction errors due to missing prompts, improves the robustness and reliability of the model when handling diverse inputs, and ultimately enhances the overall system performance.

[0150] In one embodiment, the method further includes: if a target prompt word exists in the input information, determining whether the target prompt word is the same as a reference prompt word corresponding to the reference digital content. If the target prompt word is the same as the reference prompt word, the target digital content and the reference digital content are respectively input into a preset classification model, so that the preset classification model outputs first source information of the target digital content and second source information of the reference digital content. The method then determines whether the target digital content and the reference digital content share the same source based on the first source information and the second source information.

[0151] In this embodiment, when a target prompt word exists in the input information, it is further determined whether the target prompt word is the same as the reference prompt word corresponding to the reference digital content, thus enabling accurate identification and verification of the source of the digital content. When the target prompt word is the same as the reference prompt word, the target digital content and the reference digital content are further input into a preset classification model to obtain their corresponding source information. This source information includes, but is not limited to, information about the digital content's generation model, model version information, and the platform information to which the model belongs. This process effectively identifies the source information of the digital content, thereby determining whether the target digital content and the reference digital content share the same origin. This not only improves the ability to discriminate the source of digital content but also enhances the accuracy and reliability of the system when processing similar content. By using a classification model to analyze source information, it can better support content verification and traceability, reducing the possibility of misjudgment. Furthermore, this method can provide important support in application scenarios such as content review, copyright protection, and information traceability, improving the overall system's intelligence level and application value.

[0152] In one embodiment, before inputting the target digital content and reference digital content into a preset classification model to output first source information of the target digital content and second source information of the reference digital content, the method further includes: obtaining a second sample dataset, which includes multiple sample digital contents generated by different generation models, each sample digital content being labeled with information of the corresponding generation model. Based on the second sample dataset, a preset neural network is trained using a cross-entropy loss function to obtain the preset classification model.

[0153] In this embodiment, before inputting the target digital content and reference digital content into the preset classification model, a training process for the preset classification model is also included. The second sample dataset contains multiple sample digital contents generated by different generation models, and the aforementioned method can be used here. Figure 4 The corresponding implementation uses the original dataset to generate the second sample dataset. Taking the sample image set as an example, each sample image in the second sample dataset is labeled with the generative model information of its source. Using this labeled information, a pre-defined neural network is trained using the cross-entropy loss function to obtain an optimized classification model. This training method enables the classification model to learn and distinguish the features of different generative models more accurately, thereby more effectively outputting the source information of target digital content and reference digital content in practical applications, thus improving the accuracy and reliability of determining whether short digital content is from the same source. In addition, this method also improves the model's generalization ability, enabling it to adapt to diverse generative models and content types, thereby enhancing the practicality and value of the overall system in application scenarios such as content review, copyright protection, and information traceability.

[0154] like Figure 7The diagram shown is a schematic of the training process of a preset classification model provided in an embodiment of this application. Taking a sample image dataset as the training dataset as an example, it may include the following steps:

[0155] Step 1: Configure the training input images. Define the set of real images as category 0, the set of images generated using AI generation model 1 as category 1, the set of images generated using AI generation model 2 as category 2, and so on, with the set of images generated using AI generation model n defined as category n. Real images refer to images not generated by AI generation models; for example, real images can be images captured by a camera or images drawn by a user using drawing tools. Adding real images to the sample dataset allows the classification model to learn which images are AI-generated and which are not.

[0156] Step 2: Train the AI-generated image source classification model using the cross-entropy loss function. This classification model can be a pre-defined neural network structure, such as deep models like ResNet, DenseNet, MobileNet, EffcientNet, HRNet, ConvNext, ViT, Swin Transformer, and MobileVit. By constraining the cross-entropy loss between sample images (e.g., real images, images generated by Model 1, images generated by Model 2, ..., images generated by Model n) and their corresponding category predictions (e.g., real images correspond to prediction 0, images generated by Model 1 correspond to prediction 1, images generated by Model 2 correspond to prediction 2, ..., images generated by Model n correspond to prediction n), the trained classification model can accurately output the predicted category for each input image. This predicted category can characterize the source information of the input image, such as which generation model it originated from and the model's version information.

[0157] In one embodiment, step 303 may specifically include: acquiring a second data feature of the reference digital content; calculating the feature similarity between the first data feature and the second data feature; if the feature similarity is greater than a first preset threshold, determining that the target digital content and the reference digital content are from the same source; if the feature similarity is less than or equal to the first preset threshold, determining that the target digital content and the reference digital content are from different sources.

[0158] In this embodiment, the reference digital content is a comparison object specified by the user, such as a user-specified infringing reference image. The second data feature can characterize the source information of the reference digital object, and can be extracted using the aforementioned feature extraction model. For example, if the user provides a reference image, the reference image can be input into the aforementioned feature extraction model to extract the second image features of the reference image. Then, the feature similarity between the first and second data features is calculated. The calculation of feature similarity can effectively capture the potential association between the target digital content and the reference digital content. When the feature similarity is greater than a first preset threshold, the system can determine with high confidence that the target digital content and the reference digital content are from the same source. Conversely, when the feature similarity is less than or equal to the first preset threshold, the system can accurately identify that the target digital content and the reference digital content are from different sources, thereby avoiding incorrect classification based on similarity. This feature similarity-based judgment method not only improves the system's discrimination capability but also enhances its adaptability in diverse and complex data environments, providing solid technical support for applications such as content review, copyright protection, and information tracing.

[0159] The initial preset threshold can be selected based on actual needs. For example, it can be achieved by analyzing the similarity distribution in the training dataset and observing the similarity values ​​between samples from the same source and samples from different sources. Typically, a threshold that effectively distinguishes between the two types of samples can be chosen. The threshold can also be adjusted based on the business requirements of the specific application scenario and the tolerance for false positives. For instance, in some application scenarios, there may be a greater emphasis on reducing false negatives (missed detections), while in others, the focus may be on reducing false positives (false alarms).

[0160] Optionally, the feature similarity calculation methods include, but are not limited to: Euclidean distance, cosine similarity, Manhattan distance, and Hamming distance.

[0161] like Figure 8 The diagram shown illustrates a flowchart of a digital content detection method provided in this application. Taking image detection as an example, assuming the user provides a reference image and prompts, the input information for this detection task is the image to be tested (i.e., the target product image) and its target prompts, the reference image and its prompts. The process may include the following steps:

[0162] Step 801: Determine if the input information contains prompt words. If yes, proceed to step 802; otherwise, proceed to step 805 and directly execute the source matching branch.

[0163] Step 802: Determine if the prompt words of the image to be tested are the same as those of the reference image. If they are the same, proceed to step 803 to execute the source classification branch. For example, if the prompt word of the image to be tested is "A knight in a spacesuit rides an Akhal-Teke horse on the moon," and the prompt word of the reference image is also "A knight in a spacesuit rides an Akhal-Teke horse on the moon," then the prompt words are the same, and you can proceed to step 803. Otherwise, proceed to step 808.

[0164] Step 803: The AI-generated model source tracing classification model (i.e., the preset classification model) trained by the aforementioned embodiment is used to identify the source tracing results of the image to be tested and the source tracing results of the reference image respectively.

[0165] Step 804: Determine whether the source tracing result of the image to be tested is the same as the source tracing result of the reference image. If yes, proceed to step 807; otherwise, proceed to step 808.

[0166] Step 805: Using the CLIP image feature extractor finely tuned in the aforementioned embodiment, extract the first image features of the image to be tested and the second image features of the reference image, respectively, and calculate the feature distance (i.e. feature similarity) between the first image features and the second image features, and then proceed to step 806.

[0167] Step 806: Determine whether the feature distances are similar, that is, determine whether the feature distance is greater than the first preset threshold. If yes, proceed to step 807; otherwise, proceed to step 808.

[0168] Step 807: Determine that the image to be tested and the reference image are from the same source.

[0169] Step 808: Determine that the image to be tested and the reference image are not from the same source.

[0170] The method in this embodiment, when provided by the user with reference digital content, can trace the source of the target digital content to be tested and the reference digital content, accurately matching the specific generation source or model method of the digital content. The target digital content to be tested includes, but is not limited to, AI-generated images or text, and the reference digital content includes, but is not limited to, reference images or text.

[0171] In one embodiment, there are multiple reference digital contents, and the second data feature and corresponding metadata of each reference content are stored in a preset database. Determining whether target digital content and reference digital content are from the same source based on the first data feature and the reference digital content includes: calculating the feature similarity between the first data feature and the second data feature of each reference digital content in the preset database. Based on the feature similarity and the metadata corresponding to each reference content, it is determined whether the target digital content and reference digital content are from the same source.

[0172] In this embodiment, there can be multiple reference digital contents, which can be stored in a preset database. The preset database can store the second data features and corresponding metadata of each reference digital content. The metadata includes, but is not limited to, the source information of the reference digital content, such as which model generated the reference digital content, what text prompts were used in the generation process, the version information of the model, and the platform information to which the model belongs. By pre-storing the second data features of the reference digital content in the preset database, the amount of data computation for real-time inference is reduced. During inference, the feature similarity between the first data feature of the target digital content and the second data feature of each reference digital content in the database is first calculated. The process of calculating the feature similarity one by one ensures a comprehensive analysis of all potential reference content and avoids missing possible content from the same source. Subsequently, by combining the feature similarity and the metadata corresponding to each reference content, the system can more accurately determine whether the target digital content and the reference digital content are from the same source. This strategy of combining metadata provides additional contextual information, which helps to improve the accuracy of the judgment. In addition, by utilizing the structured storage of the preset database, this method optimizes the speed of data retrieval and computation, enabling the system to operate efficiently in a large-scale data environment. It not only enhances the ability to review content and protect copyright, but also provides strong support for applications such as information tracing and content recommendation.

[0173] In one embodiment, determining whether target digital content and reference digital content are from the same source based on feature similarity and the metadata corresponding to each reference content includes: determining a set of similar digital content from multiple reference digital content whose feature similarity to the target digital content is greater than a second preset threshold; comparing the metadata of the target digital content with the metadata of each digital content in the set of similar digital content; if there is a specific digital content in the set of similar digital content that has the same metadata as the target digital content, determining that the target digital content and the specific digital content are from the same source; and / or, outputting the specific digital content; if there is no specific digital content in the set of similar digital content that has the same metadata as the target digital content, determining that the target digital content and the reference digital content are not from the same source.

[0174] In this embodiment, by combining feature similarity and metadata for homology determination, this method significantly improves the accuracy and reliability of homology identification between target digital content and reference digital content. Specifically, firstly, based on feature similarity, a set of similar digital content with a feature similarity greater than a second preset threshold to the target digital content is selected from multiple reference digital content sets. This step ensures that only highly similar content is further analyzed, reducing the computational burden. Then, by comparing the metadata of the target digital content with the metadata of each digital content in the set of similar digital content, the system can utilize the contextual information provided by the metadata to perform a more refined homology determination. If a specific digital content in the set of similar digital content has the same metadata as the target digital content, it can be determined with high confidence that the target digital content is homologous to that specific digital content, and that specific digital content can be output. This strategy of combining metadata effectively reduces the possibility of misjudgment and ensures the accuracy of homology identification.

[0175] Conversely, if no specific digital content with the same metadata as the target digital content exists in the set of similar digital content, the system can accurately determine that the target digital content and the reference digital content are from different sources. This not only improves the ability to review content and protect copyright, but also provides reliable technical support for applications such as information tracing and content recommendation, enhancing the system's adaptability and practicality in complex data environments.

[0176] The second preset threshold can be the same as or different from the first preset threshold, and can be selected according to actual needs. Refer to the description of the first preset threshold; it will not be repeated here.

[0177] like Figure 9 The diagram shown is a flowchart illustrating a digital content detection method provided in an embodiment of this application. Taking image detection as an example, assuming the user does not provide a reference image, the method is pre-selected from a preset database (i.e., ...). Figure 9The base library stores information about multiple reference images. For example, the reference digital content stored in the base library can include multiple AI-generated images with copyright protection certification. The base library saves the feature vector of each AI-generated image and its bound JSON metadata (json_meta). The json_meta information can include the generation model name (model_name) and the generation prompt. For example, the json_meta information of image 1 in the base library includes the generation model name "SDv1.4" and the prompt "A knight in a spacesuit rides a Ferghana horse and walks on the moon". The json_meta information of image 2 in the base library includes the generation model name "SDXL" and the prompt "A knight in a spacesuit rides a Ferghana horse and walks on the moon". The json_meta information of image 3 in the base library includes the generation model name "SD3" and the prompt "Sun Wukong from Journey to the West flies over Wall Street", and so on. The json_meta information of image n in the base library includes the generation model name "Midjouney" and the prompt "A wooden sailboat sails on a turbulent sea".

[0178] The feature vector of each AI-generated image in the base database can be extracted using the CLIP image feature extractor finely tuned in any of the aforementioned embodiments and stored as base database features. Assume the input information for this detection task is the image to be tested and its prompt words. The detection process for the image to be tested may include the following steps:

[0179] Step 901: Extract the feature vector of the image to be tested using the CLIP image feature extractor finely tuned in any of the foregoing embodiments.

[0180] Step 902: Calculate the feature distance between the feature vector of the image to be tested and the feature vectors of all AI-generated images in the base database.

[0181] Step 903: Filter the images in the base database based on the feature distance, and select a list of similar images from the base database whose feature distance is greater than the second preset threshold.

[0182] Step 904: Check whether the json_meta information of the image to be tested is consistent with that of all images in the list of similar images. If the model_name and prompt in the json_meta information of the base image are the same as those of the image to be tested, proceed to step 905; otherwise, proceed to step 906.

[0183] Step 905: Identify the base library images in the similar image list whose json_meta information is the same as that of the image to be tested as source images of the image to be tested. There can be multiple source images, and a list of source images can be output.

[0184] Step 906: Identify the base library images in the similar image list whose json_meta information is different from that of the image to be tested as non-homologous images of the image to be tested. There can be multiple non-homologous images, and a list of non-homologous images can be output.

[0185] The method in this embodiment can trace the source of the AI-generated image or text to be tested, even when the user does not provide reference digital content, and can accurately match its specific generation source or model method.

[0186] The aforementioned digital content detection method, targeting two AI-generated images and their corresponding text prompts, uses an AI-generated image source tracing classification algorithm and a multimodal AI-generated image source tracing matching algorithm to determine whether the two images were generated by the same AI generation model with the same prompt, while simultaneously tracing the generation method. By jointly using the AI ​​generation model source tracing classification algorithm and the AI ​​generation model source tracing matching algorithm, the AI ​​generation image source tracing task is completed. In the AI ​​generation model source tracing classification algorithm, the consistency of the classification categories of the two AI-generated images is predicted to determine whether they are from the same source. In the AI ​​generation image source tracing matching algorithm, the similarity of the feature vectors of the two AI-generated images is calculated to determine whether they are from the same source. This not only solves the copyright ownership problem of AI-generated content but also greatly improves the management and copyright protection capabilities of generated images.

[0187] The technical solutions of this application have important applications in copyright protection, and their application scenarios include, but are not limited to, the following:

[0188] 1. Platform Governance of Original Image AIGC Detection: In real-world scenarios, there are instances of content plagiarism where copyrighted model and clothing images are post-processed using AIGC technology to bypass copyright protection detection. This plagiarism is difficult to detect with the naked eye, but the detection technology in this application can effectively identify AI-generated images, thereby preventing copyright infringement and protecting the legitimate rights of creators.

[0189] 2. Platform Obligation to Label AIGC Image Content: According to relevant internet information service management regulations, the platform must clearly label content including AIGC images. Since user-posted images may originate from various sources, the platform can identify AI-generated images based on the technical solution in this application and label the corresponding content to ensure information transparency and users' right to know.

[0190] 3. Traffic generation and offline information security prevention: Through the AI-generated image source tracing and detection technology in this application embodiment, it is possible to trace and determine which model platform the AI-generated image came from, which can identify information security risks in advance and prevent potential information loss. At the same time, it can also inform the platform that provides AI-generated images to strengthen the restriction and interception of unsafe generated images.

[0191] 4. Advertising and Brand Protection: In real-world scenarios, the advertising industry can leverage AI-generated images to promote brands. However, there is a possibility of AI being used to edit advertising images of well-known brands, damaging their brand image. The multimodal AI-generated image tracing and matching method provided in this application can identify and address fraudulent advertising images early, protecting brand reputation.

[0192] 5. Anti-counterfeiting on e-commerce platforms: Online shopping platforms need to ensure the authenticity of product images. Using the AI-generated image traceability detection technology provided in this application embodiment, the source of product images can be identified and traced, protecting consumer rights and enhancing the platform's credibility.

[0193] 6. Source Matching for Malicious Plagiarism of UGC Content: User-generated content (UGC) on social media and content platforms may be maliciously plagiarized, resulting in damage to the rights and interests of original authors. The AI-generated image source detection technology provided in this application can accurately identify and trace the source of plagiarized UGC content, ensuring the rights and interests of original authors, preventing the spread of harmful content, and improving information security.

[0194] For details of each step of the above method, please refer to the relevant descriptions in the above embodiments, which will not be repeated here.

[0195] Please refer to Figure 10 This is a method for detecting product images according to an embodiment of this application. The method can be performed by... Figure 1 The electronic device 1 shown is used to perform this action and can be applied to... Figure 2 In the digital content detection application scenario shown, the method for generating watermarked product images, using terminal 220 as the execution end in this embodiment and taking the product image scenario in an e-commerce scenario as an example, includes the following steps:

[0196] Step 1001: In response to the detection task, acquire the target product image to be processed and the reference image, wherein the target product image includes product information.

[0197] In this step, product information includes, but is not limited to, product-related knowledge information, appearance information, product promotion information, product-related order information, and product-related service information.

[0198] Step 1002: Input the target product image into the preset feature extraction model so that the feature extraction model outputs the first image feature of the target product image. The feature extraction model is trained by the first sample image set, which includes at least one sample image pair. Sample image pairs with the same sample prompt words and the same generation model are labeled as homologous labels.

[0199] Step 1003: Based on the first image features and the reference image, determine whether the target product image and the reference image are from the same source.

[0200] The aforementioned product image detection method automatically extracts representative first image features from target product images using a pre-trained feature extraction model. These first image features characterize the source information of the target product image. Since this feature extraction model is trained on a first sample image set, sample image pairs with the same prompt words and the same generation model are labeled as source-related. Both the prompt words and the generation model can characterize the image's source information. Therefore, the trained feature extraction model can effectively capture the first image features that accurately characterize the source information in the target product image. This effectively verifies whether the target product image and the reference image are from the same source, improving the accuracy of image tracing and matching detection, and providing reliable data support for subsequent product image ownership protection.

[0201] For details of each step of the above method, please refer to the relevant descriptions in the above embodiments, which will not be repeated here.

[0202] Please refer to Figure 11 This is a digital content detection device 1100 according to an embodiment of this application, which can be applied to... Figure 1 The electronic device 1 shown can be applied to Figure 2 In the digital content detection application scenario shown, the device effectively verifies whether target digital content and reference digital content share the same origin, improving the accuracy of digital content tracing and matching detection, and providing reliable data support for subsequent digital content ownership protection. The device includes: an acquisition module 1101, an extraction module 1102, and a determination module 1103. The functional principles of each module are as follows:

[0203] The acquisition module 1101 is used to acquire the target digital content to be processed and the reference digital content in response to the detection task.

[0204] The extraction module 1102 is used to input the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content. The feature extraction model is trained through a first sample dataset. The first sample dataset includes at least one sample digital content pair. Sample digital content pairs with the same sample prompt words and the same generation model are labeled as homologous tags.

[0205] The determination module 1103 is used to determine whether the target digital content and the reference digital content are from the same source based on the first data characteristics and the reference digital content.

[0206] In one embodiment, the system further includes: a first judgment module, configured to determine whether a target prompt word used to generate the target digital content exists in the input information of the detection task before the target digital content is input into a preset feature extraction model to make the feature extraction model output a first data feature of the target digital content. If no target prompt word exists in the input information, the step of inputting the target digital content into the preset feature extraction model to make the feature extraction model output a first data feature of the target digital content is executed.

[0207] In one embodiment, the system further includes: a second judgment module, configured to determine whether the target prompt word is the same as the reference prompt word corresponding to the reference numeric content if the input information contains a target prompt word; and a classification module, configured to input the target numeric content and the reference numeric content into a preset classification model if the target prompt word and the reference prompt word are the same, so that the preset classification model outputs first source information of the target numeric content and second source information of the reference numeric content. The classification module is further configured to determine whether the target numeric content and the reference numeric content are from the same source based on the first source information and the second source information.

[0208] In one embodiment, the system further includes: a first training module, configured to acquire a second sample dataset before inputting target digital content and reference digital content into a preset classification model to output first source information of the target digital content and second source information of the reference digital content, respectively. The second sample dataset includes multiple sample digital contents generated by different generation models, each sample digital content being labeled with information of the corresponding generation model. Based on the second sample dataset, a preset neural network is trained using a cross-entropy loss function to obtain a preset classification model.

[0209] In one embodiment, the reference digital content is a comparison object specified by the user. The determining module 1103 is used to acquire a second data feature of the reference digital content. It calculates the feature similarity between the first data feature and the second data feature. If the feature similarity is greater than a first preset threshold, it is determined that the target digital content and the reference digital content are from the same source. If the feature similarity is less than or equal to the first preset threshold, it is determined that the target digital content and the reference digital content are from different sources.

[0210] In one embodiment, there are multiple reference digital contents, and the second data feature and corresponding metadata of each reference digital content are stored in a preset database. The determining module 1103 is used to calculate the feature similarity between the first data feature and the second data feature of each reference digital content in the preset database. Based on the feature similarity and the metadata corresponding to each reference content, it is determined whether the target digital content and the reference digital content are from the same source.

[0211] In one embodiment, the determining module 1103 is specifically configured to determine a set of similar digital content from multiple reference digital content based on feature similarity, wherein the feature similarity to the target digital content is greater than a second preset threshold. The metadata of the target digital content is compared with the metadata of each digital content in the set of similar digital content. If a specific digital content with the same metadata as the target digital content exists in the set of similar digital content, the target digital content and the specific digital content are determined to be from the same source. And / or, the specific digital content is output. If no specific digital content with the same metadata as the target digital content exists in the set of similar digital content, the target digital content and the reference digital content are determined to be from different sources.

[0212] In one embodiment, the system further includes: a data construction module, configured to acquire multiple different sample prompt words before inputting the target digital content into a preset feature extraction model to output the first data features of the target digital content; acquire multiple different content generation models; input the multiple different sample prompt words one by one into each content generation model, so that each content generation model generates corresponding sample digital content based on the multiple different sample prompt words, thereby obtaining an original sample dataset; pair the sample digital content in the original sample dataset to form multiple sample digital content pairs, and label sample digital content pairs with the same sample prompt words and the same generation model as source-related labels, thereby forming the final first sample dataset.

[0213] In one embodiment, the system further includes a second training module, configured to fine-tune and train a preset feature extractor using a contrastive loss function based on a first sample dataset before inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data features of the target digital content, thereby obtaining a feature extraction model.

[0214] In one embodiment, the first sample dataset is a set of sample images. It also includes a third training module, used to fine-tune and train a preset feature extractor based on the sample image set before inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs first data features of the target digital content. The feature extractor includes an image feature extractor and a text feature extractor. During training, the image feature extractor and the text feature extractor are fine-tuned using an inter-class contrast loss function and an image-text contrast loss function, respectively, to obtain the feature extraction model.

[0215] For a detailed description of the digital content detection device 1100 described above, please refer to the description of the relevant method steps in the above embodiments. The implementation principle and technical effect are similar, and will not be repeated here.

[0216] Figure 12 This is a schematic diagram of the structure of a cloud device 120 provided for an exemplary embodiment of this application. The cloud device 120 can be used to run the methods provided in any of the above embodiments. Figure 12 As shown, the cloud device 120 may include: a memory 1204 and at least one processor 1205. Figure 12 Let's take a processor as an example.

[0217] The memory 1204 is used to store computer programs and can be configured to store various other data to support operations on the cloud device 120. The memory 1204 may be object storage (OSS).

[0218] The memory 1204 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.

[0219] The processor 1205, coupled to the memory 1204, is used to execute the computer program in the memory 1204 to implement the solution provided in any of the above method embodiments. The specific functions and technical effects to be achieved will not be elaborated here.

[0220] Furthermore, such as Figure 12 The cloud device also includes other components such as a firewall 1201, a load balancer 1202, a communication component 1206, and a power supply component 1203. Figure 12 The diagram only shows some components and does not mean that cloud devices only include... Figure 12 The components shown.

[0221] In one embodiment, the above Figure 12 The communication component 1206 is configured to facilitate wired or wireless communication between the device containing the communication component 1206 and other devices. The device containing the communication component 1206 can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G, LTE (Long Term Evolution), 5G, or combinations thereof. In one exemplary embodiment, the communication component 1206 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication component 1206 also includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra Wide Band (UWB), Bluetooth, and other technologies.

[0222] In one embodiment, the above Figure 12 The power supply component 1203 provides power to various components of the device in which it resides. The power supply component 1203 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply component resides.

[0223] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method of any of the foregoing embodiments.

[0224] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the method of any of the foregoing embodiments.

[0225] 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 merely illustrative. For instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0226] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0227] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the application can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor. The memory may include high-speed RAM (Random Access Memory), and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk, or optical disc, etc.

[0228] The aforementioned storage media can be implemented from 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. The storage media can be any available medium accessible to general-purpose or special-purpose computers.

[0229] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. Both the processor and the storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic device or host device.

[0230] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, garment, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, garment, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, garment, or apparatus that includes that element.

[0231] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0232] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of this application.

[0233] The collection, storage, use, processing, transmission, provision, and disclosure of user data and other information involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0234] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method of detecting digital content, characterized by, include: In response to the detection task, acquire the target digital content to be processed and the reference digital content; The target digital content is input into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content. The feature extraction model is trained through a first sample dataset, which includes at least one sample digital content pair. Sample digital content pairs with the same sample prompt words and the same generation model are labeled as homologous tags. Based on the first data feature and the reference digital content, determine whether the target digital content and the reference digital content are from the same source; Before inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content, the method further includes: Obtain multiple different sample prompt words; obtain multiple different content generation models; Each of the multiple different sample prompt words is input into each of the content generation models, so that each content generation model generates corresponding sample digital content based on the multiple different sample prompt words, thus obtaining the original sample dataset. The sample digital content in the original sample dataset is paired up to form multiple sample digital content pairs, and sample digital content pairs with the same sample prompt words and the same generation model are labeled as source-related labels to form the final first sample dataset.

2. The method of claim 1, wherein, Before inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content, the method further includes: Determine whether the input information of the detection task contains a target prompt word used to generate the target digital content; If the target prompt word is not present in the input information, the step of inputting the target digital content into a preset feature extraction model, so that the feature extraction model outputs the first data feature of the target digital content, is executed.

3. The method of claim 2, wherein, Also includes: If the target prompt word exists in the input information, determine whether the target prompt word is the same as the reference prompt word corresponding to the reference number content; If the target prompt word is the same as the reference prompt word, the target digital content and the reference digital content are respectively input into a preset classification model so that the preset classification model outputs the first source information of the target digital content and the second source information of the reference digital content. Determine whether the target digital content and the reference digital content are from the same source based on the first source information and the second source information.

4. The method of claim 3, wherein, Before inputting the target digital content and the reference digital content into a preset classification model, respectively, so that the preset classification model outputs the first source information of the target digital content and the second source information of the reference digital content, the method further includes: Obtain a second sample dataset, which includes multiple sample digital contents generated by different generation models, and each sample digital content is labeled with information of the corresponding generation model; Based on the second sample dataset, a preset neural network is trained using the cross-entropy loss function to obtain the preset classification model.

5. The method of claim 1, wherein, The reference digital content is a comparison object specified by the user; the step of determining whether the target digital content and the reference digital content are from the same source based on the first data feature and the reference digital content includes: Obtain the second data feature of the reference digital content; Calculate the feature similarity between the first data feature and the second data feature; If the feature similarity is greater than a first preset threshold, it is determined that the target digital content and the reference digital content are from the same source. If the feature similarity is less than or equal to the first preset threshold, it is determined that the target digital content and the reference digital content are from different sources.

6. The method of claim 1, wherein, The reference digital content is multiple, and the second data feature and corresponding metadata of each reference digital content are stored in a preset database; the step of determining whether the target digital content and the reference digital content are from the same source based on the first data feature and the reference digital content includes: Calculate the feature similarity between the first data feature and the second data feature of each reference digital content in the preset database; Based on the feature similarity and the metadata corresponding to each of the reference digital contents, it is determined whether the target digital content and the reference digital content are from the same source.

7. The method of claim 6, wherein, The step of determining whether the target digital content and the reference digital content are from the same source based on the feature similarity and the metadata corresponding to each reference digital content includes: Based on the feature similarity, a set of similar digital content that has a feature similarity greater than a second preset threshold with the target digital content is determined from a plurality of reference digital content; The metadata of the target digital content is compared with the metadata of each digital content in the set of similar digital content; If the set of similar digital content contains specific digital content that has the same metadata as the target digital content, determine that the target digital content and the specific digital content are from the same source; and / or, output the specific digital content. If there is no specific digital content in the set of similar digital content that has the same metadata as the target digital content, it is determined that the target digital content and the reference digital content are from different sources.

8. The method according to claim 1, characterized in that, Before inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content, the method further includes: Based on the first sample dataset, a pre-set feature extractor is fine-tuned using a contrastive loss function to obtain the feature extraction model.

9. The method according to claim 1, characterized in that, The first sample dataset is a set of sample images; before inputting the target digital content into a preset feature extraction model so that the feature extraction model outputs the first data feature of the target digital content, the method further includes: The preset feature extractor is fine-tuned and trained based on the sample image set, wherein the feature extractor includes an image feature extractor and a text feature extractor; During training, the image feature extractor and the text feature extractor are fine-tuned using the inter-class contrast loss function and the image-text contrast loss function, respectively, to obtain the feature extraction model.

10. A method for detecting product images, characterized in that, include: In response to a detection task, a target product image to be processed and a reference image are acquired, wherein the target product image includes product information; The target product image is input into a preset feature extraction model so that the feature extraction model outputs the first image feature of the target product image. The feature extraction model is trained through a first sample image set, which includes at least one sample image pair. Sample image pairs with the same sample prompt words and the same generation model are labeled as homologous tags. Based on the first image features and the reference image, determine whether the target product image and the reference image are from the same source; Before inputting the target product image into a preset feature extraction model so that the feature extraction model outputs the first data features of the target product image, the method further includes: Obtain multiple different sample prompt words; obtain multiple different content generation models; Each of the multiple different sample prompt words is input into each of the content generation models, so that each content generation model generates corresponding sample image content based on the multiple different sample prompt words, thus obtaining an original sample image set; the sample image content in the original sample image set is paired up to form multiple sample image content pairs, and sample image content pairs with the same sample prompt words and the same generation model are labeled as source tags to form the final first sample image set.

11. An electronic device, characterized in that, include: At least one processor; as well as A memory that is communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, cause the electronic device to perform the method according to any one of claims 1-10.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the method as described in any one of claims 1-10.

13. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1-10.

Images