CN116012612A - Content detection method and system - Google Patents

Content detection method and system Download PDF

Info

Publication number
CN116012612A
CN116012612A CN202211729915.7A CN202211729915A CN116012612A CN 116012612 A CN116012612 A CN 116012612A CN 202211729915 A CN202211729915 A CN 202211729915A CN 116012612 A CN116012612 A CN 116012612A
Authority
CN
China
Prior art keywords
content
sample
feature
image
preset
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202211729915.7A
Other languages
Chinese (zh)
Inventor
曹佳炯
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Alipay Hangzhou Information Technology Co Ltd
Original Assignee
Alipay Hangzhou Information Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Alipay Hangzhou Information Technology Co Ltd filed Critical Alipay Hangzhou Information Technology Co Ltd
Priority to CN202211729915.7A priority Critical patent/CN116012612A/en
Publication of CN116012612A publication Critical patent/CN116012612A/en
Pending legal-status Critical Current

Links

Images

Landscapes

  • Image Analysis (AREA)

Abstract

After at least one content image of target content is acquired, inputting the at least one content image into a content detection model to obtain a fake content probability of the target content and at least one fake content feature corresponding to the fake content probability, wherein a training sample of the content detection model comprises a plurality of fake content samples generated by diffusing reference content based on a multi-expert model group, the multi-expert model group comprises a plurality of complementary image-text matching models, and a risk detection result of the target content is determined based on the fake content probability and the at least one fake content feature and is output; the scheme can improve the detection performance of content detection.

Description

Content detection method and system
Technical Field
The present disclosure relates to the field of image recognition, and in particular, to a content detection method and system.
Background
In recent years, with the rapid development of internet technology, a manner of content generation has become more and more convenient, for example, an artificial intelligence technology (AI) may be employed to generate a content image or video or the like having a specific artistic style based on an input text. For such image or video contents generated by AI technology, the counterfeit contents cannot apply for copyrights, and therefore, content detection is required for the contents in the content security link. Existing content detection methods often pass through fake content detection methods based on original images or fake content detection methods based on camera watermarks.
In the research and practice process of the prior art, the inventor of the application finds that the original image-based counterfeit content detection method needs to perform model training through a large amount of labeling data, and under the condition that the number of counterfeit content samples is limited, the detection capability of the counterfeit content detection is greatly limited, and the camera watermark-based mode needs to introduce additional watermarks in the acquisition stage, so that the coverage rate of the content detection is lower, and the method is difficult to apply to all scenes, and therefore, the detection performance of the content detection is lower.
Disclosure of Invention
The specification provides a content detection method and system with higher detection performance.
In a first aspect, the present specification provides a content detection method, including: acquiring at least one content image of target content; inputting the at least one content image into a content detection model to obtain a fake content probability of the target content and at least one fake content feature corresponding to the fake content probability, wherein a training sample of the content detection model comprises a plurality of fake content samples generated by diffusing reference content based on a multi-expert model group, and the multi-expert model group comprises a plurality of complementary image-text matching models; and determining a risk detection result of the target content based on the forged content probability and the at least one forged content feature, and outputting the risk detection result.
In some embodiments, the target content comprises at least one of image content or video content.
In some embodiments, the training process of the multiple expert model set includes the steps of: acquiring a content sample pair, and inputting the content sample pair into a preset multi-model group to obtain a feature set and a feature similarity set corresponding to the content sample pair; and converging the preset multi-model group based on the feature set and the feature similarity set to obtain the trained multi-expert model group.
In some embodiments, the content sample pair comprises an image sample and a text sample, and the preset multi-model set comprises a plurality of preset teletext matching models and a feature migration network; and inputting the content sample pair to a preset multi-model group to obtain a feature set and a feature similarity set corresponding to the content sample pair, wherein the method comprises the following steps: the content sample pairs are respectively input into a plurality of preset image-text matching models to obtain a sample feature set and a feature similarity set, wherein the sample feature set comprises sample image features of the image samples and sample text features of the text samples output by each preset image-text matching model, the feature similarity set comprises feature similarity between the sample image features and the sample text features output by each preset image-text matching model, each sample feature in the sample feature set is input into a feature migration network to obtain a migration feature set, and the migration feature set comprises migration of sample features corresponding to any one of the preset image-text matching models to sample features corresponding to other preset image-text matching models in the preset image-text matching models, and the migration feature set and the sample feature set are used as the feature set.
In some embodiments, the converging the preset multi-model set to obtain the trained multi-expert model set includes: acquiring a labeling matching result of the content sample pair, and determining feature matching loss information based on the labeling matching result, the feature similarity set and the sample feature set; determining feature complementary loss information among the plurality of preset image-text matching models based on the sample feature set; determining feature migration loss information based on the sample feature set and the migration feature set; and converging the preset multi-model group based on the feature matching loss information, the feature complementation loss information and the feature migration loss information to obtain the multi-expert model group.
In some embodiments, the determining feature matching loss information based on the annotation matching result, the feature similarity set, and the sample feature set comprises: determining a first preset similarity threshold corresponding to the content sample pair based on the annotation matching result; selecting target feature similarity corresponding to each preset image-text matching model from the feature similarity set, and comparing the target feature similarity with the first preset similarity threshold value to obtain first feature matching loss information corresponding to each preset image-text matching model; determining second feature matching loss information corresponding to the preset multi-model group based on the sample feature set and the first preset similarity threshold; and taking the first feature matching loss information and the second feature matching loss information as the feature matching loss information.
In some embodiments, the determining, based on the sample feature set and the first preset similarity threshold, second feature matching loss information corresponding to the preset multi-model group includes: selecting the sample image features from the sample feature set to obtain a sample image feature set, and selecting the sample text features from the sample feature set to obtain a sample text feature set; accumulating the sample image features in the sample image feature set to obtain target sample image features, and accumulating the sample text features in the sample text feature set to obtain target sample text features; and obtaining the current feature similarity between the target sample image features and the target sample text features, and comparing the current feature similarity with the first preset similarity threshold value to obtain the second feature matching loss information.
In some embodiments, the determining feature complementary loss information between the plurality of preset teletext matching models based on the sample feature set includes: selecting sample features of the same type output by different image-text matching models from the sample feature set to obtain at least one sample feature pair; obtaining feature similarity between features in each sample feature pair of the at least one sample feature pair to obtain a complementary feature similarity set; and comparing the feature similarity in the complementary feature similarity set with a second preset similarity threshold to obtain the feature complementary loss information, wherein the constraint condition of the feature complementary loss information is that feature similarity between sample features of the same type output by different image-text matching models is constrained to be smaller than the second preset similarity threshold.
In some embodiments, the determining feature migration loss information based on the sample feature set and the migration feature set includes: selecting candidate sample characteristics corresponding to each migration characteristic in the migration characteristic set from the sample characteristic set; obtaining feature similarity between the candidate sample features and the corresponding migration features to obtain a migration feature similarity set; and comparing the feature similarity in the migration feature similarity set with a third preset similarity threshold to obtain the feature migration loss information, wherein the constraint condition of the feature migration loss information is that the feature similarity in migration between sample features of different preset image-text matching models is constrained to be smaller than the third preset similarity threshold.
In some embodiments, the training process of the content detection model includes the steps of: acquiring a content image sample pair corresponding to a target content sample in the training sample; inputting the content image sample pair into a preset content detection model to obtain sample forged content characteristics and predicted content categories; and converging the preset content detection model based on the sample forged content characteristics and the predicted content category to obtain the trained content detection model.
In some embodiments said inputting the pair of content image samples into a preset content detection model to obtain a sample forged content feature and a predicted content class comprises: linearly superposing the content image samples in the content image sample pair to obtain a mixed content image sample; respectively carrying out feature extraction on the content image sample pair and the mixed content image sample to obtain a candidate sample image feature set; and extracting the sample forged content features from the candidate sample image feature set and determining a predicted content category for the target content sample based on the sample forged content features.
In some embodiments, the extracting the sample forged content features in the candidate sample image feature set comprises: linearly superposing sample image features in the candidate sample image feature set to obtain mixed sample image features; extracting initial forged content characteristics corresponding to each sample image characteristic from the candidate sample image characteristic set and the mixed sample image characteristic respectively; and taking the initial forged content characteristic as the sample forged content characteristic corresponding to the target content sample.
In some embodiments, the converging the preset content detection model to obtain the trained content detection model includes: obtaining the labeling content category of the target content sample, and comparing the labeling content category with the predicted content category to obtain content classification loss information; determining feature prediction loss information of the target content sample based on the labeling content category and the sample counterfeit content feature; and fusing the characteristic prediction loss information and the content classification loss information, and converging the preset content detection model based on the fused target content loss information to obtain the content detection model.
In some embodiments, the noted content categories include normal content samples and counterfeit content samples; and determining feature prediction loss information of the target content sample based on the labeling content category and the sample counterfeit content feature, comprising: based on the labeling content category, determining a preset feature category corresponding to the target content sample, determining a current feature category corresponding to the sample forged content feature, and comparing the preset feature category with the current feature category to obtain the feature prediction loss information, wherein the constraint condition of the feature prediction loss information is that the sample forged content feature corresponding to the constraint normal content sample is an all-zero feature.
In some embodiments, before the obtaining the content image sample pair corresponding to the target content sample in the training sample, the method further includes: acquiring the reference content; respectively inputting the reference content into the multi-expert model group and the diffusion model to obtain a plurality of forged content samples; and acquiring a plurality of normal content samples, and taking the plurality of normal content samples and the plurality of forged content samples as the training samples.
In some embodiments, the reference content includes a set of text and a reference image; and said inputting said reference content into said multiple expert model set and said diffusion model, respectively, to obtain a plurality of counterfeit content samples, comprising: the reference image is input into the diffusion model to obtain an initial diffusion content image, the initial diffusion content image and target texts in the text set are input into the multi-expert model group to obtain noise increment images corresponding to a plurality of diffusion directions of the initial diffusion content image, and the noise increment images are input into the diffusion model to obtain a plurality of forged content samples.
In some embodiments, the training process of the diffusion model comprises the steps of: acquiring a reference image sample and a keyword sample corresponding to the reference image sample, and inputting the reference image sample into a preset diffusion model to obtain an initial diffusion content image sample; inputting the keyword sample and the initial diffusion content image sample into the multi-expert model group to obtain image-text similarity and a plurality of sample diffusion directions corresponding to the initial diffusion content image sample, wherein the image-text similarity comprises the similarity between the keyword sample and the initial diffusion content sample; inputting noise sets corresponding to the sample diffusion directions into the preset diffusion model to obtain a current diffusion content image sample set corresponding to the sample diffusion directions; and converging the preset diffusion model based on the image-text similarity, the noise set and the current diffusion content image sample set to obtain the trained diffusion model.
In some embodiments, the converging the preset diffusion model to obtain the trained diffusion model includes: inputting the current diffusion content image set into a preset feature extraction model to obtain evaluation features of each diffusion content image sample in the current diffusion content image sample set; determining target diffusion loss information based on the image-text similarity, the noise set, the current diffusion content image sample set and the evaluation characteristics; and converging the preset diffusion model based on the target diffusion loss information to obtain the diffusion model.
In some embodiments, the determining the target diffusion loss information comprises: obtaining the marked image-text similarity between the reference image sample and the keyword sample, and comparing the marked image-text similarity with the image-text similarity to obtain image-text similarity loss information; determining sample price loss information between diffusion content image samples in the current diffusion content image sample set based on the evaluation characteristics, wherein the constraint condition of the sample price loss information is that characteristic difference values between the evaluation characteristics of different diffusion content image samples are constrained to be smaller than a preset difference value threshold; determining diffusion distance loss information between different sample diffusion directions in the plurality of sample diffusion directions based on the noise set, wherein the constraint condition of the diffusion distance loss information is to restrict diffusion distances between the different sample diffusion directions to exceed a preset distance threshold; and fusing the image-text similarity loss information, the sample evaluation loss information and the diffusion distance loss information to obtain the target diffusion loss information.
In some embodiments, the evaluation features include style features and content features; and determining sample evaluation loss information between diffuse content image samples in the current set of diffuse content image samples based on the evaluation features, comprising: and respectively acquiring feature similarity of style features and feature similarity of the content features between the diffused content image samples to obtain a style feature similarity set and a content feature similarity set, comparing the feature similarity in the style feature similarity set with a preset style feature similarity threshold to obtain style loss information, comparing the feature similarity in the content feature similarity set with a preset content feature similarity threshold to obtain content loss information, and fusing the style loss information and the content loss information to obtain the sample evaluation loss information.
In some embodiments the determining diffusion distance loss information between different sample diffusion directions of the plurality of sample diffusion directions comprises: extracting characteristics of noise in the noise set to obtain noise characteristics corresponding to each sample diffusion direction in the plurality of sample diffusion directions; acquiring feature distances among the noise features to obtain diffusion distances among the diffusion directions of the plurality of samples; and comparing the diffusion distance with a preset distance threshold value to obtain the diffusion distance loss information.
In some embodiments, the risk detection result includes one of a fake content or a normal content, the fake content being an image content or a video content with an artistic style generated based on text; and determining a risk detection result of the target content, including: and acquiring a feature average value of the at least one forged content feature, and determining a risk detection result of the target content as the forged content when the forged content probability is larger than a preset probability threshold or the feature average value is larger than a preset feature average value threshold.
In some embodiments, further comprising: and when the forged content is smaller than the preset probability threshold and the characteristic average value is smaller than the preset characteristic average value threshold, determining that the risk detection result of the target content is the normal content.
In a second aspect, the present specification also provides a content detection system, including: at least one storage medium storing at least one instruction set for content detection; and at least one processor communicatively coupled to the at least one storage medium, wherein the at least one processor reads the at least one instruction set and performs the content detection method of the first aspect of the present specification as directed by the at least one instruction set when the content detection system is operating.
As can be seen from the above technical solutions, the content detection method and system provided in the present disclosure, after obtaining at least one content image of a target content, inputs the at least one content image into a content detection model to obtain a forged content probability of the target content and at least one forged content feature corresponding to the forged content probability, wherein a training sample of the content detection model includes a plurality of forged content samples generated by diffusing a reference content based on a multi-expert model set, the multi-expert model set includes a plurality of complementary image-text matching models, and determines a risk detection result of the target content based on the forged content probability and the at least one forged content feature, and outputs the risk detection result; according to the scheme, the diffusion of the reference content is guided through the multiple expert model groups, the multiple expert models comprise multiple complementary image-text matching models, the randomness of generated data is increased, in addition, other additional factors are not required to be introduced, so that the diversity and coverage rate of generated forged content samples are improved, and further, a high-coverage rate and high-performance content detection model is obtained, and therefore the detection performance of content detection can be improved.
Additional functionality of the content detection methods and systems provided herein will be set forth in part in the description that follows. The following numbers and examples presented will be apparent to those of ordinary skill in the art in view of the description. The inventive aspects of the content detection methods and systems provided herein may be fully explained by the practice or use of the methods, devices, and combinations described in the following detailed examples.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present description, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present description, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 shows a schematic application scenario of a content detection system according to an embodiment of the present disclosure;
FIG. 2 illustrates a hardware architecture diagram of a computing device provided in accordance with an embodiment of the present description;
FIG. 3 shows a flow chart of a content detection method provided in accordance with an embodiment of the present description;
FIG. 4 shows an overall flow diagram of one content detection provided in accordance with embodiments of the present description; and
fig. 5 shows a schematic diagram for detecting counterfeit content generated for the discovery method according to an embodiment of the present disclosure.
Detailed Description
The following description is presented to enable one of ordinary skill in the art to make and use the invention, and is provided in the context of a particular application and its requirements. Various modifications to the disclosed embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of the disclosure. Thus, the present description is not limited to the embodiments shown, but is to be accorded the widest scope consistent with the claims.
The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting. For example, as used herein, the singular forms "a", "an" and "the" include plural referents unless the context clearly dictates otherwise. The terms "comprises," "comprising," "includes," and/or "including," when used in this specification, are taken to specify the presence of stated integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
These and other features of the present specification, as well as the operation and function of the related elements of structure, as well as the combination of parts and economies of manufacture, may be significantly improved upon in view of the following description. All of which form a part of this specification, reference is made to the accompanying drawings. It is to be expressly understood, however, that the drawings are for the purpose of illustration and description only and are not intended as a definition of the limits of the description. It should also be understood that the drawings are not drawn to scale.
The flowcharts used in this specification illustrate operations implemented by systems according to some embodiments in this specification. It should be clearly understood that the operations of the flow diagrams may be implemented out of order. Rather, operations may be performed in reverse order or concurrently. Further, one or more other operations may be added to the flowchart. One or more operations may be removed from the flowchart.
For convenience of description, the present specification will explain terms that will appear from the following description as follows:
multiple expert model: refers to a method of making decisions and reasoning by using multiple independent and complementary models, such as, for example, training multiple expert models, simultaneously performing content security detection, and fusing the results as final results.
Disco Diffusion: a content generation method with artistic style based on text input by user by AI technology, which is based on large-scale pre-training model CLIP to guide the optimization direction of generation model, the generated content is mostly artistic creation and difficult to be distinguished by human eyes.
Content security: the authenticity of the content is detected by an artificial intelligence algorithm, for example, the difference between the image generated by Disco Diffusion and the image created by the artist is resolved.
Before describing the specific embodiments of the present specification, the application scenario of the present specification will be described as follows:
the content detection method provided by the specification can be applied to any content detection scene, for example, in a content auditing scene, the content to be audited in the content platform can be subjected to content detection by the content detection method of the specification; in a content distribution scene, content detection can be performed on the content to be distributed through the content detection method of the specification; in a content sharing scene, content detection can be performed on the content to be shared through the content detection method of the specification; the method can also be applied to any content detection scene, and will not be described in detail herein.
Those skilled in the art will appreciate that the content detection methods and systems described herein are applicable to other usage scenarios and are also within the scope of the present disclosure.
Fig. 1 shows a schematic application scenario of a content detection system 001 according to an embodiment of the present disclosure. The content detection system 001 (hereinafter referred to as system 001) may be applied to content detection in any scenario, such as content detection in a content review scenario, content detection in a content distribution scenario, content detection in a content sharing scenario, and so on, as shown in fig. 1, the system 001 may include a target user 100, a client 200, a server 300, and a network 400 in a target space.
The target user 100 may be a user that triggers the identification of the target content, and the target user 100 may perform a content detection operation at the client 200.
The client 200 may be a device that performs content detection on target content in response to a content detection operation of the target user 100. In some embodiments, the content detection method may be performed on the client 200. At this time, the client 200 may store data or instructions to perform the content detection method described in the present specification, and may execute or be used to execute the data or instructions. In some embodiments, the client 200 may include a hardware device having a data information processing function and a program necessary to drive the hardware device to operate. As shown in fig. 1, a client 200 may be communicatively connected to a server 300. In some embodiments, the server 300 may be communicatively coupled to a plurality of clients 200. In some embodiments, client 200 may interact with server 300 over network 400 to receive or send messages, etc., such as receiving or sending targeted content or content images. In some embodiments, the client 200 may include a mobile device, a tablet, a laptop, a built-in device of a motor vehicle, or the like, or any combination thereof. In some embodiments, the mobile device may include a smart home device, a smart mobile device, a virtual reality device, an augmented reality device, or the like, or any combination thereof. In some embodiments, the smart home device may include a smart television, a desktop computer, or the like, or any combination. In some embodiments, the smart mobile device may include a smart phone, personal digital assistant, gaming device, navigation device, etc., or any combination thereof. In some embodiments, the virtual reality device or augmented reality device may include a virtual reality helmet, virtual reality glasses, virtual reality handles, an augmented reality helmet, augmented reality glasses, an augmented reality handle, or the like, or any combination thereof. For example, the virtual reality device or the augmented reality device may include google glass, head mounted display, VR, or the like. In some embodiments, the built-in devices in the motor vehicle may include an on-board computer, an on-board television, and the like. In some embodiments, the client 200 may include an image capturing device for capturing an image of the target content, thereby capturing at least one content image. In some embodiments, the image capture device may be a two-dimensional image capture device (such as an RGB camera), or may be a two-dimensional image capture device (such as an RGB camera) and a depth image capture device (such as a 3D structured light camera, a laser detector, etc.). In some embodiments, the client 200 may be a device with positioning technology for locating the position of the client 200.
In some embodiments, client 200 may be installed with one or more Applications (APP). The APP can provide the target user 100 with the ability to interact with the outside world via the network 400 as well as an interface. The APP includes, but is not limited to: web browser-like APP programs, search-like APP programs, chat-like APP programs, shopping-like APP programs, video-like APP programs, financial-like APP programs, instant messaging tools, mailbox clients, social platform software, and the like. In some embodiments, the client 200 may have a target APP installed thereon. The target APP can acquire target content or a content image of the target content for the client 200, thereby obtaining at least one content image. In some embodiments, the target user 100 may also trigger a content detection request through the target APP. The target APP may perform the content detection method described in the present specification in response to the content detection request. The content detection method will be described in detail later.
The server 300 may be a server providing various services, such as a background server providing support for content detection and at least one content image of target content acquired on the client 200. In some embodiments, the content detection method may be performed on the server 300. At this time, the server 300 may store data or instructions to perform the content detection method described in the present specification, and may execute or be used to execute the data or instructions. In some embodiments, the server 300 may include a hardware device having a data information processing function and a program necessary to drive the hardware device to operate. The server 300 may be communicatively connected to a plurality of clients 200 and receive data transmitted from the clients 200.
The network 400 is a medium used to provide communication connections between the client 200 and the server 300. The network 400 may facilitate the exchange of information or data. As shown in fig. 1, the client 200 and the server 300 may be connected to a network 400 and transmit information or data to each other through the network 400. In some embodiments, the network 400 may be any type of wired or wireless network, or a combination thereof. For example, network 400 may include a cable network, a wired network, a fiber optic network, a telecommunications network, an intranet, the Internet, a Local Area Network (LAN), a Wide Area Network (WAN), a Wireless Local Area Network (WLAN), a Metropolitan Area Network (MAN), a Wide Area Network (WAN), a Public Switched Telephone Network (PSTN), bluetooth TM Network, zigBee TM A network, a Near Field Communication (NFC) network, or the like. In some embodiments, network 400 may include one or more network access points. For example, network 400 may include a wired or wireless network access point, such as a base station or an Internet switching point, through which one or more components of client 200 and server 300 may connect to network 400 toExchanging data or information.
It should be understood that the number of clients 200, servers 300, and networks 400 in fig. 1 are merely illustrative. There may be any number of clients 200, servers 300, and networks 400, as desired for implementation.
It should be noted that, the content detection method may be performed entirely on the client 200, entirely on the server 300, or partially on the client 200 and partially on the server 300.
Fig. 2 illustrates a hardware architecture diagram of a computing device 600 provided in accordance with an embodiment of the present description. Computing device 600 may perform the content detection methods described herein. The content detection method is described elsewhere in this specification. When the content detection method is performed on the client 200, the computing device 600 may be the client 200. When the content detection method is performed on the server 300, the computing device 600 may be the server 300. When the content detection method may be partially performed on the client 200 and partially performed on the server 300, the computing device 600 may be both the client 200 and the server 300.
As shown in fig. 2, computing device 600 may include at least one storage medium 630 and at least one processor 620. In some embodiments, computing device 600 may also include a communication port 650 and an internal communication bus 610. Meanwhile, computing device 600 may also include I/O component 660.
Internal communication bus 610 may connect the various system components including storage medium 630, processor 620, and communication ports 650.
I/O component 660 supports input/output between computing device 600 and other components.
The communication port 650 is used for data communication between the computing device 600 and the outside world, for example, the communication port 650 may be used for data communication between the computing device 600 and the network 400. The communication port 650 may be a wired communication port or a wireless communication port.
The storage medium 630 may include a data storage device. The data storage device may be a non-transitory storage medium or a transitory storage medium. For example, the data storage devices may include one or more of magnetic disk 632, read Only Memory (ROM) 634, or Random Access Memory (RAM) 636. The storage medium 630 further includes at least one set of instructions stored in the data storage device. The instructions are computer program code that may include programs, routines, objects, components, data structures, procedures, modules, etc. that perform the content detection methods provided herein.
The at least one processor 620 may be communicatively coupled with at least one storage medium 630 and a communication port 650 via an internal communication bus 610. The at least one processor 620 is configured to execute the at least one instruction set. When the computing device 600 is running, the at least one processor 620 reads the at least one instruction set and performs the content detection methods provided herein according to the instructions of the at least one instruction set. The processor 620 may perform all the steps involved in the content detection method. The processor 620 may be in the form of one or more processors, and in some embodiments, the processor 620 may include one or more hardware processors, such as microcontrollers, microprocessors, reduced Instruction Set Computers (RISC), application Specific Integrated Circuits (ASICs), application specific instruction set processors (ASIPs), central Processing Units (CPUs), graphics Processing Units (GPUs), physical Processing Units (PPUs), microcontroller units, digital Signal Processors (DSPs), field Programmable Gate Arrays (FPGAs), advanced RISC Machines (ARM), programmable Logic Devices (PLDs), any circuit or processor capable of executing one or more functions, or the like, or any combination thereof. For illustrative purposes only, only one processor 620 is depicted in the computing device 600 in this specification. It should be noted, however, that computing device 600 may also include multiple processors, and thus, operations and/or method steps disclosed in this specification may be performed by one processor as described herein, or may be performed jointly by multiple processors. For example, if the processor 620 of the computing device 600 performs steps a and B in this specification, it should be understood that steps a and B may also be performed by two different processors 620 in combination or separately (e.g., a first processor performs step a, a second processor performs step B, or the first and second processors perform steps a and B together).
Fig. 3 shows a flowchart of a content detection method P100 provided according to an embodiment of the present specification. As before, the computing device 600 may perform the content detection method P100 of the present specification. Specifically, the processor 620 may read an instruction set stored in its local storage medium and then execute the content detection method P100 of the present specification according to the specification of the instruction set. As shown in fig. 3, the method P100 may include:
s110: at least one content image of the target content is acquired.
Wherein the target content comprises at least one of image content or video content. The content image may be an image containing the target content, for example, when the target content is an image content, the content image may be the image content itself, when the target content is a video content, the content image may be at least one frame of video frame in the video content, and so on.
The manner of acquiring at least one content image of the target content may be various, and specifically may be as follows:
for example, the processor 620 may directly acquire at least one content image of the target content uploaded by the target user 100 through the client 200 or the terminal, or may acquire the target content, then acquire the at least one content image of the target content through the image acquisition device, or may acquire the at least one content image of the target content from a network or a content database, or may further receive a content detection request when the number of the target content is large or the memory is large, the content detection request carrying a storage address of the target content or the at least one content image of the target content, acquire the at least one content image of the target content based on the storage address, and so on.
S120: at least one piece of content image is input to the content detection model to obtain a spurious content probability of the target content and at least one spurious content feature corresponding to the spurious content probability.
The content detection model is a model for detecting risk of target content or detecting fake content. The content detection model may include an image blending network, a feature encoding network, a feature blending network, and a counterfeit content prediction network. The image blending network is configured to linearly superimpose two content images to obtain a blended content image. The feature encoding network is configured to image feature encode the content image and the blended content image to obtain image features of the content image and image features of the blended content image. The feature blending network is configured to linearly superimpose image features of the content image and image features of the blended content image to obtain blended image features. The spurious content prediction network is configured to determine spurious content characteristics corresponding to the image characteristics and the hybrid image characteristics, and determine spurious content probabilities of the target content based on the spurious content characteristics. The counterfeit content feature may be feature information characterizing the target content as counterfeit content. The counterfeit content may be image content or video content generated based on text and having an artistic style, etc. The probability of the forged content is the probability of the target content being the forged content. The training samples of the content detection model comprise a plurality of forged content samples generated by diffusing the reference content based on a multi-expert model group, and the multi-expert model group comprises a plurality of complementary image-text matching models. The reference content includes a text set and a reference image, which may be an original image before diffusion.
The method for inputting at least one content image into the content detection model to obtain the probability of the forged content of the target content and at least one forged content feature corresponding to the probability of the forged content may be various, and specifically may be as follows:
for example, the processor 620 may select a content image pair from at least one content image, input the content image pair to an image blending network to obtain a blended content image, input the content image pair and the blended content image to a feature encoding network, input an image feature set including image features of the content image pair and image features of the blended content image, and then input the image feature set to the feature blending network to obtain a blended feature. The mixed features and the image feature set are input to a counterfeit content prediction network to obtain a counterfeit content probability and at least one counterfeit content feature.
The training process of the content detection model may include the following: the processor 620 may obtain a sample content image pair corresponding to the target content sample in the training sample, input the sample content image pair to a preset content detection model to obtain a sample counterfeit content feature and a predicted content category, and converge the preset content detection model based on the sample counterfeit content feature and the predicted content category to obtain a trained content detection model.
The manner of acquiring the content image sample pair corresponding to the target content sample in the training sample may be various, for example, the processor 620 may randomly select at least one content sample in the training sample as the target content sample, then acquire a content image sample set of the target content sample, and randomly select two content image samples in the content image sample set, thereby obtaining the content image sample pair, or may also select a content sample of a specific category as the target content sample based on selecting the content sample of the specific category, then acquire a content image sample set of the target content sample, and randomly select two content image samples in the content image sample set, thereby obtaining the content image sample pair, and so on.
After acquiring the content image sample pair corresponding to the target content sample in the training sample, the processor 620 may input the content image sample pair into a preset content detection model to obtain a sample forged content feature and a predicted content category. The manner in which the content image sample pair is input to the preset content detection model to obtain the sample forged content feature and the predicted content category may be various, for example, the processor 620 may linearly superimpose the content image samples in the content image sample pair to obtain a mixed content image sample, and perform feature extraction on the content image sample and the mixed content image sample respectively to obtain a candidate sample image feature set, extract the sample forged content feature from the candidate sample image feature set, and determine the predicted content category of the target content sample based on the sample forged content feature.
The manner of linearly superimposing the content image samples in the content image sample pair may be various, for example, the processor 620 may directly linearly superimpose the content image sample pair, and then linearly superimpose the content label corresponding to the content image sample pair, so as to obtain a mixed content image sample, where the content label may be a labeling content type.
After the content image samples are linearly superimposed and feature extracted, the processor 620 may extract sample spurious content features from the extracted candidate sample image feature set. There may be various ways of extracting the sample forged content features from the candidate sample image feature set, for example, the processor 620 may linearly superimpose the sample image features in the candidate sample image feature set to obtain a mixed sample image feature, extract initial forged content features corresponding to each sample image feature from the candidate sample image feature set and the mixed sample image feature, and use the initial forged content features as sample forged content features corresponding to the target content sample. Then, based on the sample forged content characteristics, a predicted content category of the target content sample is determined.
After determining the sample spurious content characteristics and the predicted content categories of the target content sample, the processor 620 may converge the preset content detection model based on the sample spurious content characteristics and the predicted content characteristics, thereby obtaining a trained content detection model. There may be various ways to converge the preset content detection model, for example, the processor 620 may obtain a labeling content type of the target content sample, compare the labeling content type with a predicted content type to obtain content classification loss information, determine feature prediction loss information of the target content sample based on the labeling content type and sample counterfeit content features, fuse the feature prediction loss information and the content classification loss information, and converge the preset content detection model based on the fused target content loss information to obtain the content detection model.
The content classification loss information may be loss information generated when classifying the target content sample. There may be various ways to compare the labeling content category with the predicted content category, for example, the processor 620 may compare the labeling content category with the predicted content category based on a cross entropy loss function to obtain content classification loss information, or may also compare the labeling content category with the predicted content category based on other loss functions to obtain content classification loss information, and so on.
Wherein the labeling content categories include normal content samples and counterfeit content samples. The counterfeit content sample may be a content sample corresponding to the counterfeit content. The feature prediction loss information may be loss information generated when predicting counterfeit content features. The constraint condition of the feature prediction loss information is to restrict the sample forged content features corresponding to the normal content samples to all-zero features. There may be various ways to determine the feature prediction loss information, for example, the processor 620 may determine a preset feature class corresponding to the target content sample based on the labeling content class, determine a current feature class corresponding to the counterfeit content feature of the sample, and compare the preset feature class with the current feature class to obtain the feature prediction loss information. The feature class may include all zero features, which may be features with feature elements of all 0, and non-all zero features.
After determining the content classification loss information and the feature prediction loss information, the processor 620 may fuse the content classification loss information and the feature prediction loss information to obtain fused target content loss information. There may be various ways to fuse the content classification loss information and the feature prediction loss information, for example, the processor 620 may directly add the content classification loss information to obtain the target content loss information, or may acquire a content weight, respectively weight the content classification loss information and the feature prediction loss information based on the content weight, fuse the weighted content classification loss information and the weighted feature prediction loss information to obtain the target content loss information, or the like.
After fusing the content classification loss information and the feature prediction loss information, the processor 620 may converge the preset content detection model based on the fused target content loss information, thereby obtaining a trained content detection model. The method of converging the preset content detection model may be various, for example, the processor 620 may update the network parameters of each network layer of the preset content detection model by using a gradient descent algorithm based on the target content loss information and/or the content classification loss information and the feature prediction loss information in the target content loss information, and return to execute the step of acquiring the content image sample pair corresponding to the target content sample in the training sample until the preset content detection model converges, thereby obtaining the trained content detection model, or may update the network parameters of each network layer of the preset content detection model by using other network parameter update algorithms based on the target content loss information and/or the content classification loss information and the feature prediction loss information in the target content loss information, return to execute the step of acquiring the content image sample pair corresponding to the target content sample in the training sample until the preset content detection model converges, thereby obtaining the trained content detection model, and so on.
In some embodiments, the processor 620 may also generate or acquire a training sample before acquiring the pair of content image samples corresponding to the target content sample in the training sample. There are various ways in which the processor 620 generates or obtains the training samples, for example, the processor 620 may obtain the reference content, input the reference content into the multiple expert model groups and the diffusion model to obtain a plurality of falsified content samples, and obtain a plurality of normal content samples and take the plurality of normal content samples and the plurality of falsified content samples as the training samples.
Wherein the reference content includes a text set and a reference image. There are various ways to obtain the reference content, for example, the processor 620 may randomly generate at least one text to obtain a text set, randomly select an image in the network or the image database as the reference image, or may also receive the reference content uploaded by the target user 100 through the client 200 or the terminal, and so on.
After the reference content is obtained, the processor 620 may input the reference content into a plurality of expert model groups and diffusion models, respectively, to obtain a plurality of counterfeit content samples. The reference content may be input to the multiple expert model groups and the diffusion model respectively to obtain multiple falsified content samples, for example, the processor 620 may input the reference image to the diffusion model to obtain an initial diffuse content image, input the initial diffuse content and the target text in the text set to the multiple expert model groups to obtain noise-added images corresponding to multiple diffusion directions of the initial diffuse content image, and input the noise-added images to the diffusion model to obtain multiple falsified content samples.
The noise-added image may be an added image generated based on noise corresponding to each of a plurality of diffusion directions. The diffusion direction can be understood as a direction of adjusting, modifying or diffusing the reference image (or the initial diffusion content image), the diffusion direction can be represented by noise, for example, noise corresponding to a certain diffusion direction is generated into a noise-increased image, and the noise-increased image is fused with the reference image or the initial diffusion content image, so that the diffusion content image diffused in the diffusion direction can be obtained. In addition, the diffuse content image may be understood as a content image of a counterfeit content sample. Through a plurality of expert model groups until diffusion in a plurality of diffusion directions is carried out, the randomness of the generated forged content samples can be increased, and the diversity and coverage rate of the generated content samples are further improved.
The training process of the diffusion model may include the following steps: the processor 620 may obtain a reference image sample and a keyword sample corresponding to the reference image sample, input the reference image sample to a preset diffusion model to obtain an initial diffusion content image sample, input the keyword sample and the initial diffusion content image sample to a multi-expert model group to obtain a plurality of sample diffusion directions corresponding to the image-text similarity and the initial diffusion content image sample, input a noise set corresponding to the plurality of sample diffusion directions to the preset diffusion model to obtain a current diffusion content image sample set corresponding to the plurality of sample diffusion directions, and converge the preset diffusion model based on the image-text similarity, the noise set and the current diffusion content image sample set to obtain a trained diffusion model.
The keyword sample is a text for describing the style or art of the image, for example, the text can be described by a 'Sanskyline style'. The image-text similarity comprises similarity between the keyword sample and the initial diffusion content image sample. And the initial diffusion content image sample is a diffusion content image sample obtained by randomly selecting one direction for diffusing the reference content by using an initialized preset diffusion model.
Based on the image-text similarity, the noise set and the current diffusion content image sample set, the preset diffusion model is converged, and therefore the trained diffusion model is obtained. There may be various ways to converge the preset diffusion model, for example, the processor 620 may input the current diffusion content image set into the preset feature extraction model to obtain an evaluation feature of each diffusion content image sample in the current diffusion content image sample set, determine target diffusion loss information based on the image-text similarity, the noise set, the current diffusion content image sample set and the evaluation feature, and converge the preset diffusion model based on the target diffusion loss information, so as to obtain the diffusion model.
The evaluation feature may be a feature extracted by the feature extraction model for performing sample evaluation on the diffusion content image sample. The evaluation features may include style features (style) and content features (content). The network structure of the feature extraction model may be various, and for example, may include a CNN network (convolutional neural network), a DNN network (deep learning neural network), an RNN network (cyclic neural network), or other network structure, and so on. The feature extraction model may be a local feature extraction model integrated within the processor 620 or may be a feature extraction model of a third party.
After extracting the evaluation feature of each diffuse content image sample, the processor 620 may determine the target diffuse loss information based on the teletext similarity, the noise set, the current diffuse content image sample set, and the evaluation feature. The target diffusion loss information may be loss information generated by a preset diffusion model when diffusing the reference image sample or the initial content diffusion image sample. The method for determining the target diffusion loss information may be various, for example, the processor 620 may obtain a labeled image-text similarity between the reference image sample and the keyword sample, and compare the labeled image-text similarity with the image-text similarity to obtain image-text similarity loss information, determine sample evaluation loss information between diffusion content image samples in the current diffusion content image sample set based on the evaluation feature, determine diffusion distance loss information between diffusion directions of different samples in the diffusion directions of the plurality of samples based on the noise set, and fuse the image-text similarity loss information, the sample evaluation loss information and the diffusion distance loss information to obtain the target diffusion loss information.
The sample evaluation loss information may be characteristic information for evaluating a style and content of a diffuse content image sample generated from a plurality of sample diffusion directions. The constraint condition of the sample evaluation loss information is that the feature difference value between the evaluation features of the samples of different diffusion content images is constrained to be smaller than a preset difference threshold value, which means that the sample evaluation loss information is used for constraining the style features and the content features of the diffusion content sample images generated in different sample diffusion directions to be consistent. There may be various ways of determining the sample evaluation loss information, for example, the processor 620 may obtain the feature similarity of the style features and the feature similarity of the content features between the diffused content image samples, so as to obtain a style feature similarity set and a content feature similarity set, compare the feature similarity in the style feature similarity set with a preset style feature similarity threshold to obtain style loss information, compare the feature similarity in the content feature similarity set with a preset content feature similarity threshold to obtain content loss information, and fuse the style loss information and the content loss information to obtain sample evaluation loss information.
The diffusion distance loss information is loss information of distances between diffusion directions of different samples, and the constraint condition of the diffusion distance loss information is that the diffusion distances between diffusion directions of different samples are constrained to exceed a preset distance threshold, namely that the diffusion distance loss information is used for constraining the distances of a plurality of diffusion directions to be as large as possible. Various manners of determining the diffusion distance loss information may be used, for example, the processor 620 may perform feature extraction on noise in the noise set to obtain noise features corresponding to each of the diffusion directions of the plurality of samples, obtain feature distances between the noise features to obtain diffusion distances between the diffusion directions of the plurality of samples, and compare the diffusion distances with a preset distance threshold to obtain the diffusion distance loss information.
The feature distance may be an L1/L2 distance, a Euclidean distance, or the like.
After determining the image-text similarity loss information, the sample evaluation loss information, and the diffusion distance loss information, the processor 620 may fuse the image-text similarity loss information, the sample evaluation loss information, and the diffusion distance loss information, thereby obtaining target diffusion loss information. The manner of fusing the image-text similarity loss information, the sample evaluation loss information and the diffusion distance loss information may be various, for example, the processor 620 may directly add the image-text loss information, the sample evaluation loss information and the diffusion distance loss information, so as to obtain the target diffusion loss information, which may be specifically shown in formula (1):
Loss total1 ＝Loss sim +Loss style-content +Loss mse (1)
Wherein, loss total1 Loss information for target diffusion sim Loss of information for image-text similarity style-content Evaluating Loss information for samples, loss mse Information is lost for diffusion distance.
In some embodiments, the processor 620 may further obtain a diffusion weight, and weight the image-text similarity loss information, the sample evaluation loss information, and the diffusion distance loss information based on the diffusion weight, and fuse the weighted image-text similarity loss information, the weighted sample evaluation loss information, and the weighted diffusion distance loss information to obtain the target diffusion loss information.
After determining the target diffusion loss information, the processor 620 may converge the predetermined diffusion model based on the target diffusion loss information. The manner of converging the preset diffusion model is similar to that of converging the preset content detection model, and detailed description is omitted herein.
Before training the preset diffusion model, the processor 620 may train the multiple expert model group, train the preset diffusion model based on the trained multiple expert model group, thereby obtaining a trained diffusion model, and then instruct the trained diffusion model based on the multiple expert model group, thereby enabling the diffusion model to generate a plurality of forged content samples, and use the plurality of forged content samples and the plurality of normal content samples as training samples of the preset content detection model, train the preset content detection model through the training samples, and perform content detection on the target content by adopting the trained content detection model.
Wherein the training process of the multi-expert model group may comprise the steps of: the processor 620 may acquire a content sample pair, input the content sample pair to a preset multi-model group, obtain a feature set and a feature similarity set corresponding to the content sample pair, and converge the preset multi-model group based on the feature set and the feature similarity set to obtain a trained multi-expert model group.
Wherein the content sample pair includes an image sample and a text sample. The preset multi-model group comprises a plurality of preset image-text matching models and a characteristic migration network. The teletext matching model is configured to predict a degree of matching between an image sample and a text sample. The image-text matching model may further comprise at least one image encoder and at least one text encoder, the image encoder being configured to perform feature encoding on the image samples to obtain image features, the text encoder being configured to perform feature encoding on the text samples to obtain text features. The input of the graph-text matching model is a matched/unmatched image and text pair, and the output is the corresponding characteristics and the similarity of the two. The feature migration network is configured to predict features output by other graph-text matching models through features output by a certain graph-text matching model. The input of the feature migration network is the feature (image feature or text feature) output by a certain image-text matching model, and the output is the feature (image feature or text feature) output by another (other) image-text matching model. The manner of inputting the content sample pair into the preset multi-model group to obtain the feature similarity set of the feature set corresponding to the content sample pair may be various, for example, the processor 620 may respectively value the content sample pair into a plurality of preset image matching models to obtain a sample feature set and a feature similarity set, input each sample feature in the sample feature set into the feature migration network to obtain a migration feature set, and use the migration feature set and the sample feature set as feature sets.
The sample feature set comprises sample image features of image samples and sample text features of text samples output by each preset image-text matching model in the plurality of preset image-text matching models. The feature similarity set comprises feature similarity between the sample image features and the sample text features output by each preset image-text matching model. The migration feature set includes sample features corresponding to any one of the preset image-text matching models and sample features corresponding to other preset image-text matching models in the preset image-text matching models, for example, taking the number of the preset image-text matching models as 3 as an example, inputting the sample image features output by the preset image-text matching model A into the feature migration network, so that sample image features corresponding to the preset image-text matching model B and sample image features corresponding to the preset image-text matching model C can be obtained, and the like, so that a migration feature set can be obtained after feature migration between different preset image-text matching models.
After obtaining the feature set and the feature similarity set corresponding to the content sample pair, the processor 620 may converge the preset multi-model set based on the feature set and the feature similarity set to obtain a trained multi-expert model set. There may be various ways to converge the preset multi-model set, for example, the processor 620 may obtain a label matching result of the content sample pair, determine feature matching loss information based on the label matching result, the feature similarity set and the sample feature set, determine feature complementary loss information between a plurality of preset graphic models based on the sample feature set, determine feature migration loss information based on the sample feature set and the migration feature set, and converge the preset multi-model set based on the feature matching loss information, the special complementary loss information and the feature migration loss information to obtain a multi-expert model set.
The label matching result may be a label that is matched between the image sample and the text sample that are labeled in the content sample pair. The feature matching penalty information may be penalty information generated by a difference between a degree of feature matching between the sample image feature and the sample text feature and a degree of sample matching between the image sample and the text sample. The method for determining the feature matching loss information may be various, for example, the processor 620 may determine a first preset similarity threshold corresponding to the content sample pair based on the labeling matching result, select a target feature similarity corresponding to each preset image matching model in the feature similarity set, compare the target feature similarity with the first preset similarity threshold to obtain first feature matching loss information corresponding to each preset image matching model, determine second feature matching loss information corresponding to the preset multi-model set based on the sample feature set and the first preset similarity threshold, and use the first feature matching loss information and the second feature matching loss information as feature matching loss information.
The first feature matching loss information may be loss information generated by a difference between feature similarity between an image sample and a text sample output by each preset image-text matching model and a labeling matching result of a content sample pair. The constraint condition of the first feature matching loss information is: for matched pairs of content samples, the similarity of features between the constrained image sample and the text sample is greater than a first preset similarity threshold, or for unmatched pairs of content samples, the similarity of features between the constrained image sample and the text sample is less than a first preset similarity threshold. That is, for matched pairs of content samples, the features between the image and text samples are as consistent as possible, otherwise, as inconsistent as possible is required.
The second feature matching loss information may be a matching loss of a sum of features input by the plurality of image-text matching models, and may also be understood as a matching loss after the image features are added and the text features are added. The constraint condition of the second feature matching loss information may be that the similarity of the sum of constraint features is greater than a first preset similarity threshold for the matched image sample and the text sample, or the similarity of the sum of constraint features is less than the first preset similarity threshold for the unmatched image sample and the text sample. There may be various ways of determining the second feature matching loss information, for example, the processor 620 may select a sample image feature from the sample feature set to obtain a sample image feature set, select a sample text feature from the sample feature set to obtain a sample text feature set, accumulate sample image features in the sample image feature set to obtain a target sample image feature, accumulate sample text features in the sample text feature set to obtain a target sample text feature, and obtain a current feature similarity between the target sample image feature and the target sample text feature, and compare the current feature similarity with a first preset similarity threshold to obtain the second feature matching loss information.
The feature complementary loss information may be loss information generated by differences between sample image features or sample text features output by a plurality of image-text matching models for the same image sample or text sample. The constraint condition of the feature complementation loss information is that feature similarity between sample features (sample image features or sample text features) of the same type output by different image-text matching models in the plurality of image-text matching models is constrained to be smaller than a second preset similarity threshold. For example, taking the second preset similarity threshold value as 0 as an example, the constraint condition may be that cosine similarity between sample image features or sample text features output by different image-text matching models needs to be as close to 0 as possible. The method for determining the feature complementary loss information may be various, for example, the processor 620 may select the same type of sample features output by different image-text matching models from the sample feature set, obtain at least one sample feature pair, obtain feature similarities between features in each sample feature pair in the at least one sample feature pair, obtain a complementary feature similarity set, and compare feature similarities in the complementary feature similarity set with a second preset similarity threshold to obtain the feature complementary loss information.
The characteristic migration loss information is loss information generated when the characteristic migration network migrates sample characteristics output by different image-text matching models. The constraint condition of the feature migration loss information is that feature similarity when migration between sample features of different preset image-text matching models is constrained to be smaller than a third preset similarity threshold, and the feature migration loss information can also be understood that when the sample image feature A output by the preset image-text matching model A is migrated to the sample image feature B output by the preset image-text matching model B, the similarity (cosine similarity) between the migrated image feature B and the sample image feature B is as low as possible. The manner of determining the feature migration loss information may be various, for example, the processor 620 may select a candidate sample feature corresponding to each migration feature in the migration feature set from the sample feature set, obtain feature similarity between the candidate sample feature and the corresponding migration feature, obtain a migration feature similarity set, and compare feature similarity in the migration feature similarity set with a third preset similarity threshold to obtain feature migration loss information.
After determining the feature matching loss information, the extra-complementary loss information, and the feature migration loss information, the processor 620 may converge the preset multi-model set based on the feature matching loss information, the extra-complementary loss information, and the feature migration loss information, thereby obtaining a multi-expert model set. There may be various ways to converge the preset multi-model set, for example, the processor 620 may fuse the feature matching loss information, the special complementary loss information, and the feature migration loss information to obtain the target loss information, and converge the preset multi-model set based on the target loss information, so as to obtain the multi-expert model set.
The manner of fusing the feature matching loss information, the extra complementary loss information, and the feature migration loss information may be various, for example, the processor 620 may directly add the feature matching loss information, the extra complementary loss information, and the feature migration loss information, so as to obtain the target loss information, which may be specifically shown in formula (2):
Loss total2 ＝Loss matc h+Loss ensemble-matc h+Loss complementaty +Loss transfer (2)
wherein, loss total2 Loss of information for targets, loss matc h is first feature matching Loss information, loss ensemble-matc h is second characteristic matching Loss information, loss complementaty To characterize complementary Loss information, loss transfer Loss information for feature migration.
In some embodiments, the processor 620 may further obtain a target loss weight, weight the feature matching loss information, the feature complementary loss information, and the feature migration loss information based on the target loss weight, and fuse the weighted feature matching loss information, the weighted feature complementary loss information, and the weighted feature migration loss information, respectively, to obtain the target loss information.
After fusing the feature matching loss information, the extra complementary loss information, and the feature migration loss information, the processor 620 may converge the preset multi-model set based on the fused target loss information, thereby obtaining a trained multi-expert model set. The manner of converging the preset multi-model set and the manner of converging the content detection model are detailed in the above description, and will not be described in detail herein.
It should be noted that, in training the preset multi-model set, multiple complementary large-scale pre-training models (preset image-text matching models) are trained, so as to cover multiple text and content scenes.
S130: and determining a risk detection result of the target content based on the forged content probability and the at least one forged content feature, and outputting the risk detection result.
The risk detection result may include one of a fake content or a normal content, the fake content being an image content or a video content generated based on text and having an artistic style.
The manner of determining the risk detection result of the target content may be various, and specifically may be as follows:
for example, the processor 620 may acquire a feature average of at least one of the spurious content features, and determine that the risk detection result of the target content is spurious content when the spurious content probability is greater than a preset probability threshold, or the feature average is greater than a preset feature average threshold; or when the forged content is smaller than the preset probability threshold and the characteristic average value is smaller than the preset characteristic average value threshold, determining that the risk detection result of the target content is normal content.
After determining the risk detection result of the target content, the processor 620 may output the risk detection result. There may be various ways to output the risk detection result, for example, the processor 620 may directly send the risk detection result to at least one of the client 200, the terminal, or the server corresponding to the target user 100, so that the client 200, the terminal, or the server responds to the target content or the request corresponding to the target content based on the risk detection result, or may directly visually display the risk detection result, or the like.
The manner of visually displaying the risk detection result may be various, for example, the processor 620 may directly display the risk detection result, or may display the risk detection result through an acousto-optic manner (for example, by broadcasting the risk detection result through voice, or may display the risk detection result of different types through explicit different colors of light, or may display the risk detection result through an acousto-optic linkage manner), or may display the risk detection result of a specific type (for example, display the risk detection result of a type of counterfeit content only, display the risk detection result of a type of normal content only, etc.), or the like.
In some embodiments, the processor 620 may further respond to the target content or the request corresponding to the target content based on the risk detection result after determining the risk detection result of the target content or outputting the risk detection result, and the responding manner may be various, for example, the processor 620 may directly intercept the target content or the request corresponding to the target content, or the processor 620 may directly perform two or more times of verification on the target content, and based on the secondary verification result, perform a final response on the target content or the request corresponding to the target content, or the like.
Taking the content generated by using the falsified content as a base and using the Disco distribution technology as an example, the overall flow of the content detection method may be as shown in fig. 4, and mainly includes multi-expert model group training, falsified content generation, content detection model training, model deployment and reasoning, and may specifically be as follows:
(1) Training of multiple expert model groups: multiple complementary large-scale pre-training models are trained to cover a variety of text, content scenarios, and specific training procedures can be found as described above.
(2) Counterfeit content generation: the existing fake content generation method is generally based on a diffusion process (namely a Markov process) known by a single pre-training model, so that the generated content mode is relatively single and can influence the performance of a subsequent content detection model, and therefore, the scheme adopts a multi-expert model group obtained through training to conduct joint guidance, the randomness of generated data is increased, and the diversity and coverage rate of the content are improved.
(3) Training a content detection model: and training the preset content detection model by using the generated forged content sample (forged image) and the existing data as training data to obtain a content detection model with high coverage rate and high safety performance, and introducing double-aspect mixing (mixup) of the image and the characteristic layer in the training stage, so that the performance of the content detection model is improved.
(4) Model deployment and reasoning: the trained fake content detection model is deployed to a server (local or remote), the content image acquired by the client 200 is uploaded to the server, the content image is input to the content detection model, the fake content probability and the corresponding fake text characteristics of the content image are obtained, when the fake content probability p is greater than a threshold value T1 or the mean value of the fake content characteristics is greater than a threshold value T2, the fake content is determined, otherwise, the fake content is determined to be normal content, and the method can be specifically shown in fig. 5.
According to the scheme, a plurality of complementary large-scale pre-training models are trained to obtain a multi-expert model group covering various texts and content scenes, then, under the guidance of the multi-expert model group, a forged content sample with diversity and high coverage rate is generated after large-scale diffusion of reference content, and a preset content detection model is trained by forged content sample set existing data, so that the text with uneven training data distribution is effectively solved, a content security model with high coverage rate and high security capability is obtained, and the detection performance of content detection is further improved.
In summary, the content detection method P100 and the system 001 provided in the present disclosure, after obtaining at least one content image of a target content, input the at least one content image into a content detection model to obtain a forged content probability of the target content and at least one forged content feature corresponding to the forged content probability, where a training sample of the content detection model includes a plurality of forged content samples generated by diffusing a reference content based on a multi-expert model set, where the multi-expert model set includes a plurality of complementary image-text matching models, and determine a risk detection result of the target content based on the forged content probability and the at least one forged content feature, and output the risk detection result; according to the scheme, the diffusion of the reference content is guided through the multiple expert model groups, the multiple expert models comprise multiple complementary image-text matching models, the randomness of generated data is increased, in addition, other additional factors are not required to be introduced, so that the diversity and coverage rate of generated forged content samples are improved, and further, a high-coverage rate and high-performance content detection model is obtained, and therefore the detection performance of content detection can be improved.
Another aspect of the present disclosure provides a non-transitory storage medium storing at least one set of executable instructions for content detection. When executed by a processor, the executable instructions direct the processor to perform the steps of the content detection method P100 described herein. In some possible implementations, aspects of the specification can also be implemented in the form of a program product including program code. The program code is for causing the computing device 600 to perform the steps of the content detection method P100 described in the present specification when the program product is run on the computing device 600. The program product for implementing the methods described above may employ a portable compact disc read only memory (CD-ROM) comprising program code and may run on computing device 600. However, the program product of the present specification is not limited thereto, and in the present specification, the readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system. The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. The computer readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, with readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A readable storage medium may also be any readable medium that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing. Program code for carrying out operations of the present specification may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on computing device 600, partly on computing device 600, as a stand-alone software package, partly on computing device 600, partly on a remote computing device, or entirely on a remote computing device.
The foregoing describes specific embodiments of the present disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
In view of the foregoing, it will be evident to a person skilled in the art that the foregoing detailed disclosure may be presented by way of example only and may not be limiting. Although not explicitly described herein, those skilled in the art will appreciate that the present description is intended to encompass various adaptations, improvements, and modifications of the embodiments. Such alterations, improvements, and modifications are intended to be proposed by this specification, and are intended to be within the spirit and scope of the exemplary embodiments of this specification.
Furthermore, certain terms in the present description have been used to describe embodiments of the present description. For example, "one embodiment," "an embodiment," and/or "some embodiments" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the present description. Thus, it is emphasized and should be appreciated that two or more references to "an embodiment" or "one embodiment" or "an alternative embodiment" in various portions of this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined as suitable in one or more embodiments of the invention.
It should be appreciated that in the foregoing description of embodiments of the present specification, various features have been combined in a single embodiment, the accompanying drawings, or description thereof for the purpose of simplifying the specification in order to assist in understanding one feature. However, this is not to say that a combination of these features is necessary, and it is entirely possible for a person skilled in the art to label some of the devices as separate embodiments to understand them upon reading this description. That is, embodiments in this specification may also be understood as an integration of multiple secondary embodiments. While each secondary embodiment is satisfied by less than all of the features of a single foregoing disclosed embodiment.
Each patent, patent application, publication of patent application, and other materials, such as articles, books, specifications, publications, documents, articles, etc., cited herein are hereby incorporated by reference. All matters are to be interpreted in a generic and descriptive sense only and not for purposes of limitation, except for any prosecution file history associated therewith, any and all matters not inconsistent or conflicting with this document or any and all matters not complaint file histories which might have a limiting effect on the broadest scope of the claims. Now or later in association with this document. For example, if there is any inconsistency or conflict between the description, definition, and/or use of terms associated with any of the incorporated materials, the terms in the present document shall prevail.
Finally, it is to be understood that the embodiments of the application disclosed herein are illustrative of the principles of the embodiments of the present specification. Other modified embodiments are also within the scope of this specification. Accordingly, the embodiments disclosed herein are by way of example only and not limitation. Those skilled in the art can adopt alternative arrangements to implement the application in the specification based on the embodiments in the specification. Therefore, the embodiments of the present specification are not limited to the embodiments precisely described in the application.

Claims (24)

1. A content detection method, comprising:
acquiring at least one content image of target content;
inputting the at least one content image into a content detection model to obtain a fake content probability of the target content and at least one fake content feature corresponding to the fake content probability, wherein a training sample of the content detection model comprises a plurality of fake content samples generated by diffusing reference content based on a multi-expert model group, and the multi-expert model group comprises a plurality of complementary image-text matching models; and
and determining a risk detection result of the target content based on the forged content probability and the at least one forged content feature, and outputting the risk detection result.
2. The content detection method of claim 1, wherein the target content comprises at least one of image content or video content.
3. The content detection method according to claim 1, wherein the training process of the multi-expert model group comprises the steps of:
acquiring a content sample pair, and inputting the content sample pair into a preset multi-model group to obtain a feature set and a feature similarity set corresponding to the content sample pair; and
and converging the preset multi-model group based on the feature set and the feature similarity set to obtain the trained multi-expert model group.
4. The content detection method according to claim 3, wherein the content sample pair comprises an image sample and a text sample, and the preset multi-model group comprises a plurality of preset image-text matching models and a feature migration network; and
the step of inputting the content sample pair to a preset multi-model group to obtain a feature set and a feature similarity set corresponding to the content sample pair, including:
respectively inputting the content sample pairs into a plurality of preset image-text matching models to obtain a sample feature set and a feature similarity set, wherein the sample feature set comprises sample image features of the image samples and sample text features of the text samples output by each preset image-text matching model in the plurality of preset image-text matching models, the feature similarity set comprises feature similarity between the sample image features and the sample text features output by each preset image-text matching model,
Inputting each sample feature in the sample feature set to the feature migration network to obtain a migration feature set, wherein the migration feature set comprises migration of sample features corresponding to any one of the preset image-text matching models to sample features corresponding to other preset image-text matching models in the preset image-text matching models, and
and taking the migration feature set and the sample feature set as the feature set.
5. The content detection method according to claim 4, wherein the converging the preset multi-model group to obtain the trained multi-expert model group includes:
acquiring a labeling matching result of the content sample pair, and determining feature matching loss information based on the labeling matching result, the feature similarity set and the sample feature set;
determining feature complementary loss information among the plurality of preset image-text matching models based on the sample feature set;
determining feature migration loss information based on the sample feature set and the migration feature set; and
and converging the preset multi-model group based on the feature matching loss information, the feature complementation loss information and the feature migration loss information to obtain the multi-expert model group.
6. The content detection method of claim 5, wherein the determining feature matching loss information based on the annotation matching result, the feature similarity set, and the sample feature set comprises:
determining a first preset similarity threshold corresponding to the content sample pair based on the annotation matching result;
selecting target feature similarity corresponding to each preset image-text matching model from the feature similarity set, and comparing the target feature similarity with the first preset similarity threshold value to obtain first feature matching loss information corresponding to each preset image-text matching model;
determining second feature matching loss information corresponding to the preset multi-model group based on the sample feature set and the first preset similarity threshold; and
and taking the first characteristic matching loss information and the second characteristic matching loss information as the characteristic matching loss information.
7. The content detection method according to claim 6, wherein the determining, based on the sample feature set and the first preset similarity threshold, second feature matching loss information corresponding to the preset multi-model group includes:
Selecting the sample image features from the sample feature set to obtain a sample image feature set, and selecting the sample text features from the sample feature set to obtain a sample text feature set;
accumulating the sample image features in the sample image feature set to obtain target sample image features, and accumulating the sample text features in the sample text feature set to obtain target sample text features; and
and acquiring the current feature similarity between the target sample image features and the target sample text features, and comparing the current feature similarity with the first preset similarity threshold value to obtain the second feature matching loss information.
8. The content detection method according to claim 5, wherein the determining feature complementary loss information between the plurality of preset teletext matching models based on the sample feature set includes:
selecting sample features of the same type output by different image-text matching models from the sample feature set to obtain at least one sample feature pair;
obtaining feature similarity between features in each sample feature pair of the at least one sample feature pair to obtain a complementary feature similarity set; and
And comparing the feature similarity in the complementary feature similarity set with a second preset similarity threshold to obtain the feature complementary loss information, wherein the constraint condition of the feature complementary loss information is that feature similarity among sample features of the same type output by different image-text matching models is constrained to be smaller than the second preset similarity threshold.
9. The content detection method of claim 5, wherein the determining feature migration loss information based on the sample feature set and the migration feature set comprises:
selecting candidate sample characteristics corresponding to each migration characteristic in the migration characteristic set from the sample characteristic set;
obtaining feature similarity between the candidate sample features and the corresponding migration features to obtain a migration feature similarity set; and
and comparing the feature similarity in the migration feature similarity set with a third preset similarity threshold to obtain the feature migration loss information, wherein the constraint condition of the feature migration loss information is that the feature similarity in migration between sample features of different preset image-text matching models is constrained to be smaller than the third preset similarity threshold.
10. The content detection method according to claim 1, wherein the training process of the content detection model comprises the steps of:
acquiring a content image sample pair corresponding to a target content sample in the training sample;
inputting the content image sample pair into a preset content detection model to obtain sample forged content characteristics and predicted content categories; and
and converging the preset content detection model based on the sample forged content characteristics and the predicted content category to obtain the trained content detection model.
11. The content detection method according to claim 10, wherein the inputting the pair of content image samples into a preset content detection model to obtain a sample counterfeit content feature and a predicted content category comprises:
linearly superposing the content image samples in the content image sample pair to obtain a mixed content image sample;
respectively carrying out feature extraction on the content image sample pair and the mixed content image sample to obtain a candidate sample image feature set; and
and extracting the sample forged content characteristics from the candidate sample image characteristic set, and determining the predicted content category of the target content sample based on the sample forged content characteristics.
12. The content detection method of claim 11, wherein the extracting the sample counterfeit content feature from the candidate sample image feature set comprises:
linearly superposing sample image features in the candidate sample image feature set to obtain mixed sample image features;
extracting initial forged content characteristics corresponding to each sample image characteristic from the candidate sample image characteristic set and the mixed sample image characteristic respectively; and
and taking the initial forged content characteristic as the sample forged content characteristic corresponding to the target content sample.
13. The content detection method according to claim 10, wherein the converging the preset content detection model to obtain the trained content detection model includes:
obtaining the labeling content category of the target content sample, and comparing the labeling content category with the predicted content category to obtain content classification loss information;
determining feature prediction loss information of the target content sample based on the labeling content category and the sample counterfeit content feature; and
and fusing the characteristic prediction loss information and the content classification loss information, and converging the preset content detection model based on the fused target content loss information to obtain the content detection model.
14. The content detection method of claim 13, wherein the annotated content categories include normal content samples and counterfeit content samples; and
the determining the feature prediction loss information of the target content sample based on the labeling content category and the sample counterfeit content feature comprises the following steps:
based on the labeling content category, determining a preset feature category corresponding to the target content sample,
determining the current feature class corresponding to the sample forged content features, and
and comparing the preset characteristic category with the current characteristic category to obtain the characteristic prediction loss information, wherein the constraint condition of the characteristic prediction loss information is that the sample forged content characteristics corresponding to the constraint normal content sample are all zero characteristics.
15. The content detection method according to claim 10, wherein before the obtaining the content image sample pair corresponding to the target content sample in the training sample, further includes:
acquiring the reference content;
respectively inputting the reference content into the multi-expert model group and the diffusion model to obtain a plurality of forged content samples; and
and acquiring a plurality of normal content samples, and taking the plurality of normal content samples and the plurality of forged content samples as the training samples.
16. The content detection method of claim 15, wherein the reference content comprises a set of text and a reference image; and
said inputting said reference content into said multiple expert model set and said diffusion model, respectively, to obtain a plurality of counterfeit content samples, comprising:
inputting the reference image into the diffusion model to obtain an initial diffusion content image,
inputting the initial diffusion content image and target text in the text set into the multi-expert model group to obtain noise increment images corresponding to a plurality of diffusion directions of the initial diffusion content image, and
the noise-plus-image is input to the diffusion model to obtain the plurality of counterfeit content samples.
17. The content detection method according to claim 15, wherein the training process of the diffusion model comprises the steps of:
acquiring a reference image sample and a keyword sample corresponding to the reference image sample, and inputting the reference image sample into a preset diffusion model to obtain an initial diffusion content image sample;
inputting the keyword sample and the initial diffusion content image sample into the multi-expert model group to obtain image-text similarity and a plurality of sample diffusion directions corresponding to the initial diffusion content image sample, wherein the image-text similarity comprises the similarity between the keyword sample and the initial diffusion content sample;
Inputting noise sets corresponding to the sample diffusion directions into the preset diffusion model to obtain a current diffusion content image sample set corresponding to the sample diffusion directions; and
and converging the preset diffusion model based on the image-text similarity, the noise set and the current diffusion content image sample set to obtain the trained diffusion model.
18. The content detection method according to claim 17, wherein the converging the preset diffusion model to obtain the trained diffusion model includes:
inputting the current diffusion content image set into a preset feature extraction model to obtain evaluation features of each diffusion content image sample in the current diffusion content image sample set;
determining target diffusion loss information based on the image-text similarity, the noise set, the current diffusion content image sample set and the evaluation characteristics; and
and converging the preset diffusion model based on the target diffusion loss information to obtain the diffusion model.
19. The content detection method of claim 18, wherein the determining target diffusion loss information comprises:
Obtaining the marked image-text similarity between the reference image sample and the keyword sample, and comparing the marked image-text similarity with the image-text similarity to obtain image-text similarity loss information;
determining sample price loss information between diffusion content image samples in the current diffusion content image sample set based on the evaluation characteristics, wherein the constraint condition of the sample price loss information is that characteristic difference values between the evaluation characteristics of different diffusion content image samples are constrained to be smaller than a preset difference value threshold;
determining diffusion distance loss information between different sample diffusion directions in the plurality of sample diffusion directions based on the noise set, wherein the constraint condition of the diffusion distance loss information is to restrict diffusion distances between the different sample diffusion directions to exceed a preset distance threshold; and
and fusing the image-text similarity loss information, the sample evaluation loss information and the diffusion distance loss information to obtain the target diffusion loss information.
20. The content detection method of claim 19, wherein the evaluation features include style features and content features; and
the determining sample evaluation loss information between diffuse content image samples in the current set of diffuse content image samples based on the evaluation features comprises:
Respectively obtaining the feature similarity of the style features and the feature similarity of the content features between the diffusion content image samples to obtain a style feature similarity set and a content feature similarity set,
comparing the feature similarity in the style feature similarity set with a preset style feature similarity threshold to obtain style loss information, and comparing the feature similarity in the content feature similarity set with a preset content feature similarity threshold to obtain content loss information, and
and fusing the style loss information and the content loss information to obtain the sample evaluation loss information.
21. The content detection method according to claim 19, wherein the determining diffusion distance loss information between different sample diffusion directions among the plurality of sample diffusion directions includes:
extracting characteristics of noise in the noise set to obtain noise characteristics corresponding to each sample diffusion direction in the plurality of sample diffusion directions;
acquiring feature distances among the noise features to obtain diffusion distances among the diffusion directions of the plurality of samples; and
And comparing the diffusion distance with a preset distance threshold value to obtain the diffusion distance loss information.
22. The content detection method according to claim 1, wherein the risk detection result includes one of a falsified content or a normal content, the falsified content being an image content or a video content having an artistic style generated based on text; and
the determining the risk detection result of the target content comprises the following steps:
acquiring a feature mean of the at least one counterfeit content feature, and
and when the probability of the forged content is larger than a preset probability threshold value or the characteristic average value is larger than a preset characteristic average value threshold value, determining that the risk detection result of the target content is the forged content.
23. The content detection method of claim 22, further comprising:
and when the forged content is smaller than the preset probability threshold and the characteristic average value is smaller than the preset characteristic average value threshold, determining that the risk detection result of the target content is the normal content.
24. A content detection system, comprising:
at least one storage medium storing at least one instruction set for content detection; and
At least one processor communicatively coupled to the at least one storage medium,
wherein the at least one processor reads the at least one instruction set and performs the content detection method of any one of claims 1-23 as directed by the at least one instruction set when the content detection system is running.
CN202211729915.7A 2022-12-30 2022-12-30 Content detection method and system Pending CN116012612A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211729915.7A CN116012612A (en) 2022-12-30 2022-12-30 Content detection method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211729915.7A CN116012612A (en) 2022-12-30 2022-12-30 Content detection method and system

Publications (1)

Publication Number Publication Date
CN116012612A true CN116012612A (en) 2023-04-25

Family

ID=86027906

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211729915.7A Pending CN116012612A (en) 2022-12-30 2022-12-30 Content detection method and system

Country Status (1)

Country Link
CN (1) CN116012612A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117235534A (en) * 2023-11-13 2023-12-15 支付宝(杭州)信息技术有限公司 Method and device for training content understanding model and content generating model

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117235534A (en) * 2023-11-13 2023-12-15 支付宝(杭州)信息技术有限公司 Method and device for training content understanding model and content generating model
CN117235534B (en) * 2023-11-13 2024-02-20 支付宝(杭州)信息技术有限公司 Method and device for training content understanding model and content generating model

Similar Documents

Publication Publication Date Title
CN111709408B (en) Image authenticity detection method and device
CN111324874B (en) Certificate authenticity identification method and device
CN117079299B (en) Data processing method, device, electronic equipment and storage medium
CN116012612A (en) Content detection method and system
CN114972016A (en) Image processing method, image processing apparatus, computer device, storage medium, and program product
CN111898544B (en) Text image matching method, device and equipment and computer storage medium
CN116468113A (en) Living body detection model training method, living body detection method and living body detection system
CN116311546A (en) Living body detection method and system
CN117252947A (en) Image processing method, image processing apparatus, computer, storage medium, and program product
CN115578768A (en) Training method of image detection network, image detection method and system
CN115984977A (en) Living body detection method and system
CN115880530A (en) Detection method and system for resisting attack
CN114581978A (en) Face recognition method and system
CN111291640B (en) Method and apparatus for recognizing gait
CN114676705A (en) Dialogue relation processing method, computer and readable storage medium
Ahmed GPark: Vehicle parking management system using smart glass
CN116246152A (en) Content detection method and system
CN116343346A (en) Living body detection method and system
CN116665315A (en) Living body detection model training method, living body detection method and living body detection system
CN117370679B (en) Method and device for verifying false messages of multi-mode bidirectional implication social network
CN116259116A (en) Living body detection method and system
CN115953709A (en) Detection method and system for resisting attack
CN116777914B (en) Data processing method, device, equipment and computer readable storage medium
CN115909511A (en) Living body detection method and system
CN116110136A (en) Living body detection method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination