Image authentic identification and training method based on structural causality and related equipment

Abstract

The invention relates to an image authentic identification and training method based on structure causality and related equipment, and the method comprises the steps: for each training image in two training images, extracting the depth feature of the training image through a backbone network, and carrying out the decoupling of the depth feature of the training image through an FBEM based on the coordinate position corresponding to the training image; any first feature and any second feature extracted from the two training images are spliced through an FBEM; inputting the splicing features into a classifier; calculating loss based on the authenticity prediction result of the image and the authenticity label of the image; and adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss. Thus, through a training mode of decoupling and splicing the overall features of the training images with different contents, it can be ensured that the trained image authentic identification model can accurately identify the tampered area in the image, and thus the accuracy of image authentic identification can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and more specifically, to an image forgery identification and training method based on structural causality and related equipment. Background Art

[0002] With the rapid development of technology, it has become increasingly common to generate various highly realistic images and videos by artificial means, which makes the creation of high-quality content more convenient. However, this also brings the risk of technology being abused, for example, it may lead to the proliferation of false information. Therefore, it is crucial to develop image forgery detection technology.

[0003] In the related art, when an image is authenticated, the authenticity of the image is usually determined based on the overall features of all the areas contained in the image. However, the tampered area in the image is often only a small part, and most of the areas contained in the image are actually real areas that have not been tampered with. For example, for a face image, the tampered area may be only the nose area or the mouth area, while most of the other areas of the face image have not actually been tampered with. At this time, if the image authentication is performed based on the overall features of the image, most of the natural areas in the image that have not been tampered with will interfere with the identification of the forged areas in the image, which will reduce the accuracy of the image authentication. Summary of the invention

[0004] The present disclosure provides a structural causal-based image authentication and training method and related equipment to at least solve the problem in the above-mentioned related technologies that most of the natural areas in the image that have not been tampered with interfere with the identification of the forged areas, thereby reducing the accuracy of image authentication.

[0005] According to a first aspect of an embodiment of the present disclosure, a training method for an image counterfeit detection model based on structural causality is provided, wherein the image counterfeit detection model comprises a backbone network, a forgery bias elimination module (FBEM) and a classifier, and the training method comprises: obtaining a training image sample, wherein the training image sample comprises two training images with different contents, each of the training images corresponding to an image authenticity label and a coordinate position for indicating an image tampering area; for each of the two training images, extracting a deep feature of the training image through the backbone network, and decoupling the deep feature of the training image based on the coordinate position corresponding to the training image through the FBEM to obtain a decoupled first feature and a second feature, wherein the The first feature is a forgery-related feature related to the image tampering area, and the second feature is a forgery-irrelevant feature unrelated to the image tampering area; any of the first feature and the second feature extracted from the two training images are spliced ​​through the FBEM to obtain a spliced ​​feature; the spliced ​​feature is input into the classifier to obtain an authenticity prediction result of the image; based on the authenticity prediction result of the image and the image authenticity label corresponding to the spliced ​​feature, the loss is calculated, wherein the image authenticity label corresponding to the spliced ​​feature is the image authenticity label of the training image corresponding to the first feature contained in the spliced ​​feature; the parameters of the backbone network, the FBEM and the classifier are adjusted according to the loss to train the image authentication model.

[0006] Optionally, the step of splicing any of the first features and the second features extracted from the two training images through the FBEM to obtain a spliced ​​feature comprises: splicing the first features and the second features from the same training image through the FBEM to obtain a first spliced ​​feature; splicing the first features and the second features from different training images through the FBEM to obtain a second spliced ​​feature; wherein, the step of inputting the spliced ​​feature into the classifier to obtain a true or false prediction result of the image; inputting the first spliced ​​feature into the classifier to obtain a first true or false prediction result of the image; and inputting the second spliced ​​feature into the classifier to obtain a first true or false prediction result of the image. Input the classifier to obtain a second authenticity prediction result of the image; wherein, the calculating the value of the loss function based on the authenticity prediction result of the image and the image authenticity label corresponding to the splicing feature includes: calculating a first loss based on the first authenticity prediction result and the image authenticity label corresponding to the first splicing feature; calculating a second loss based on the second authenticity prediction result and the image authenticity label corresponding to the second splicing feature; and adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss includes: adjusting the parameters of the backbone network, the FBEM and the classifier according to the first loss and the second loss.

[0007] Optionally, adjusting the parameters of the backbone network, the FBEM and the classifier according to the first loss and the second loss includes: performing a weighted summation of the first loss and the second loss to obtain a weighted loss; and adjusting the parameters of the backbone network, the FBEM and the classifier according to the weighted loss.

[0008] Optionally, the training method further includes: calculating the conditional mutual information between the first feature and the second feature decoupled from the two training images, wherein the conditional mutual information is used to characterize the degree of decoupling between the first feature and the second feature; and adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss includes: adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss and the conditional mutual information.

[0009] Optionally, adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss and the conditional mutual information includes: calculating a weighted sum of the loss and the conditional mutual information; and adjusting the parameters of the backbone network, the FBEM and the classifier according to the weighted sum.

[0010] Optionally, the obtaining of the training image samples includes: obtaining two original images with different contents; and tampering with the two original images respectively to obtain the training image samples.

[0011] Optionally, the tampering method includes at least one of the following items: adding Gaussian noise, adding jitter, splicing, copying and pasting, and erasing.

[0012] According to a second aspect of an embodiment of the present disclosure, a method for image authentication based on structural causality is provided. The method for image authentication based on structural causality is implemented based on an image authentication model. The image authentication model includes a backbone network, a forgery bias elimination module (FBEM) and a classifier. The method for image authentication based on structural causality includes: acquiring a target image to be authenticated; extracting a target depth feature of the target image through the backbone network, and decoupling the target depth feature through the FBEM to obtain a decoupled first target feature and a second target feature, wherein the first target feature is a forgery-related feature related to an image tampering area, and the second target feature is a forgery-irrelevant feature unrelated to the image tampering area; splicing the first target feature and the second target feature through the FBEM to obtain a target splicing feature; and inputting the target splicing feature into the classifier to obtain a authenticity prediction result of the target image.

[0013] According to a third aspect of an embodiment of the present disclosure, a training device for an image counterfeit detection model based on structural causality is provided, wherein the image counterfeit detection model comprises a backbone network, a forgery bias elimination module (FBEM) and a classifier, and the training device comprises: an image sample acquisition module, configured to acquire training image samples, wherein the training image samples comprise two training images with different contents, each of the training images corresponding to an image authenticity label and a coordinate position for indicating an image tampering area; a decoupling module, configured to extract a depth feature of each of the two training images through the backbone network, and decouple the depth feature of the training image based on the coordinate position corresponding to the training image through the FBEM to obtain a decoupled first feature and a second feature, wherein the first feature is related to the image authenticity label. a forgery-related feature related to the tampered area, wherein the second feature is a forgery-independent feature that is unrelated to the image tampered area; a splicing module, configured to splice any of the first feature and the second feature extracted from the two training images through the FBEM to obtain a spliced ​​feature; a feature input module, configured to input the spliced ​​feature into the classifier to obtain an authenticity prediction result of the image; a loss calculation module, configured to calculate the loss based on the authenticity prediction result of the image and the image authenticity label corresponding to the splicing feature, wherein the image authenticity label corresponding to the splicing feature is the image authenticity label of the training image corresponding to the first feature contained in the splicing feature; a training module, configured to train the image authentication model by adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss.

[0014] Optionally, the stitching module is configured to: stitch the first feature and the second feature from the same training image through the FBEM to obtain a first stitching feature; stitch the first feature and the second feature from different training images through the FBEM to obtain a second stitching feature; the feature input module is configured to: input the first stitching feature into the classifier to obtain a first authenticity prediction result of the image; input the second stitching feature into the classifier to obtain a second authenticity prediction result of the image; the loss calculation module is configured to: calculate a first loss based on the first authenticity prediction result and the image authenticity label corresponding to the first stitching feature; calculate a second loss based on the second authenticity prediction result and the image authenticity label corresponding to the second stitching feature; the training module is configured to: adjust the parameters of the backbone network, the FBEM and the classifier according to the first loss and the second loss.

[0015] Optionally, the training module is configured to: perform a weighted summation of the first loss and the second loss to obtain a weighted loss; and adjust parameters of the backbone network, the FBEM and the classifier according to the weighted loss.

[0016] Optionally, the training device also includes: a conditional mutual information calculation module, configured to calculate the conditional mutual information between the first feature and the second feature decoupled from the two training images, wherein the conditional mutual information is used to characterize the degree of decoupling between the first feature and the second feature; the training module is configured to: adjust the parameters of the backbone network, the FBEM and the classifier according to the loss and the conditional mutual information.

[0017] Optionally, the training module is configured to: calculate a weighted sum of the loss and the conditional mutual information; and adjust parameters of the backbone network, the FBEM and the classifier according to the weighted sum.

[0018] Optionally, the image sample acquisition module is configured to: acquire two original images with different contents; and tamper with the two original images respectively to obtain the training image samples.

[0019] Optionally, the tampering method includes at least one of the following items: adding Gaussian noise, adding jitter, splicing, copying and pasting, and erasing.

[0020] According to a fourth aspect of an embodiment of the present disclosure, there is provided an image authentication device based on structural causality, wherein the image authentication device based on structural causality is implemented based on an image authentication model, wherein the image authentication model comprises a backbone network, a forgery bias elimination module (FBEM) and a classifier, and the image authentication device based on structural causality comprises: a target image acquisition module, configured to acquire a target image to be authenticated; a feature decoupling module, configured to extract a target depth feature of the target image through the backbone network, and decouple the target depth feature through the FBEM to obtain a decoupled first target feature and a second target feature, wherein the first target feature is a forgery-related feature related to an image tampering area, and the second target feature is a forgery-irrelevant feature unrelated to the image tampering area; a feature splicing module, configured to splice the first target feature and the second target feature through the FBEM to obtain a target splicing feature; and a authenticity result prediction module, configured to input the target splicing feature into the classifier to obtain an authenticity prediction result of the target image.

[0021] According to a fifth aspect of an embodiment of the present disclosure, there is provided an electronic device, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement a training method for an image counterfeit identification model based on structural causality according to the present disclosure, or to implement an image counterfeit identification method based on structural causality according to the present disclosure.

[0022] According to the sixth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the training method of an image counterfeit identification model based on structural causality according to the present disclosure, or to execute the image counterfeit identification method based on structural causality according to the present disclosure.

[0023] According to a seventh aspect of an embodiment of the present disclosure, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the training method of the image counterfeit detection model based on structural causality according to the present disclosure is implemented, or the image counterfeit detection method based on structural causality according to the present disclosure is implemented.

[0024] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects: In the present disclosure, by decoupling the overall features of training images with different contents respectively, and splicing the decoupled forgery-related features and forgery-irrelevant features to train the image authentication model, it can be ensured that the trained image authentication model can accurately identify the tampered areas in the image, that is, it can effectively eliminate the confusion caused by natural areas irrelevant to image authentication, thereby improving the accuracy of image authentication.

[0025] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0027] Figure 1 is a schematic diagram showing the structure of an image counterfeit identification model according to an exemplary embodiment of the present disclosure; Figure 2 is a logic diagram illustrating a structural causal model according to an exemplary embodiment of the present disclosure; Figure 3 is a flowchart illustrating a method for training a structural causal-based image counterfeit detection model according to an exemplary embodiment of the present disclosure; Figure 4is a flow chart showing a method for detecting counterfeit images based on structural causality according to an exemplary embodiment of the present disclosure; Figure 5 is a block diagram showing a training device for a structural causal-based image counterfeit identification model according to an exemplary embodiment of the present disclosure; Figure 6 is a block diagram showing an image counterfeit identification device based on structural causality according to an exemplary embodiment of the present disclosure; Figure 7 is a block diagram illustrating an electronic device according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0028] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation methods described in the following examples do not represent all implementation methods consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the attached claims.

[0030] It should be noted that the phrase "at least one of the items" in the present disclosure includes three types of parallel situations: "any one of the items", "a combination of any number of the items", and "all of the items". For example, "including at least one of A and B" includes the following three types of parallel situations: (1) including A; (2) including B; (3) including A and B. Another example is "executing at least one of step 1 and step 2" which means the following three types of parallel situations: (1) executing step 1; (2) executing step 2; (3) executing step 1 and step 2.

[0031] With the rapid development of visual-language models, it has become increasingly common to generate highly realistic visual media such as images and videos, making high-quality content creation more convenient. However, this also brings the risk of technology being abused, which may lead to the proliferation of false information. Therefore, it is crucial to develop models for detecting visual media forgeries.

[0032] Research in related technologies mainly focuses on identifying the authenticity of content and its forgery location, and emphasizes improving the accuracy and generalization of detection. However, these methods often ignore the importance of explaining why certain results are judged to be true or false, which is crucial to ensure reliable traceability and forensic analysis. In view of this, it is urgent to build a forgery detection model that is both credible and explainable.

[0033] In addition, it should be noted that causal reasoning has gained significant attention in fields such as natural language generation and information retrieval because it can model causal relationships and make interpretable and generalizable decisions by addressing confounding biases in data. In the field of forgery detection, causal learning can clearly demonstrate the impact of confounding variables and help to unravel spurious correlations, thus becoming a reliable tool for improving visual media forgery detection.

[0034] In order to solve the above-mentioned problems existing in the related technology, the image authentication and training method and related equipment based on structural causality provided by the present invention decouple the overall features of training images with different contents respectively, and splice the decoupled forgery-related features and forgery-irrelevant features to train the image authentication model. It can ensure that the trained image authentication model can accurately identify the tampered areas in the image, that is, it can effectively eliminate the confusion interference caused by natural areas irrelevant to image authentication, thereby improving the accuracy of image authentication.

[0035] It should be noted that the data and images involved in this disclosure are information that has been fully authorized by all parties. Figure 1 1 is a schematic diagram showing the structure of an image authentication model according to an exemplary embodiment of the present disclosure. The image authentication model may be a model obtained by improving an existing structural causal model (SCM). The existing structural causal model mainly uses a directed graph to model the causal relationship between variables, and the directed graph may generally be a directed acyclic graph (DAG).

[0036] Figure 2 is a logical diagram illustrating a structural causal model according to an exemplary embodiment of the present disclosure. Figure 2 In the structural causal model, the causal effect from the input image X to the predicted label Y can be defined, and the influence of domain knowledge K on the input image X and the predicted label Y can also be considered, that is, domain knowledge K can be introduced to affect the relationship between input and output. "Domain knowledge K" can refer to different tampering methods, that is, it can refer to various disturbances added to the image.

[0037] It should be noted that the structural causal model in the relevant technology can mainly include a backbone network (BackboneEncoder) and a classifier (classifier). Specifically, the "backbone network" is mainly used to extract the deep features of the image; the "classifier" is mainly used to perform binary classification prediction of the image, that is, the "classifier" is mainly used to predict whether the image is a real image that has not been tampered with or a fake image that has been tampered with.

[0038] In the present disclosure, a "Forgery Bias Eliminating Module (FBEM)" is mainly added on the basis of the existing structural causal model. The "Forgery Bias Eliminating Module" is mainly used to decouple the deep features extracted from the image to obtain forgery-related features related to the image tampering area and forgery-irrelevant features unrelated to the image tampering area. In addition, the "Forgery Bias Eliminating Module" is also used to splice the decoupled forgery-related features and forgery-irrelevant features to obtain splicing features.

[0039] Thus, in the present disclosure, by setting up a forgery bias elimination module (FBEM), effective decomposition of deep features can be achieved, that is, effective decoupling of forgery-related features and forgery-irrelevant features can be ensured, thereby improving the practicality and flexibility of the image forgery detection model in different tampering scenarios.

[0040] Figure 3 is a flow chart showing a training method of a structural causal-based image counterfeit detection model according to an exemplary embodiment of the present disclosure. As mentioned above, the image counterfeit detection model may include a backbone network, a forgery bias elimination module (FBEM) and a classifier.

[0041] Reference Figure 3 In step 301, a training image sample may be obtained, wherein the training image sample may include two training images with different contents, and each training image may correspond to an image authenticity label and a coordinate position for indicating an image tampering area.

[0042] For example, assuming that the training image is a face image, each face image may correspond to an image authenticity label indicating whether the face image has been tampered with, and a coordinate position indicating a specific tampered area. For example, the tampered area in the face image may be, but is not limited to, a nose area, a mouth area, or an eye area, etc.

[0043] According to an exemplary embodiment of the present disclosure, two original images with different contents may be obtained. Then, the two original images may be tampered with respectively to obtain training image samples. Exemplarily, as mentioned above, the original images may be face images. In this case, the face images of two people may be tampered with respectively to obtain training image samples for training an image counterfeit detection model.

[0044] It should be noted that the tampering methods and tampering areas for two original images with different contents can be the same or different. For example, the same tampering method can be used to tamper with the same area of ​​the two original images; or different tampering methods can be used to tamper with the same area of ​​the two original images; or different tampering methods can be used to tamper with different areas of the two original images, etc., and the present disclosure does not make specific restrictions on this.

[0045] In this way, by tampering with the original image to train the image authentication model, it can be ensured that the trained image authentication model can effectively perform binary classification prediction of the image, that is, it can be ensured that the trained image authentication model can effectively distinguish whether the image to be authenticated is a tampered fake image or a non-tampered real image.

[0046] According to an exemplary embodiment of the present disclosure, the aforementioned tampering method may include at least one of the following items: adding Gaussian noise, adding jitter, splicing, copying and pasting, and erasing. It should be noted that the tampering method provided by the present disclosure may be other methods in addition to the above-mentioned methods. The present disclosure does not specifically limit this, and the aforementioned implementation method is merely an exemplary description. In this way, by setting a variety of different image tampering methods, it can be ensured that the trained image counterfeit detection model can cope with various tampering scenarios, and the image counterfeit detection model has good generalization.

[0047] In step 302, for each of the two training images, the depth feature Z of the training image can be extracted through the aforementioned backbone network, and the depth feature Z of the training image can be decoupled based on the coordinate position corresponding to the training image through the aforementioned FBEM to obtain the decoupled first feature and second feature. Specifically, the two projection layers (Projection layer) included in the FBEM: φ and ψ can be used to decouple the depth feature Z into the first feature and the second feature. In addition, the first feature can be a forgery-related feature related to the image tampering area. , the second feature can be a forgery-independent feature that is irrelevant to the image tampering area .

[0048] For example, refer to Figure 1, decoupling the deep feature Z of the i-th training image can obtain the forgery-related features of the i-th training image Features not related to forgery ; Decoupling the deep feature Z of the j-th training image can obtain the forgery-related features of the j-th training image Features not related to forgery .

[0049] In step 303, any first feature and second feature extracted from two training images can be spliced ​​by FBEM to obtain a spliced ​​feature. That is, forgery-related features and forgery-irrelevant features extracted from two training images can be spliced ​​by FBEM to obtain a spliced ​​feature.

[0050] For example, refer to Figure 1 , the forgery-related features can be extracted from two training images through FBEM Features not related to forgery Perform splicing to obtain splicing features: CAT( , ); and, the forgery-related features extracted from the two training images by FBEM Features not related to forgery Perform splicing to obtain splicing features: CAT ( , ), where "CAT()" means splicing along the channel dimension, that is, it can represent connecting features along the depth direction. It should be noted that in the present disclosure, in addition to using the "CAT()" method for feature splicing, other splicing methods can also be used. The present disclosure does not make specific restrictions on this, and the aforementioned splicing method is only an exemplary description.

[0051] In step 304, the splicing features may be input into a classifier to obtain a prediction result of the authenticity of the image. That is, the splicing features may be input into a classifier to determine whether the corresponding image is a tampered fake image or a non-tampered real image.

[0052] In step 305, the loss can be calculated based on the authenticity prediction result of the image and the image authenticity label corresponding to the stitching feature, wherein the image authenticity label corresponding to the stitching feature can be the image authenticity label of the training image corresponding to the first feature contained in the stitching feature, that is, the image authenticity label corresponding to the stitching feature can be the image authenticity label of the training image from which the forgery-related feature contained in the stitching feature comes.

[0053] Exemplary, for the aforementioned splicing features: , the image authenticity label corresponding to the splicing feature can be the forgery-related features contained in the splicing feature The image authenticity label from the i-th training image; or, for the splicing feature: , the image authenticity label corresponding to the splicing feature can be the forgery-related features contained in the splicing feature The image authenticity label from the i-th training image.

[0054] In step 306, the image counterfeit detection model may be trained by adjusting the parameters of the backbone network, FBEM, and classifier according to the loss.

[0055] According to an exemplary embodiment of the present disclosure, the first feature and the second feature from the same training image can be spliced ​​by FBEM to obtain a first spliced ​​feature. Features not related to forgery Perform splicing to obtain the first splicing feature: Furthermore, the first feature and the second feature from different training images can be spliced ​​together by FBEM to obtain a second spliced ​​feature. For example, as mentioned above, the forgery-related feature from the i-th training image can be spliced ​​together by FBEM. and the forged irrelevant features from the jth training image Perform splicing to obtain the second splicing feature: .

[0056] Then, the first splicing feature can be input into the classifier to obtain the first authenticity prediction result of the image. For example, the first splicing feature: Input the classifier to obtain the first authenticity prediction result for the i-th training image; and the second splicing feature can also be input into the classifier to obtain the second authenticity prediction result of the image. Exemplarily, the second splicing feature can be: Input the classifier to obtain the second true or false prediction result of the i-th training image.

[0057] Next, the first loss may be calculated based on the first authenticity prediction result and the image authenticity label corresponding to the first splicing feature. Exemplarily, the first authenticity prediction result and the first splicing feature may be: Corresponding image authenticity labels, calculate the first loss, where the first loss can be a cross entropy loss ; Moreover, the second loss may be calculated based on the second authenticity prediction result and the image authenticity label corresponding to the second splicing feature. Exemplarily, the second authenticity prediction result and the second splicing feature may be: Corresponding image authenticity labels, calculate the second loss, where the second loss can be a decoupling loss .

[0058] Then, the parameters of the backbone network, FBEM and classifier can be adjusted according to the first loss and the second loss.

[0059] It should be noted that the above cross entropy loss It can be calculated by the following formula:

[0060] in, is the decoupled forgery-related features of the i-th training image, is the decoupled forgery-irrelevant feature of the i-th training image, is the image authenticity label corresponding to the i-th training image, BCELoss is the binary cross entropy loss function (Binary Cross Entropy Loss), Represents a "CAT()" connection operation, where, as mentioned above, "CAT()" represents concatenation along the channel dimension, that is, it can represent connecting features along the depth direction.

[0061] The above decoupling loss It can be calculated by the following formula:

[0062] in, is the decoupled forgery-related features of the i-th training image, is the decoupled forgery-irrelevant feature of the j-th training image, is the true or false image label corresponding to the i-th training image, BCELoss is the binary cross entropy loss function (Binary Cross Entropy Loss), Represents a "CAT()" connection operation, where, as mentioned above, "CAT()" represents concatenation along the channel dimension, that is, it can represent connecting features along the depth direction.

[0063] According to an exemplary embodiment of the present disclosure, the first loss and the second loss may be weighted and summed to obtain a weighted loss. and decoupling loss A weighted sum is performed to obtain a weighted loss. Then, the parameters of the backbone network, FBEM, and classifier can be adjusted according to the weighted loss.

[0064] According to an exemplary embodiment of the present disclosure, the conditional mutual information between the first feature and the second feature decoupled from the two training images can also be calculated. , where the conditional mutual information It can be used to characterize the degree of decoupling between the first feature and the second feature, that is, the conditional mutual information It can be used to characterize the degree of decoupling between forgery-related features and forgery-irrelevant features. The smaller it is, the higher the degree of decoupling between features. Therefore, one of the goals of this disclosure is to minimize the conditional mutual information To ensure the full decoupling of forgery-related features and forgery-irrelevant features.

[0065] Exemplary, conditional mutual information It can be calculated by the following formula:

[0066] Where N is the number of training images contained in the training image sample, is the decoupled forgery-related features of the i-th training image, is the decoupled forgery-irrelevant feature of the i-th training image, is the decoupled forgery-independent feature of the j-th training image. When the image authenticity label corresponding to the i-th training image contained in the training image sample is the same as the image authenticity label corresponding to the j-th training image, = 1; otherwise, that is, when the image authenticity label corresponding to the i-th training image contained in the training image sample is different from the image authenticity label corresponding to the j-th training image, = 0.

[0067] Then, we can calculate the conditional mutual information by Adjusting the parameters of the backbone network, FBEM and classifier. Specifically, the weighted summation result L of the aforementioned loss and conditional mutual information can be calculated, and then the parameters of the backbone network, FBEM and classifier can be adjusted according to the weighted summation result L.

[0068] Exemplarily, L can be calculated by the following formula: L= in, To decouple the loss The corresponding weight, Conditional Mutual Information The corresponding weight.

[0069] In this way, in the present disclosure, by introducing the forgery bias elimination loss function and the conditional mutual information regularization term, while ensuring the detection performance of the image counterfeit detection model, the computational cost can also be reduced, so that the structural causal-based image counterfeit detection method provided by the present disclosure can be seamlessly integrated into the existing system.

[0070] In addition, refer to Figure 1 FBEM decouples a total of 4 features, namely: the forgery-related features from the i-th training image , the forged irrelevant features from the i-th training image , the forgery-related features from the jth training image , the forged irrelevant features from the jth training image When these four features are combined, two results can be obtained. The first result is Figure 1 The splicing results are shown in: and ; The second splicing result can be: CAT ( , ) and CAT ( , ).

[0071] For the first concatenation result, it can correspond to its own cross entropy loss , decoupling loss And conditional mutual information ; For the second splicing result, it can also correspond to its own cross entropy loss , decoupling loss And conditional mutual information At this point, the cross entropy loss corresponding to the first concatenation result can be Cross entropy loss corresponding to the second concatenation result Add together to get the total cross entropy loss; similarly, the decoupling loss corresponding to the first concatenation result can be Decoupling loss corresponding to the second splicing result Add together to get the total decoupling loss; you can also use the conditional mutual information corresponding to the first splicing result Conditional mutual information corresponding to the second splicing result Add them together to get the total conditional mutual information.

[0072] Next, a weighted summation may be performed on the total cross entropy loss, the total decoupling loss, and the total conditional mutual information, and then the parameters of the image counterfeit detection model may be adjusted based on the weighted summation result, thereby training the image counterfeit detection model.

[0073] In this disclosure, a method for image forgery detection based on a structural causal model is proposed to explain the inference target of forgery detection from a causal perspective, which improves the accuracy of traditional forgery detection methods in estimating the required causal effects. Specifically, this disclosure introduces a forgery bias elimination module (FBEM), which is mainly used to decouple forgery-related features in feature representation. Features not related to forgery . Furthermore, FBEM can decouple features from the encoder through two projection layers φ and ψ, thus avoiding the addition of additional complexity. Moreover, FBEM is a lightweight feature decoupling plug-in, which is designed to be used in a plug-and-play manner, making it easy to integrate into existing systems, that is, it can be flexibly and seamlessly applied to various scenarios, allowing the corresponding functions to be realized without major modifications to the current model.

[0074] On this basis, the present disclosure also proposes a forgery bias elimination loss function and the conditional mutual information regularization term To ensure that the image forgery detection model can effectively distinguish the aforementioned forgery-related features during the learning process Features not related to forgery , which can enhance the interpretability and robustness of the image forgery detection model while ensuring the lowest computational cost.

[0075] Figure 4 The present invention is a flowchart showing a structural causal-based image authentication method according to an exemplary embodiment of the present invention. The structural causal-based image authentication method can be implemented based on an image authentication model trained by a training method according to the present invention, wherein the image authentication model can include a backbone network, a forgery bias elimination module (FBEM) and a classifier.

[0076] Reference Figure 4 In step 401, a target image to be authenticated may be obtained. For example, the target image to be authenticated may be a person image, an animal image, a natural landscape image, etc. The present disclosure does not specifically limit the type of the target image to be authenticated.

[0077] In step 402, the target depth feature of the target image can be extracted through the backbone network, and the target depth feature can be decoupled through FBEM to obtain the decoupled first target feature and second target feature. Specifically, the target depth feature can be decoupled into the first target feature and the second target feature using the two projection layers φ and ψ included in FBEM. In addition, the first target feature can be a forgery-related feature related to the image tampering area, and the second target feature can be a forgery-irrelevant feature that is irrelevant to the image tampering area.

[0078] In step 403, the first target feature and the second target feature may be spliced ​​together through FBEM to obtain a target spliced ​​feature, that is, the forgery-related features and the forgery-irrelevant features in the target image may be spliced ​​together through FBEM to obtain a target spliced ​​feature.

[0079] In step 404, the target splicing feature can be input into the classifier to obtain the authenticity prediction result of the target image. That is, the target splicing feature can be input into the classifier to perform binary classification on the target image through the classifier, that is, predict whether the target image is a real image that has not been tampered with or a tampered forged image.

[0080] It should be noted that causal inference can effectively identify confounding factors in the data, which can help explain and reliably trace the forgery detection results. Therefore, the present disclosure proposes a forgery detection method based on a structural causal model (SCM), which mainly uses causal relationships to capture the causal effect between the input image and the predicted label, thereby improving the accuracy and interpretability of forgery detection. In addition, the forgery detection method provided by the present disclosure also has high credibility and traceability. Therefore, it can provide a more interpretable and reliable basis for forgery evidence collection, thereby coping with the increasingly serious problem of false information dissemination.

[0081] Figure 5 is a block diagram showing a training apparatus 500 for a structural causal-based image counterfeit detection model according to an exemplary embodiment of the present disclosure. The image counterfeit detection model may include a backbone network, a forgery bias elimination module (FBEM), and a classifier.

[0082] Reference Figure 5 The training device 500 of the structural causal-based image counterfeit detection model may include an image sample acquisition module 501, a decoupling module 502, a splicing module 503, a feature input module 504, a loss calculation module 505 and a training module 506.

[0083] The image sample acquisition module 501 can acquire training image samples, wherein the training image samples can include two training images with different contents, and each training image can correspond to an image authenticity label and a coordinate position for indicating an image tampering area.

[0084] For example, assuming that the training image is a face image, each face image may correspond to an image authenticity label indicating whether the face image has been tampered with, and a coordinate position indicating a specific tampered area. For example, the tampered area in the face image may be, but is not limited to, a nose area, a mouth area, or an eye area, etc.

[0085] According to an exemplary embodiment of the present disclosure, the image sample acquisition module 501 can acquire two original images with different contents. Then, the image sample acquisition module 501 can tamper with the two original images respectively to obtain training image samples. Exemplarily, as mentioned above, the original images can be face images. In this case, the face images of two people can be tampered with respectively, and then the training image samples for training the image counterfeit detection model can be obtained.

[0086] It should be noted that the tampering methods and tampering areas for two original images with different contents can be the same or different. For example, the same tampering method can be used to tamper with the same area of ​​the two original images; or different tampering methods can be used to tamper with the same area of ​​the two original images; or different tampering methods can be used to tamper with different areas of the two original images, etc., and the present disclosure does not make specific restrictions on this.

[0087] In this way, by tampering with the original image to train the image authentication model, it can be ensured that the trained image authentication model can effectively perform binary classification prediction of the image, that is, it can be ensured that the trained image authentication model can effectively distinguish whether the image to be authenticated is a tampered fake image or a non-tampered real image.

[0088] According to an exemplary embodiment of the present disclosure, the aforementioned tampering method may include at least one of the following items: adding Gaussian noise, adding jitter, splicing, copying and pasting, and erasing. It should be noted that the tampering method provided by the present disclosure may be other methods in addition to the above-mentioned methods. The present disclosure does not specifically limit this, and the aforementioned implementation method is merely an exemplary description. In this way, by setting a variety of different image tampering methods, it can be ensured that the trained image counterfeit detection model can cope with various tampering scenarios, and the image counterfeit detection model has good generalization.

[0089] For each of the two training images, the decoupling module 502 can extract the depth feature Z of the training image through the aforementioned backbone network, and can decouple the depth feature Z of the training image based on the coordinate position corresponding to the training image through the aforementioned FBEM to obtain the decoupled first feature and second feature. Specifically, the two projection layers (Projection layer) included in the FBEM: φ and ψ can be used to decouple the depth feature Z into the first feature and the second feature. In addition, the first feature can be a forgery-related feature related to the image tampering area. , the second feature can be a forgery-independent feature that is irrelevant to the image tampering area .

[0090] Thus, in the present disclosure, by setting up a forgery bias elimination module (FBEM), effective decomposition of deep features can be achieved, that is, effective decoupling of forgery-related features and forgery-irrelevant features can be ensured, thereby improving the practicality and flexibility of the image forgery detection model in different tampering scenarios.

[0091] The stitching module 503 can stitch any first feature and second feature extracted from two training images through FBEM to obtain a stitching feature, that is, it can stitch forgery-related features and forgery-irrelevant features extracted from two training images through FBEM to obtain a stitching feature.

[0092] The feature input module 504 can input the spliced ​​features into the classifier to obtain the authenticity prediction result of the image. That is, the spliced ​​features can be input into the classifier to determine whether the corresponding image is a tampered fake image or an untampered real image.

[0093] The loss calculation module 505 can calculate the loss based on the authenticity prediction result of the image and the image authenticity label corresponding to the splicing feature, wherein the image authenticity label corresponding to the splicing feature can be the image authenticity label of the training image corresponding to the first feature contained in the splicing feature, that is, the image authenticity label corresponding to the splicing feature can be the image authenticity label of the training image from which the forgery-related feature contained in the splicing feature comes.

[0094] The training module 506 can train the image counterfeit detection model by adjusting the parameters of the backbone network, FBEM and classifier according to the loss.

[0095] According to an exemplary embodiment of the present disclosure, the stitching module 503 may stitch the first feature and the second feature from the same training image through FBEM to obtain a first stitching feature. In addition, the stitching module 503 may also stitch the first feature and the second feature from different training images through FBEM to obtain a second stitching feature.

[0096] Then, the feature input module 504 may input the first splicing feature into the classifier to obtain a first authenticity prediction result of the image; and the feature input module 504 may also input the second splicing feature into the classifier to obtain a second authenticity prediction result of the image.

[0097] Next, the loss calculation module 505 can calculate the first loss based on the first authenticity prediction result and the image authenticity label corresponding to the first splicing feature; and the loss calculation module 505 can also calculate the second loss based on the second authenticity prediction result and the image authenticity label corresponding to the second splicing feature. Then, the training module 506 can adjust the parameters of the backbone network, FBEM and classifier according to the first loss and the second loss.

[0098] According to an exemplary embodiment of the present disclosure, the training module 506 may perform a weighted summation on the first loss and the second loss to obtain a weighted loss. Then, the training module 506 may adjust the parameters of the backbone network, FBEM and classifier according to the weighted loss.

[0099] According to an exemplary embodiment of the present disclosure, the training device 500 of the structural causal-based image counterfeit detection model may further include a conditional mutual information calculation module. The conditional mutual information calculation module may calculate the conditional mutual information between the first feature and the second feature decoupled from the two training images. , where the conditional mutual information It can be used to characterize the degree of decoupling between the first feature and the second feature, that is, the conditional mutual information It can be used to characterize the degree of decoupling between forgery-related features and forgery-irrelevant features. The smaller it is, the higher the degree of decoupling between features. Therefore, one of the goals of this disclosure is to minimize the conditional mutual information To ensure the full decoupling of forgery-related features and forgery-irrelevant features.

[0100] Then, the training module 506 can be trained by using the aforementioned loss and conditional mutual information Adjusting the parameters of the backbone network, FBEM and classifier. Specifically, the training module 506 can calculate the weighted sum of the aforementioned loss and conditional mutual information, and then adjust the parameters of the backbone network, FBEM and classifier according to the weighted sum.

[0101] In this way, in the present disclosure, by introducing the forgery bias elimination loss function and the conditional mutual information regularization term, while ensuring the detection performance of the image counterfeit detection model, the computational cost can also be reduced, so that the structural causal-based image counterfeit detection method provided by the present disclosure can be seamlessly integrated into the existing system.

[0102] Figure 6 is a block diagram showing an image counterfeit detection device 600 based on structural causality according to an exemplary embodiment of the present disclosure. The image counterfeit detection device 600 based on structural causality can be implemented based on an image counterfeit detection model trained by the training method according to the present disclosure, wherein the image counterfeit detection model can include a backbone network, a forgery bias elimination module (FBEM) and a classifier.

[0103] Reference Figure 6 The image authentication device 600 based on structural causality may include a target image acquisition module 601 , a feature decoupling module 602 , a feature splicing module 603 and a true-false result prediction module 604 .

[0104] The target image acquisition module 601 can acquire the target image to be authenticated. Exemplarily, the target image to be authenticated can be a person image, an animal image, a natural landscape image, etc. The present disclosure does not specifically limit the type of the target image to be authenticated.

[0105] The feature decoupling module 602 can extract the target depth feature of the target image through the backbone network, and can decouple the target depth feature through FBEM to obtain the decoupled first target feature and second target feature. Specifically, the target depth feature can be decoupled into the first target feature and the second target feature using the two projection layers: φ and ψ included in FBEM. In addition, the first target feature can be a forgery-related feature related to the image tampering area, and the second target feature can be a forgery-irrelevant feature that is irrelevant to the image tampering area.

[0106] The feature stitching module 603 can stitch the first target feature and the second target feature through FBEM to obtain a target stitching feature, that is, it can stitch the forgery-related features and the forgery-irrelevant features in the target image through FBEM to obtain a target stitching feature.

[0107] The authenticity prediction module 604 can input the target splicing features into the classifier to obtain the authenticity prediction result of the target image. That is, the target splicing features can be input into the classifier to perform binary classification on the target image through the classifier, that is, predict whether the target image is a real image that has not been tampered with or a tampered forged image.

[0108] Figure 7 is a block diagram illustrating an electronic device 700 according to an exemplary embodiment of the present disclosure.

[0109] Reference Figure 7 The electronic device 700 includes at least one memory 701 and at least one processor 702, wherein the at least one memory 701 stores instructions. When the instructions are executed by the at least one processor 702, a training method for an image counterfeit identification model based on structural causality or an image counterfeit identification method based on structural causality according to an exemplary embodiment of the present disclosure is executed.

[0110] As an example, the electronic device 700 may be a PC, a tablet device, a personal digital assistant, a smart phone, or other device capable of executing the above instructions. Here, the electronic device 700 is not necessarily a single electronic device, but may also be any device or circuit collection capable of executing the above instructions (or instruction sets) individually or in combination. The electronic device 700 may also be part of an integrated control system or system manager, or may be configured as a portable electronic device interconnected with a local or remote (e.g., via wireless transmission) interface.

[0111] In the electronic device 700, the processor 702 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example and not limitation, the processor may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0112] The processor 702 may execute instructions or codes stored in the memory 701, which may also store data. Instructions and data may also be sent and received over a network via a network interface device, which may employ any known transmission protocol.

[0113] The memory 701 may be integrated with the processor 702, for example, by placing RAM or flash memory within an integrated circuit microprocessor or the like. In addition, the memory 701 may include a separate device, such as an external disk drive, a storage array, or any other storage device that can be used by a database system. The memory 701 and the processor 702 may be operatively coupled, or may communicate with each other, such as through an I / O port, a network connection, etc., so that the processor 702 can read files stored in the memory.

[0114] In addition, the electronic device 700 may further include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, a mouse, a touch input device, etc.) All components of the electronic device 700 may be connected to each other via a bus and / or a network.

[0115] According to an exemplary embodiment of the present disclosure, a computer-readable storage medium may also be provided. When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the above-mentioned training method of the image authentication model based on structural causality or the image authentication method based on structural causality. Examples of computer-readable storage media here include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk and any other device, any other device is configured to store computer programs and any associated data, data files and data structures in a non-transitory manner and provide the computer programs and any associated data, data files and data structures to a processor or computer so that the processor or computer can execute the computer program. The computer program in the above-mentioned computer-readable storage medium can be run in an environment deployed in a computer device such as a client, a host, an agent device, a server, etc. In addition, in one example, the computer program and any associated data, data files and data structures are distributed on a networked computer system, so that the computer program and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.

[0116] According to an exemplary embodiment of the present disclosure, a computer program product may also be provided, including a computer program, which, when executed by a processor, implements the training method of the image counterfeit identification model based on structural causality or the image counterfeit identification method based on structural causality according to the present disclosure.

[0117] According to the structural causal-based image authentication and training method and related equipment disclosed in the present invention, by decoupling the overall features of training images with different contents respectively, and splicing the decoupled forgery-related features and forgery-irrelevant features to train the image authentication model, it can be ensured that the trained image authentication model can accurately identify the tampered areas in the image, that is, it can effectively eliminate the confusion caused by natural areas irrelevant to image authentication, thereby improving the accuracy of image authentication.

[0118] According to an exemplary embodiment of the present disclosure, by tampering with the original image to train the image authentication model, it can be ensured that the trained image authentication model can effectively perform binary classification prediction of the image, that is, it can be ensured that the trained image authentication model can effectively distinguish whether the image to be authenticated is a tampered fake image or a non-tampered real image.

[0119] According to an exemplary embodiment of the present disclosure, by setting a variety of different image tampering methods, it can be ensured that the trained image authentication model can cope with various tampering scenarios, and the image authentication model has good generalization.

[0120] According to an exemplary embodiment of the present disclosure, by setting up a forgery bias elimination module (FBEM), effective decomposition of deep features can be achieved, that is, effective decoupling of forgery-related features and forgery-irrelevant features can be ensured, thereby improving the practicality and flexibility of the image forgery detection model in different tampering scenarios.

[0121] According to an exemplary embodiment of the present disclosure, by introducing a forgery bias elimination loss function and a conditional mutual information regularization term, while ensuring the detection performance of the image counterfeit identification model, the computational cost can also be reduced, so that the structural causal-based image counterfeit identification method provided by the present disclosure can be seamlessly integrated into the existing system.

[0122] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0123] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A training method for an image counterfeit detection model based on structural causality, characterized in that: The image forgery detection model comprises a backbone network, a forgery deviation elimination module FBEM and a classifier, and the training method comprises: Acquire training image samples, wherein the training image samples include two training images with different contents, and each of the training images corresponds to an image authenticity label and a coordinate position for indicating an image tampering area; For each of the two training images, extracting the depth features of the training image through the backbone network, and decoupling the depth features of the training image based on the coordinate position corresponding to the training image through the FBEM to obtain a decoupled first feature and a second feature, wherein the first feature is a forgery-related feature related to the image tampering area, and the second feature is a forgery-irrelevant feature unrelated to the image tampering area; splicing any first feature and the second feature extracted from the two training images through the FBEM to obtain a spliced ​​feature; Inputting the splicing features into the classifier to obtain a prediction result of the authenticity of the image; Calculating the loss based on the authenticity prediction result of the image and the image authenticity label corresponding to the splicing feature, wherein the image authenticity label corresponding to the splicing feature is the image authenticity label of the training image corresponding to the first feature contained in the splicing feature; The image counterfeit detection model is trained by adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss.

2. The training method according to claim 1, characterized in that: The step of splicing any first feature and the second feature extracted from the two training images by the FBEM to obtain a spliced ​​feature includes: Splicing the first feature and the second feature from the same training image through the FBEM to obtain a first splicing feature; The first feature and the second feature from different training images are spliced ​​by the FBEM to obtain a second spliced ​​feature; wherein, the splicing feature is input into the classifier to obtain a prediction result of the authenticity of the image; Inputting the first splicing feature into the classifier to obtain a first authenticity prediction result of the image; Inputting the second splicing feature into the classifier to obtain a second authenticity prediction result of the image; The step of calculating the value of the loss function based on the authenticity prediction result of the image and the authenticity label of the image corresponding to the splicing feature includes: Calculating a first loss based on the first authenticity prediction result and the image authenticity label corresponding to the first splicing feature; Calculating a second loss based on the second authenticity prediction result and the image authenticity label corresponding to the second splicing feature; The method of adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss comprises: By adjusting the parameters of the backbone network, the FBEM and the classifier according to the first loss and the second loss.

3. The training method according to claim 2, characterized in that: The method of adjusting the parameters of the backbone network, the FBEM and the classifier according to the first loss and the second loss comprises: Performing a weighted summation on the first loss and the second loss to obtain a weighted loss; By adjusting the parameters of the backbone network, the FBEM and the classifier according to the weighted loss.

4. The training method according to claim 1, characterized in that: The training method further comprises: Calculating conditional mutual information between the first feature and the second feature decoupled from the two training images, wherein the conditional mutual information is used to characterize a degree of decoupling between the first feature and the second feature; The method of adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss comprises: By adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss and the conditional mutual information.

5. The training method according to claim 4, characterized in that: The method of adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss and the conditional mutual information comprises: Calculating a weighted sum of the loss and the conditional mutual information; The parameters of the backbone network, the FBEM and the classifier are adjusted according to the weighted summation result.

6. The training method according to claim 1, characterized in that: The obtaining of training image samples comprises: Get two original images with different contents; The two original images are tampered with respectively to obtain the training image samples.

7. The training method according to claim 6, characterized in that: The tampering method includes at least one of the following items: Add Gaussian noise, add jitter, splice, copy and paste, erase.

8. A method for detecting fake images based on structural causality, characterized in that: The image authentication method based on structural causality is implemented based on an image authentication model, wherein the image authentication model includes a backbone network, a forgery deviation elimination module FBEM and a classifier. The image authentication method based on structural causality includes: Acquire a target image to be authenticated; Extracting a target depth feature of the target image through the backbone network, and decoupling the target depth feature through the FBEM to obtain a decoupled first target feature and a second target feature, wherein the first target feature is a forgery-related feature related to the image tampering area, and the second target feature is a forgery-irrelevant feature unrelated to the image tampering area; Splicing the first target feature and the second target feature by using the FBEM to obtain a target splicing feature; The target splicing features are input into the classifier to obtain a true or false prediction result of the target image.

9. A training device for an image counterfeit detection model based on structural causality, characterized in that: The image forgery detection model comprises a backbone network, a forgery deviation elimination module FBEM and a classifier, and the training device comprises: An image sample acquisition module is configured to acquire training image samples, wherein the training image samples include two training images with different contents, and each of the training images corresponds to an image authenticity label and a coordinate position for indicating an image tampering area; a decoupling module configured to extract, for each of the two training images, a deep feature of the training image through the backbone network, and decouple the deep feature of the training image based on the coordinate position corresponding to the training image through the FBEM to obtain a decoupled first feature and a second feature, wherein the first feature is a forgery-related feature related to the image tampering area, and the second feature is a forgery-irrelevant feature unrelated to the image tampering area; A splicing module is configured to splice any first feature and any second feature extracted from the two training images through the FBEM to obtain a splicing feature; A feature input module, configured to input the splicing feature into the classifier to obtain a prediction result of the authenticity of the image; A loss calculation module is configured to calculate the loss based on the authenticity prediction result of the image and the image authenticity label corresponding to the splicing feature, wherein the image authenticity label corresponding to the splicing feature is the image authenticity label of the training image corresponding to the first feature contained in the splicing feature; The training module is configured to train the image counterfeit detection model by adjusting the parameters of the backbone network, the FBEM and the classifier according to the loss.

10. An image counterfeit detection device based on structural causality, characterized in that: The image counterfeit detection device based on structural causality is implemented based on an image counterfeit detection model, wherein the image counterfeit detection model includes a backbone network, a forgery deviation elimination module FBEM and a classifier. The image counterfeit detection device based on structural causality includes: A target image acquisition module is configured to acquire a target image to be subjected to image authentication; a feature decoupling module, configured to extract a target depth feature of the target image through the backbone network, and decouple the target depth feature through the FBEM to obtain a decoupled first target feature and a second target feature, wherein the first target feature is a forgery-related feature related to the image tampering area, and the second target feature is a forgery-irrelevant feature unrelated to the image tampering area; a feature splicing module, configured to splice the first target feature and the second target feature through the FBEM to obtain a target splicing feature; The authenticity prediction module is configured to input the target splicing feature into the classifier to obtain the authenticity prediction result of the target image.

11. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the training method of the image counterfeit identification model based on structural causality as described in any one of claims 1 to 7, or to implement the image counterfeit identification method based on structural causality as described in claim 8.

12. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the training method of the image counterfeit identification model based on structural causality as described in any one of claims 1 to 7, or to execute the image counterfeit identification method based on structural causality as described in claim 8.

13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the training method of the image counterfeit identification model based on structural causality as claimed in any one of claims 1 to 7 is implemented, or the image counterfeit identification method based on structural causality as claimed in claim 8 is implemented.

Images