Image detection method and device, storage medium and electronic equipment
By using an improved YOLO model and a bidirectional complementary cross-attention fusion method, the challenge of high-quality image tampering detection was solved, achieving accurate detection of images tampered with by generative models.
Patent Information
- Application Number
- CN202510920631.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-03
- Publication Date
- 2025-11-11
AI Technical Summary
Existing image tampering detection methods are ineffective at detecting high-quality tampered images.
An improved YOLO model is used to extract image features through horizontal and vertical axis convolutional networks, and combined with a bidirectional complementary cross-attention fusion method to detect tampered regions in the image.
It accurately captures the edges and details of tampered areas in images, improving the ability to detect images tampered with by generative models.
Smart Images
Figure CN120932077A_ABST
Abstract
Description
Technical Field
[0001] This specification relates to the field of computer technology, and in particular to an image detection method, apparatus, storage medium, and electronic device. Background Technology
[0002] With the development of AI technology, the quality of images generated by AI is getting higher and higher. For example, high-quality images can be generated by machine learning models such as generators in Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and diffusion models.
[0003] However, this has also led to increasingly higher quality image manipulation using these machine learning models. Currently, traditional image manipulation detection methods struggle to detect such high-quality manipulation. Therefore, effectively detecting image manipulation is a pressing issue that needs to be addressed. Summary of the Invention
[0004] This specification provides an image detection method, apparatus, storage medium, and electronic device to partially solve the problems existing in the prior art.
[0005] The embodiments in this specification adopt the following technical solutions:
[0006] This specification provides an image detection method, the method comprising:
[0007] Acquire the image to be detected;
[0008] The image to be detected is input into a pre-trained image detection model;
[0009] The image detection model is used to extract the first image feature and the second image feature from the image to be detected.
[0010] Perform horizontal convolution on the first image features to obtain horizontal axis features; and perform vertical convolution on the second image features to obtain vertical axis features;
[0011] The horizontal axis features and the vertical axis features are fused to obtain the fused features;
[0012] Based on the fusion features, the tampered regions in the image to be detected are detected.
[0013] This specification provides an image detection device, the device comprising:
[0014] The acquisition module is used to acquire the image to be detected;
[0015] The input module is used to input the image to be detected into a pre-trained image detection model;
[0016] The extraction module is used to extract the first image feature and the second image feature from the image to be detected using the image detection model.
[0017] The convolution module is used to perform horizontal convolution on the first image features to obtain horizontal axis features; and to perform vertical convolution on the second image features to obtain vertical axis features;
[0018] The fusion module is used to fuse the horizontal axis features and the vertical axis features to obtain fused features;
[0019] The detection module is used to detect tampered regions in the image to be detected based on the fusion features.
[0020] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described image detection method.
[0021] This specification provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the image detection method described above.
[0022] The above-described at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects:
[0023] This specification discloses an image detection method. The method extracts first image features and second image features from an image to be detected. The first image features are convolved along the horizontal axis to obtain horizontal-axis features, and the second image features are convolved along the vertical axis to obtain vertical-axis features. The horizontal-axis features and vertical-axis features are fused to obtain fused features. The fused features are then used to detect tampered regions in the image to be detected. By performing horizontal and vertical convolutions on the image features of the image to be detected, this method can accurately capture the edges and details of tampered regions in the image, thereby effectively detecting images tampered with by generative models. Attached Figure Description
[0024] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and are used to explain this specification, but do not constitute an undue limitation thereof. In the drawings:
[0025] Figure 1 This is a flowchart of an image detection method provided in an embodiment of this specification;
[0026] Figure 2 This is a schematic diagram of the channel segmentation process provided in the embodiments of this specification;
[0027] Figure 3This is a schematic diagram illustrating the working process of any convolutional network provided in the embodiments of this specification;
[0028] Figure 4 This is a schematic diagram of an image detection device provided in an embodiment of this specification;
[0029] Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this specification. Detailed Implementation
[0030] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.
[0031] The technical solutions provided in the various embodiments of this specification are described in detail below with reference to the accompanying drawings.
[0032] Figure 1 A flowchart of an image detection method provided in this specification includes the following steps:
[0033] S100: Acquire the image to be detected.
[0034] In the embodiments of this specification, the following are used: Figure 1 The device used to detect whether an image has been tampered with, as shown in the method, can be any electronic device, including computers, mobile phones, tablets, servers, etc. The following explanation uses a server as an example only.
[0035] The server can first obtain the image that needs to be detected, as the image to be detected.
[0036] S102: Input the image to be detected into a pre-trained image detection model.
[0037] The image detection model described in the embodiments of this specification may include the YOLO model, or other models. The following description uses the YOLO model as an example only.
[0038] After the server obtains the image to be detected, it can input the image into the YOLO model.
[0039] S104: Extract the first image feature and the second image feature from the image to be detected using the image detection model.
[0040] S106: Perform horizontal convolution on the first image features to obtain horizontal axis features; and perform vertical convolution on the second image features to obtain vertical axis features.
[0041] S108: The horizontal axis features and the vertical axis features are fused to obtain the fused features.
[0042] Since the backbone network of the traditional YOLO model is mainly composed of several convolutional networks, and these convolutional networks cannot capture the edges and details in images tampered with by modern generative models, the embodiments in this specification change each convolutional network in the YOLO model to a multi-branch horizontal and vertical axis convolutional network. After the server inputs the image to be detected into the image detection model, each improved convolutional network in the image detection model performs the processing shown in steps S104 to S108 on the image to be detected.
[0043] Specifically, in step S104, the image detection model can convolve the image to be detected using different channel convolution kernels in the image detection model to obtain the image features of the image to be detected in each channel, and determine the first image feature and the second image feature of the image to be detected based on the image features of the image to be detected in each channel.
[0044] In this specification, the channel convolution kernel is a 1×1 kernel. Convolving the image to be detected using a 1×1 channel convolution kernel does not change the size of the image, but each channel convolution kernel can generate an image feature. These image features are the image features of different channels, such as... Figure 2 As shown.
[0045] After the server obtains the image features of the image to be detected in different channels using convolutional kernels across several channels, it can determine the image features within the first channel range as the first image features of the image to be detected, and the image features within the second channel range as the second image features of the image to be detected. The first and second channel ranges do not overlap. For example, image features within the range of channels 1 to n can be used as the first image features, and image features within the range of channels n+1 to m can be used as the second image features.
[0046] In step S106, the image detection model can use at least one horizontal convolution kernel to perform horizontal convolution on the first image features, and use at least one vertical convolution kernel to perform vertical convolution on the second image features. In this embodiment, the horizontal convolution kernel is a 1×i kernel, and the horizontal features obtained by performing horizontal convolution on the image to be detected are mainly used to characterize the edge features of the image to be detected on the horizontal axis. The vertical convolution kernel is a j×1 kernel, and the vertical features obtained by performing vertical convolution on the image to be detected are mainly used to characterize the edge features of the image to be detected on the vertical axis.
[0047] Since each improved convolutional network in the YOLO model of this specification has more than one horizontal axis convolutional kernel and more than one vertical axis convolutional kernel, when performing horizontal axis convolution, the image detection model can use each horizontal axis convolutional kernel in the convolutional network to convolve the first image features, obtaining horizontal axis sub-features corresponding to each horizontal axis convolutional kernel, and then fuse the horizontal axis sub-features to obtain the horizontal axis features. Similarly, when performing vertical axis convolution, the image detection model can also use each vertical axis convolutional kernel in the convolutional network to convolve the second image features, obtaining vertical axis sub-features corresponding to each vertical axis convolutional kernel, and then fuse the vertical axis sub-features to obtain the vertical axis features. The fusion of horizontal axis sub-features or vertical axis features can be achieved by direct addition.
[0048] After obtaining the horizontal and vertical axis features, in step S108, the image detection model can fuse the horizontal and vertical axis features to obtain the fused features. Specifically, when fusing the horizontal and vertical axis features, the horizontal and vertical axis features can be directly concatenated along the channel dimension, and the concatenated features are the fused features.
[0049] S110: Detect the tampered region in the image to be detected based on the fusion features.
[0050] For any convolutional network in the image detection model, after the convolutional network obtains the fusion feature through the above steps S104 to S108, the fusion feature is further input into the next convolutional network cascaded with the convolutional network. The next convolutional network continues to process the input fusion feature through the above steps S104 to S108 until the fusion feature obtained by the last convolutional network is input into the fully connected network of the image detection model. Through the fully connected network, the tampered area of the image to be detected is detected based on the fusion feature.
[0051] The above method performs horizontal and vertical convolution on the image features of the image to be detected, which can accurately capture the edges and details of the tampered areas in the image, thus effectively detecting images tampered with by generative models.
[0052] To further improve the detection accuracy of tampered images, each convolutional network of the image detection model in this embodiment, in addition to fusing the horizontal and vertical features obtained from horizontal and vertical axis convolutions, also fuses the original image features and global features of the image to be detected. Fusing the original image features allows the fused features to retain as many features as possible from the original image to be detected, avoiding the gradient vanishing problem that may occur during training as the network grows. Fusing global features ensures that the fused features include not only suspicious edge features in the horizontal and vertical directions of the image to be detected, but also the overall features of the image to be detected (such as semantic features), such as... Figure 3 As shown.
[0053] exist Figure 3 In step S102, after obtaining the image features of the image to be detected in each channel, in step S104, in addition to using the image features within the first and second channel ranges as the first and second image features, the image features within the third channel range can also be determined as the third image features of the image to be detected, and the image features within the fourth channel range can be determined as the fourth image features of the image to be detected. The first, second, third, and fourth channel ranges do not overlap with each other.
[0054] Accordingly, in step S108, a standard two-dimensional convolution can be performed on the fourth image feature to obtain the global feature. Specifically, at least one two-dimensional convolution kernel from the convolutional network can be used to perform a two-dimensional convolution on the fourth image feature, obtaining global sub-features corresponding to each two-dimensional convolution kernel. These global sub-features are then fused using a direct addition method to obtain the global feature. This is similar to the operation of horizontal and vertical axis convolution, the only difference being that the two-dimensional convolution kernel is not 1×i or j×1 in size, but rather j×i, where i and j are both non-1.
[0055] After obtaining the global features, the horizontal axis features, vertical axis features, third image features, and global features can be fused. Specifically, the above four features can still be directly spliced along the channel dimension to obtain the fused features.
[0056] In addition, in order to enable the image detection model to more accurately focus on the edge details in the image to be detected, in the embodiments of this specification, when fusing horizontal and vertical axis features, the KQV mechanism can be used in each convolutional network to fuse the horizontal and vertical axis features based on the bidirectional complementary cross-attention fusion method.
[0057] Specifically, when any convolutional network fuses horizontal and vertical axis features, the horizontal key feature (K, representing the key), horizontal value feature (V, representing the value), and horizontal query feature (Q, representing the query) can be determined based on the horizontal feature. Similarly, the same applies to the vertical feature; the vertical key feature, vertical value feature, and vertical query feature are determined based on the vertical feature. Before determining the horizontal and vertical KQV features, the dimensions of the horizontal and vertical features in the width and height dimensions of the image to be detected can be reduced to 1, keeping only the dimensions in the channel dimension unchanged. Then, the horizontal and vertical KQV features are determined based on the transformed horizontal and vertical features.
[0058] Then, for the horizontal axis features, based on the similarity between the vertical axis query features and the horizontal axis key features, the horizontal axis attention weight matrix is determined, and the horizontal axis value features are then weighted using the horizontal axis attention weight matrix to obtain the attention-weighted horizontal axis features. That is, by querying the horizontal axis K through the vertical axis Q, the attention weight matrix is determined, and attention weighting is applied to the horizontal axis V; this is one direction.
[0059] Similarly, for the vertical axis features, based on the similarity between the horizontal axis query features and the vertical axis key features, a vertical axis attention weight matrix is determined. This matrix is then used to weight the vertical axis value features, resulting in attention-weighted vertical axis features. In other words, by querying the vertical axis K through the horizontal axis Q, the attention weight matrix is determined, and attention weights are applied to the vertical axis V. This represents another approach.
[0060] By using complementary and cross-weighted attention in the two directions mentioned above, the horizontal and vertical features can interact. For example, by increasing the attention weight of the horizontal axis at the same locations where there are edges in the vertical axis, the image detection model will pay more attention to the horizontal features at the locations where there are edges in the vertical axis. Similarly, it will pay more attention to the vertical features at the locations where there are edges in the horizontal axis, thereby enhancing the image detection model's ability to capture edges and details in the image to be detected.
[0061] After obtaining the attention-weighted horizontal and vertical axis features, since these features are only present in the channel dimension and have a value of 1 in the width and height dimensions of the image to be detected, we can first restore these attention-weighted horizontal and vertical axis features in the width and height dimensions of the image to be detected. That is, we restore these attention-weighted horizontal and vertical axis features to the same dimensions as the width and height of the image to be detected. Then, we directly concatenate the restored attention-weighted horizontal and vertical axis features in the channel dimension to obtain the fused features.
[0062] Of course, in addition to the above-mentioned fusion of the original image features and global features of the image to be detected, the attention-weighted horizontal axis features, attention-weighted vertical axis features, third image features and global features can also be spliced together in the channel dimension to obtain fused features, which will not be elaborated here.
[0063] In the embodiments of this specification, the above-mentioned image detection model can be trained using a supervised training method. Specifically, the server can first obtain the image detection model to be trained and sample images, determine the known tampered regions in the sample images as the annotations of the sample images, input the sample images into the image detection model to be trained, and extract the first image features and the second image features from the sample images through the image detection model to be trained. The first image features of the sample images are convolved along the horizontal axis to obtain the sample horizontal axis features, and the second image features of the sample images are convolved along the vertical axis to obtain the sample vertical axis features. The sample horizontal axis features and the sample vertical axis features are fused to obtain the sample fused features. Based on the sample fused features, tampered regions in the sample images are detected. Based on the difference between the detected tampered regions in the sample images and the annotations of the sample images, the loss value of the image detection model to be trained is determined, wherein the difference is positively correlated with the loss value. Finally, the model parameters of the image detection model to be trained are adjusted with the goal of reducing the loss value.
[0064] The server can employ the aforementioned training method, first using a general image dataset as the sample set (i.e., using samples from the general image dataset as sample images) to pre-train the image detection model. Then, it can use the task dataset from the specific image detection scenario as the sample set (i.e., using images from the task dataset as sample images) to fine-tune the pre-trained image detection model. The aforementioned task scenarios can include document image detection scenarios, face image detection scenarios, etc.
[0065] It should be noted that, in the above methods, each convolutional network has more than one horizontal convolution kernel for horizontal convolution, one vertical convolution kernel for vertical convolution, and one two-dimensional convolution kernel for two-dimensional convolution. Therefore, after training the image detection model, using this image detection model to perform... Figure 1 Prior to the image detection method shown, for any convolutional network, each horizontal axis convolutional kernel, each vertical axis convolutional kernel, and each 2D convolutional kernel in the convolutional network can be reparameterized, retaining only one of each of the horizontal axis convolutional kernel, vertical axis convolutional kernel, and 2D convolutional kernel in the convolutional network, to improve performance. Figure 1 The efficiency of the image detection method shown is as follows. Specifically, the horizontal axis convolution kernels in the convolutional network can be scaled down to the same size and directly added together to obtain the retained horizontal axis convolution kernels. The same reparameterization method is used for the vertical axis convolution kernels and the two-dimensional convolution kernels.
[0066] The above is an image detection method provided by the embodiments of this specification. Based on the same idea, this specification also provides corresponding devices, storage media and electronic devices.
[0067] Figure 4 This is a schematic diagram of an image detection device provided in an embodiment of this specification. The device includes:
[0068] The acquisition module 401 is used to acquire the image to be detected;
[0069] Input module 402 is used to input the image to be detected into a pre-trained image detection model;
[0070] Extraction module 403 is used to extract first image features and second image features from the image to be detected through the image detection model;
[0071] The convolution module 404 is used to perform horizontal convolution on the first image features to obtain horizontal axis features; and to perform vertical convolution on the second image features to obtain vertical axis features;
[0072] The fusion module 405 is used to fuse the horizontal axis features and the vertical axis features to obtain fused features;
[0073] The detection module 406 is used to detect the tampered region in the image to be detected based on the fusion features.
[0074] Optionally, the extraction module 403 is specifically used to convolve the image to be detected using different channel convolution kernels in the image detection model to obtain the image features of the image to be detected in each channel; and to determine the first image features and the second image features of the image to be detected based on the image features of the image to be detected in each channel.
[0075] Optionally, the extraction module 403 is specifically used to determine the image features within the first channel range as the first image features of the image to be detected; and to determine the image features within the second channel range as the second image features of the image to be detected.
[0076] Optionally, the convolution module 404 is specifically used to: convolve the first image features using each horizontal axis convolution kernel in the image detection model to obtain horizontal axis sub-features corresponding to each horizontal axis convolution kernel; and fuse the horizontal axis sub-features to obtain horizontal axis features; and convolve the first image features using each vertical axis convolution kernel in the image detection model to obtain vertical axis sub-features corresponding to each vertical axis convolution kernel; and fuse the vertical axis sub-features to obtain vertical axis features.
[0077] Optionally, the fusion module 405 is specifically used to: determine the image features within the third channel range as the third image features of the image to be detected, and determine the image features within the fourth channel range as the fourth image features of the image to be detected; perform two-dimensional convolution on the fourth image features to obtain global features; and fuse the horizontal axis features, the vertical axis features, the third image features, and the global features to obtain fused features.
[0078] Optionally, the fusion module 405 is specifically configured to: determine the horizontal axis key feature, horizontal axis value feature, and horizontal axis query feature corresponding to the horizontal axis feature based on the horizontal axis feature; determine the vertical axis key feature, vertical axis value feature, and vertical axis query feature corresponding to the vertical axis feature based on the vertical axis feature; determine the horizontal axis attention weight matrix based on the similarity between the vertical axis query feature and the horizontal axis key feature, and apply the horizontal axis attention weight matrix to the horizontal axis value feature for attention weighting; determine the vertical axis attention weight matrix based on the similarity between the horizontal axis query feature and the vertical axis key feature, and apply the vertical axis attention weight matrix to the vertical axis value feature for attention weighting; and fuse the attention-weighted horizontal axis value feature and the attention-weighted vertical axis value feature.
[0079] Optionally, the device further includes:
[0080] Training module 407 is used to acquire sample images, identify tampered regions in the sample images as annotations of the sample images; input the sample images into an image detection model to be trained; extract first image features and second image features from the sample images using the image detection model to be trained; perform horizontal convolution on the first image features of the sample images to obtain sample horizontal axis features; and perform vertical convolution on the second image features of the sample images to obtain sample vertical axis features; fuse the sample horizontal axis features and sample vertical axis features to obtain sample fused features; detect tampered regions in the sample images based on the sample fused features; determine the loss value of the image detection model to be trained based on the difference between the detected tampered regions in the sample images and the annotations of the sample images; and adjust the model parameters of the image detection model to be trained with the goal of reducing the loss value.
[0081] Optionally, the image detection model includes the YOLO model.
[0082] This specification also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can be used to perform the image detection method provided above.
[0083] based on Figure 1 The image detection method shown in this specification also provides embodiments that provide... Figure 5 The diagram shows the structure of the electronic device. Figure 5 At the hardware level, the electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to implement the image detection method described above.
[0084] The above description is merely an embodiment of this specification and is not intended to limit this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification.
Claims
1. An image detection method, the method comprising: Acquire the image to be detected; The image to be detected is input into a pre-trained image detection model; The image detection model is used to extract the first image feature and the second image feature from the image to be detected. Perform horizontal convolution on the first image features to obtain horizontal axis features; and perform vertical convolution on the second image features to obtain vertical axis features; The horizontal axis features and the vertical axis features are fused to obtain the fused features; Based on the fusion features, the tampered regions in the image to be detected are detected.
2. The method as described in claim 1, wherein extracting the first image feature and the second image feature from the image to be detected specifically includes: The image to be detected is convolved using different channel convolution kernels in the image detection model to obtain the image features of the image to be detected in each channel. Based on the image features of the image to be detected in each channel, the first image feature and the second image feature of the image to be detected are determined.
3. The method as described in claim 2, wherein determining the first image feature and the second image feature of the image to be detected based on the image features in each channel of the image to be detected, specifically includes: The image features within the first channel range are determined as the first image features of the image to be detected; The image features within the second channel range are determined as the second image features of the image to be detected.
4. The method as described in claim 1, wherein the first image features are convolved along the horizontal axis to obtain horizontal axis features, specifically including: Each horizontal axis convolution kernel in the image detection model is used to convolve the first image feature to obtain the horizontal axis sub-feature corresponding to each horizontal axis convolution kernel. The features of each horizontal axis are fused together to obtain the horizontal axis features; Performing a vertical convolution on the second image features yields vertical axis features, specifically including: Each vertical axis convolution kernel in the image detection model is used to convolve the first image feature to obtain the vertical axis sub-feature corresponding to each vertical axis convolution kernel. The vertical axis features are obtained by fusing the features of each vertical axis.
5. The method as described in claim 2, wherein the horizontal axis features and the vertical axis features are fused to obtain fused features, specifically including: Image features within the third channel range are determined as the third image features of the image to be detected, and image features within the fourth channel range are determined as the fourth image features of the image to be detected. Perform a two-dimensional convolution on the fourth image features to obtain global features; The horizontal axis features, the vertical axis features, the third image features, and the global features are fused to obtain the fused features.
6. The method as described in claim 1, wherein fusing the horizontal axis features and the vertical axis features specifically includes: Based on the horizontal axis features, determine the horizontal axis key features, horizontal axis value features, and horizontal axis query features corresponding to the horizontal axis features; Based on the vertical axis features, determine the vertical axis key features, vertical axis value features, and vertical axis query features corresponding to the vertical axis features; Based on the similarity between the vertical axis query features and the horizontal axis key features, a horizontal axis attention weight matrix is determined, and the horizontal axis value features are weighted by the attention weight matrix. Based on the similarity between the horizontal axis query features and the vertical axis key features, a vertical axis attention weight matrix is determined, and the vertical axis value features are then weighted by the vertical axis attention weight matrix. The attention-weighted horizontal axis features and the attention-weighted vertical axis features are fused together.
7. The method as described in claim 1, wherein the image detection model is pre-trained, specifically comprising: Obtain a sample image, and identify the tampered areas in the sample image as annotations for the sample image; Input the sample image into the image detection model to be trained; The first image feature and the second image feature are extracted from the sample image using the image detection model to be trained. Perform horizontal convolution on the first image feature of the sample image to obtain the sample horizontal axis feature; and perform vertical convolution on the second image feature of the sample image to obtain the sample vertical axis feature; The sample horizontal axis features and sample vertical axis features are fused to obtain the sample fused features; Based on the sample fusion features, the tampered regions in the sample image are detected; The loss value of the image detection model to be trained is determined based on the difference between the detected tampered region in the sample image and the annotation of the sample image. The model parameters of the image detection model to be trained are adjusted with the goal of reducing the loss value.
8. The method according to any one of claims 1 to 7, wherein the image detection model includes the YOLO model.
9. An image detection apparatus, the apparatus comprising: The acquisition module is used to acquire the image to be detected; The input module is used to input the image to be detected into a pre-trained image detection model; The extraction module is used to extract the first image feature and the second image feature from the image to be detected using the image detection model. The convolution module is used to perform horizontal convolution on the first image features to obtain horizontal axis features; and to perform vertical convolution on the second image features to obtain vertical axis features; The fusion module is used to fuse the horizontal axis features and the vertical axis features to obtain fused features; The detection module is used to detect tampered regions in the image to be detected based on the fusion features.
10. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in any one of claims 1-8.
11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method according to any one of claims 1-8.