Feature fusion-based AIGC image detection method, device and equipment
By adopting feature fusion method in AIGC image detection, combining global and local feature extraction, and using multi-head attention fusion network to generate fusion feature vectors, the problem that the existing technology cannot take into account both global and local artifact feature analysis, and more accurate and reliable detection effects.
Patent Information
- Application Number
- CN202411129492.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-11-15
AI Technical Summary
The prior art cannot take into account the analysis of global and local artifact characteristics of AIGC images, resulting in poor detection effects and difficult to meet actual needs.
Using a feature fusion-based method, the global feature map and global feature embedding vector of the target AIGC image are extracted through the global feature extraction branch, and the local patch information is extracted using the IPSM network layer, and feature fusion analysis is performed in combination with the multi-head attention fusion network to generate fusion features. Vectors are subject to classification detection.
It ensures the accuracy and reliability of the detection results, and can more comprehensively express various characteristics of AIGC images, taking into account global and local artifact feature information, and meeting actual detection needs.
Smart Images

Figure CN118968247A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of image detection technology, and in particular to an AIGC image detection method, device and equipment based on feature fusion. Background Art
[0002] AIGC, or artificial intelligence generated content, mainly includes pictures and videos. In the past few years, generative models have made remarkable progress, and various models, including VAE, GAN, GLOW, Diffusion and their variants have flourished in the field of image synthesis. With the rapid progress of generative models, AIGC images have become more and more realistic and visually indistinguishable, which has also exacerbated people's concerns about the spread of AIGC-free images. Therefore, it is crucial to develop an effective detector that can identify fake images generated by modern generative models. AIGC images can be roughly divided into two categories, namely AI-generated images and manipulation-based images.
[0003] However, existing technologies either focus on global feature extraction but are not sensitive enough to subtle local artifacts; or emphasize local artifacts but ignore global artifact features, which cannot meet the current AIGC image detection requirements and have poor actual implementation effects. Summary of the invention
[0004] The present application provides an AIGC image detection method, device and equipment based on feature fusion, which is used to solve the technical problem that the existing technology cannot take into account both global and local artifact feature analysis, resulting in poor AIGC image detection effect and difficulty in meeting actual needs.
[0005] In view of this, the first aspect of the present application provides an AIGC image detection method based on feature fusion, comprising:
[0006] The target AIGC image is input into the preset global feature extraction branch for global feature extraction analysis to obtain a global feature map and a global feature embedding vector;
[0007] Extracting local patch information from the global feature map through an IPSM network layer and generating a local feature map, wherein the local feature map includes a plurality of local patch coordinates;
[0008] Based on a preset multi-head attention fusion network, feature fusion analysis is performed according to the global feature embedding vector and the local feature map to obtain a fused feature vector;
[0009] The fused feature vector is input into a preset classifier for classification detection to obtain a detection result.
[0010] Preferably, extracting local patch information from the global feature map through the IPSM network layer and generating a local feature map includes:
[0011] The global feature graph is input into the IPSM network layer and aggregated into a feature aggregation graph;
[0012] Using a window of a preset size to calculate the amount of information of each window block on the feature aggregation graph, and obtaining a window block score;
[0013] After sorting the scores of the window blocks, a preset number of high-scoring window blocks are selected as patch information blocks;
[0014] After mapping the coordinates of the patch information block to the target AIGC image, an image cropping operation is performed to obtain a local feature map.
[0015] Preferably, after sorting the scores of the window blocks, selecting a preset number of high-scoring window blocks as patch information blocks further includes:
[0016] The non-maximum suppression method is used to remove redundant window blocks on the feature aggregation graph.
[0017] Preferably, the method further comprises: performing feature fusion analysis based on the global feature embedding vector and the local feature map to obtain a fused feature vector based on a preset multi-head attention fusion network, and the method further comprises:
[0018] Using a preset feature extractor to extract features from the local feature map to obtain a local feature embedding vector;
[0019] Then, the feature fusion analysis is performed based on the preset multi-head attention fusion network according to the global feature embedding vector and the local feature map to obtain a fused feature vector, including:
[0020] Based on a preset multi-head attention fusion network, feature fusion analysis is performed according to the global feature embedding vector and the local feature embedding vector to obtain a fused feature vector.
[0021] Preferably, the method of performing feature fusion analysis based on the preset multi-head attention fusion network according to the global feature embedding vector and the local feature map to obtain a fused feature vector includes:
[0022] Connecting the global feature embedding vector and the embedding vector of the local feature map into a linear embedding vector;
[0023] Based on the scaled dot product attention parameters in the preset multi-head attention fusion network, the fusion features of the linear embedding vector are extracted to obtain a fusion feature vector.
[0024] The second aspect of the present application provides an AIGC image detection device based on feature fusion, comprising:
[0025] A global feature extraction unit, used for inputting the target AIGC image into a preset global feature extraction branch for global feature extraction analysis to obtain a global feature map and a global feature embedding vector;
[0026] A local feature extraction unit, used to extract local patch information from the global feature map through an IPSM network layer, and generate a local feature map, wherein the local feature map includes a plurality of local patch coordinates;
[0027] A feature embedding fusion unit, configured to perform feature fusion analysis based on the global feature embedding vector and the local feature map based on a preset multi-head attention fusion network to obtain a fused feature vector;
[0028] The image classification detection unit is used to input the fused feature vector into a preset classifier for classification detection to obtain a detection result.
[0029] Preferably, the local feature extraction unit comprises:
[0030] A feature aggregation subunit, used for inputting the global feature graph into the IPSM network layer and aggregating it into a feature aggregation graph;
[0031] A block calculation subunit, used to calculate the information amount of each window block on the feature aggregation graph using a window of a preset size to obtain a window block score;
[0032] A sorting and selecting subunit, used for sorting the scores of the window blocks and selecting a preset number of high-scoring window blocks as patch information blocks;
[0033] The mapping and cropping subunit is used to map the coordinates of the patch information block to the target AIGC image and then perform an image cropping operation to obtain a local feature map.
[0034] Preferably, it also includes:
[0035] A local embedding extraction unit, used to extract features from the local feature map using a preset feature extractor to obtain a local feature embedding vector;
[0036] Then, the feature embedding fusion unit is specifically:
[0037] Based on a preset multi-head attention fusion network, feature fusion analysis is performed according to the global feature embedding vector and the local feature embedding vector to obtain a fused feature vector.
[0038] Preferably, the feature embedding fusion unit is specifically used for:
[0039] Connecting the global feature embedding vector and the embedding vector of the local feature map into a linear embedding vector;
[0040] Based on the scaled dot product attention parameters in the preset multi-head attention fusion network, the fusion features of the linear embedding vector are extracted to obtain a fusion feature vector.
[0041] The third aspect of the present application provides an AIGC image detection device based on feature fusion, the device comprising a processor and a memory;
[0042] The memory is used to store program code and transmit the program code to the processor;
[0043] The processor is used to execute the AIGC image detection method based on feature fusion described in the first aspect according to the instructions in the program code.
[0044] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:
[0045] In the present application, an AIGC image detection method based on feature fusion is provided, including: inputting the target AIGC image into a preset global feature extraction branch for global feature extraction analysis to obtain a global feature map and a global feature embedding vector; extracting local patch information in the global feature map through an IPSM network layer, and generating a local feature map, the local feature map including multiple local patch coordinates; based on a preset multi-head attention fusion network, performing feature fusion analysis according to the global feature embedding vector and the local feature map to obtain a fused feature vector; inputting the fused feature vector into a preset classifier for classification detection to obtain a detection result.
[0046] The AIGC image detection method based on feature fusion provided by the present application uses a preset global feature extraction branch to extract the global features of the target AIGC image; and uses the IPSM network layer to extract local patch information in the image; the fused feature vector determined based on this can more comprehensively express the various characteristics of the AIGC image. Image detection is performed while taking into account global and local artifact feature information, which can ensure the accuracy and reliability of the detection results and better meet actual needs. Therefore, the present application can solve the technical problem that the prior art cannot take into account global and local artifact feature analysis, resulting in poor AIGC image detection effect and difficulty in meeting actual needs. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 A flowchart of an AIGC image detection method based on feature fusion provided in an embodiment of the present application;
[0048] Figure 2 A schematic diagram of the structure of an AIGC image detection device based on feature fusion provided in an embodiment of the present application;
[0049] Figure 3A schematic diagram of a feature fusion network operation based on multi-head attention provided in an embodiment of the present application;
[0050] Figure 4 Schematic diagram of the dot product scaling attention calculation process provided in an embodiment of the present application. DETAILED DESCRIPTION
[0051] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0052] For easier understanding, see Figure 1 , an embodiment of an AIGC image detection method based on feature fusion provided by the present application includes:
[0053] Step 101: Input the target AIGC image into a preset global feature extraction branch for global feature extraction analysis to obtain a global feature map and a global feature embedding vector.
[0054] The target AIGC image is an image generated by a generative model, which can be an image generated by artificial intelligence or an image based on manipulation, without limitation here; a manipulated image refers to an image created by tampering with a real image, such as Deepfake.
[0055] The preset global feature extraction branch of this embodiment is constructed based on ResNet-50, which is used to implement advanced global spatial feature learning. The target AIGC image is input into this branch network, and the global feature map can be extracted in the last convolutional layer based on the CNN feature extractor, which is specifically expressed as ,in, represents the extracted global feature map, represents the input target AIGC image, , , They represent the number of channels, image length and width of the input image, that is, the image space size. Global feature map By inputting the average pooling layer and the fully connected layer, the global feature embedding vector can be obtained. The global feature embedding vector contains the spatial and structural information from the input image.
[0056] Step 102: extract local patch information from the global feature map through the IPSM network layer, and generate a local feature map, where the local feature map includes multiple local patch coordinates.
[0057] Furthermore, step 102 includes:
[0058] The global feature map is input into the IPSM network layer and aggregated into a feature aggregation map;
[0059] A window of preset size is used to calculate the information content of each window block on the feature aggregation graph to obtain the window block score;
[0060] After sorting the scores of the window blocks, a preset number of high-scoring window blocks are selected as patch information blocks;
[0061] The coordinates of the patch information block are mapped to the target AIGC image and then image cropping is performed to obtain a local feature map.
[0062] Furthermore, after sorting the scores of the window blocks, a preset number of high-scoring window blocks are selected as patch information blocks, which also includes:
[0063] The non-maximum suppression method is used to remove redundant window blocks on the feature aggregation graph.
[0064] It should be noted that the IPSM network layer is essentially an information block selection module, which is used to extract local artifact information blocks; the local feature extraction branch aims to locate the most informative blocks and extract discriminative local features from them. Assuming the global feature map Regions with higher energy are usually more informative for classification tasks. Based on this, this embodiment locates more useful information from the global feature map. Specifically, the global feature map is first converted into a feature aggregation map, and the global feature maps of C channels are aggregated:
[0065]
[0066] in, Indicates Then, a window of a preset size is moved on the feature aggregation map to obtain different window blocks, and each window block can calculate the corresponding window average score, that is, the window block score. The window block score can reflect the amount of information in the image block, and the preset size of the window can be set according to the actual situation. In this embodiment, it is expressed as Then the score calculation process of each window block is expressed as:
[0067]
[0068] in, The preset size of the window in this embodiment is = ,or = ; The sliding step size of the window can also be set adaptively. The step size selected in this embodiment is 1.
[0069] The window block scores are sorted from small to large or from large to small, as long as they are arranged in a certain regular pattern to facilitate score selection. In order to reduce redundant patches, the present embodiment can also use a non-maximum suppression algorithm to remove redundant window blocks. Moreover, the selected patch blocks are high-score blocks, and the preset number of selected ones can be set as needed, which is not limited here. The preset number of selected patch information blocks are mapped back to the original input image, that is, the target AIGC image, and the image can be cropped according to the coordinates of the information blocks to obtain a local patch with more information, which is the local feature map that needs to be extracted in this embodiment.
[0070] Step 103: Based on the preset multi-head attention fusion network, feature fusion analysis is performed according to the global feature embedding vector and the local feature map to obtain a fused feature vector.
[0071] Furthermore, step 103, before that, also includes:
[0072] A preset feature extractor is used to extract features from the local feature map to obtain a local feature embedding vector;
[0073] Then, step 103 includes:
[0074] Based on the preset multi-head attention fusion network, feature fusion analysis is performed according to the global feature embedding vector and the local feature embedding vector to obtain the fused feature vector.
[0075] Furthermore, step 103 includes:
[0076] Concatenate the global feature embedding vector and the embedding vector of the local feature map into a linear embedding vector;
[0077] Based on the scaled dot product attention parameters in the preset multi-head attention fusion network, the fusion features of the linear embedding vector are extracted to obtain the fusion feature vector.
[0078] It should be noted that the objects of the feature fusion operation are the global feature embedding vector and the embedding vector of the local feature map, so the embedding vector of the local feature map needs to be extracted before the fusion operation. The preset feature extractor can be the feature extractor used in the global feature embedding vector extraction process above, or it can be other designed efficient feature extractors, which are not limited here.
[0079] See also Figure 3In this embodiment, a feature fusion network of a multi-head attention mechanism is designed to perform feature fusion analysis on the global feature embedding vector and the local feature embedding vector, which can effectively combine global and local feature information. The global feature embedding vector and the local feature embedding vector are now connected into a linear embedding vector; then, the linear embedding vector is subjected to feature learning through the multi-head attention feature fusion network constructed in this embodiment, the correlation between the embedding vectors is determined, and they are combined as overall features with the learning weights to obtain a fused feature vector.
[0080] Specifically, the linear embedding vector is input into the multi-head attention feature fusion network as the query Q, key value K and value V; a multi-head attention mechanism with n heads is used; then the i-th attention head Scaling the attention calculation by the dot product yields:
[0081]
[0082] in, , , are parameter matrices, is the normalized exponential function, is the dimension of the key value K; this embodiment uses three attention network layers with four heads to fuse all the extracted features to obtain a fused feature vector. In addition, the detailed process of dot product scaling attention calculation in this embodiment can be found in Figure 4 .
[0083] Connect the attention scores of all heads as the input of the linear transformation to get the value of multi-head attention, ,in is also a parameter matrix. Finally, The one-dimensional vector fused in is used as the required fused feature vector.
[0084] Step 104: Input the fused feature vector into a preset classifier for classification detection to obtain a detection result.
[0085] The fused feature vector is input into a preset classifier to perform classification detection, and the detection result includes true or false; that is, the preset classifier of this embodiment is a binary classification; and the preset classifier can be designed according to actual conditions and is not limited here.
[0086] The AIGC image detection method based on feature fusion provided in the embodiment of the present application adopts a preset global feature extraction branch to extract the global features of the target AIGC image; and adopts the IPSM network layer to extract the local patch information in the image; the fusion feature vector determined based on this can more comprehensively express the various characteristics of the AIGC image. Image detection is performed while taking into account global and local artifact feature information, which can ensure the accuracy and reliability of the detection results and better meet actual needs. Therefore, the embodiment of the present application can solve the technical problem that the prior art cannot take into account global and local artifact feature analysis, resulting in poor AIGC image detection effect and difficulty in meeting actual needs.
[0087] For easier understanding, see Figure 2 , the present application provides an embodiment of an AIGC image detection device based on feature fusion, comprising:
[0088] A global feature extraction unit 201 is used to input the target AIGC image into a preset global feature extraction branch to perform global feature extraction analysis to obtain a global feature map and a global feature embedding vector;
[0089] A local feature extraction unit 202 is used to extract local patch information from the global feature map through the IPSM network layer and generate a local feature map, where the local feature map includes multiple local patch coordinates;
[0090] A feature embedding fusion unit 203 is used to perform feature fusion analysis based on a preset multi-head attention fusion network according to the global feature embedding vector and the local feature map to obtain a fused feature vector;
[0091] The image classification detection unit 204 is used to input the fused feature vector into a preset classifier for classification detection to obtain a detection result.
[0092] Furthermore, the local feature extraction unit 202 includes:
[0093] The feature aggregation subunit 2021 is used to input the global feature map into the IPSM network layer and aggregate it into a feature aggregation map;
[0094] The block calculation subunit 2022 is used to calculate the information amount of each window block on the feature aggregation graph using a window of a preset size to obtain a window block score;
[0095] The sorting and selecting subunit 2023 is used to sort the scores of the window blocks and select a preset number of high-scoring window blocks as patch information blocks;
[0096] The mapping and cropping subunit 2024 is used to map the coordinates of the patch information block to the target AIGC image and then perform an image cropping operation to obtain a local feature map.
[0097] Furthermore, it also includes:
[0098] A local embedding extraction unit 205 is used to extract features from the local feature map using a preset feature extractor to obtain a local feature embedding vector;
[0099] Then, the feature embedding fusion unit 203 is specifically:
[0100] Based on the preset multi-head attention fusion network, feature fusion analysis is performed according to the global feature embedding vector and the local feature embedding vector to obtain the fused feature vector.
[0101] Furthermore, the feature embedding fusion unit 203 is specifically used for:
[0102] Concatenate the global feature embedding vector and the embedding vector of the local feature map into a linear embedding vector;
[0103] Based on the scaled dot product attention parameters in the preset multi-head attention fusion network, the fusion features of the linear embedding vector are extracted to obtain the fusion feature vector.
[0104] The present application also provides an AIGC image detection device based on feature fusion, the device comprising a processor and a memory;
[0105] The memory is used to store the program code and transmit the program code to the processor;
[0106] The processor is used to execute the AIGC image detection method based on feature fusion in the above method embodiment according to the instructions in the program code.
[0107] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0108] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0109] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0110] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions for executing all or part of the steps of the method described in each embodiment of the present application through a computer device (which can be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (full name in English: Read-Only Memory, English abbreviation: ROM), random access memory (full name in English: Random Access Memory, English abbreviation: RAM), disk or optical disk and other media that can store program codes.
[0111] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. An AIGC image detection method based on feature fusion, characterized in that: include: The target AIGC image is input into the preset global feature extraction branch for global feature extraction analysis to obtain a global feature map and a global feature embedding vector; Extracting local patch information from the global feature map through an IPSM network layer and generating a local feature map, wherein the local feature map includes a plurality of local patch coordinates; Based on a preset multi-head attention fusion network, feature fusion analysis is performed according to the global feature embedding vector and the local feature map to obtain a fused feature vector; The fused feature vector is input into a preset classifier for classification detection to obtain a detection result.
2. The AIGC image detection method based on feature fusion according to claim 1 is characterized in that: The extracting local patch information from the global feature map through the IPSM network layer and generating a local feature map includes: The global feature graph is input into the IPSM network layer and aggregated into a feature aggregation graph; Using a window of a preset size to calculate the amount of information of each window block on the feature aggregation graph, and obtaining a window block score; After sorting the scores of the window blocks, a preset number of high-scoring window blocks are selected as patch information blocks; After mapping the coordinates of the patch information block to the target AIGC image, an image cropping operation is performed to obtain a local feature map.
3. The AIGC image detection method based on feature fusion according to claim 2 is characterized in that: After sorting the scores of the window blocks, selecting a preset number of high-scoring window blocks as patch information blocks also includes: The non-maximum suppression method is used to remove redundant window blocks on the feature aggregation graph.
4. The AIGC image detection method based on feature fusion according to claim 1 is characterized in that: Based on the preset multi-head attention fusion network, feature fusion analysis is performed according to the global feature embedding vector and the local feature map to obtain a fused feature vector, and the above also includes: Using a preset feature extractor to extract features from the local feature map to obtain a local feature embedding vector; Then, the feature fusion analysis is performed based on the preset multi-head attention fusion network according to the global feature embedding vector and the local feature map to obtain a fused feature vector, including: Based on a preset multi-head attention fusion network, feature fusion analysis is performed according to the global feature embedding vector and the local feature embedding vector to obtain a fused feature vector.
5. The AIGC image detection method based on feature fusion according to claim 1 is characterized in that: The method of performing feature fusion analysis based on the preset multi-head attention fusion network according to the global feature embedding vector and the local feature map to obtain a fused feature vector includes: Connecting the global feature embedding vector and the embedding vector of the local feature map into a linear embedding vector; Based on the scaled dot product attention parameters in the preset multi-head attention fusion network, the fusion features of the linear embedding vector are extracted to obtain a fusion feature vector.
6. An AIGC image detection device based on feature fusion, characterized in that: include: A global feature extraction unit, used for inputting the target AIGC image into a preset global feature extraction branch for global feature extraction analysis to obtain a global feature map and a global feature embedding vector; A local feature extraction unit, used to extract local patch information from the global feature map through an IPSM network layer, and generate a local feature map, wherein the local feature map includes a plurality of local patch coordinates; A feature embedding fusion unit, configured to perform feature fusion analysis based on the global feature embedding vector and the local feature map based on a preset multi-head attention fusion network to obtain a fused feature vector; The image classification detection unit is used to input the fused feature vector into a preset classifier for classification detection to obtain a detection result.
7. The AIGC image detection device based on feature fusion according to claim 6 is characterized in that: The local feature extraction unit comprises: A feature aggregation subunit, used for inputting the global feature graph into the IPSM network layer and aggregating it into a feature aggregation graph; A block calculation subunit, used to calculate the information amount of each window block on the feature aggregation graph using a window of a preset size to obtain a window block score; A sorting and selecting subunit, used for sorting the scores of the window blocks and selecting a preset number of high-scoring window blocks as patch information blocks; The mapping and cropping subunit is used to map the coordinates of the patch information block to the target AIGC image and then perform an image cropping operation to obtain a local feature map.
8. The AIGC image detection device based on feature fusion according to claim 6, characterized in that: Also includes: A local embedding extraction unit, used to extract features from the local feature map using a preset feature extractor to obtain a local feature embedding vector; Then, the feature embedding fusion unit is specifically: Based on a preset multi-head attention fusion network, feature fusion analysis is performed according to the global feature embedding vector and the local feature embedding vector to obtain a fused feature vector.
9. The AIGC image detection device based on feature fusion according to claim 6, characterized in that: The feature embedding fusion unit is specifically used for: Connecting the global feature embedding vector and the embedding vector of the local feature map into a linear embedding vector; Based on the scaled dot product attention parameters in the preset multi-head attention fusion network, the fusion features of the linear embedding vector are extracted to obtain a fusion feature vector.
10. An AIGC image detection device based on feature fusion, characterized in that: The device comprises a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the AIGC image detection method based on feature fusion according to any one of claims 1-5 according to the instructions in the program code.
Citation Information
Patent Citations
Double-flow network image forgery detection method and system based on image block feature extraction
CN113361474A
Image retrieval method and device based on joint learning, storage medium and electronic equipment
CN114610942A
Forgery detection method and device, storage medium and electronic equipment
CN115830487A
Cited By
Digital fingerprint extraction and traceability method for diffusion model generated image
CN120766321A
