Image propagation chain detection method and device based on dynamic information compensation and computer

Abstract

The application discloses a kind of based on dynamic information compensation's image propagation chain detection method, device and computer, comprising: obtaining image to be handled, respectively to the feature extraction of image obtains noise domain feature and frequency domain feature;After noise domain feature and frequency domain feature are fused, input encoder and extract the composite feature of image;From the dynamic information compensation mechanism of preestablished extraction compensation feature, and the detection precision of image in platform propagation is improved to realize the enhancement of antecedent platform trace by combining compensation feature with composite feature, and the combined composite feature is input to decoder to detect next platform.The application establishes memory bank by dynamic information compensation mechanism, and adaptively learns the part of antecedent platform trace lost in the propagation process of image due to subsequent platform processing, to improve the detection precision of image in platform propagation.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of image processing, and particularly relates to an image propagation chain detection method and device based on dynamic information compensation and a computer device. BACKGROUND

[0002] With the popularity of social platforms (such as WeChat, Facebook, WhatsApp, etc.), image propagation among multiple platforms has become a common phenomenon. Each platform usually performs a series of customized processing operations on uploaded images, such as renaming, re-compression, size adjustment, etc., which will leave unique "traces" in the images. These traces provide important clues for image propagation chain detection, enabling the complete reconstruction of the propagation path of images on social platforms in chronological order. Image propagation chain detection is an important research task in the field of multimedia forensics. In the current network environment, illegal propagation of images poses a serious security threat, including copyright infringement, privacy leakage, reputation damage, etc. Therefore, monitoring the entire process of image propagation is crucial for maintaining network security.

[0003] There are mainly the following difficulties in studying the problem of image propagation chain: 1) concealment: the traces left by the previous platform processing in the propagation process of the image are often covered or weakened by the compression, size adjustment, etc. of the subsequent platform, resulting in unclear platform processing traces. 2) Incompleteness: images usually undergo compression operations by platforms during propagation, however, existing algorithms usually only rely on DCT histograms to extract compression information. This method is difficult to fully perceive compression features on the one hand; on the other hand, due to the lack of effective use of spatial information, it is difficult to capture the correlation characteristics of pixel-level modifications and compression changes between pixel blocks in the compression process. SUMMARY

[0004] The present application provides an image propagation chain detection method based on dynamic information compensation, comprising:

[0005] Obtaining an image to be processed, and extracting noise domain features and frequency domain features from the image respectively;

[0006] Fusing the noise domain features and the frequency domain features and inputting them into an encoder to extract composite features of the image;

[0007] Extracting compensation features from a preset dynamic information compensation mechanism, combining the compensation features with the composite features to enhance the traces of the previous platform to improve the detection accuracy of the image during platform propagation, and inputting the combined composite features into a decoder to detect the next platform.

[0008] Further, the extracting noise domain features and frequency domain features from the image respectively comprises:

[0009] extract noise information of the image through an SRM conversion model, and perform feature extraction on the noise information through a Backbone_N model to generate the noise domain feature;

[0010] extract binary stereo DCT information of the image through a DCT conversion model, and perform feature extraction through a Backbone_D model to generate the frequency domain feature.

[0011] Further, the fusion of the noise domain feature and the frequency domain feature after the fusion is input into an encoder to extract the composite feature of the image, comprising:

[0012] The noise domain feature and the frequency domain feature are spliced according to the channel and input into the fusion model for feature fusion;

[0013] The fused feature is spliced with the container feature and input into the Transformer encoder to obtain the composite feature of the image.

[0014] Further, the dynamic information compensation mechanism comprises a dynamic memory bank for storing context and compensation features, and an addressing module for accessing the dynamic memory bank, the compensation features are used to capture the feature information lost in the processing of the image during the propagation process on the previous platform, and the context represents the encoding information of the propagation chain.

[0015] Further, the compensation feature is extracted from the preset dynamic information compensation mechanism, and the compensation feature is combined with the composite feature, comprising:

[0016] The addressing module takes the propagation sub-chain input by the decoder as a query condition, retrieves the most relevant context information from the storage unit of the dynamic memory bank, and extracts the corresponding compensation feature;

[0017] The compensation feature is combined with the composite feature set to enhance the composite feature.

[0018] Further, the compensation feature is extracted from the preset dynamic information compensation mechanism, and the compensation feature is combined with the composite feature, further comprising:

[0019] The gradient descent method is used to minimize the cross-entropy loss function, and the dynamic information compensation mechanism is continuously optimized and updated.

[0020] The application provides an image propagation chain detection device based on dynamic information compensation, comprising:

[0021] The feature acquisition module is used for acquiring an image to be processed, and respectively extracting noise domain features and frequency domain features from the image;

[0022] a feature processing module, configured to input the noise domain feature and the frequency domain feature into an encoder after fusion to extract a composite feature of the image;

[0023] a feature detection module, configured to extract a compensation feature from a preset dynamic information compensation mechanism, combine the compensation feature with the composite feature to realize pre-sequence platform trace enhancement to improve detection accuracy of the image when propagating on a platform, and input the combined composite feature to a decoder to detect a next platform.

[0024] Further, the feature acquisition module comprises:

[0025] a first feature acquisition submodule, configured to extract noise information of the image through an SRM conversion model, and extract features of the noise information through a Backbone_N model to generate the noise domain feature;

[0026] a second feature acquisition submodule, configured to extract binary stereo DCT information of the image through a DCT conversion model, and extract features through a Backbone_D model to generate the frequency domain feature.

[0027] Further, the feature processing module comprises:

[0028] a first feature processing submodule, configured to input the noise domain feature and the frequency domain feature into a fusion model for feature fusion according to a channel splicing manner;

[0029] a second feature processing submodule, configured to splice the fused feature with a container feature, and input into a Transformer encoder to obtain the composite feature of the image.

[0030] Further, the dynamic information compensation mechanism comprises a dynamic memory bank for storing context and compensation features, and an addressing module for accessing the dynamic memory bank, wherein the compensation feature is used to capture feature information lost in the process of propagation on a previous platform, and the context represents encoding information of a propagation chain.

[0031] Further, the feature detection module comprises:

[0032] a first feature detection submodule, configured to search for the most relevant context information from a storage unit of the dynamic memory bank by taking a propagation sub-chain input by the decoder as a query condition, and extract a corresponding compensation feature;

[0033] a second feature detection submodule, configured to combine the compensation feature with the composite feature to enhance the composite feature.

[0034] Further, the feature detection module further comprises:

[0035] The third feature detection submodule is configured to continuously optimize and update the dynamic information compensation mechanism by using a gradient descent method to minimize a cross-entropy loss function.

[0036] The present application provides a computer device comprising a memory and a processor, wherein the memory stores computer readable instructions, and the computer readable instructions are executed by the processor to enable the processor to perform the steps of the image propagation chain detection method based on dynamic information compensation.

[0037] The present application provides a storage medium storing computer readable instructions, and the computer readable instructions are executed by one or more processors to enable the one or more processors to perform the steps of the image propagation chain detection method based on dynamic information compensation.

[0038] The beneficial effects of the embodiments of the present application are as follows: the present application establishes a memory bank through a dynamic information compensation mechanism, and adaptively learns the part of the previous platform traces lost in the propagation process of the image due to subsequent platform processing. In the reasoning stage, the stored compensation information is extracted from the memory bank to enhance the image features, thereby effectively strengthening the previous platform traces in the image. To solve the problem of incompleteness, the present application further proposes a double-flow image feature extraction framework, which extracts the compression information of the image in the frequency domain and the residual domain at the same time, and comprehensively captures the compression features. In addition, the binary stereo DCT is used to replace the traditional DCT histogram, focusing on the pixel-level modification generated in the compression process, enhancing the perception of spatial changes caused by compression operations, and improving the extraction effect of platform compression features. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0040] Figure 1 A flow chart of an image propagation chain detection method based on dynamic information compensation provided by the present application is shown in the figure.

[0041] Figure 2 A network framework schematic diagram of an image propagation chain detection method based on dynamic information compensation provided by the present application is shown in the figure.

[0042] Figure 3 A Backbone_D model structure schematic diagram provided by the present application is shown in the figure.

[0043] Figure 4A Backbone_N model structure schematic diagram provided by the present application is shown in the figure;

[0044] Figure 5 An SRM structure schematic diagram provided by the present application is shown in the figure;

[0045] Figure 6 A ResBlock structure schematic diagram provided by the present application is shown in the figure;

[0046] Figure 7 A basic structure block diagram of an image propagation chain detection device based on dynamic information compensation provided by the present application is shown in the figure;

[0047] Figure 8 A basic structure block diagram of a computer device provided by the present application is shown in the figure. DETAILED DESCRIPTION

[0048] In order to enable personnel in the art to better understand the present application scheme, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present application.

[0049] As shown in the figure, Figure 1 The present application provides an image propagation chain detection method based on dynamic information compensation, comprising:

[0050] S1, obtaining an image to be processed, and respectively extracting features of the image to obtain noise domain features and frequency domain features;

[0051] S2, inputting the noise domain features and the frequency domain features after fusion into an encoder to extract composite features of the image;

[0052] S3, extracting compensation features from a preset dynamic information compensation mechanism, combining the compensation features with the composite features to realize pre-sequence platform trace enhancement to improve the detection accuracy of the image when propagating on a platform, and inputting the combined composite features into a decoder to detect a next platform.

[0053] In S1 of the present application, respectively extracting features of the image to obtain noise domain features and frequency domain features, comprising:

[0054] Step one, extracting noise information of the image through an SRM conversion model, and extracting features of the noise information through a Backbone_N model to generate the noise domain features;

[0055] Step two, extracting binary stereo DCT information of the image through a DCT conversion model, and extracting features through a Backbone_D model to generate the frequency domain features.

[0056] Please refer to Figure 2As shown, this invention extracts noise and DCT information from the input image using SRM and DCT conversion models, respectively. The DCT conversion model also converts the DCT information into binary stereo DCT information to comprehensively capture the compression features during image propagation. These features are then extracted using Backbone_N and Backbone_D, respectively, to generate noise domain features and frequency domain features. Specifically, the dual-stream image feature extraction process is as follows:

[0057] 1. Extraction of frequency domain flow features:

[0058] The image to be processed is converted into a DCT representation using a DCT transformation model. Then will Converted to binary stereo DCT form The specific steps are as follows: First, for... The cropping is performed, with the range limited to a threshold [-T, T], where T is 20. Next, the cropped value is binary-coded and transformed into... The specific formula is as follows:

[0059]

[0060] in, , Represents the absolute value of each element. Indicates will Operations constrained within the range [-T, T]:

[0061] Then And the quantization table is passed through Backbone_D, such as Figure 3 As shown, we obtain The characteristics of Backbone_D. Backbone_D first uses a... A dilated convolution with an inflation rate of 8 and a Convolution to obtain features Next, copy the quantization table Q to dimension 256. 256 1, and again Multiplication yields characteristics ,then and They were reshaped into The features are then concatenated, and finally processed through a regular convolutional layer to obtain... Frequency domain characteristics .

[0062] 2. Extraction of flow features in the noise domain:

[0063] First, convert the three channels of the image to be processed, such as... Figure 5 As shown, nine feature maps are generated by processing through three convolutional layers in the SRM model. These feature maps are then concatenated by channel and input into Backbone_N, as shown. Figure 4 As shown, it first goes through 4 layers of convolution to obtain Then the input is fed into the ResBlock and max pooling layers to obtain... Continue to obtain through ResBlock Finally, it goes through a max pooling layer to obtain ,in , , and All dimensions are Then, the noise features are concatenated according to channels, passed through two convolutional layers, then through a cSE layer, and finally obtained. Its dimensions are 64, of which ResBlock is such as Figure 6 As shown.

[0064] In embodiment S2 of the present invention, the composite features of the image are extracted by fusing the noise domain features and the frequency domain features and then inputting the fused features into the encoder. This includes:

[0065] Step 1: Input the noise domain features and the frequency domain features into the fusion model by concatenating them according to channels for feature fusion;

[0066] Step 2: The fused features are concatenated with the container features and input into the Transformer encoder to obtain the composite features of the image.

[0067] In one embodiment of the present invention, frequency domain features are... and noise characteristics After concatenation by channel, the data is input into the fusion module. The fusion module consists of fully connected layers, which convert the feature dimensions from... Mapped to 64, finally flattened as 64, then concatenate it with the container features of the image to obtain Finally, it is fed into the encoder to obtain the composite feature. It should be noted that the Transformer encoder used is an attention-free Transformer encoder.

[0068] In the embodiment of the present application, the dynamic information compensation mechanism comprises a dynamic memory bank for storing context and compensation features, and an addressing module for accessing the dynamic memory bank, wherein the compensation features are used to capture feature information lost in the process of propagation of the image on the previous platform, and the context represents encoding information of the propagation chain.

[0069] The compensation features are extracted from the preset dynamic information compensation mechanism, and the compensation features are combined with the composite features, comprising:

[0070] Step one, the addressing module takes the propagation sub-chain input by the decoder as a query condition, retrieves the most relevant context information from the storage unit of the dynamic memory bank, and extracts the corresponding compensation features;

[0071] Step two, the compensation features are combined with the composite feature set to enhance the composite features.

[0072] In one embodiment of the present application, as shown in the figure, Figure 2 the dynamic information compensation mechanism comprises a dynamic memory bank and an addressing module.

[0073] The dynamic memory bank is composed of storage units, each of which contains two types of information: compensation features and context , the compensation features are used to capture feature information lost in the process of propagation of the image on the previous platform, and the context represents encoding information of the propagation sub-chain.

[0074] Suppose there are propagation chains, the th propagation sub-chain of the th propagation chain is allocated to the th index of the storage unit, wherein , the th propagation sub-chain is as follows:

[0075]

[0076] wherein, represents the number of propagation of the th propagation chain, represents the th platform on the propagation sub-chain . In this embodiment, the embedding vector of is randomly initialized , and the zero vector is used to initialize .

[0077] The addressing module uses the propagation subchain output by the decoder as a query to retrieve the most relevant context information from the storage units of the memory bank and extract the corresponding compensation features.

[0078] For the image to be processed, the encoder generates composite features. And in time t Decode the propagation subchain of the corresponding propagation chain. :

[0079]

[0080] in, This indicates that the propagation subchain has been decoded. The first One platform.

[0081] Enhancement using compensating features from the memory bank Specifically, regarding time t The decoding process will combine features and decoded propagation subchain Input into the following equation to generate the query code. :

[0082]

[0083]

[0084] in, , , , This indicates the average in the first dimension. Indicates position code, , and Represents the weight matrix. This indicates the transpose operation. express Dimensions.

[0085] Subsequently, the encoding is selected and queried from the memory. Most similar front K Enhancement based on compensation features corresponding to each context Their weights are used Softmax The function is determined as follows:

[0086]

[0087] in, Represents cosine similarity. The first two lines represent the highest similarity. K The context of each storage unit. Then, the decoded propagation subchain is computed in time. Compensation features :

[0088]

[0089] wherein is a compensation feature corresponding to .

[0090] The is weighted summed with to obtain an enhanced composite feature . The memory bank is adaptively learned during the training process and .

[0091] The composite feature and the decoded propagation sub-chain are fed into an attention-free Transformer decoder for predicting the next platform. One embodiment of the invention further comprises:

[0092] The dynamic information compensation mechanism is continuously optimized and updated using gradient descent method to minimize the cross-entropy loss function.

[0093] wherein the loss function is , the neural network is optimized through the cross-entropy loss function, so that the predicted propagation chain approximates the real propagation chain .

[0094] As shown in Table 1, the R-SMUD and V-SMUD datasets used are existing image propagation chain datasets. Each column shows the accuracy of the model evaluated on subsets of the propagation chain length from 1 to 3 for both datasets. For example, C3 represents a picture propagating across 3 platforms. The metric used in the table is accuracy, which measures the accuracy of the model's detection results.

[0095] Table 1. Comparison of experimental results of different image propagation chain detection methods (accuracy%)

[0096]

[0097] Table 1 compares the experimental results of different image propagation chain detection methods, based on these results, the following conclusions are drawn: the method proposed in the present application (Ours) is superior to the current advanced method on all propagation sub-chains of each dataset. Compared with the algorithm P-CNN-FF of the convolutional neural network and the algorithm Ve method based on the random forest classifier, the method proposed in the present application achieves significant improvement in detection performance. For example, for the C3 subset of the R-SMUD and V-SMUD datasets, P-CNN-FF and Ve method cannot exceed 50% accuracy, while the accuracy of the method proposed in the present application is 60.35% and 65.42% respectively, which shows that the double-flow image propagation chain detection method based on the dynamic information compensation mechanism designed in the present application can better capture the correlation between platforms compared with the traditional classification method, thereby improving the detection performance. Compared with the sequence-to-sequence image propagation chain detection algorithm Seq2Seq, the method proposed in the present application achieves greater performance improvement. For example, the accuracy of Seq2Seq on the C3 subset of R-SMUD and V-SMUD is 50.68% and 59.68% respectively, while the prediction accuracy of the present application is 60.35% and 65.42% respectively, which increases by 9.67% and 5.74% respectively, indicating that the double-flow image propagation chain detection method based on the dynamic information compensation mechanism designed in the present application can better enhance the trace loss of the previous platform in the image propagation process compared with Seq2Seq, thereby improving the detection performance.

[0098] Therefore, the present application proposes and designs a double-flow image feature extraction method to extract the compression traces of the image from the frequency domain and the residual domain at the same time, so as to realize comprehensive capture of the compression features. In the frequency domain, a binary stereo DCT is used to replace the traditional DCT histogram, which focuses on the pixel-level modification caused by the compression operation and is conducive to perceiving the spatial changes in compression.

[0099] In addition, the dynamic information compensation mechanism proposed in the present application dynamically enhances the trace of the previous platform by using the decoded propagation sub-chain information, thereby improving the accuracy of the propagation chain detection.

[0100] As Figure 7As shown, the application provides an image propagation chain detection device based on dynamic information compensation, comprising: a feature acquisition module 2100, configured to acquire an image to be processed, and perform feature extraction on the image respectively to obtain noise domain features and frequency domain features; a feature processing module 2200, configured to input the noise domain features and the frequency domain features into an encoder after fusion to extract composite features of the image; and a feature detection module 2300, configured to extract compensation features from a preset dynamic information compensation mechanism, combine the compensation features with the composite features to realize pre-sequence platform trace enhancement to improve detection accuracy of the image in platform propagation, and input the combined composite features into a decoder to detect a next platform.

[0101] In some embodiments, the feature acquisition module 2100 comprises: a first feature acquisition submodule, configured to extract noise information of the image through an SRM conversion model, and perform feature extraction on the noise information through a Backbone_N model to generate the noise domain features; and a second feature acquisition submodule, configured to extract binary stereo DCT information of the image through a DCT conversion model, and perform feature extraction through a Backbone_D model to generate the frequency domain features.

[0102] In some embodiments, the feature processing module 2200 comprises: a first feature processing submodule, configured to input the noise domain features and the frequency domain features into a fusion model for feature fusion according to channels; and a second feature processing submodule, configured to splice the fused features with container features, and input the same into a Transformer encoder to obtain the composite features of the image.

[0103] In some embodiments, the dynamic information compensation mechanism comprises: a dynamic memory bank configured to store contexts and compensation features, and an addressing module configured to access the dynamic memory bank, wherein the compensation features are used to capture feature information lost in the process of propagation of the image in a previous platform, and the contexts represent encoding information of a propagation chain.

[0104] In some embodiments, the feature detection module 2300 comprises: a first feature detection submodule, configured to take a propagation sub-chain input by the decoder as a query condition, retrieve the most relevant context information from a storage unit of the dynamic memory bank, and extract corresponding compensation features; and a second feature detection submodule, configured to combine the compensation features with the composite features to enhance the composite features.

[0105] In some embodiments, the feature detection module 2300 further comprises: a third feature detection submodule, configured to minimize a cross-entropy loss function by using a gradient descent method to continuously optimize and update the dynamic information compensation mechanism.

[0106] The application establishes a memory bank through a dynamic information compensation mechanism, and adaptively and dynamically learns the part of the previous platform traces lost in the propagation process of the image due to subsequent platform processing. In the reasoning stage, the stored compensation information is extracted from the memory bank for enhancing the image features, thereby effectively strengthening the previous platform traces in the image. To solve the incompleteness problem, the application further proposes a double-flow image feature extraction framework, which fully captures the compression features by extracting the compression information of the image in the frequency domain and the residual domain at the same time. In addition, the binary stereo DCT is used to replace the traditional DCT histogram, which focuses on the pixel-level modification generated in the compression process, enhances the perception of the spatial changes caused by the compression operation, and improves the extraction effect of the platform compression features.

[0107] To solve the above technical problems, the embodiment of the application further provides a computer device. Figure 8 , Figure 8 The basic structure block diagram of the computer device is shown in the figure.

[0108] As shown in the figure, Figure 8 the internal structure schematic diagram of the computer device. As shown in the figure, Figure 8 the computer device comprises a processor, a non-volatile storage medium, a memory and a network interface connected through a system bus. The non-volatile storage medium of the computer device stores an operating system, a database and computer readable instructions, the database can store a control information sequence, and the computer readable instructions can make the processor realize an image propagation chain detection method based on dynamic information compensation when executed by the processor. The processor of the computer device is used to provide computing and control ability to support the operation of the whole computer device. The memory of the computer device can store computer readable instructions, which can make the processor execute an image processing method when executed by the processor. The network interface of the computer device is used to connect and communicate with the terminal. Those skilled in the art can understand that the structure shown in the figure, Figure 8 only the block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or less components than those shown in the figure, or combine certain components, or have different component arrangement.

[0109] In the embodiment, the processor is used to execute Figure 7The specific content of the middle feature acquisition module 2100, the feature processing module 2200 and the feature detection module 2300, the memory stores the program code and various data required for executing the above-mentioned modules. The network interface is used for data transmission between the user terminal or the server. The memory in the embodiment stores the program code and data required for executing all sub-modules in the image propagation chain detection method based on dynamic information compensation, and the server can call the program code and data of the server to execute the functions of all sub-modules.

[0110] The computer device provided by the embodiment of the application establishes a memory bank through a dynamic information compensation mechanism, and adaptively and dynamically learns part of the previous platform traces lost in the propagation process of an image due to subsequent platform processing. In the reasoning stage, the stored compensation information is extracted from the memory bank to enhance the image features, thereby effectively strengthening the previous platform traces in the image. To solve the problem of incompleteness, the embodiment further proposes a double-flow image feature extraction framework, which comprehensively captures the compression features by extracting the compression information of the image in the frequency domain and the residual domain at the same time. In addition, a binary stereo DCT is used to replace the traditional DCT histogram, which focuses on the pixel-level modification generated in the compression process, enhances the perception of the spatial changes caused by the compression operation, and improves the extraction effect of the platform compression features.

[0111] The application further provides a storage medium storing computer readable instructions, which are executed by one or more processors to enable the one or more processors to perform the steps of the image propagation chain detection method based on dynamic information compensation according to any of the above-mentioned embodiments.

[0112] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments. The storage medium can be a non-volatile storage medium such as a magnetic disc, an optical disc, a read-only memory (ROM) or a random access memory (RAM).

[0113] It should be understood that although the steps in the flowcharts of the drawings are shown in sequence according to the indication of the arrows, the steps are not necessarily executed in sequence according to the indication of the arrows. Unless explicitly stated herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other sequences. Moreover, at least part of the steps in the flowcharts of the drawings can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or sub-steps or stages of other steps.

[0114] The above only describes some embodiments of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.

Claims

1. A method for image propagation chain detection based on dynamic information compensation, characterized in that, The method comprises the following steps: acquiring an image to be processed, and extracting noise domain features and frequency domain features from the image respectively; inputting the noise domain features and the frequency domain features into an encoder after fusion to extract composite features of the image; extracting compensation features from a preset dynamic information compensation mechanism, combining the compensation features with the composite features to realize pre-sequence platform trace enhancement, and improving detection accuracy of the image in platform propagation, and inputting the combined composite features into a decoder to detect a next platform; wherein the dynamic information compensation mechanism comprises a dynamic memory bank for storing context and compensation features, and an addressing module for accessing the dynamic memory bank, the compensation features are used to capture feature information lost in the process of image propagation in the previous platform, and the context represents the encoding information of the propagation chain; wherein the extraction of compensation features from the preset dynamic information compensation mechanism and the combination of the compensation features with the composite features comprise: the addressing module uses the decoded propagation sub-chain output by the decoder as a query condition to retrieve the most relevant context information from the storage unit of the dynamic memory bank, and extracts the corresponding compensation features; the compensation features are combined with the composite features to enhance the composite features.

2. The detection method according to claim 1, characterized in that, The extraction of noise domain features and frequency domain features from the image respectively comprises: extracting noise information of the image through an SRM conversion model, and passing the noise information through a Backbone_N model through 4 convolution layers to obtain , input into a ResBlock and a maximum pooling layer to obtain , through the ResBlock to obtain , and then through the maximum pooling layer to obtain , and , , and are spliced according to the channel, then passed through two convolution layers, and then generated through a cSElayer to obtain the noise domain feature. extracting binary stereo DCT information of the image through a DCT conversion model, and performing feature extraction through a Backbone_D model to generate the frequency domain feature, specifically comprising: performing binary stereo DCT information of the image through dilated convolution with a dilation rate of 8 and convolution to obtain feature Copy the quantization table Q to the dimension 256 256 1, and multiply it by to obtain feature , and are respectively reshaped into features of and spliced, and the spliced features are processed through a normal convolution layer to obtain the frequency domain feature of .

3. The method of claim 1, wherein The inputting of the noise domain features and the frequency domain features into the encoder after fusion to extract the composite features of the image comprises: the noise domain features and the frequency domain features are spliced according to channels and input into a fusion model for feature fusion; the fused features are spliced with container features and input into a Transformer encoder to obtain the composite features of the image.

4. The method of claim 1, wherein The extraction of compensation features from the preset dynamic information compensation mechanism and the combination of the compensation features with the composite features further comprise: the gradient descent method is used to minimize the cross-entropy loss function to continuously optimize and update the dynamic information compensation mechanism.

5. An image propagation chain detection apparatus based on dynamic information compensation, characterized by, The method comprises the following steps: a feature acquisition module is configured to acquire an image to be processed, and extract noise domain features and frequency domain features from the image respectively; a feature processing module is configured to input the noise domain features and the frequency domain features into an encoder after fusion to extract composite features of the image; a feature detection module is configured to extract compensation features from a preset dynamic information compensation mechanism, combine the compensation features with the composite features to realize pre-sequence platform trace enhancement, improve detection accuracy of the image in platform propagation, and input the combined composite features into a decoder to detect a next platform; wherein the dynamic information compensation mechanism comprises a dynamic memory bank for storing context and compensation features, and an addressing module for accessing the dynamic memory bank, the compensation features are used to capture feature information lost in the process of image propagation in the previous platform, and the context represents the encoding information of the propagation chain; the feature detection module comprises: The first feature detection submodule is configured to retrieve the most relevant context information from the storage unit of the dynamic memory bank by taking the decoded propagation sub-chain output by the decoder as a query condition, and extract a compensation feature corresponding to the context information. The second feature detection submodule is configured to combine the compensation feature with the composite feature to enhance the composite feature.

6. The apparatus of claim 5, wherein, The feature acquisition module comprises: The first feature acquisition submodule is configured to extract noise information of the image through an SRM conversion model, and pass the noise information through a Backbone_N model for 4-layer convolution to obtain , input into a ResBlock and a maximum pooling layer to obtain , pass through the ResBlock to obtain , pass through the maximum pooling layer again to obtain , and pass , , and through two convolution layers after channel splicing, and then generate the noise domain feature through a cSElayer. The second feature acquisition submodule is used to extract the binary stereo DCT information of the image through a DCT conversion model, and to generate the frequency domain features through feature extraction using a Backbone_D model. Specifically, this includes: extracting the binary stereo DCT information of the image through... dilated convolution with a dilation rate of 8 Convolution to obtain features Copy the quantization table Q to dimension 256. 256 1, and again Multiplication yields characteristics , and They were reshaped into The features are concatenated and then processed through a regular convolutional layer to obtain... The frequency domain characteristics. 7.A computer device comprising a memory and a processor, wherein the memory stores computer readable instructions, and the computer readable instructions, when executed by the processor, cause the processor to perform the steps of the image propagation chain detection method based on dynamic information compensation according to any one of claims 1 to 4. 8.A storage medium storing computer readable instructions, and the computer readable instructions, when executed by one or more processors, cause the one or more processors to perform the steps of the image propagation chain detection method based on dynamic information compensation according to any one of claims 1 to 4.

Images