Video steganography method based on deep face forgery replacement technology

Abstract

The invention discloses a video steganography method based on a deep face forgery replacement technology. The video steganography method comprises the following steps: acquiring a carrier video to be processed, a reference image and secret information; the method comprises the following steps: respectively carrying out video coding on a carrier video to be processed, coding secret information and carrying out identity feature extraction on a reference image through an information coding module to correspondingly obtain video coding information, secret coding information and identity feature representation; and performing embedding processing on the secret coding information and the identity feature representation through the secret information embedding module, injecting the secret-containing identity feature into the video coding information to obtain secret-containing video coding information, and performing decoding processing by using the secret-containing video coding information to obtain the steganography video. The method provided by the invention can improve the accuracy and reliability of secret information transmission.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a video steganography method based on deep face forgery replacement technology. Background Art

[0002] In the past few years, with the booming development of online social networks (OSNs), videos have become a suitable medium for steganography.

[0003] Traditional video steganography methods are based on direct pixel value processing, coding mapping or adaptive distortion functions, which exploit the redundancy of video data to achieve information hiding. Although these methods have good security performance and embedding capacity, their robustness is limited, making the hidden secret message easily lost inadvertently during the subsequent compression and encoding process. In addition, traditional video steganography methods also face the challenge of mitigating the effects of various distortions that may occur during transmission over lossy channels. Therefore, more robust video steganography methods are needed. Summary of the invention

[0004] The present invention provides a video steganography method based on deep face forgery replacement technology, aiming to effectively solve the above technical problems.

[0005] According to a first aspect of the present invention, the present invention provides a video steganography method based on deep face forgery replacement technology, the method comprising:

[0006] Obtain a carrier video to be processed, a reference image and secret information, and process the carrier video to be processed, the reference image and the secret information using a pre-trained video steganography model, wherein the video steganography model includes an information encoding module and a secret information embedding module;

[0007] The information encoding module is used to respectively encode the carrier video to be processed, encode the secret information, and extract identity features from the reference image, thereby obtaining video encoding information, secret encoding information, and identity feature representation;

[0008] The secret information embedding module embeds the secret coding information and the identity feature representation, and injects the secret identity feature into the video coding information to obtain the secret video coding information, and then uses the secret video coding information to perform decoding processing to obtain the steganographic video.

[0009] Furthermore, injecting the confidential identity feature into the video coding information to obtain the confidential video coding information includes:

[0010] Obtaining a first channel mean and a first standard deviation of the video coding information, and a second channel mean and a second standard deviation of the encrypted identity feature;

[0011] The encrypted video coding information is obtained by calculation according to the first channel mean and the first standard deviation, and the second channel mean and the second standard deviation of the encrypted identity feature.

[0012] Further, the calculating and obtaining the encrypted video coding information according to the first channel mean and the first standard deviation, and the second channel mean and the second standard deviation of the encrypted identity feature, includes:

[0013] The encrypted video encoding information is obtained by calculating the encrypted video embedding formula according to the first channel mean and the first standard deviation, the second channel mean and the second standard deviation of the encrypted identity feature

[0014]

[0015] In the formula, and Respectively represent video coding information The first channel mean and first standard deviation of and μ(F′ id ) represent the secret identity features F′ id The second channel mean and second standard deviation.

[0016] Furthermore, the embedding process of the secret coded information and the identity feature representation includes:

[0017] According to the secret coding information, the preset hyperparameters and the identity feature representation, the secret identity feature F′ is calculated using the secret identity embedding formula id :

[0018] F′ id =F id +λ*F m

[0019] In the formula, F id is the identity feature representation, F m is the secret encoding information, λ is the preset hyperparameter, and λ is used to adjust the impact of secret information on the identity feature representation.

[0020] Furthermore, the video steganography model further includes a secret information extraction module. After obtaining the steganographic video, the method further includes:

[0021] The secret information extraction module is used to extract secret information from the steganographic video to obtain new secret information.

[0022] Furthermore, the video steganography model further includes a distortion module. After obtaining the steganographic video, the method further includes:

[0023] Using the distortion module to perform distortion processing on the steganographic video to obtain a processed steganographic video;

[0024] The distortion processing is one or more of quantization step, pixel linear transformation, and noise simulation.

[0025] According to a second aspect of the present invention, the present invention also provides a video steganography device based on deep face forgery replacement technology, comprising:

[0026] A data acquisition module, used to acquire a carrier video to be processed, a reference image and secret information, and process the carrier video to be processed, the reference image and the secret information using a pre-trained video steganography model, wherein the video steganography model includes an information encoding module and a secret information embedding module;

[0027] An encoding module, used to respectively perform video encoding on the carrier video to be processed, encode the secret information, and extract identity features from the reference image through the information encoding module, and correspondingly obtain video encoding information, secret encoding information, and identity feature representation;

[0028] The information embedding module is used to embed the secret coded information and the identity feature representation through the secret information embedding module, and inject the secret identity feature into the video coding information to obtain the secret video coding information, and then use the secret video coding information to perform decoding processing to obtain the steganographic video.

[0029] According to a third aspect of the present invention, the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above method.

[0030] According to a fourth aspect of the present invention, the present invention further provides a computer program product, comprising a computer program, wherein the computer program implements the above method when executed by a processor.

[0031] According to a fifth aspect of the present invention, the present invention further provides a storage medium on which a computer program is stored, and when the program is executed by a processor, the method as described above is implemented.

[0032] Through one or more of the above embodiments of the present invention, at least the following technical effects can be achieved: the information encoding module is used to respectively encode the carrier video to be processed, encode the secret information, and extract the identity features of the reference image, and correspondingly obtain video encoding information, secret encoding information, and identity feature representation; the secret information embedding module is used to embed the secret encoding information and the identity feature representation, and the secret identity feature is injected into the video encoding information to obtain the secret video encoding information, and then the secret video encoding information is used for decoding processing to obtain the steganographic video. The method provided by the present application can accurately and robustly transmit secret information, and improve the accuracy and reliability of secret information transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The technical solutions and other beneficial effects of the present invention will be made apparent by describing in detail the specific embodiments of the present invention in conjunction with the accompanying drawings.

[0034] Figure 1 One of the flow charts of the video steganography method based on deep face forgery replacement technology provided by an embodiment of the present invention;

[0035] Figure 2 This is the second flowchart of the video steganography method based on deep face fake replacement technology provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The technical scheme in the embodiment of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiment of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0037] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the term "and / or" herein is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " herein, unless otherwise specified, generally indicates that the associated objects before and after are in an "or" relationship.

[0038] The following is an introduction to a video steganography method based on deep face forgery replacement technology provided by the present invention in conjunction with the accompanying drawings.

[0039] Figure 1 One of the flow charts of the video steganography method based on deep face forgery replacement technology provided by an embodiment of the present invention, Figure 2 The second flowchart of the video steganography method based on deep face forgery replacement technology provided by the embodiment of the present invention; Figure 1-2 As shown, a video steganography method based on deep face forgery replacement technology includes the following steps:

[0040] S101, obtaining a carrier video to be processed, a reference image and secret information, and using a pre-trained video steganography model to process the carrier video to be processed, the reference image and the secret information.

[0041] Wherein, the video steganography model includes an information encoding module and a secret information embedding module.

[0042] S102, respectively performing video encoding on the carrier video to be processed, encoding the secret information, and extracting identity features from the reference image through the information encoding module, and correspondingly obtaining video encoding information, secret encoding information, and identity feature representation.

[0043] Among them, the information encoding module includes a video feature extractor E φ , secret information encoder E m , Identity Extractor E id The video feature extractor is used to extract the cover video V C ) that is, performing video encoding to obtain video encoding information; the secret information encoder is used to encode the secret information to obtain secret encoding information; the identity extractor is used to extract identity features of the reference image to obtain identity feature representation.

[0044] Specifically, the video feature extractor is a convolutional neural network consisting of four convolutional blocks, each of which increases the number of channels by two times, including a convolutional layer, a normalization layer, and an activation unit. The spatial size of the features is reduced by a pooling layer with a stride of 2. Therefore, the output feature size of the extractor is 1 / 3 of the original image. Each frame of the carrier video to be processed is sent to the feature extractor, and after downsampling processing by four convolution blocks, the features of each frame are extracted to obtain the video coding information. ( is the feature of the i-th frame image).

[0045] Secret information encoder E m Used to encode secret information, it is a neural network composed of a multi-layer perceptron (Multi-LayerPerceptron) with a dense layer. The function of this encoder is to change the dimension of the secret information to make it the same as the dimension of the identity feature.

[0046] Identity ExtractorE idUsed to extract the reference image I R The identity feature representation is for the reference image I R In one implementation of the present application, the identity extractor E id It can be the face recognition network ArcFace. ArcFace uses an angle cosine similarity loss function to map the face feature vector to the sphere and optimize the model by maximizing the angle cosine similarity. The advantage of ArcFace is that it can further improve the robustness and generalization ability of the model, and the scaling of the cosine similarity can be controlled by adjusting the learnable parameters.

[0047] The above video feature extractor E φ , secret information encoder E m , Identity Extractor E id Specifically, the corresponding information is obtained through the following formula:

[0048] F id =E id (I R ) (1)

[0049]

[0050] F m =W m M+b m (3)

[0051] in, represents the i-th frame image of the cover video, Represents the feature representation of the i-th frame of the video (that is, video coding information). id ∈R 1*512 is the reference image (I R )’s identity feature matrix (that is, identity feature representation). 1*m (m=9 or 18) is the secret information. m ∈R m*512 and b m ∈R 1*512 Represents learnable weights and biases.

[0052] S103, embedding the secret coding information and the identity feature representation through the secret information embedding module, and injecting the secret identity feature into the video coding information to obtain the secret video coding information, and then using the secret video coding information to perform decoding processing to obtain the steganographic video.

[0053] The video steganography method based on deep face fake replacement technology provided in the embodiment of the present application can accurately and robustly transmit secret information, thereby improving the accuracy and reliability of secret information transmission.

[0054] In some embodiments of the present application, injecting the confidential identity feature into the video coding information to obtain the confidential video coding information includes:

[0055] A first channel mean and a first standard deviation of the video coding information, and a second channel mean and a second standard deviation of the encrypted identity feature are obtained.

[0056] The encrypted video coding information is obtained by calculation according to the first channel mean and the first standard deviation, and the second channel mean and the second standard deviation of the encrypted identity feature.

[0057] Specifically, the calculating and obtaining the encrypted video coding information according to the first channel mean and the first standard deviation, the second channel mean and the second standard deviation of the encrypted identity feature includes:

[0058] The encrypted video encoding information is obtained by calculating the encrypted video embedding formula according to the first channel mean and the first standard deviation, the second channel mean and the second standard deviation of the encrypted identity feature

[0059]

[0060] In the formula, and Respectively represent video coding information The first channel mean and first standard deviation of and μ(F i ′ d ) represent the secret identity features F i ′ d The second channel mean and second standard deviation.

[0061] In this embodiment, since the identity feature and the attribute feature are highly coupled, it is not possible to directly obtain the identity feature from the video encoding information. In order to better preserve the attributes of the face in the video, this application designs a Secret-ID block composed of a modified version of the residual block and Adaptive Instance Normalization (AdaIN), which converts F i ′ d Inject into The modified Secret-ID block is expressed as follows:

[0062]

[0063] in, and Respectively represent video coding information The channel mean and standard deviation of and μ(F i ′ d ) corresponds to the secret identity feature F i ′ d The channel mean and standard deviation.

[0064] After N Secret-ID blocks, The identity feature in is replaced by F i ′ d , get the secret video coding information

[0065] Then, through the video decoder D φ From the encrypted video encoding information The I-th frame image of the steganographic video stego video is recovered Video Decoder D φ Contains four upsampling blocks, a reflection padding layer (ReflectionPad2d) and a convolution layer. Each upsampling block consists of an upsampling layer, a convolution layer and a regularization layer (BatchNorm). The process can be expressed as:

[0066]

[0067] In some embodiments of the present application, the embedding process of the secret coded information and the identity feature representation includes:

[0068] According to the secret coding information, the preset hyperparameters and the identity feature representation, the secret identity feature F′ is calculated using the secret identity embedding formula id :

[0069] F′ id =F id +λ*F m

[0070] In the formula, F id is the identity feature representation, F m is the secret encoding information, λ is the preset hyperparameter, and λ is used to adjust the impact of secret information on the identity feature representation.

[0071] In this embodiment, the purpose of the secret message embedding module is to embed a secret message in the face swapping process. The key issue is how to add secret information so that the secret information can guide the face swapping. The video encoding information extracted from the carrier video to be processed (i.e., the cover video) can be regarded as the coupling of identity features and attribute features. The essence of face swapping is to replace the identity features of the cover video with the reference image (I R ) identity features. In order to complete the identity feature replacement while ensuring the embedding of secret information, this application proposes to embed the secret information into the reference image I R The embedding is realized by using the identity feature, which is expressed as:

[0072] F′ id =F id +λ*F m (4)

[0073] Among them, λ is a hyperparameter with a value range of [0,1], which is used to adjust the impact of secret messages on identity features. In the present invention, by conducting experiments with different values ​​of λ, it is concluded that when λ is closer to 1, the accuracy of secret information extraction is higher, but the visual quality of the generated video is worse; when λ is closer to 0, the visual quality of the generated video is better, but the accuracy of secret information extraction is lower. Therefore, in order to balance the visual quality and the accuracy of secret information extraction, λ=0.01 is set in this embodiment. It should be noted that in other embodiments, λ can also be set to other values ​​within [0,1], and this application does not limit this.

[0074] In some embodiments of the present application, the video steganography model further includes a secret information extraction module. After obtaining the steganographic video, the method further includes:

[0075] The secret information extraction module is used to extract secret information from the steganographic video to obtain new secret information.

[0076] In this embodiment, the video steganography model also includes a secret information extraction module, which is used to extract secret information from the generated stego video. The secret information extraction module contains 7 convolutional layers with ReLU activation function. Finally, the sigmoid activation function and binarization processing are added to extract the embedded secret information. The secret information extraction module is expressed as follows:

[0077] M′=E ext (V s ) (9)

[0078] Among them, M′ is the extracted secret information, E ext represents the secret information extraction module, V s It's a stego video.

[0079] In some embodiments of the present application, the video steganography model further includes a distortion module. After obtaining the steganographic video, the method further includes:

[0080] The distortion module is used to perform distortion processing on the steganographic video to obtain a processed steganographic video.

[0081] The distortion processing is one or more of quantization step, pixel linear transformation, and noise simulation.

[0082] In this embodiment, the video steganography model also includes a distortion module, which is used to improve the robustness of video steganography to cope with the scenarios of real video usage. The method includes several common distortions on most social networking platforms, including JPEG compression, color distortion, blur, noise and saturation.

[0083] Distortion caused by JPEG compression: JPEG compression is a commonly used image compression method in the process of photo storage. However, due to the rounding process, the quantization step is not differentiable. In order to solve this distortion problem, formula (7) is used to approximate the quantization step size close to zero:

[0084]

[0085] Where x represents a specific pixel of the input image. The JPEG quality is uniformly sampled in the range [50,100].

[0086] Color distortion: Color distortion includes two common types of distortion: brightness distortion and contrast distortion. These two types of distortion are usually caused by Internet transmission and artificial processing of videos. This can be simulated by performing a linear transformation on the pixels of each channel, as shown in formula (8):

[0087] p(x)=a*f(x)+c (8)

[0088] Where p(x) is the deformed image and f(x) is the original image. a and c are hyperparameters that control contrast and brightness, respectively.

[0089] Blur distortion: Blur is a common distortion produced by video processing software when it is spread on social networking platforms. To simulate blur, a linear blur kernel with a width ranging from 3 to 7 pixels is generated by randomly selecting an angle.

[0090] In addition, random linear interpolation is performed between the RGB image and its grayscale equivalent to simulate distortion to saturation. Gaussian noise is used to simulate any other distortion not considered in the distortion module. A Gaussian noise model (with sampling standard deviation δ~U[0,0.2]) is also used to simulate imaging noise.

[0091] The steganographic video is distorted by the above-mentioned distortion processing method to improve the robustness.

[0092] In other embodiments of the present application, the video steganography method based on deep face forgery replacement technology also includes a training process of a video steganography model, and the video steganography model is trained using the VGGFace face dataset. During the training process, in addition to the above-mentioned information encoding module, secret information embedding module, secret information extraction module and distortion module, the video steganography model also includes a discriminator.

[0093] The discriminator solves a binary classification problem and learns to distinguish between true and false generated face-swapped videos. It does this by predicting whether the observation was generated by the generator (fake) or from the original data distribution (real). In the process, it learns a set of parameters or weights. As training progresses, the weights are constantly updated.

[0094] In other embodiments of the present application, applicable evaluation indicators are introduced to evaluate the extraction accuracy, security and robustness of secret information.

[0095] The process of extracting secret information can be regarded as a binary classification problem. Let x represent the number of bits of real secret information embedded in the generated stego video, x ′ Indicates the number of bits of the secret correctly extracted from the stego video. The accuracy of secret information extraction is calculated by dividing the number of bits of the correctly extracted secret information by the number of bits of the embedded secret information. The calculation formula is as follows:

[0096]

[0097] To measure the robustness, a series of video encoding and distortion processing that may be encountered in real applications are added to the video to obtain various types of distorted videos, and then secret information is extracted from these distorted videos to calculate the extraction accuracy. The higher the accuracy, the more robust the present invention is.

[0098] In addition, the video steganography method provided by this application is compared with the traditional steganography method in a robustness test. The robustness test results are shown in Table 1:

[0099] Table 1 Robustness test results

[0100]

[0101] It can be seen from Table 1 that the video steganography method provided by the present application has better effects in PNG, H.264ABR, H.264CRF, saturation, contrast, motion blur, bit errors, etc. compared with the traditional PWRN (which is a high-efficiency video steganography method proposed in "An Anti-Steganalysis HEVC Video Steganography With High Performance Based on CNN and PU Partition Modes") and HiDDeN (i.e. Hiding Data With DeepNetworks). Therefore, the video steganography method provided by the present application has better robustness.

[0102] For the analysis of security, this embodiment uses the F1 score, and the closer this value is to 0.5, the safer the present invention is. For classification problems in the field of machine learning, confusion matrices (Confusion Matrix), also known as error matrices, are often used to measure algorithm performance. The confusion matrix is ​​a specific table layout, in which each row of the matrix represents all instances of a class in the classification problem, and each column represents all prediction results of a class in the prediction. Through the confusion matrix, it can be seen whether the algorithm mistakenly marks a class as another class. Therefore, the confusion matrix can produce four different results by comparing the actual category set with the predicted category set. 1) True positive (True Positive, TP): the actual classification is positive and the predicted classification is positive, 2) False negative (False Negative, FN): the actual classification is positive and the predicted classification is negative, 3) False positive (False Positive, FP) is actually classified as negative and the predicted classification is positive, 4) True negative (True Negative, TN): the actual classification is negative and the predicted classification is negative, and the above confusion matrix is ​​shown in Table 2.

[0103] Table 2 Matrix for calculating F1 parameters

[0104]

[0105] Precision and recall are also important indicators for measuring the quality of a classifier. Precision measures how many samples are correctly predicted among all samples predicted as positive. The higher the precision, the better the algorithm is at finding positive examples compared to negative examples. The calculation formula is as follows:

[0106]

[0107] Recall measures how many of the actually positive samples are correctly predicted by the algorithm, and the calculation formula is:

[0108]

[0109] In classification tasks, a precision of 1.0 for a class means that every sample predicted to be in that class does belong to that class, but does not take into account the number of samples in that class that were not correctly predicted. A recall of 1.0 means that every item in that class was predicted correctly, but does not indicate how many samples in other classes were also incorrectly predicted to be in that class. Therefore, precision and recall only have a good sense of measuring classifier performance when used together.

[0110] In order to further consider the accuracy and recall of the algorithm at the same time and make a normalized comparison, the F-score is calculated from the accuracy and recall, which can better comprehensively judge the significance of accuracy and recall as evaluation indicators. Among them, the F1 score is a commonly used indicator, which is the harmonic mean of accuracy and recall. The highest F1 score is 1.0, indicating that both accuracy and recall are optimal. When it is 0, it means that the accuracy or recall is 0. The calculation formula is as follows:

[0111]

[0112] It can be seen from the test results of the above-mentioned various aspects that the video steganography method based on deep face forgery replacement technology (Deepfake) proposed in the present invention has strong robustness.

[0113] In some other embodiments of the present application, a video steganography device based on deep face forgery replacement technology is also provided, the device comprising:

[0114] The data acquisition module is used to obtain the carrier video to be processed, the reference image and the secret information, and use the pre-trained video steganography model to process the carrier video to be processed, the reference image and the secret information. The video steganography model includes an information encoding module and a secret information embedding module.

[0115] The encoding module is used to respectively perform video encoding on the carrier video to be processed, encode the secret information, and extract identity features from the reference image through the information encoding module, and correspondingly obtain video encoding information, secret encoding information, and identity feature representation.

[0116] The information embedding module is used to embed the secret coded information and the identity feature representation through the secret information embedding module, and inject the secret identity feature into the video coding information to obtain the secret video coding information, and then use the secret video coding information to perform decoding processing to obtain the steganographic video.

[0117] It should be noted that the video steganography device based on deep face fake replacement technology corresponds to the above-mentioned video steganography method based on deep face fake replacement technology, which will not be repeated here.

[0118] Based on any of the above embodiments, another embodiment of the present invention further provides an electronic device, which may include: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor may call logic instructions in the memory to execute the above method.

[0119] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0120] On the other hand, an embodiment of the present invention further provides a storage medium on which a plurality of instructions are stored, and the instructions are suitable for being loaded by a processor to execute a video steganography method based on deep face fake replacement technology as provided in the above embodiments.

[0121] On the other hand, the technical solution of the present invention, in essence or in other words, the part that contributes to the prior art, or the whole or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0122] On the other hand, an embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the video steganography method based on deep face fake replacement technology as described above.

[0123] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0124] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0125] In summary, although the present invention has been disclosed as above in terms of preferred embodiments, the above preferred embodiments are not intended to limit the present invention. A person skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be based on the scope defined in the claims.

Claims

1. A video steganography method based on deep face forgery replacement technology, characterized in that: include: Obtain a carrier video to be processed, a reference image and secret information, and process the carrier video to be processed, the reference image and the secret information using a pre-trained video steganography model, wherein the video steganography model includes an information encoding module and a secret information embedding module; The information encoding module is used to respectively encode the carrier video to be processed, encode the secret information, and extract identity features from the reference image, thereby obtaining video encoding information, secret encoding information, and identity feature representation; The secret information embedding module embeds the secret coding information and the identity feature representation, and injects the secret identity feature into the video coding information to obtain the secret video coding information, and then uses the secret video coding information to perform decoding processing to obtain the steganographic video.

2. The method according to claim 1, characterized in that The step of injecting the confidential identity feature into the video coding information to obtain the confidential video coding information includes: Obtaining a first channel mean and a first standard deviation of the video coding information, and a second channel mean and a second standard deviation of the encrypted identity feature; The encrypted video coding information is obtained by calculation according to the first channel mean and the first standard deviation, and the second channel mean and the second standard deviation of the encrypted identity feature.

3. The method according to claim 2, characterized in that The step of calculating and obtaining the encrypted video coding information according to the first channel mean and the first standard deviation and the second channel mean and the second standard deviation of the encrypted identity feature comprises: The encrypted video encoding information is obtained by calculating the encrypted video embedding formula according to the first channel mean and the first standard deviation, the second channel mean and the second standard deviation of the encrypted identity feature In the formula, and Respectively represent video coding information The first channel mean and first standard deviation of and μ(F' id ) represent the secret identity features F' id The second channel mean and second standard deviation.

4. The method according to claim 1, characterized in that: The embedding process of the secret coded information and the identity feature representation comprises: According to the secret coding information, the preset hyperparameters and the identity feature representation, the secret identity feature F' is calculated using the secret identity embedding formula id : F’ id =F id +λ*F m In the formula, F id is the identity feature representation, F m is the secret encoding information, λ is the preset hyperparameter, and λ is used to adjust the impact of secret information on the identity feature representation.

5. The method according to claim 1, characterized in that The video steganography model also includes a secret information extraction module. After obtaining the steganographic video, the method further includes: The secret information extraction module is used to extract secret information from the steganographic video to obtain new secret information.

6. The method according to claim 1, characterized in that The video steganography model further includes a distortion module. After obtaining the steganographic video, the method further includes: Using the distortion module to perform distortion processing on the steganographic video to obtain a processed steganographic video; The distortion processing is one or more of quantization step, pixel linear transformation, and noise simulation.

7. A video steganography device based on deep face forgery replacement technology, characterized in that: include: A data acquisition module, used to acquire a carrier video to be processed, a reference image and secret information, and process the carrier video to be processed, the reference image and the secret information using a pre-trained video steganography model, wherein the video steganography model includes an information encoding module and a secret information embedding module; An encoding module, used to respectively perform video encoding on the carrier video to be processed, encode the secret information, and extract identity features from the reference image through the information encoding module, and correspondingly obtain video encoding information, secret encoding information, and identity feature representation; The information embedding module is used to embed the secret coded information and the identity feature representation through the secret information embedding module, and inject the secret identity feature into the video coding information to obtain the secret video coding information, and then use the secret video coding information to perform decoding processing to obtain the steganographic video.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor runs the computer program to implement the method according to any one of claims 1 to 6.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images