Method and device for hiding dynamic image information based on depth generation model, processor and computer readable storage medium thereof

Abstract

The invention relates to a method for hiding dynamic image information based on a depth generation model, which comprises the following steps that: a sender processes secret information and encrypts the secret information into a ciphertext by using a key; mapping the ciphertext to standard Gaussian distribution; the sender performs iterative denoising on the standard Gaussian distribution containing the secret information in combination with a controllable text and a motion module to generate a GIF containing the secret information; and the receiver decrypts the information through inter-frame comparison after receiving the information. By adopting the method and the device for hiding the dynamic image information based on the depth generation model, the processor and the computer readable storage medium, the security of information hiding is remarkably improved, and the embedded secret information is ensured not to be easily detected by a third party; the robustness of information hiding is enhanced, and it is ensured that the embedded information can resist common distortion operation; high-concealment information embedding and reliable information extraction are realized, and the user experience and the system practicability are improved.

Description

Technical Field

[0001] This invention relates to the field of deep learning, and more particularly to the field of GIF steganography technology, specifically to a method, apparatus, processor, and computer-readable storage medium for hiding dynamic image information based on a deep generative model. Background Technology

[0002] GIF (Graphics Interchange Format), a widely used animated image format on the internet, has become a common medium for information hiding (steganography) due to its ubiquity and multi-frame structure. GIF steganography aims to use GIF images as cover to secretly transmit information without alerting third parties.

[0003] Traditional GIF steganography techniques primarily focus on directly modifying the image data itself. One of the most representative methods is based on indexed color modification. This type of method embeds secret information by changing the index values ​​of pixels in the GIF color palette. For example, by fine-tuning the color index of a pixel to point to a new color that is similar in color but has a different index value, the information bits are encoded into the change in the index value. However, this type of traditional steganography method suffers from two major, insurmountable drawbacks: First, there is insufficient security. Directly modifying index values ​​inevitably disrupts the statistical properties of the original image. For example, statistical features such as the histogram distribution of index values ​​and the transition probability between adjacent indices will undergo detectable changes. In recent years, with the development of steganalysis technology, detection algorithms for such statistical anomalies (such as chi-square test and spatial rich model analysis) have become very mature and can detect the presence of hidden information with high accuracy, thus greatly reducing the stealth of these methods.

[0004] Second, it has poor robustness. Steganized GIF images are frequently subjected to lossy or lossless compression (e.g., frame dropping, palette optimization, re-encoding) during network transmission and social media sharing. Traditional index-based modification methods are extremely sensitive to these operations. Once a GIF file is compressed or transcoded, minute changes embedded in the index values ​​can easily be corrupted or erased, making it impossible for the recipient to accurately extract the secret information, or even causing complete information loss. This greatly limits its application value in real-world network environments.

[0005] To overcome these problems, some research has begun to explore the application of deep learning, especially generative models, to the field of information hiding. However, how to design a GIF steganography framework that can guarantee statistical indistinguishability (i.e., high security), resist common image compression (i.e., high robustness), and does not require a large amount of training data remains a pressing technical challenge in the field.

[0006] Therefore, developing a novel GIF steganography method that significantly improves robustness to distortion operations such as compression while maintaining high security has significant theoretical research value and practical application prospects. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, apparatus, processor and computer-readable storage medium for dynamic image information hiding based on a deep generative model that is highly secure, robust and widely applicable.

[0008] To achieve the above objectives, the present invention provides a method, apparatus, processor, and computer-readable storage medium for dynamic image information hiding based on a deep generative model, as follows: The main feature of this method for dynamic image information hiding based on a deep generative model is that the method includes the following steps: (1) The sender processes the secret information and encrypts it into ciphertext using a key; (2) Map the ciphertext to a standard Gaussian distribution; (3) The sender combines controllable text and motion modules to iteratively denoise the standard Gaussian distribution containing secret information and generate a GIF containing secret information; (4) After receiving the information, the receiver decrypts it by comparing the frames.

[0009] Preferably, step (1.1) specifically comprises: The secret information is redundantly expanded based on the number of frames in the GIF, and each copy of the information is encrypted with an independent key of equal length to become ciphertext.

[0010] Preferably, step (2) specifically includes the following steps: (2.1) Obtain a binary bit information to be embedded; (2.2) Obtain an original variable sampled from a preset probability distribution; (2.3) Determine a symbolic variable based on the value of the binary information bits; (2.4) Calculate the absolute value of the original variable; (2.5) Multiply the symbolic variable determined in (2.3) with the absolute value calculated in (2.4) to generate a modified variable embedded with the binary information bits.

[0011] Preferably, step (2.3) specifically includes: When the binary information bit is in the first state, the symbol variable is a first preset value; when the binary information bit is in the second state, the symbol variable is a second preset value.

[0012] Preferably, step (3) specifically includes the following steps: (3.1) Use a pre-trained personalized diffusion model, combined with controllable text and motion modules, to perform personalized denoising; (3.2) The denoised latent space is decoded by the decoder to generate a steganalytic GIF.

[0013] Preferably, step (4) specifically includes the following steps: (4.1) After receiving the steganographic GIF, the receiver divides it into single frames and converts each frame into the latent space through an encoder; (4.2) The latent space of each frame is combined with DDIM inversion to reconstruct the original standard Gaussian distribution containing secret information; (4.3) Perform a sign-reverse mapping on the standard Gaussian distribution of each frame to convert it into an encrypted ciphertext sequence, and use the corresponding key to decrypt it to obtain the distribution of the secret information; (4.4) For the decrypted secret information, an inter-frame voting mechanism is used to select the final secret information for each frame; (4.5) Use a cross-frame voting mechanism to recover the original secret information from the obtained results.

[0014] The device for dynamic image information hiding based on a deep generative model is characterized in that the device comprises: A processor is configured to execute computer-executable instructions; The memory stores one or more computer-executable instructions, which, when executed by the processor, implement the various steps of the method for dynamic image information hiding based on a deep generative model.

[0015] The processor for dynamic image information hiding based on a deep generative model is characterized in that the processor is configured to execute computer-executable instructions, and when the computer-executable instructions are executed by the processor, the various steps of the above-mentioned method for dynamic image information hiding based on a deep generative model are implemented.

[0016] The computer-readable storage medium is characterized in that it stores a computer program thereon, which can be executed by a processor to implement the various steps of the above-described method for dynamic image information hiding based on a deep generative model.

[0017] The method, apparatus, processor, and computer-readable storage medium for dynamic image information hiding based on a deep generative model significantly improve the security of information hiding, ensuring that the embedded secret information is not easily detected by third parties; enhance the robustness of information hiding, ensuring that the embedded information can resist common distortion operations; achieve highly concealed information embedding and reliable information extraction, improving user experience and system usability; and bring significant benefits at both theoretical and practical levels, including efficiency improvement and broad application potential. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method for dynamic image information hiding based on a deep generative model according to the present invention.

[0019] Figure 2 This is a flowchart of the method for dynamic image information hiding based on a deep generative model according to the present invention.

[0020] Figure 3 This diagram illustrates image generation under different embedding capacities using the method for dynamic image information hiding based on a deep generative model according to the present invention.

[0021] Figure 4 This is a flowchart illustrating the steps of the method for dynamic image information hiding based on a deep generative model according to the present invention. Detailed Implementation

[0022] To more clearly describe the technical content of the present invention, the following description is provided in conjunction with specific embodiments.

[0023] The method for dynamic image information hiding based on a deep generative model of the present invention includes the following steps: (1) The sender processes the secret information and encrypts it into ciphertext using a key; (2) Map the ciphertext to a standard Gaussian distribution; (3) The sender combines controllable text and motion modules to iteratively denoise the standard Gaussian distribution containing secret information and generate a GIF containing secret information; (4) After receiving the information, the receiver decrypts it by comparing the frames.

[0024] In a preferred embodiment of the present invention, step (1.1) specifically comprises: The secret information is redundantly expanded based on the number of frames in the GIF, and each copy of the information is encrypted with an independent key of equal length to become ciphertext.

[0025] In a preferred embodiment of the present invention, step (2) specifically includes the following steps: (2.1) Obtain a binary bit information to be embedded; (2.2) Obtain an original variable sampled from a preset probability distribution; (2.3) Determine a symbolic variable based on the value of the binary information bits; (2.4) Calculate the absolute value of the original variable; (2.5) Multiply the symbolic variable determined in (2.3) with the absolute value calculated in (2.4) to generate a modified variable embedded with the binary information bits.

[0026] In a preferred embodiment of the present invention, step (2.3) specifically comprises: When the binary information bit is in the first state, the symbol variable is a first preset value; when the binary information bit is in the second state, the symbol variable is a second preset value.

[0027] In a preferred embodiment of the present invention, step (3) specifically includes the following steps: (3.1) Use a pre-trained personalized diffusion model, combined with controllable text and motion modules, to perform personalized denoising; (3.2) The denoised latent space is decoded by the decoder to generate a steganalytic GIF.

[0028] In a preferred embodiment of the present invention, step (4) specifically includes the following steps: (4.1) After receiving the steganographic GIF, the receiver divides it into single frames and converts each frame into the latent space through an encoder; (4.2) The latent space of each frame is combined with DDIM inversion to reconstruct the original standard Gaussian distribution containing secret information; (4.3) Perform a sign-reverse mapping on the standard Gaussian distribution of each frame to convert it into an encrypted ciphertext sequence, and use the corresponding key to decrypt it to obtain the distribution of the secret information; (4.4) For the decrypted secret information, an inter-frame voting mechanism is used to select the final secret information for each frame; (4.5) Use a cross-frame voting mechanism to recover the original secret information from the obtained results.

[0029] The apparatus for dynamic image information hiding based on a deep generative model of the present invention includes: A processor is configured to execute computer-executable instructions; The memory stores one or more computer-executable instructions, which, when executed by the processor, implement the various steps of the method for dynamic image information hiding based on a deep generative model.

[0030] The processor of the present invention for implementing dynamic image information hiding based on a deep generative model is configured to execute computer-executable instructions, which, when executed by the processor, implement the various steps of the above-described method for implementing dynamic image information hiding based on a deep generative model.

[0031] The computer-readable storage medium of the present invention stores a computer program thereon, which can be executed by a processor to implement the various steps of the above-described method for dynamic image information hiding based on a deep generative model.

[0032] In a specific embodiment of the present invention, in view of the problems mentioned above, the purpose of the present invention is to overcome the aforementioned security and robustness issues, and to provide a dynamic image (GIF) information hiding method based on a deep generative model, comprising the following steps: S1: The sender processes the secret information and encrypts it into ciphertext using a key; Furthermore, the specific steps of S1 are as follows: S1.1: Redundant expansion of the secret information is performed based on the number of frames in the GIF, and each copy of the information is matched with an independent equal-length key for encryption into ciphertext; S2: Map the ciphertext to a standard Gaussian distribution; Furthermore, the specific steps of S2 are as follows: S2.1: Obtain a binary bit information to be embedded; S2.2: Obtain an original variable sampled from a predefined probability distribution (e.g., a standard Gaussian distribution); S2.3: Determine a symbolic variable based on the value of the binary information bit, wherein when the binary information bit is in a first state (e.g., 1), the symbolic variable is a first preset value (e.g., +1), and when the binary information bit is in a second state (e.g., 0), the symbolic variable is a second preset value (e.g., -1). S2.4: Calculate the absolute value of the original variable; S2.5: Multiply the symbolic variable determined in step S2.3 with the absolute value calculated in step S2.4 to generate a modified variable that embeds the binary information bits.

[0033] S3: The sender combines controllable text and motion modules to iteratively denoise a standard Gaussian distribution containing secret information to generate a GIF containing secret information; Furthermore, the specific steps of S3 are as follows: S3.1: Use a pre-trained personalized diffusion model and combine it with controllable text and motion modules to perform personalized denoising, and then decode the denoised latent space to generate a steganalytic GIF.

[0034] S4: After receiving the information, the receiver decrypts it by comparing frames.

[0035] Furthermore, the specific steps of S4 are as follows: S4.1: After receiving the steganographic GIF, the receiver divides it into single frames and converts each frame into the latent space through an encoder; S4.2: The latent space of each frame is combined with DDIM inversion to reconstruct the original standard Gaussian distribution containing secret information, without the need for any text cue words in this process; S4.3: Perform a sign-reverse mapping on the standard Gaussian distribution of each frame to convert it into an encrypted ciphertext sequence, and then use the corresponding key to decrypt it to obtain the distribution of the secret information; S4.4: For the decrypted secret information, an inter-frame voting mechanism is first used to select the final secret information for each frame; S4.5: Use a cross-frame voting mechanism to recover the original secret information from the obtained results.

[0036] The specific steps of step S3.1 above are as follows: (3.1.1) Prepare a pre-trained personalized T2I diffusion model (such as Stable Diffusion). Personalization refers to fine-tuning the pre-trained T2I model on a specific small dataset (such as pictures containing specific people, objects or styles) through techniques such as DreamBooth and LoRA, so that it can generate images with specific personalized concepts.

[0037] (3.1.2) A typical T2I diffusion model consists of a VAE (Variational AutoEncoder), a U-Net, and a text encoder. First, dense standard Gaussian noise is encoded into a low-dimensional latent space by the VAE's encoder. Then, the U-Net receives the noisy latent representation, time step, and text conditions and predicts the added noise. The text encoder is used to encode the text into a sequence of conditional vectors that the U-Net can understand.

[0038] (3.1.3) Use a pre-trained motion module and insert it into the U-Net structure of the personalized model. This module is general and is responsible for producing smooth and natural dynamic changes between consecutive frames.

[0039] (3.1.4) The model begins multi-step iterative denoising. The personalized model generates the correct content and style based on the text prompts, and the motion module coordinates the smoothness between frames. After dozens of iterations, the initial random noise will become a clear, dense latent space sequence containing complete animation information.

[0040] (3.1.5) Use the VAE decoder to decode each frame in the latent space into the final steganalytic GIF.

[0041] The specific steps of step S4.4 above are as follows: When decoding the secret information of each frame, a majority voting mechanism is used to determine the final secret information recovered for each frame, since the secret information is redundantly expanded before transmission. For example, if more than half of the positions in a frame are 1, then the original secret information is considered to be 1.

[0042] The specific steps of step S4.5 above are as follows: When the secret information of each frame is obtained, since a GIF has multiple frames, each frame in this method carries secret information. Therefore, by comparing the same positions between frames, a majority voting mechanism is still used to recover the original secret information.

[0043] The technical solution of this invention first redundantly expands the secret information, then maps the secret information to standard Gaussian noise and encodes it into the latent space for information hiding. Encoding into the latent space aims to reduce computational load. In this solution, a motion module is used to generate multiple consecutive GIF frames from the secret information. The image is generated as a secret-containing static image driven by category labels and random noise. Furthermore, in the decoding process, DDIM inversion is used to recover the secret standard Gaussian distribution from the secret GIF, thus obtaining the secret information. Moreover, this solution adds a cross-frame voting mechanism to recover the original secret information, thus exhibiting good robustness. This solution does not use a carrier image; instead, it directly generates the secret-containing GIF image from the secret information.

[0044] The technical solution of this invention generates continuous animated GIF images by converting secret information into standard Gaussian noise. Furthermore, the secret information embedding method of this solution directly maps the secret information to standard Gaussian noise to generate the animated GIF. The receiver performs extraction processing by dividing the animated GIF into single frames, performing DDIM inversion on each frame to obtain the original noise distribution, and then mapping the noise distribution back to the secret information. During this process, this solution requires no prompts for image generation. This solution also adds a cross-frame voting mechanism to enhance robustness.

[0045] For the specific implementation scheme of this embodiment, please refer to the relevant descriptions in the above embodiments, which will not be repeated here.

[0046] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0047] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means at least two.

[0048] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0049] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution device. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0050] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The corresponding program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

[0051] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0052] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0053] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0054] The method, apparatus, processor, and computer-readable storage medium for dynamic image information hiding based on a deep generative model significantly improve the security of information hiding, ensuring that the embedded secret information is not easily detected by third parties; enhance the robustness of information hiding, ensuring that the embedded information can resist common distortion operations; achieve highly concealed information embedding and reliable information extraction, improving user experience and system usability; and bring significant benefits at both theoretical and practical levels, including efficiency improvement and broad application potential.

[0055] In this specification, the invention has been described with reference to specific embodiments thereof. However, it will be apparent that various modifications and variations can be made without departing from the spirit and scope of the invention. Therefore, the specification and drawings should be considered illustrative rather than restrictive.

Claims

1. A method for realizing dynamic image information hiding based on a deep generative model, characterized in that, The method comprises the following steps: (1) The sender processes the secret information and encrypts it into ciphertext using a key; (2) The ciphertext is mapped to a standard Gaussian distribution; (3) The sender combines the controllable text and the motion module to iteratively denoise the standard Gaussian distribution containing the secret information, generating a GIF containing the secret information; (4) The receiver decrypts the information received by inter-frame comparison.

2. The method of claim 1, wherein the deep generative model is a variational autoencoder (VAE) or a generative adversarial network (GAN). Step (1.1) is specifically: According to the number of GIF frames, the secret information is redundantly expanded, and each information copy is matched with an independent equal-length key for encryption into ciphertext.

3. The method of claim 1, wherein the deep generative model is a variational autoencoder (VAE). Step (2) specifically comprises the following steps: (2.1) Obtain a binary bit information to be embedded; (2.2) Obtain an original variable sampled from a preset probability distribution; (2.3) Determine a sign variable according to the value of the binary information bit; (2.4) Calculate the absolute value of the original variable; (2.5) Multiply the sign variable determined in (2.3) and the absolute value calculated in (2.4) to generate a modified variable embedded with the binary information bit.

4. The method of claim 3, wherein the deep generative model is a variational autoencoder (VAE). Step (2.3) is specifically: When the binary information bit is in a first state, the sign variable is a first preset value; when the binary information bit is in a second state, the sign variable is a second preset value.

5. The method of claim 1, wherein the deep generative model is a variational autoencoder (VAE). Step (3) specifically comprises the following steps: (3.1) Use a pre-trained personalized diffusion model to combine the controllable text and the motion module for personalized denoising; (3.2) Decode the denoised latent space through a decoder to generate a steganographic GIF.

6. The method of claim 1, wherein the deep generative model is a variational autoencoder (VAE). Step (4) specifically comprises the following steps: (4.1) The receiver splits the steganographic GIF into single frames after receiving it, and converts each frame to the latent space through an encoder; (4.2) The latent space of each frame is combined with DDIM inversion to reconstruct the original standard Gaussian distribution containing the secret information; (4.3) Perform sign reverse mapping on each frame of the standard Gaussian distribution to convert it to an encrypted ciphertext sequence, and use the corresponding key to decrypt to obtain the distribution of the secret information; (4.4) For the decrypted secret information, a frame-by-frame voting mechanism is used to select the final secret information of each frame; (4.5) The result obtained is recovered to the original secret information using a cross-frame voting mechanism.

7. An apparatus for realizing dynamic image information hiding based on a deep generative model, characterized in that, The device comprises: a processor configured to execute computer executable instructions; a memory storing one or more computer executable instructions, which, when executed by the processor, implement the steps of the method for realizing dynamic image information hiding based on a deep generative model according to any one of claims 1 to 6. 8.A processor for realizing dynamic image information hiding based on a deep generative model, characterized in that, The processor is configured to execute computer executable instructions, which, when executed by the processor, implement the steps of the method for realizing dynamic image information hiding based on a deep generative model according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, A computer program product, comprising a computer readable medium having stored thereon a computer program, the computer program being executable by a processor to implement each of the steps of the method of claim 1 to 6.

Images