An information camouflage method and system based on combined generated adversarial samples

Abstract

The present application relates to the technical field of information security, and especially relates to a method and system for information camouflage based on combined generation of adversarial samples, the method comprising: obtaining a carrier image; generating random noise embedding adversarial samples, calculating gradient values of a neural network; obtaining adversarial noise according to the gradient values and embedding the carrier image to generate a strengthened carrier image; testing the strengthened carrier image, hiding a plurality of random information in the strengthened carrier image to generate a plurality of information camouflage images, and judging whether the probability of the original information camouflage image misleading the neural network is greater than a preset threshold; embedding secret information in the strengthened carrier image after testing to generate an information camouflage target image. The present application combines the generation method of adversarial samples with the information camouflage method, and designs a corresponding information camouflage method, which can effectively counter the information camouflage analysis method based on deep learning, and can effectively improve the security, practicality and applicability of the information camouflage technology.

Description

Technical Field

[0001] This invention relates to the field of information security technology, and in particular to an information camouflage method and system based on the combined generation of adversarial samples. Background Technology

[0002] With the widespread application of deep learning technology in the field of information camouflage analysis, the security of traditional information hiding and camouflage methods is greatly threatened, making them easily cracked by information camouflage analysis techniques with deep learning capabilities. However, in information warfare, the spear and the shield always complement each other and promote each other's development.

[0003] Currently, traditional information spoofing methods are unable to counter deep learning-based information spoofing analysis mechanisms. This is primarily because current analysis mechanisms focus only on analyzing statistical features, neglecting to consider featureless input methods and failing to introduce adversarial mechanisms. Consequently, existing analytical skills are ineffective against deep learning-based information spoofing analysis methods. Furthermore, existing single-sample generation methods rely on large amounts of data, have high computational time complexity, and generate adversarial samples applicable only to single samples, making it difficult to capture commonalities among adversarial samples. Finally, when facing information spoofing analysis techniques with deep learning capabilities, current information spoofing techniques suffer from insufficient security and cannot effectively counter deep learning-based information spoofing analysis methods. Summary of the Invention

[0004] This invention provides an information camouflage method and system based on the combined generation of adversarial examples, which addresses the shortcomings of existing technologies that cannot effectively counter deep learning-based information camouflage analysis methods and are not conducive to improving the security of information camouflage technology. It achieves an organic combination of adversarial example generation and information camouflage methods, which can effectively improve the security of information camouflage technology and enhance its practicality and applicability in real-world applications.

[0005] This invention provides an information spoofing method based on combined generation of adversarial examples, comprising:

[0006] S1 acquires the carrier image used for generating adversarial examples and information camouflage analysis;

[0007] S2 generates random noise that embeds adversarial examples, embeds the random noise into the carrier image, and inputs the generated new carrier image into the adversarial neural network to calculate the gradient value of the neural network; based on the gradient value, it obtains adversarial noise, embeds the adversarial noise into the carrier image to generate an enhanced carrier image;

[0008] S3 tests the enhanced carrier image by hiding some random information into the enhanced carrier image to generate several information camouflage images. When the probability of the information camouflage image being judged as a carrier image by the neural network is greater than that of being judged as an information camouflage image, it is determined that the original information camouflage image misled the neural network. When the probability of the original information camouflage image misleading the neural network is greater than a preset threshold, the original enhanced carrier image passes the test.

[0009] S4 embeds secret information into the enhanced carrier image after the test, generating an information camouflage target image.

[0010] According to the information camouflage method based on combined generation of adversarial examples provided by the present invention, step S2 includes:

[0011] Define a noise vector n;

[0012] The noise vector n is added to the carrier image C to obtain a new carrier image C. n The gradient η of the neural network is obtained through backpropagation algorithm. This gradient value is then multiplied by a preset coefficient ε to obtain the adversarial noise Na. d Embedding adversarial noise into the carrier image C n An enhanced carrier image C′ is generated in the middle.

[0013] According to the information camouflage method based on combined generative adversarial examples provided by the present invention, step S2, generating the enhanced carrier image C′ includes:

[0014] Perform several iterations on each carrier image Cn and the neural network with coefficients θ;

[0015] To simulate information camouflage, random noise Ni is added to Ci-1 to obtain a noisy image Cn. i The data is transmitted to the neural network.

[0016] Calculate and obtain the neural network gradient η for the i-th iteration. i Obtain the noisy image Cn i Corresponding adversarial noise Na di =εη i , will Cn i The i-th iteration update is completed on the corresponding adversarial noise embedding carrier image Ci-1, i.e., Ci = Ci-1 + Na. di ;

[0017] The correction vector Δ obtained in the Kth iteration is:

[0018]

[0019] Where Δ iIt is the cumulative correction vector obtained after i rounds of iterative training, i.e., Δ i =n ad j.

[0020] According to the present invention, an information spoofing method based on combined generative adversarial examples is used to perform robustness testing on the enhanced carrier image C′, including:

[0021] After K iterations, the pixel values ​​of the enhanced carrier image C′ are integers between 0 and 255;

[0022] V sets of random information are embedded into the enhanced carrier image C′ to obtain V information camouflage images {S1, S2, ..., S}. V};

[0023] When the probability of Si being judged by the neural network as a carrier image is greater than that of it being judged as an information spoofing image, it is determined that the original information spoofing image Si misled the neural network.

[0024] If the probability that the corresponding information spoofing image misleads the neural network is greater than a preset threshold T, then the original enhanced carrier image C′ is judged to have passed the test; otherwise, the enhanced carrier image C′ obtained in the Kth iteration is used. K Assign it to C, and proceed with the next K-round iteration.

[0025] According to the present invention, an information spoofing method based on combinatorial generation of adversarial examples is provided, in which secret information is embedded into an enhanced carrier image C′ that has passed a robustness test to generate an information spoofing target image S, and the information spoofing target image S is sent to the recipient.

[0026] Optionally, according to the information camouflage method based on combined generative adversarial examples provided by the present invention, an information camouflage image of an adversarial neural network is generated by combining a distortion-minimized information camouflage architecture with a generated enhanced adversarial vector.

[0027] On the other hand, the present invention also provides an information camouflage system based on combined adversarial sample generation, including a data acquisition module, an adversarial sample generation module, and a steganography module, wherein,

[0028] The data acquisition module is used to acquire carrier images for generating adversarial samples and information camouflage analysis;

[0029] The adversarial example generation module is used to generate random noise that is embedded in the adversarial example, embed the random noise into the carrier image, input the generated new carrier image into the adversarial neural network, calculate the gradient value of the neural network, obtain the adversarial noise based on the gradient value, and embed the adversarial noise into the carrier image to generate an enhanced carrier image.

[0030] The steganography module is used to test the enhanced carrier image by hiding some random information into the enhanced carrier image to generate several information camouflage images. When the probability of the information camouflage image being judged as a carrier image by the neural network is greater than that of being judged as an information camouflage image, it is determined that the original information camouflage image misled the neural network. When the probability of the original information camouflage image misleading the neural network is greater than a preset threshold, the original enhanced carrier image passes the test.

[0031] The steganography module is also used to embed secret information into the enhanced carrier image after it has passed the test, thereby generating an information-disguised target image.

[0032] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the information spoofing method based on combinatorial generation of adversarial examples as described above.

[0033] The present invention provides an information camouflage method and system based on combined generation of adversarial examples, which, compared with the prior art, has at least the following technical advantages:

[0034] This invention addresses the practical application needs of information camouflage technology, aiming to improve the ability of information camouflage to counter information camouflage analysis based on deep learning. By organically combining the generation method of adversarial samples with the information camouflage method, and designing corresponding information camouflage methods, it can more effectively counter information camouflage analysis methods based on deep learning, thereby effectively improving the security of information camouflage technology and enhancing its practicality and applicability in real-world applications.

[0035] Compared to other non-adaptive information camouflage algorithms, it has a lower detection rate under the same information camouflage analysis and detection environment, and can better resist the detection of information camouflage analysis mechanisms. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0037] Figure 1 This is a flowchart illustrating the information camouflage method based on the combined generation of adversarial examples provided by the present invention.

[0038] Figure 2 This is a schematic diagram of the information camouflage system based on combined generation of adversarial samples provided by the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0040] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or modules is not limited to the steps or modules listed, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to such process, method, product, or apparatus.

[0041] It should be noted that the terms "first" and "second" used in this invention merely distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first" and "second" can be interchanged in a specific order or sequence where permissible. It should be understood that the objects distinguished by "first" and "second" can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those described or illustrated herein.

[0042] In one embodiment, such as Figure 1 As shown, the present invention provides an information camouflage method based on combined generation of adversarial examples, comprising:

[0043] S1 acquires the carrier image used for generating adversarial examples and information camouflage analysis;

[0044] S2 generates random noise that embeds adversarial examples, embeds the random noise into the carrier image, and inputs the generated new carrier image into the adversarial neural network to calculate the gradient value of the neural network; based on the gradient value, it obtains adversarial noise, embeds the adversarial noise into the carrier image to generate an enhanced carrier image;

[0045] S3 tests the enhanced carrier image by hiding some random information into the enhanced carrier image to generate several information camouflage images. When the probability of the information camouflage image being judged as a carrier image by the neural network is greater than that of being judged as an information camouflage image, it is determined that the original information camouflage image misled the neural network. When the probability of the original information camouflage image misleading the neural network is greater than a preset threshold, the original enhanced carrier image passes the test.

[0046] S4 embeds secret information into the enhanced carrier image after the test, generating an information camouflage target image.

[0047] Specifically, step S2 includes:

[0048] Define a noise vector n. Specifically, the intensity of the noise should be comparable to the intensity of the information spoofing noise that needs to be resisted.

[0049] The noise vector n is added to the carrier image C to obtain a new carrier image C. n The gradient η of the neural network is obtained through backpropagation algorithm. This gradient value is then multiplied by a preset coefficient ε to obtain the adversarial noise Na. d Embedding adversarial noise into the carrier image C n An enhanced carrier image C′ is generated in the middle.

[0050] In step S2, to successfully generate the carrier image C′ and ensure its robustness, several iterations are typically required to generate the enhanced carrier image. Generating the enhanced carrier image C′ specifically includes:

[0051] For each carrier image Cn and the neural network with coefficients θ, several iterations are performed to ensure the stability and strength against noise.

[0052] To simulate information camouflage, random noise Ni is added to Ci-1 to obtain a noisy image Cn. i The data is transmitted to the neural network.

[0053] Calculate and obtain the neural network gradient η for the i-th iteration. i Obtain the noisy image Cn i Corresponding adversarial noise Na di =εη i , will Cn i The i-th iteration update is completed on the corresponding adversarial noise embedding carrier image Ci-1, i.e., Ci = Ci-1 + Na. di ;

[0054] The correction vector Δ obtained in the Kth iteration is:

[0055]

[0056] Where Δ i It is the cumulative correction vector obtained after i rounds of iterative training, i.e., Δ i =n adj ;

[0057] Specifically, as the number of iterations increases, C+Δ i Gradually adapt to noise interference, η i It converges to a vector that approaches zero.

[0058] Furthermore, the enhanced carrier image C′ obtained after iteration needs to be robustly tested, specifically including:

[0059] After K iterations, the pixel values ​​of the enhanced carrier image C′ are integers between 0 and 255;

[0060] A certain information camouflage algorithm and corresponding padding information amount can be used to hide V sets of random information into C′ to obtain V information camouflage images {S1, S2, ..., SV};

[0061] When the probability of Si being judged by the neural network as a carrier image is greater than that of it being judged as an information spoofing image, it is determined that the original information spoofing image Si misled the neural network.

[0062] When the probability that the corresponding information spoofing image misleads the neural network is greater than the preset threshold T, the original enhanced carrier image C′ is judged to have passed the test, and C will converge into a stable and robust carrier image; otherwise, the enhanced carrier image CK obtained in the Kth iteration is assigned as C, and the next K-round iteration cycle is carried out.

[0063] Furthermore, for the robust carrier image C′ that has passed the robustness test, the secret information is embedded into the robust carrier image C′ to generate the information disguised target image S, and the information disguised target image S is sent to the recipient.

[0064] This invention utilizes the stronger anti-information camouflage analysis and detection capabilities of adaptive algorithms compared to non-adaptive information camouflage algorithms. It organically combines the generation method of adversarial examples with the information camouflage method, enabling the invention to achieve a lower detection rate under the same information camouflage analysis and detection environment. It can better resist the detection of information camouflage analysis mechanisms and more effectively combat information camouflage analysis methods based on deep learning.

[0065] Specifically, when generating the information camouflage target image S, the information camouflage image of the adversarial neural network is generated by combining the information camouflage architecture with minimized distortion and the generated enhanced adversarial vector.

[0066] Optionally, let ni represent the adversarial noise of the i-th pixel in the image, and Pi. + Pi and 'i' represent the +-1 modification on the i-th pixel of the adaptive information camouflage algorithm, respectively. If ni is positive, Pi will be reduced according to the order of magnitude of the noise. + The value is the same as the value, and vice versa.

[0067] In another embodiment, the present invention also provides an information camouflage system based on combined generative adversarial examples. The system described below and the method described above can be referred to in correspondence. Specifically, as shown below... Figure 2As shown, the system includes a data acquisition module, an adversarial example generation module, and a steganography module, wherein,

[0068] The data acquisition module is used to acquire carrier images for generating adversarial samples and information camouflage analysis;

[0069] The adversarial example generation module is used for key steps such as random noise generation, gradient calculation, adversarial noise generation, and adversarial noise embedding. It includes: generating random noise to embed adversarial examples, embedding the random noise into the carrier image, inputting the generated new carrier image into the adversarial neural network, calculating the gradient value of the neural network, obtaining adversarial noise based on the gradient value, and embedding the adversarial noise into the carrier image to generate an enhanced carrier image.

[0070] The steganography module is used to embed information camouflage information into adversarial examples, including calculation of adversarial information camouflage capability and noise capability, and embedding of information camouflage information. It is used to test the enhanced carrier image by hiding several random pieces of information into the enhanced carrier image to generate several information camouflage images. When the probability of the information camouflage image being judged as a carrier image by the neural network is greater than that of being judged as an information camouflage image, it is determined that the original information camouflage image misled the neural network. When the probability of the original information camouflage image misleading the neural network is greater than a preset threshold, the original enhanced carrier image passes the test. Secret information is embedded into the enhanced carrier image that has passed the test to generate an information camouflage target image, i.e., a hidden sample.

[0071] The present invention also provides an electronic device, which may include a processor, a communications interface, a memory, and a communication bus, wherein the processor, the communications interface, and the memory communicate with each other via the communication bus. The processor can invoke logical instructions in the memory to execute the steps of the information spoofing method based on the combined generative adversarial examples provided by the above method.

[0072] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0073] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, and when the program instructions are executed by a computer, the computer is able to perform the steps of the information spoofing method based on the combined generation of adversarial samples provided by the above methods.

[0074] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the information spoofing method based on the combined generation of adversarial examples provided by the above methods.

[0075] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0076] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An information camouflage method based on combined generative adversarial examples, characterized in that, include: S1 acquires the carrier image used for generating adversarial examples and information camouflage analysis; S2 defines the noise vector n; The noise vector n is added to the carrier image C to obtain a new carrier image C. n Gradients of a neural network are obtained through backpropagation algorithm. The gradient value is compared with a preset coefficient. Multiplication to obtain adversarial noise Embedding adversarial noise into the carrier image C n Generate enhanced carrier images Among them, generating enhanced carrier images Includes: for each carrier image Cn and coefficients of The neural network is iterated several times; simulated information camouflage is performed to disguise random noise. Join Obtain a noisy image The data is transmitted to the neural network; the gradient of the neural network in the i-th iteration is calculated. Acquire noisy images Corresponding anti-noise ,Will Corresponding adversarial noise embedding carrier image Complete the first The next iteration update, i.e. ; Obtain the correction vector obtained in the Kth iteration for: ; in It was through The cumulative correction vector obtained after rounds of iterative training, i.e. ; S3 enhances the carrier image. Robustness testing was performed, including: after K iterations, the enhanced carrier image... The pixel values ​​are integers between 0 and 255; V sets of random information are embedded into the enhanced carrier image. Obtain V images of camouflaged information from the middle ;when If the probability of the neural network identifying the image as a carrier image is greater than that of it identifying it as an information spoofing image, then the original information spoofing image is identified. The neural network was misled; when the probability of the corresponding spoofed image misleading the neural network exceeds a preset threshold. When the original enhanced carrier image is determined, it is then judged. Pass the test; otherwise, use the enhanced carrier image obtained in the Kth iteration. Assigned to To proceed to the next time A loop of iteration; S4 embeds secret information into the enhanced carrier image after the test, generating an information-camouflaged target image; using To represent the first in the image Anti-noise per pixel, using and These represent the i-th pixel in the adaptive information camouflage algorithm. Modify; if If it is a positive number, it will be reduced according to the order of magnitude of the noise. The value is the same as the value, and vice versa.

2. The information camouflage method based on combined generation of adversarial examples according to claim 1, characterized in that, Disguising secret information and embedding it into a robust carrier image that has passed robustness tests. In the process, information is generated to disguise the target image. and disguise the information as a target image. Send to the recipient.

3. The information camouflage method based on combined generative adversarial examples according to any one of claims 1-2, characterized in that, By combining a distortion-minimized information camouflage architecture with a generated enhanced adversarial vector, an information camouflage image of an adversarial neural network is generated.

4. An information camouflage system based on combined generative adversarial examples, used to implement the information camouflage method based on combined generative adversarial examples as described in any one of claims 1 to 3, characterized in that, It includes a data acquisition module, an adversarial example generation module, and a steganography module, among which: The data acquisition module is used to acquire carrier images for generating adversarial samples and information camouflage analysis; The adversarial example generation module is used to generate random noise that is embedded in the adversarial example, embed the random noise into the carrier image, input the generated new carrier image into the adversarial neural network, calculate the gradient value of the neural network, obtain the adversarial noise based on the gradient value, and embed the adversarial noise into the carrier image to generate an enhanced carrier image. The steganography module is used to test the enhanced carrier image by hiding some random information into the enhanced carrier image to generate several information camouflage images. When the probability of the information camouflage image being judged as a carrier image by the neural network is greater than that of being judged as an information camouflage image, it is determined that the original information camouflage image misled the neural network. When the probability of the original information camouflage image misleading the neural network is greater than a preset threshold, the original enhanced carrier image passes the test. The steganography module is also used to embed secret information into the enhanced carrier image after it has passed the test, thereby generating an information-disguised target image.

5. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the information camouflage method based on the combined generation of adversarial examples as described in any one of claims 1 to 3.

Images