Image steganography method, program product, storage medium, and electronic device

Abstract

The application relates to the technical field of information security, and provides an image steganography method, a program product, a storage medium and an electronic device. The image steganography method comprises the following steps: encrypting an original text in a predetermined manner, and determining first ciphertext according to an encryption result; processing the first ciphertext to obtain second ciphertext, and recording restoration information for restoring the second ciphertext into the first ciphertext; generating first target text based on the second ciphertext by using the text-to-text capability of a first large model; generating a first carrier image based on the first target text by using the text-to-image capability of a second large model; and writing the restoration information into the first carrier image by using an image steganography algorithm to obtain a first steganography image. The method uses a large model to dynamically generate a carrier image, theoretically can generate an infinite number of carrier images, thereby greatly reducing the probability that the steganography image is suspected, intercepted and recognized, and is beneficial to improving the security of information transmission.

Description

Technical Field

[0001] This application relates to the field of information security technology, and more specifically, to an image steganography method, a program product, a storage medium, and an electronic device. Background Technology

[0002] Information hiding technology is an important area of ​​information security research. Compared with the WYSIWYG nature of text and the WYSIWYG nature of audio, images often hide more rich and redundant information. This information is hidden in the image and will not cause visual deviation from the image itself.

[0003] Image steganography hides messages within an image without altering its perceptual properties, ensuring that the image containing the embedded information looks visually very similar to the original image, thus avoiding being noticed, suspected, intercepted, or attacked by others.

[0004] However, in existing technologies, one or more carrier images are usually used for image steganography, which leads to these images being suspected and subsequently intercepted and identified. Summary of the Invention

[0005] The purpose of this application is to provide an image steganography method, program product, storage medium, and electronic device to improve at least some of the above-mentioned technical problems.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] In a first aspect, embodiments of this application provide an image steganography method, comprising: encrypting original text according to a pre-defined method, and determining a first ciphertext based on the encryption result, wherein the first ciphertext is readable but non-fluent text; processing the first ciphertext to obtain a second ciphertext, and recording restoration information for restoring the second ciphertext to the first ciphertext; wherein the second ciphertext is readable but non-fluent text; generating a first target text based on the second ciphertext using the text-to-text capability of a first large model, wherein the first target text is fluent text; generating a first carrier image based on the first target text using the text-to-image capability of a second large model; and writing the restoration information into the first carrier image using an image steganography algorithm to obtain a first steganographic image.

[0008] The above method has at least the following advantages:

[0009] Firstly, by using large models to dynamically generate carrier images, theoretically an infinite number of carrier images can be produced, thereby greatly reducing the probability of steganographic images being suspected, intercepted, or identified, which is conducive to improving the security of information transmission.

[0010] Secondly, large models exhibit a degree of randomness in image generation; even with identical input prompts, the generated images are usually different. This characteristic further enriches the number of carrier images, which helps reduce the probability of steganographic images being suspected, intercepted, or detected.

[0011] Third, large models have developed rapidly in recent years, and the quality and content of the images they generate are far superior to those of small models in the traditional field of computer vision. This helps to improve the quality of carrier images and further reduces the probability of steganographic images being suspected, intercepted, or identified.

[0012] Fourth, the carrier image is not generated out of thin air by the large model, but has a certain connection with the original text (the first target text is generated based on the original text). This makes the generated carrier image somewhat controllable in terms of content, further reducing the probability that the steganographic image will be suspected, intercepted, or identified.

[0013] Fifth, the object of steganography is not the first ciphertext, but the information that restores the second ciphertext to the first ciphertext. Thus, this method actually performs double encryption during the steganography process (once for the first ciphertext and once for the second ciphertext), which makes it highly secure.

[0014] Sixth, this method is also conducive to the commercialization of multimodal large model technology, as well as the development of natural language processing technology and artificial intelligence technology.

[0015] In one implementation of the first aspect, the step of processing the first ciphertext to obtain the second ciphertext and recording the restoration information of restoring the second ciphertext to the first ciphertext includes: segmenting the first ciphertext into words, randomly shuffling the word order in the segmentation result to obtain the second ciphertext, and recording each word in the second ciphertext and the position of that word in the first ciphertext.

[0016] In the above implementation, since the second ciphertext is randomly generated based on the word segmentation results of the first ciphertext, even if the first ciphertext is the same, the second ciphertext may differ significantly, resulting in substantial differences in the subsequently generated carrier images, which helps to enrich the number of carrier images. Furthermore, the second ciphertext merely reorganizes the words in the first ciphertext without compromising its readability, thus facilitating the subsequent generation of the first target text.

[0017] In one implementation of the first aspect, if the encryption result is unreadable text, then determining the first ciphertext based on the encryption result includes: mapping the characters in the encryption result to readable characters to obtain the first ciphertext.

[0018] In the above implementation, if the encryption result is unreadable text, it can be converted into readable first ciphertext, which is beneficial for using more encryption algorithms to generate the first ciphertext (because many encryption algorithms produce unreadable ciphertext), thus improving the practicality of the method.

[0019] In one implementation of the first aspect, the step of generating a first carrier image based on the first target text using the text-to-image capability of the second model includes: determining a first target prompt word based on the first target text, wherein the first target prompt word requires the generation of a meaningful image based on the first target text; and inputting the first target prompt word into the second model to obtain the first carrier image generated by the second model.

[0020] In the above implementation, the first target prompt word limits the generated carrier image to a meaningful image, avoiding the generation of images with incomprehensible content, which could lead to the steganographic image being suspected, intercepted, or even detected (because meaningless images are usually not transmitted).

[0021] In one implementation of the first aspect, the first steganographic image is sent to a receiver, and the first target prompt requires the generation of a meaningful image containing only conventional content based on the first target text.

[0022] In the above implementation, the first target prompt word limits the generated carrier image to contain only conventional content (e.g., human face, animal, landscape), avoiding the generation of images with rare content, which could lead to the steganographic image being suspected, intercepted, or even detected (because images with conventional content are usually transmitted).

[0023] In one implementation of the first aspect, a plurality of first carrier images generated sequentially contain associated content.

[0024] In the above implementation, if multiple carrier images are generated continuously, the content of these images can be associated with each other through the first target prompt word, so as to avoid generating images that are completely isolated in content, which would lead to the steganographic images being suspected or even intercepted and identified (because images transmitted continuously are usually related in content).

[0025] Secondly, embodiments of this application provide an image steganography method, comprising: encrypting the original text according to a pre-agreed method, and determining a first ciphertext based on the encryption result, wherein the first ciphertext is readable but non-fluent text; generating a second target text based on the first ciphertext using the text-to-text capability of a first large model, wherein the second target text is fluent text; generating a second carrier image based on the second target text using the text-to-image capability of a second large model; and writing the first ciphertext into the second carrier image using an image steganography algorithm to obtain a second steganographic image.

[0026] The above method has at least the following advantages:

[0027] Firstly, by using large models to dynamically generate carrier images, theoretically an infinite number of carrier images can be produced, thereby greatly reducing the probability of steganographic images being suspected, intercepted, or identified, which is conducive to improving the security of information transmission.

[0028] Secondly, large models exhibit a degree of randomness in image generation; even with identical input prompts, the generated images are usually different. This characteristic further enriches the number of carrier images, which helps reduce the probability of steganographic images being suspected, intercepted, or detected.

[0029] Third, large models have developed rapidly in recent years, and the quality and content of the images they generate are far superior to those of small models in the traditional field of computer vision. This helps to improve the quality of carrier images and further reduces the probability of steganographic images being suspected, intercepted, or identified.

[0030] Fourth, the carrier image is not generated out of thin air by the large model, but has a certain connection with the original text (the second target text is generated based on the original text). This makes the generated carrier image somewhat controllable in terms of content, further reducing the probability that the steganographic image will be suspected or even intercepted and identified.

[0031] Fifth, the method has relatively simple steps and a relatively efficient steganography process, which is conducive to batch steganography.

[0032] Sixth, this method is also conducive to the commercialization of multimodal large model technology, as well as the development of natural language processing technology and artificial intelligence technology.

[0033] Thirdly, embodiments of this application provide an image steganography device, comprising: a first encryption module, configured to encrypt original text according to a pre-defined method and determine a first ciphertext based on the encryption result, wherein the first ciphertext is readable but non-fluent text; a first processing module, configured to process the first ciphertext to obtain a second ciphertext and record restoration information for restoring the second ciphertext to the first ciphertext; wherein the second ciphertext is readable but non-fluent text; a first text-to-text module, configured to generate a first target text based on the second ciphertext using the text-to-text capability of a first large model, wherein the first target text is fluent text; a first text-to-image module, configured to generate a first carrier image based on the first target text using the text-to-image capability of a second large model; and a first steganography module, configured to write the restoration information into the first carrier image using an image steganography algorithm to obtain a first steganography image.

[0034] Fourthly, embodiments of this application provide an image steganography method, comprising: a second encryption module, configured to encrypt the original text according to a pre-defined method and determine a first ciphertext based on the encryption result, wherein the first ciphertext is readable but non-fluent text; a second text-to-text module, configured to generate a second target text based on the first ciphertext using the text-to-text generation capability of a first large model, wherein the second target text is fluent text; a second text-to-image module, configured to generate a second carrier image based on the second target text using the text-to-image generation capability of a second large model; and a second steganography module, configured to write the first ciphertext into the second carrier image using an image steganography algorithm to obtain a second steganography image.

[0035] Fifthly, embodiments of this application provide a computer program product, including computer program instructions, which, when read and executed by a processor, perform the method provided by the first aspect, the second aspect, or any one of the two aspects.

[0036] Sixthly, embodiments of this application provide a computer-readable storage medium storing computer program instructions, which, when read and executed by a processor, perform the method provided by the first aspect, the second aspect, or any one of the two aspects.

[0037] In a seventh aspect, embodiments of this application provide an electronic device, including: a memory and a processor, wherein the memory stores computer program instructions, and the computer program instructions are read and executed by the processor to perform the method provided by any implementation of the first aspect, the second aspect, or both aspects. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 The sending flow of the first image steganography method provided in the embodiments of this application is shown;

[0040] Figure 2 The receiving flow of the first image steganography method provided in the embodiments of this application is illustrated;

[0041] Figure 3 The sending flow of the second image steganography method provided in this application embodiment is shown;

[0042] Figure 4 The receiving flow of the second image steganography method provided in the embodiments of this application is illustrated;

[0043] Figure 5 This application illustrates the functional modules included in the first image steganography apparatus provided in this embodiment;

[0044] Figure 6 This application illustrates the functional modules included in the second image steganography apparatus provided in an embodiment of the present application;

[0045] Figure 7 The possible structure of the electronic device provided in the embodiments of this application is illustrated. Detailed Implementation

[0046] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. It should be noted that similar reference numerals and letters in the following drawings indicate similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0047] The terms “comprising,” “including,” or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0048] The terms “first,” “second,” etc., are used only to distinguish one entity or operation from another, and should not be construed as indicating or implying relative importance, nor as requiring or implying any such actual relationship or order between these entities or operations.

[0049] In general, this application provides two image steganography methods (of course, these two methods can also include several implementations). The first image steganography method calculates both the first and second ciphertexts, while the second image steganography method only calculates the first ciphertext. Both image steganography methods include processes executed by the sender and processes executed by the receiver. Figure 1 The sender flow of the first image steganography method is shown. Figure 2 The receiver flow of the first image steganography method is shown. Figure 3 The sender flow of the second image steganography method is shown. Figure 4 The receiver flow of the second image steganography method is shown.

[0050] In other words, both image steganography methods involve a sender and a receiver. The sender is responsible for embedding the information to be transmitted into a carrier image, obtaining a steganographic image, and then sending the steganographic image to the receiver. The receiver is responsible for reconstructing the information to be transmitted from the steganographic image. Thus, image steganography technology enables the covert transmission of information without human detection. Here, the sender and receiver can be understood as the device and / or software on that device used by the user sending the information, and the device and / or software on that device used by the user receiving the information.

[0051] However, it should be understood that after obtaining the steganographic image, the sender is not necessarily obligated to send it to the receiver. For example, it could be stored as a regular image, displayed, etc. (If the image is not sent, the sender may not be called a "sender" in this case). Therefore, Figure 1 Step S160 and Figure 3 Step S350 in the text are all optional. However, considering that information transmission is one of the main application scenarios of image steganography, for the sake of simplicity, the following explanation will use the information transmission scenario as an example. Other scenarios can be understood and analyzed in a similar way.

[0052] (1) The first image steganography method

[0053] Reference Figure 1 The steps performed by the sender in this method include:

[0054] Step S110: Encrypt the original text according to the agreed method, and determine the first ciphertext based on the encryption result.

[0055] The original text contains the information that the sender wants to convey to the receiver. Of course, it doesn't mean that the original text is directly transmitted to the receiver; it can simply be some equivalent information of the original text, as long as the receiver can reconstruct the original text based on that equivalent information.

[0056] To enhance security, the original text can be encrypted first, according to a method agreed upon by the sender and receiver, to obtain the encrypted result. This application does not limit the encryption algorithm agreed upon by the sender and receiver; for example, it can be symmetric encryption, asymmetric encryption, etc.

[0057] The encryption result yields the first ciphertext, which is readable but not entirely coherent text. "Readable" means the first ciphertext consists entirely of readable characters. These readable characters can be defined in different ways:

[0058] One approach is to define readable characters as visible characters, such as Chinese characters, numbers, operators, punctuation marks, English letters, Greek letters, etc. Some characters are invisible and are therefore defined as unreadable characters in this way.

[0059] One approach is to define readable characters as those that are visible and appear in ordinary conversational contexts. For example, some special symbols whose meanings are almost unknown to ordinary people and do not appear in ordinary conversations or articles are not considered readable characters, even though they are visible.

[0060] Of course, other definition methods are not excluded.

[0061] "Incoherent text" refers to the first ciphertext not consisting of fluent and meaningful sentences. For example, the first ciphertext may consist of a series of Chinese characters, but if these characters are randomly arranged and cannot express clear semantics, then it falls under the category of "incoherent text" as described here. However, it should be noted that "incoherent" here is only in an algorithmic or probabilistic sense, not in an absolute sense. That is to say, after executing the algorithm in step S110, a readable text, i.e., the first ciphertext, will be obtained, but the algorithm will not deliberately ensure that the generated first ciphertext is fluent. Therefore, the first ciphertext is likely to be incoherent, but at the probabilistic level, the possibility of generating a fluent text by chance cannot be ruled out.

[0062] In one implementation, if the encryption result is itself readable but incoherent text, it can be directly used as the first ciphertext. If the encryption result is unreadable text (i.e., contains at least one unreadable character), it can be converted into readable text to obtain the first ciphertext. For example, characters in the encryption result can be mapped to readable characters according to a certain preset mapping rule to obtain the first ciphertext. Mapping can be performed only on unreadable characters in the encryption result, or on both unreadable and readable characters. Furthermore, mapping can be performed on each character individually or on several characters together. Note that the process of converting the encryption result into readable text does not consider the fluency of the text, thus the resulting text conforms to the incoherent nature of the first ciphertext.

[0063] Considering that many encryption algorithms generate unreadable ciphertext, the above implementation method allows for the use of more encryption algorithms in the generation of the first ciphertext, increasing the freedom in the selection of encryption algorithms and thus improving the practicality of image steganography methods.

[0064] Step S120: Process the first ciphertext to obtain the second ciphertext, and record the restoration information of restoring the second ciphertext to the first ciphertext.

[0065] The second ciphertext is also readable but not entirely coherent text; that is, processing the first ciphertext does not impair readability. The process from the first ciphertext to the second is reversible. The restored information can be some kind of information generated based on the second ciphertext. By processing the restored information in a predetermined manner, the first ciphertext can be obtained. Therefore, the restored information can replace the first ciphertext for subsequent information transmission. Examples regarding the processing of the first ciphertext and the specific content of the restored information will be given later.

[0066] Step S130: Utilize the text-to-text capability of the first model to generate the first target text based on the second ciphertext.

[0067] Step S140: Utilize the text-to-image capability of the second model to generate the first carrier image based on the first target text.

[0068] The above two steps are explained together.

[0069] Large models are artificial intelligence models with massive amounts of parameters (e.g., billions or more). Trained through deep learning on large-scale data, they possess powerful language understanding, generation, and multi-task processing capabilities, and can be widely applied in many fields such as natural language processing and image recognition, providing core support for intelligent applications. Here, "large" can be understood as the scale of the model compared to traditional artificial intelligence models (e.g., image classification models).

[0070] Multimodal large models are trained by integrating multiple modal data (such as text, images, audio, video, etc.) on the basis of large models. They can understand and generate multiple modal information, realize cross-modal perception and interaction, and have more powerful comprehensive intelligent processing capabilities. They can be applied to complex scenarios such as intelligent driving, intelligent security, and content creation.

[0071] The most common use of large models is to provide user-input (or automated input from a program) prompts (which can be phrases or sentences). The large model then processes these prompts to generate the target content and output it, such as text, images, or music. The output of large models has a degree of randomness; even if the user inputs the exact same prompts, the large model will usually output different results (although these results may be semantically similar or identical). Furthermore, some large models also allow users to directly input images, voice recordings, etc.

[0072] Currently, there are many platforms on the market based on large models that can be called by the methods in this application, such as ChatGPT, DeepSeek, and Kimi. Of course, you can also implement the large model that you want to call in the method yourself.

[0073] The first major model is a text-to-text model, where the user inputs text-based prompts, and the first major model outputs text. The second major model is a text-to-image model, where the user inputs text-based prompts, and the second major model outputs an image. In different implementations, the first and second major models can be two different models (e.g., the first major model uses the ChatGPT model, and the second major model uses the Stable Diffusion model), or they can be the same model (e.g., a multimodal model).

[0074] The primary function of the first model here is to transform the second ciphertext from an incoherent text into a coherent text (coherent text must be readable), i.e., the first target text, to facilitate the generation of images by the second model. For example, the prompt word "Please use your imagination to add necessary words and transform the following disordered sentence into a coherent paragraph: [Fill in the second ciphertext here]" can be generated first. Then, this prompt word can be input into the first model, which will output the first target text that meets the prompt word requirements.

[0075] Optionally, more restrictions can be added to the prompt words to meet the needs of text-to-text generation, such as requiring the generated text to contain all the words in the second ciphertext, etc.

[0076] The second major model here functions to generate one or more images based on the first target text, serving as the primary carrier image for image steganography. For example, the prompt phrase "Please generate an image based on the following statement: [Enter the first target text here]" can be generated first, and then this prompt phrase can be input into the second major model, which will output the primary carrier image that meets the requirements of the prompt phrase.

[0077] Optionally, more restrictions can be added to the prompts based on the needs of the text-to-image generation, such as requiring the generated image to contain a certain theme, or to have a 4:3 aspect ratio, or to have a realistic image style, etc.

[0078] Step S150: Using an image steganography algorithm, the restored information is written into the first carrier image to obtain the first steganographic image.

[0079] This application does not limit the image steganography algorithm used; for example, it can be an algorithm such as least significant bit (LSB) replacement. The first steganographic image and the first carrier can be visually identical or highly similar.

[0080] Step S160: Send the first steganographic image to the receiver.

[0081] The sender and receiver can agree on a time for sending the first steganographic image so that the receiver can prepare to receive it. Optionally, for security reasons, this time can be just an approximate time period, such as 22:00-23:00, rather than a precise moment. Furthermore, in addition to sending the first steganographic image, the sender may also simultaneously send some unsteganized images to the receiver to prevent the first steganographic image from being too obvious and thus detected by malicious actors.

[0082] Reference Figure 2 The steps performed by the receiver in this method include:

[0083] Step S210: Receive suspicious images.

[0084] When sending images, the sender may send not only the first steganographic image but also a regular image. Furthermore, besides the sender of the first steganographic image, there may be other senders sending regular images to the receiver. Therefore, the images received by the receiver are currently only suspicious images and need to be further filtered to identify the first steganographic image.

[0085] For example, referring to step S160, the receiver can receive images between 22:00 and 23:00, and all received images will be considered as suspicious images.

[0086] Step S220: Use an image steganography algorithm to perform a reverse operation and extract the reconstructed information from the image.

[0087] The image steganography algorithm in step S220 should correspond to the image steganography algorithm in step S150. If the suspected image is the first steganography image, then the image steganography algorithm can be used to reverse the operation to obtain the restoration information (or if the restoration information can be parsed, then the image can be determined to be the first steganography image); if the suspected image is not the first steganography image, then the image steganography algorithm cannot be used to reverse the operation to obtain the restoration information, and such images can be skipped.

[0088] Step S230: Restore the first ciphertext based on the restored information.

[0089] Step S240: Decrypt the first ciphertext according to the agreed method to obtain the original text.

[0090] The decryption method in step S240 should correspond to the encryption method in step S110, such as using the same algorithm or the same (or the same pair of) keys.

[0091] The first image steganography method has at least the following advantages:

[0092] Firstly, by using large models to dynamically generate carrier images, theoretically an infinite number of carrier images can be produced, thereby greatly reducing the probability of steganographic images being suspected, intercepted, or identified, which is conducive to improving the security of information transmission.

[0093] Secondly, as mentioned earlier, large models exhibit a degree of randomness in image generation. Even with identical input prompts, the generated images are usually not the same. This characteristic further enriches the number of carrier images, which helps reduce the probability of steganographic images being suspected, intercepted, or detected.

[0094] Third, large models have developed rapidly in recent years, and the quality and content of the images they generate are far superior to those of small models in the traditional field of computer vision. This helps to improve the quality of carrier images and further reduces the probability of steganographic images being suspected, intercepted, or identified.

[0095] Fourth, the carrier image is not generated out of thin air by the large model, but rather has a certain connection with the original text (the first target text is generated based on the original text). This makes the content of the generated carrier image somewhat controllable, further reducing the probability of the steganographic image being suspected, intercepted, or detected. For example, if the sender and receiver pretend to be chatting and send images, the images are usually related to daily life. However, if the carrier image is generated completely randomly by the large model, it may generate some images with content unrelated to daily life, or even bizarre and illogical images, making it easy to detect anomalies. But if some daily life-related content is mixed into the original text, the carrier image generated according to this method is usually related to daily life, greatly reducing the probability of being suspected.

[0096] Fifth, the object of steganography is not the first ciphertext, but the information that restores the second ciphertext to the first ciphertext. Thus, this method actually performs double encryption during the steganography process (once for the first ciphertext and once for the second ciphertext), which makes it highly secure.

[0097] Sixth, this method is also conducive to the commercialization of multimodal large model technology, as well as the development of natural language processing technology and artificial intelligence technology.

[0098] Optionally, step S120 may include the following implementations:

[0099] First, the first ciphertext is segmented to obtain the segmentation results; then, the word order in the segmentation results is randomly shuffled to obtain the second ciphertext; finally, each word in the second ciphertext and its position in the first ciphertext are recorded.

[0100] The "each word in the second ciphertext and the position of that word in the first ciphertext" is the restoration information mentioned in step S120. For example, the restoration information can be represented as a key-value pair Pairs(index, voc), where voc represents a word in the second ciphertext and index represents the position of that word in the first ciphertext. In step S230, based on the key-value pair, each voc is placed at the position indicated by index to restore the first ciphertext.

[0101] In the above implementation, since the second ciphertext is randomly generated based on the word segmentation results of the first ciphertext, even if the first ciphertext is the same (for example, the original text to be sent is the same in two separate instances), the second ciphertext may differ significantly, resulting in substantial differences in the subsequently generated carrier images, which helps to enrich the number of carrier images. Furthermore, the second ciphertext merely reorganizes the words in the first ciphertext without compromising its readability, thus facilitating the subsequent generation of the first target text.

[0102] It should be understood that step S120 can also be implemented in other ways. For example, the first ciphertext can be split into characters, then randomly shuffled to obtain the second ciphertext, and the information can be restored to each character in the second ciphertext and the position of that character in the first ciphertext.

[0103] For example, first, the first ciphertext (assuming it is Chinese) is segmented to obtain the segmentation results; then, each word is mapped to an English word according to a preset mapping relationship to obtain the second ciphertext; finally, each word in the second ciphertext and the position of that word in the second ciphertext are recorded as restoration information (when restoring, the second ciphertext can be restored first according to the restoration information, and then the mapping relationship can be applied in reverse to restore the first ciphertext).

[0104] Optionally, step S140 may include the following implementations:

[0105] First, a first target prompt word is determined based on the first target text. The first target prompt word requires the generation of a meaningful image based on the first target text. Here, "meaningful" can be defined as something that a normal person can understand. Then, the first target prompt word is input into the second model to obtain the first carrier image generated by the second model. For example, the first prompt word could be "Please generate an image based on the following statement, requiring the image content to be meaningful and logical: [Fill in the first target text here]".

[0106] In the above implementation, the first target prompt specifies that the generated carrier image must be a meaningful image to avoid generating images with incomprehensible content, which could lead to suspicion, interception, or detection of the steganographic image, since meaningless images are generally not transmitted. For example, without this restriction, the content of the carrier image might be meaningless jumbled lines or meaningless symbols, which would easily be noticed as abnormal.

[0107] Optionally, in addition to requiring the carrier image to be meaningful, the first target prompt can be further limited to the carrier image containing only conventional content, such as faces, animals, and landscapes. This content can be obtained by identifying and statistically analyzing a large number of images transmitted on the network. For example, the 100 most frequently occurring objects in network images can be defined as conventional content, and the carrier image can be one or a combination of these contents. For example, the first prompt could be "Please generate an image based on the following statement, requiring the image content to be meaningful, logical, and containing only one or more of the following three contents: faces, animals, and landscapes: [Fill in the first target text here]".

[0108] This approach helps avoid generating images with unusual content, which could lead to suspicion, interception, or detection of the steganographic images, since images with conventional content are usually transmitted. For example, the sender and receiver might pretend to be chatting casually, but if the images contain grand themes like the universe or stars, it could easily arouse suspicion.

[0109] Optionally, if multiple carrier images are generated consecutively, the first target prompt word can be used to make the generated carrier images contain related content, so as to avoid generating images that are completely isolated in content, which would lead to the steganographic images being suspected or even intercepted and identified, because the images transmitted consecutively are usually related in content.

[0110] For example, the first prompt could be "Please generate an image based on the following statement. The image content should be meaningful, logical, and related to daily life: [Fill in the first target text here]". Similar prompts would be used in all carrier images generated within half an hour (except for the content in brackets, which should be replaced). Then, the image theme would be changed to generate a new prompt template.

[0111] (2) The second image steganography method

[0112] Reference Figure 3 The steps performed by the sender in this method include:

[0113] Step S310: Encrypt the original text according to the agreed method, and determine the first ciphertext based on the encryption result. The first ciphertext is a readable but non-fluent text.

[0114] Step S310 can be understood by referring to step S110, and will not be repeated here.

[0115] Step S320: Utilize the text-to-text capability of the first model to generate the second target text based on the first ciphertext. The second target text is a fluent text.

[0116] Step S320 can be understood with reference to step S130. However, in step S320, a second ciphertext is not generated. Instead, the first ciphertext is processed directly using the large model. Other similarities will not be repeated.

[0117] Step S330: Utilize the text-to-image capability of the second model to generate a second carrier image based on the second target text.

[0118] Step S330 can be understood by referring to step S140, and will not be repeated here.

[0119] Step S340: Using an image steganography algorithm, the first ciphertext is written into the second carrier image to obtain the second steganographic image.

[0120] Step S340 can be understood with reference to step S150. However, in step S340, the information written into the carrier image is not restoration information, but the first ciphertext. Other similarities will not be repeated.

[0121] Step S350: Send the second steganographic image to the receiver.

[0122] Step S350 can be understood by referring to step S160, and will not be repeated here.

[0123] Reference Figure 4 The steps performed by the receiver in this method include:

[0124] Step S410: Receive suspicious images.

[0125] Step S410 can be understood by referring to step S210, and will not be repeated here.

[0126] Step S420: Perform a reverse operation using an image steganography algorithm to parse out the first ciphertext in the image.

[0127] Step S420 can be understood with reference to step S220. However, since the first ciphertext is written in step S340, the parsed information in step S420 is also the first ciphertext, not the restored information. Other similarities will not be repeated.

[0128] Step S430: Decrypt the first ciphertext according to the agreed method to obtain the original text.

[0129] Step S430 can be understood by referring to step S240, and will not be repeated here.

[0130] For the remaining details of the second image steganography method, those not mentioned herein, please refer to the corresponding content of the first image steganography method. The second image steganography method has at least the following advantages:

[0131] Firstly, by using large models to dynamically generate carrier images, theoretically an infinite number of carrier images can be produced, thereby greatly reducing the probability of steganographic images being suspected, intercepted, or identified, which is conducive to improving the security of information transmission.

[0132] Secondly, large models exhibit a degree of randomness in image generation; even with identical input prompts, the generated images are usually different. This characteristic further enriches the number of carrier images, which helps reduce the probability of steganographic images being suspected, intercepted, or detected.

[0133] Third, large models have developed rapidly in recent years, and the quality and content of the images they generate are far superior to those of small models in the traditional field of computer vision. This helps to improve the quality of carrier images and further reduces the probability of steganographic images being suspected, intercepted, or identified.

[0134] Fourth, the carrier image is not generated out of thin air by the large model, but has a certain connection with the original text (the second target text is generated based on the original text). This makes the generated carrier image somewhat controllable in terms of content, further reducing the probability that the steganographic image will be suspected or even intercepted and identified.

[0135] Fifth, the method has relatively simple steps and a relatively efficient steganography process, which is conducive to batch steganography.

[0136] Sixth, this method is also conducive to the commercialization of multimodal large model technology, as well as the development of natural language processing technology and artificial intelligence technology.

[0137] Figure 5 The illustration shows the functional modules that may be included in the image steganography apparatus 500 provided in an embodiment of this application. (Refer to...) Figure 5 The image steganography device 500 includes:

[0138] The first encryption module 510 is used to encrypt the original text according to the agreed method, and determine the first ciphertext according to the encryption result. The first ciphertext is readable but non-fluent text.

[0139] The first processing module 520 is used to process the first ciphertext to obtain the second ciphertext, and record the restoration information of restoring the second ciphertext to the first ciphertext; wherein the second ciphertext is readable but non-fluent text;

[0140] The first text-to-text module 530 is used to generate a first target text based on the second ciphertext by utilizing the text-to-text capability of the first large model. The first target text is a fluent text.

[0141] The first text-to-image module 540 is used to generate a first carrier image based on the first target text by utilizing the text-to-image capability of the second large model.

[0142] The first steganography module 550 is used to write the restored information into the first carrier image using an image steganography algorithm to obtain a first steganography image.

[0143] In one implementation of the image steganography device 500, the first processing module 520 processes the first ciphertext to obtain the second ciphertext and records the restoration information of restoring the second ciphertext to the first ciphertext, including: segmenting the first ciphertext into words, randomly shuffling the word order in the segmentation result to obtain the second ciphertext, and recording each word in the second ciphertext and the position of the word in the first ciphertext.

[0144] In one implementation of the image steganography device 500, if the encryption result is unreadable text, the first encryption module 510 determines the first ciphertext based on the encryption result, including: mapping the characters in the encryption result to readable characters to obtain the first ciphertext.

[0145] In one implementation of the image steganography device 500, the first text-generated image module 540 utilizes the text-generated image capability of the second large model to generate a first carrier image based on the first target text, including: determining a first target prompt word based on the first target text, wherein the first target prompt word requires the generation of a meaningful image based on the first target text; and inputting the first target prompt word into the second large model to obtain the first carrier image generated by the second large model.

[0146] In one implementation of the image steganography device 500, the first steganographic image is used to send to a receiver, and the first target prompt requires the generation of a meaningful image containing only conventional content based on the first target text.

[0147] In one implementation of the image steganography device 500, a plurality of first carrier images generated in succession contain associated content.

[0148] The image steganography apparatus 500 provided in this application embodiment can be used to execute the first image steganography method provided in this application embodiment. Its implementation principle and the resulting technical effects have been described in the foregoing method embodiments. For the sake of brevity, any part not mentioned in the apparatus embodiment can be referred to the corresponding content in any of the foregoing method embodiments.

[0149] Figure 6 The illustration shows the functional modules that may be included in the image steganography apparatus 600 provided in an embodiment of this application. (Refer to...) Figure 6 The image steganography device 600 includes:

[0150] The second encryption module 610 is used to encrypt the original text according to the agreed method, and determine the first ciphertext based on the encryption result. The first ciphertext is readable but non-fluent text.

[0151] The second text-to-text module 620 is used to generate a second target text based on the first ciphertext by utilizing the text-to-text capability of the first model. The second target text is fluent text.

[0152] The second text-to-image module 630 is used to generate a second carrier image based on the second target text by utilizing the text-to-image capability of the second large model.

[0153] The second steganography module 640 is used to write the first ciphertext into the second carrier image using an image steganography algorithm to obtain a second steganography image.

[0154] The image steganography apparatus 600 provided in this application embodiment can be used to execute the second image steganography method provided in this application embodiment. Its implementation principle and the resulting technical effects have been described in the foregoing method embodiments. For the sake of brevity, any part not mentioned in the apparatus embodiment can be referred to the corresponding content in any of the foregoing method embodiments.

[0155] Figure 7 This illustration shows a possible structure of the electronic device 700 provided in an embodiment of this application. (Refer to...) Figure 7 The electronic device 700 includes a processor 710 and a memory 720, which are interconnected and communicate with each other via a communication bus 730 and / or other forms of connection mechanism (not shown).

[0156] The processor 710 includes one or more (only one is shown in the figure), which can be an integrated circuit chip with signal processing capabilities. The processor 710 can be a general-purpose processor, including a Central Processing Unit (CPU), a Microcontroller Unit (MCU), a Network Processor (NP), or other conventional processors; it can also be a special-purpose processor, including a Graphics Processing Unit (GPU), a Neural-network Processing Unit (NPU), a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Furthermore, when there are multiple processors 710, some can be general-purpose processors and others can be special-purpose processors.

[0157] The memory 720 includes one or more (only one is shown in the figure), which may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0158] The processor 710 and other possible components can access the memory 720, reading and / or writing data therein. In particular, one or more computer program instructions can be stored in the memory 720, and the processor 710 can read and execute these computer program instructions to implement the image steganography method provided in the embodiments of this application.

[0159] Understandable. Figure 7 The structure shown is for illustrative purposes only; the electronic device 700 may also include more than [other components]. Figure 7 The more or fewer components shown, or having the same Figure 7 Different configurations are shown. For example, if electronic device 700 needs to communicate with other devices, it may include a communication unit, which can be a wired and / or wireless communication module. For instance, both the sending and receiving devices can include communication units for sending or receiving stegographic images.

[0160] Figure 7 The components shown can be implemented using hardware, software, or a combination thereof. Electronic device 700 may be a physical device, such as a mobile phone, PC, server, robot, or industrial equipment, or a virtual device, such as a virtual machine or container. Furthermore, electronic device 700 is not limited to a single device; it can also be a combination of multiple devices or a cluster of numerous devices.

[0161] This application also provides a computer-readable storage medium storing computer program instructions. These instructions are read and executed by a processor to perform the image steganography method provided in this application. For example, the computer-readable storage medium can be implemented as follows: Figure 7 The memory 720 in the electronic device 700. For example, a computer-readable storage medium can be implemented as a standalone storage device or medium, such as a USB flash drive, portable hard drive, or optical disc.

[0162] This application also provides a computer program product, which includes computer program instructions. These computer program instructions are read and executed by a processor to perform the image steganography method provided in this application. For example, these computer program instructions can be stored in... Figure 7 The memory 320 is located inside the electronic device 300. Alternatively, these computer program instructions can be stored on a separate storage device or medium, such as a USB flash drive, external hard drive, or optical disc.

[0163] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. An image steganography method, characterized in that, include: The original text is encrypted according to the agreed method, and the first ciphertext is determined based on the encryption result. The first ciphertext is readable but not coherent text. The first ciphertext is processed to obtain the second ciphertext, and the restoration information of restoring the second ciphertext to the first ciphertext is recorded; wherein, the second ciphertext is readable but non-fluent text; Using the text-to-text capability of the first major model, a first target text is generated based on the second ciphertext. The first target text is a fluent text. Using the text-to-image capability of the second model, a first carrier image is generated based on the first target text; Using an image steganography algorithm, the restored information is written into the first carrier image to obtain the first steganographic image.

2. The image steganography method according to claim 1, characterized in that, The process of processing the first ciphertext to obtain the second ciphertext and recording the restoration information of restoring the second ciphertext to the first ciphertext includes: The first ciphertext is segmented into words, and the word order in the segmentation result is randomly shuffled to obtain the second ciphertext. Each word in the second ciphertext and its position in the first ciphertext are recorded.

3. The image steganography method according to claim 1, characterized in that, If the encryption result is unreadable text, then determining the first ciphertext based on the encryption result includes: The characters in the encryption result are mapped to readable characters to obtain the first ciphertext.

4. The image steganography method according to claim 1, characterized in that, The step of generating a first carrier image based on the first target text using the text-to-image capability of the second model includes: A first target prompt word is determined based on the first target text, and the first target prompt word requires the generation of a meaningful image based on the first target text; The first target prompt word is input into the second large model to obtain the first carrier image generated by the second large model.

5. The image steganography method according to claim 4, characterized in that, The first steganographic image is used to send to the receiver, and the first target prompt requires the generation of a meaningful image containing only regular content based on the first target text.

6. The image steganography method according to any one of claims 1-5, characterized in that, Multiple first carrier images generated in succession contain related content.

7. An image steganography method, characterized in that, include: The original text is encrypted according to the agreed method, and the first ciphertext is determined based on the encryption result. The first ciphertext is readable but not coherent text. Using the text-to-text capability of the first major model, a second target text is generated based on the first ciphertext. The second target text is fluent text. Utilizing the text-to-image generation capability of the second major model, a second carrier image is generated based on the second target text; Using an image steganography algorithm, the first ciphertext is written into the second carrier image to obtain the second steganographic image.

8. A computer program product, characterized in that, It includes computer program instructions, which, when read and executed by a processor, perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when read and executed by a processor, perform the method as described in any one of claims 1-7.

10. An electronic device, characterized in that, include: A memory and a processor, wherein the memory stores computer program instructions, which are read and executed by the processor to perform the method of any one of claims 1-7.

Images