High capacity provably secure steganographic embedding, extraction method and system based on concatenated sampling
The high-capacity proven-secure steganography method using cascaded sampling solves the problems of low embedding capacity and poor quality of generated content in existing steganography methods by utilizing recursive uniform grouping and cascaded sampling mechanisms, and achieves a high-capacity, high-quality and high-security steganography process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INSTITUTE OF INFORMATION ENGINEERING CHINESE ACADEMY OF SCIENCES
- Filing Date
- 2026-02-11
- Publication Date
- 2026-06-05
AI Technical Summary
Existing steganography methods suffer from low embedding capacity, poor quality of generated content, and low security, failing to meet the stringent security requirements of practical applications.
We employ a high-capacity, provably secure steganography method based on cascaded sampling. By recursively uniform grouping and cascaded sampling mechanisms, we optimize the embedding capability of the candidate word set, maintain the original probability distribution characteristics of the generation model, and improve the steganography capacity and the quality of the generated content.
While ensuring security and resistance to detection, it significantly improves steganography capacity and generation quality, achieving a more efficient steganography process.
Smart Images
Figure CN122160043A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of message steganography technology, and in particular to a high-capacity provably secure steganography embedding and extraction method and system based on cascaded sampling. Background Technology
[0002] With the development and application of computer network and multimedia technologies, information hiding technology has been greatly promoted. Steganography, as an important information hiding technique, utilizes the redundancy of carrier information in terms of time or space to hide secret information within the carrier information, thus obtaining a covert carrier. Commonly used covert carriers include multimedia files such as images, audio, and video. Covert carriers can be transmitted over public channels, and it can be guaranteed that during transmission, unauthorized third parties cannot perceive whether information is hidden in the covert carrier, thereby achieving the purpose of covert communication.
[0003] Traditional steganography methods often employ a distortion cost minimization framework, but this introduces detectable distributional discrepancies. With the rapid iteration of deep learning steganalyzers, traditional methods can no longer meet stringent security requirements. In recent years, with the rapid development of artificial intelligence technologies such as deep learning and large models, researchers have begun to utilize deep learning-based generative models to directly construct or generate secret-carrying files based on the secret information. This method, through precise sampling in a controlled distribution, avoids any modification to the carrier file during the steganography process, thus exhibiting strong resistance to steganalyzers.
[0004] To achieve secure steganographic communication, the design and implementation of steganographic methods must comprehensively consider factors such as embedding capacity, undetectability, and imperceptibility, ensuring that the carrier file maintains a relatively stable sensory quality, undetectability, and statistical characteristics after information embedding compared to before steganography. However, the security of existing steganographic methods still essentially relies on empirical judgment. With the continuous development of steganalysis, purely empirical security is no longer sufficient to meet higher levels of security requirements. Therefore, it is necessary to theoretically prove the security and reliability of steganographic methods.
[0005] Numerous patents exist regarding generative provably secure steganography. One such method (patent number: CN109886857A) utilizes adaptive arithmetic coding to encode the secret message according to the probabilities predicted by the model, ensuring it conforms to the original pixel distribution. Rejection sampling is then used instead of the original random sampling to embed the secret message. Another method (patent number: CN117997530A) employs a shared prefix-based encoding method to optimize the distribution order without modifying the probability distribution predicted by the generative model. This optimizes the probability interval corresponding to the selected marker at each time step by maximizing the shared prefix, thereby increasing the embedding capacity of the steganography. However, current methods all have shortcomings and limitations, such as low security, low embedding capacity, or poor quality of generated content, making them unsuitable for practical applications. Summary of the Invention
[0006] To address the problems of low embedding capacity, poor quality of generated content, and low security in existing technologies, this invention proposes a high-capacity provably secure steganography embedding and extraction method and system based on cascaded sampling. The aim is to improve steganography capacity and quality of generated content while proving steganography security, thereby enabling better application in real-world scenarios.
[0007] To achieve the above objectives, the technical solution adopted by the present invention includes the following:
[0008] A high-capacity provably secure steganography embedding method based on cascaded sampling, the method comprising: Get input content and secret messages and initialize the data set. ; Time step Input content below The input is used in the generative model for inference and prediction to obtain a set of predicted candidate words. and the set of predicted candidate words Corresponding probability set ; Combining probability sets For candidate word set Perform recursive uniform grouping and based on the secret message. Select a target group; if the probability difference between the highest-probability candidate word and the second-highest-probability candidate word in the target group is greater than a set value, then cascade sampling is performed based on the highest-probability candidate word, and the results of the cascade sampling are determined according to the secret message. Reselect target group; the secret message From secret messages Extract from the middle; Grouping the data into dense sets based on the selected target Update the data and use the updated data set. Generate input content ; make And re-execute the time step. Input content below The input is used to perform inference and prediction in the generative model until the secret message is obtained. After all data is extracted, the final data set is output.
[0009] Furthermore, the secret message From secret messages Extracted from the middle, including: secret message Convert to a binary sequence; Based on probability sets Calculate the maximum number of groupable groups based on the probability of each candidate word in the given data. ; Based on the maximum number of groups This gives the cut-off length; The secret message is obtained by extracting content from the binary sequence based on the specified truncation length. .
[0010] Furthermore, combining probability sets For candidate word set Perform recursive uniform grouping and for secret messages Select the target group, including: Based on probability sets Calculate the maximum number of groupable groups based on the probability of each candidate word in the given data. ; predict candidate word set Classified according to the probability of candidate words One group; secret message Convert the binary to decimal number to an index value using a binary-to-decimal function; Grouping is selected for this index value based on the sampling function; If a group contains only one candidate word, then that group is the target group. If the group contains multiple candidate words, then the probabilities of the candidate words in the group are renormalized, and the process based on the renormalized candidate word probabilities is re-executed. Calculate the maximum number of groupable groups based on the probability of each candidate word in the given data. This continues until the uniform grouping condition is met, which includes: the probability of any candidate word within the group is greater than 50%.
[0011] Furthermore, if the probability difference between the highest-probability candidate word and the second-highest-probability candidate word in the target group is greater than a set value, then cascaded sampling is performed based on the highest-probability candidate word, and the result of the cascaded sampling is determined according to the secret message. Reselect the target group, including: If the candidate word with the highest probability in the target group With the second highest probability candidate words If the probability difference is greater than a set value, then the candidate word with the highest probability will be selected. The candidate word is obtained by feeding it into the generative model. Predicted candidate word set and the set of predicted candidate words The corresponding probability set; Based on candidate words and the candidate word The predicted candidate word set, constructing a candidate word pair set. ; Based on candidate words The probability of candidate words and the predicted candidate word set The corresponding probability set yields the candidate word pair set. The set of probabilities; Set of candidate words The predicted candidate word set is obtained by merging it with the remaining candidate words in the target group. ; Based on the probability of target grouping combined with the candidate word pair set The probability set is used to obtain the predicted candidate word set. The set of probabilities; Based on this predicted candidate word set and the predicted candidate word set The probability set, re-execute the combination of the probability set. For candidate word set Perform recursive uniform grouping and based on the secret message. Select the target group.
[0012] A high-capacity provably secure steganography extraction method based on cascaded sampling, the method comprising: Get input content and the final set of encrypted data obtained based on any of the methods described above; At time step Next, extract one piece of classified content sequentially from the final classified content set. ; Time step Input content below The input is used in the generative model for inference and prediction to obtain a set of predicted candidate words. and the set of predicted candidate words Corresponding probability set ; Combining probability sets For candidate word set Perform recursive uniform grouping and assign confidential content. Locate the target packet to obtain the recovered message fragment. If the probability difference between the highest-probability candidate word and the second-highest-probability candidate word in the target group is greater than a set value, then cascaded sampling is performed based on the highest-probability candidate word, and the results of the cascaded sampling are determined according to the confidential content. Reposition the target group; Confidential content Append to input content In the middle, the input content is obtained. After that, And re-execute the above at time step Next, extract one piece of classified content sequentially from the final classified content set. ; Based on each time step The recovered message fragment obtained below Received secret information .
[0013] A high-capacity provably secure steganography embedding system based on cascaded sampling, the system comprising: The initialization module is used to obtain input content. and secret messages and initialize the data set. ; The message steganography module is used to: [decode] time steps Input content below The input is used in the generative model for inference and prediction to obtain a set of predicted candidate words. and the set of predicted candidate words Corresponding probability set Combined with probability sets For candidate word set Perform recursive uniform grouping and based on the secret message. Select a target group; if the probability difference between the highest-probability candidate word and the second-highest-probability candidate word in the target group is greater than a set value, then cascade sampling is performed based on the highest-probability candidate word, and the results of the cascade sampling are determined according to the secret message. Reselect target group; the secret message From secret messages Extract from the middle; group the data into dense sets according to the selected target. Update the data and use the updated data set. Generate input content ;make And re-execute the time step. Input content below The input is used to perform inference and prediction in the generative model until the secret message is obtained. After all data is extracted, the final data set is output.
[0014] A high-capacity provably secure steganalysis system based on cascaded sampling, the system comprising: The data acquisition module is used to acquire input content. And the final encrypted set obtained based on the method of any one of claims 1 to 4; The message extraction module is used to extract messages at time steps. Next, extract one piece of classified content sequentially from the final classified content set. ; take time steps Input content below The input is used in the generative model for inference and prediction to obtain a set of predicted candidate words. and the set of predicted candidate words Corresponding probability set Combined with probability sets For candidate word set Perform recursive uniform grouping, and based on the encrypted content... Locate the target packet to obtain the recovered message fragment. If the probability difference between the highest-probability candidate word and the second-highest-probability candidate word in the target group is greater than a set value, then cascaded sampling is performed based on the highest-probability candidate word, and the results of the cascaded sampling are determined according to the confidential content. Relocate the target group; redirect the classified content. Append to input content In the middle, the input content is obtained. After that, And re-execute the above at time step Next, extract one piece of classified content sequentially from the final classified content set. Based on each time step The recovered message fragment obtained below Received secret information .
[0015] A computer device, the computer device comprising: a processor and a memory storing computer program instructions; wherein the processor, when executing the computer program instructions, implements the high-capacity provably secure steganalysis embedding method based on cascaded sampling or the high-capacity provably secure steganalysis extraction method based on cascaded sampling described above.
[0016] A computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the high-capacity provably secure steganalysis embedding method based on cascaded sampling or the high-capacity provably secure steganalysis extraction method based on cascaded sampling described above.
[0017] A computer program product, characterized in that, when the computer program product is run on a computer device, the computer device executes the high-capacity provably secure steganography embedding method based on cascaded sampling or the high-capacity provably secure steganography extraction method based on cascaded sampling described above.
[0018] Compared with the prior art, the present invention has at least the following beneficial effects.
[0019] 1. This invention proposes a high-capacity provably secure steganography method based on cascaded sampling. This method achieves steganography while maintaining the original probability distribution characteristics of the generative model, thereby ensuring the security and resistance to detection of the steganography process.
[0020] 2. This invention utilizes the concept of cascaded sampling, and improves the embedding capability of candidate word sets by introducing multi-level deep sampling and joint sampling mechanisms on the basis of group sampling, thereby effectively increasing the capacity of steganography.
[0021] 3. This invention preserves the generative capabilities and naturalness of a large language model by precisely sampling the candidate word set. This method can improve the generation quality and speed of steganography while maintaining steganography capacity. Attached Figure Description
[0022] Figure 1 This is a schematic diagram of the steganography embedding process proposed in this invention.
[0023] Figure 2 This is a schematic diagram of the steganography extraction process proposed in this invention. Detailed Implementation
[0024] The present invention will now be described in further detail with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.
[0025] In this embodiment, a high-capacity, provably secure steganography embedding and extraction method based on cascaded sampling achieves efficient steganography by utilizing recursive uniform grouping and cascaded sampling mechanisms without compromising the predictive probability distribution characteristics of the generative model. This method first generates a uniform candidate set at each time step through group sampling and selects the target group based on the secret message. Then, a cascaded sampling mechanism is introduced. When the groups are unbalanced, the sampling depth and candidate pair construction are expanded, allowing the steganography process to utilize model entropy more finely, thereby significantly improving the steganography capacity. The cascaded sampling and recursive grouping algorithm proposed in this embodiment ensures the uniformity and controllability of the probability intervals corresponding to the selected steganographic candidate words, improving the capacity and security of the embedded information. Simultaneously, while maintaining the naturalness and fluency of the generative model, this method can stably generate high-quality steganographic text and optimizes sampling efficiency through joint sampling, thereby achieving a comprehensive improvement in capacity, security, and speed.
[0026] Message embedding phase.
[0027] In this embodiment of the invention, a generative model is used to embed secret messages within the generated content based on the ideas of recursive grouping and cascaded sampling. To embed as many messages as possible at each time step, this invention proposes, based on the probability distribution of candidate words, first performing recursive uniform grouping and selecting target groups according to message-driven principles; subsequently, in cases of imbalanced grouping, a cascaded sampling mechanism is introduced to expand the candidate set and refine the grouping by increasing the sampling depth, making the probabilities of each group more balanced. Thus, without changing the original probability distribution characteristics of the generative model, the steganography capacity and entropy utilization are effectively improved. Figure 1 As shown, the message embedding process includes the following steps.
[0028] 1) Define the current time step as and initialize Get messages Convert the secret message into a binary sequence; define and initialize the set of secret content. It is an empty set; 2) Set the time step Input content below Inputting the data into the generative model for inference and prediction, we obtain the time steps. The following is the predicted candidate word set and its corresponding probability set .
[0029] 3) For the candidate word set Perform recursive uniform grouping and, based on the secret message to be embedded... Select the target group. If the selected group still contains multiple candidate words, renormalize their probabilities and continue uniform grouping until uniform division is no longer possible. Specifically, follow these steps: Step 3.1: Based on candidate word probability The maximum number of groups can be calculated using the following formula. : .
[0030] Step 3.2: Set the candidate words Classified according to probability magnitude Grouping, resulting in a group set. Each group The cumulative probability satisfies .
[0031] Step 3.3: Embed the secret message Convert to a binary sequence and take the length of . The fragment is embedded as the content at the current time step, denoted as .
[0032] Step 3.4: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require Using binary to decimal conversion function Convert to index value, based on sampling function Select group: .
[0033] Step 3.5: If grouped If the group contains multiple candidate words, the probabilities within each group are renormalized, and sub-steps 3.1–3.4 are repeated until the uniform grouping condition is met and further processing is not possible (i.e., ...). Finally, a candidate word is randomly sampled from this group as the time step. The candidate words for the encrypted data are added to the set. .
[0034] 4) If, after recursive uniform grouping, the candidate word with the highest probability is found in the final selected target group. With the second highest probability candidate words The probability difference satisfies If so, the group is determined to be unbalanced. At this point, The data is then fed back into the generative model as input to obtain the prediction set. Next, construct a set of candidate word pairs. .
[0035] 5) For the candidate word pair set The probabilities are normalized so that the probability of each candidate word pair is . .Will With the current group The remaining candidate words are merged to obtain a new set. .
[0036] For this set Repeat step 3) to achieve fine-grained uniform grouping.
[0037] 6) Use the final selected candidate words as time steps. Confidential content Add to collection and add it to the input sequence. , obtain the input for the next time step If message If not fully embedded, then let Return to step 2 to continue execution; if message embedding is complete, output the final set of encrypted content. As an embedded result.
[0038] 7) Experiments were conducted on the IMDB text dataset, using 1000 sample texts as the carrier. Results show that the average embedding capacity of this invention is 2.7 bits / token, and the generation speed is approximately 42 tokens / s. In security tests, the detection accuracy of CNN and Transformer steganalyzers were 50.8% and 49.5%, respectively. These experimental results fully demonstrate that this invention has high capacity and strong security.
[0039] II. Message Extraction Stage.
[0040] The sender and receiver share the generative model, initial input content, encrypted content, and threshold. The receiver can completely recreate the process by which the sender generates the encrypted content, thereby extracting the hidden secret message. The extraction process at each time step is as follows: Figure 2 As shown, the main process is as follows: 1) Define the current time step as and initialize The recipient obtains the final collection of encrypted content. and initial input content ; at each time step Below, from the set Extract a confidential content in sequence .
[0041] 2) Set the time step Input content below Inputting the data into the generative model for inference and prediction, we obtain the time steps. candidate word set and the corresponding probability set .
[0042] 3) For the candidate set Perform the same recursive uniform grouping as in the embedding phase to obtain the group set. And maintain the probability of each group. .
[0043] 4) Based on the confidential content obtained by the recipient Locate the group containing the candidate word. And use the group index value as the time step The following message fragment was restored. .
[0044] 5) If the group being located Internal imbalance, i.e., satisfying the highest probability candidate word and the second highest candidate words probability difference Then, cascade sampling consistent with the embedding phase is performed: The data is fed into the generative model as input to obtain an expanded set. Then, construct words for the set Set of word pairs Merge with the current group to form a new set and to Regroup the materials evenly. (Based on the classified content) Extract the corresponding message fragment from the location in this set. .
[0045] 6) Restore the message fragment Add to message sequence At the same time, Append to input content In the middle, the input for the next time step is obtained. .
[0046] 7) Judgment Is this the last encrypted content in set S? If yes, it indicates that message extraction is complete, and the message fragments recovered from all time steps are retrieved. By piecing together the information, we can obtain the final complete message. Otherwise, Return to step 2.2 and continue execution.
[0047] 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 it. Although the present invention has been described in detail using examples, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A high-capacity provably secure steganography embedding method based on cascaded sampling, characterized in that, The method includes: Get input content and secret messages and initialize the data set. ; Time step Input content below The input is used in the generative model for inference and prediction to obtain a set of predicted candidate words. and the set of predicted candidate words The corresponding probability set ; Combining probability sets For candidate word set Perform recursive uniform grouping and based on the secret message. Select a target group; if the probability difference between the candidate word with the highest probability and the candidate word with the second highest probability in the target group is greater than a set value, then cascade sampling is performed based on the candidate word with the highest probability, and the secret message is determined according to the cascade sampling result. Reselect target group; the secret message From secret messages Extract from the middle; Grouping the data into dense sets based on the selected target Update the data and use the updated data set. Generate input content ; make And re-execute the time step. Input content below The input is used to perform inference and prediction in the generative model until the secret message is obtained. After all data is extracted, the final data set is output.
2. The method according to claim 1, characterized in that, The secret message From secret messages Extracted from the middle, including: secret message Convert to a binary sequence; Based on probability sets Calculate the maximum number of groupable groups based on the probability of each candidate word in the given data. ; Based on the maximum number of groups This gives the cut-off length; The secret message is obtained by extracting content from the binary sequence based on the specified truncation length. .
3. The method according to claim 1, characterized in that, Combining probability sets For candidate word set Perform recursive uniform grouping and based on the secret message. Select the target group, including: Based on probability sets Calculate the maximum number of groupable groups based on the probability of each candidate word in the given data. ; predict candidate word set Classified according to the probability of candidate words One group; secret message Convert the binary to decimal number to an index value using a binary-to-decimal function; The target group is selected based on the index value according to the sampling function; If a group contains only one candidate word, then that group is the target group. If the group contains multiple candidate words, then the probabilities of the candidate words in the group are renormalized, and the process based on the renormalized candidate word probabilities is re-executed. Calculate the maximum number of groupable groups based on the probability of each candidate word in the given data. This continues until the uniform grouping condition is met, which includes: the probability of any candidate word within the group is greater than 50%.
4. The method according to claim 1, characterized in that, If the probability difference between the highest-probability candidate word and the second-highest-probability candidate word in the target group is greater than a set value, then cascaded sampling is performed based on the highest-probability candidate word, and the results of the cascaded sampling are determined according to the secret message. Reselect the target group, including: If the candidate word with the highest probability in the target group With the second highest probability candidate words If the probability difference is greater than a set value, then the candidate word with the highest probability will be selected. The candidate word is obtained by feeding it into the generative model. Predicted candidate word set and the set of predicted candidate words The corresponding probability set; Based on candidate words And the candidate word The predicted candidate word set is used to construct a candidate word pair set. ; Based on candidate words The probability of candidate words and the predicted candidate word set The corresponding probability set yields the candidate word pair set. The set of probabilities; Set of candidate words The predicted candidate word set is obtained by merging it with the remaining candidate words in the target group. ; Based on the probability of target grouping combined with the candidate word pair set The probability set is used to obtain the predicted candidate word set. The set of probabilities; Based on this predicted candidate word set and the predicted candidate word set The probability set, re-execute the combination of the probability set. For candidate word set Perform recursive uniform grouping and based on the secret message. Select the target group.
5. A high-capacity provably secure steganalysis extraction method based on cascaded sampling, characterized in that, The method includes: Get input content And the final encrypted set obtained based on the method of any one of claims 1 to 4; At time step Next, extract one piece of classified content sequentially from the final classified content set. ; Time step Input content below The input is used in the generative model for inference and prediction to obtain a set of predicted candidate words. and the set of predicted candidate words Corresponding probability set ; Combining probability sets For candidate word set Perform recursive uniform grouping, and based on the encrypted content... Locate the target packet to obtain the recovered message fragment. If the probability difference between the highest-probability candidate word and the second-highest-probability candidate word in the target group is greater than a set value, then cascaded sampling is performed based on the highest-probability candidate word, and the results of the cascaded sampling are determined according to the confidential content. Reposition the target group; Confidential content Append to input content In the middle, the input content is obtained. After that, And re-execute the above at time step Next, extract one piece of classified content sequentially from the final classified content set. ; Based on each time step The recovered message fragment obtained below Received secret information .
6. A high-capacity provably secure steganography embedding system based on cascaded sampling, characterized in that, The system includes: The initialization module is used to obtain input content. and secret messages and initialize the data set. ; The message steganography module is used to: [decode] time steps Input content below The input is used in the generative model for inference and prediction to obtain a set of predicted candidate words. and the set of predicted candidate words Corresponding probability set Combined with probability sets For candidate word set Perform recursive uniform grouping and based on the secret message. Select a target group; if the probability difference between the highest-probability candidate word and the second-highest-probability candidate word in the target group is greater than a set value, then cascade sampling is performed based on the highest-probability candidate word, and the results of the cascade sampling are determined according to the secret message. Reselect target group; the secret message From secret messages Extract from the middle; group the data into dense sets according to the selected target. Update the data and use the updated data set. Generate input content ;make And re-execute the time step. Input content below The input is used to perform inference and prediction in the generative model until the secret message is obtained. After all data is extracted, the final data set is output.
7. A high-capacity provably secure steganalysis system based on cascaded sampling, characterized in that, The system includes: The data acquisition module is used to acquire input content. And the final encrypted set obtained based on the method of any one of claims 1 to 4; The message extraction module is used to extract messages at time steps. Next, extract one piece of classified content sequentially from the final classified content set. ; take time steps Input content below The input is used in the generative model for inference and prediction to obtain a set of predicted candidate words. and the set of predicted candidate words Corresponding probability set Combined with probability sets For candidate word set Perform recursive uniform grouping, and based on the encrypted content... Locate the target packet to obtain the recovered message fragment. If the probability difference between the highest-probability candidate word and the second-highest-probability candidate word in the target group is greater than a set value, then cascade sampling is performed based on the highest-probability candidate word, and the cascade sampling is based on the confidential content. Relocate the target group; redirect the classified content. Append to input content In the middle, the input content is obtained. After that, And re-execute the above at time step Next, extract one piece of classified content sequentially from the final classified content set. Based on each time step The recovered message fragment obtained below Received secret information .
8. A computer device, characterized in that, The computer device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the high-capacity provably secure steganography embedding method based on cascaded sampling as described in any one of claims 1-4 or the high-capacity provably secure steganography extraction method based on cascaded sampling as described in claim 5.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the high-capacity provably secure steganalysis embedding method based on cascaded sampling as described in any one of claims 1-4 or the high-capacity provably secure steganalysis extraction method based on cascaded sampling as described in claim 5.
10. A computer program product, characterized in that, When the computer program product is run on a computer device, the computer device performs the high-capacity provably secure steganography embedding method based on cascaded sampling as described in any one of claims 1-4 or the high-capacity provably secure steganography extraction method based on cascaded sampling as described in claim 5.