Information Steganography Embedding and Extraction Methods and Systems Based on Semi-Decoding of AAC Bitstreams
By performing semi-decoding on the AAC bitstream, filtering QMDCT coefficients and Huffman information, and combining STC encoding and scrambling algorithms, the problems of low capacity, slow speed, and insufficient anti-analysis capability in AAC audio format steganography methods are solved, achieving higher steganography capacity, speed, and security.
Patent Information
- Application Number
- CN202311385300.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2022-12-26
- Filing Date
- 2023-10-24
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2043-10-24
AI Technical Summary
Existing AAC audio format steganography methods have low capacity, concentrated embedding locations, slow embedding speed, and insufficient resistance to steganalysis. Existing compressed domain audio steganography algorithms increase detection risks and time costs.
A semi-decoding-based AAC code stream information steganography method is adopted. By semi-decoding the AAC code stream file, QMDCT coefficients and Huffman information are extracted, and coefficients that meet the conditions are selected for embedding. Combined with STC encoding and scrambling algorithm, the best embedding path is selected to minimize modification distortion, thereby realizing the embedding and extraction of secret information.
It significantly improves steganography capacity and embedding speed, enhances resistance to steganalysis, achieves higher security and less distortion during modification, and meets the needs of practical applications.
Smart Images

Figure CN117496986B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information technology, and in particular to a method and system for embedding and extracting information from AAC code streams based on semi-decoding. 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] Audio, as one of the most common multimedia file contents on the internet today, is widely used. Examples include content sharing platforms like QQ Music and NetEase Cloud Music, FM radio stations, instant messaging apps like WeChat, and video audio streams. Therefore, the abundance of online audio carriers provides a favorable environment and conditions for information hiding. Among them, Advanced Audio Coding (AAC) compressed audio format, as the most popular audio coding format, has characteristics such as large steganography capacity and complex coding structure, making it a relatively ideal steganography embedding carrier. Therefore, steganography methods for AAC compressed audio have become a research hotspot for scholars both domestically and internationally in recent years.
[0004] To achieve secure steganographic communication, the design and implementation of steganography methods need to 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. Currently, there are few publicly available results on steganography algorithms for AAC audio streams, but many steganography algorithms for other compressed audio formats such as MP3 exist, mainly divided into steganography methods based on encoding parameters and steganography methods based on encoded data. Among them, MP3 steganography methods based on encoding parameters generally have low steganography capacity and make significant modifications to the audio content, making them easily detectable. Compared with steganography methods based on encoding parameters, MP3 steganography methods based on encoded data have a significantly higher steganography capacity, but most algorithms still do not consider the impact of embedding operations on the sensory and statistical characteristics of the carrier during the steganography process, nor do they employ reasonable methods to avoid steganography in locations that have a significant impact on the carrier. Therefore, they are still insufficient in terms of anti-steganography analysis.
[0005] To improve the security load of steganography methods, modern steganography, represented by adaptive steganography, has been proposed in recent years. Adaptive steganography mainly studies how to select the embedding position based on the audio content while maintaining a certain security load. It employs STC (Syndrome-Trellis Codes) encoding combined with a distortion cost function to adaptively embed the data onto the carrier. This method can select the optimal embedding position and path in the carrier based on the modification cost, reducing the number of modifications to the carrier, lowering distortion perturbations, and minimizing carrier distortion caused by the embedding operation. Therefore, adaptive steganography algorithms generally have a higher security load. Currently, there are few achievements in adaptive steganography methods based on the AAC audio format. Furthermore, from an algorithm implementation perspective, existing compressed domain audio steganography algorithms are all based on recoding techniques, which significantly increases the risk of steganographic analysis and detection, as well as the time cost of steganographic embedding. Therefore, designing and implementing a steganography algorithm based on AAC semi-decoding technology is a critical problem that urgently needs to be solved. Summary of the Invention
[0006] To address the problems of low capacity, concentrated embedding locations, and slow embedding speed of existing compressed audio steganography methods such as AAC, this invention proposes a method and system for embedding and extracting information from AAC bitstreams based on semi-decoding.
[0007] The "semi-decoding" described in this invention refers to a method that decodes only the AAC bitstream file to QMDCT (Quantized Modified Discrete Cosine Transform) coefficients, obtains the QMDCT coefficient values and some additional Huffman information, and then embeds secret information into the AAC bitstream by directly modifying some bit values. Compared to existing methods that decode compressed audio into WAV files before steganography embedding, the semi-decoding steganography method can significantly improve the speed of secret information embedding and extraction.
[0008] The technical solution adopted in this invention is as follows:
[0009] A method for embedding information steganography in AAC bitstreams based on semi-decoding includes the following steps:
[0010] S1: Perform semi-decoding on the AAC bitstream file to extract the QMDCT coefficients Q and Huffman information;
[0011] S2: Traverse the QMDCT coefficients Q of each audio frame, extract coefficients q that meet the filtering conditions from Q. The filtering conditions include that the Huffman codeword corresponding to q contains an overflow word, the magnitude of q is not greater than the filtering threshold T, and q does not change the length of the corresponding Huffman codeword after sign inversion; map and concatenate the overflow word LSB and sign bit in coefficient q to obtain the carrier bit string C, and record the source frame number A and QMDCT coefficient index number K of each bit;
[0012] S3: For each frame's carrier bit string C, calculate the modification distortion cost ρ corresponding to the carrier bit string C based on the QMDCT coefficients Q and their index numbers K.
[0013] S4: Repeat steps S2 and S3 to perform frame-by-frame calculations until all audio frames have been traversed, obtain the carrier bit string C of all frames, the modification distortion cost ρ of all bits, and record the source frame number A and QMDCT coefficient index number K corresponding to all bits in the carrier bit string C.
[0014] S5: Scramble the carrier bit string C and the distortion cost ρ using the same scrambling algorithm and parameter configuration to obtain the carrier C′ and ρ′;
[0015] S6: Based on the modified distortion cost ρ and the secret information M, the secret information is embedded using the check trellis code STC embedding carrier C′ to obtain the out-of-order secret bit string S′.
[0016] S7: Use the reverse scrambling algorithm and parameter configuration corresponding to step S5 to reverse scramble the bit string S′ to obtain the ordered dense bit string S;
[0017] S8: Based on the ordered encrypted bit string S obtained in step S7 and the source frame number A and QMDCT coefficient index number K obtained in step S4, and based on the QMDCT coefficient Q and Huffman information recorded in step S1, determine the codeword bit modification scheme, which consists of two parts: bit reversal and bit substitution.
[0018] S9: Modify some bits in the original AAC bitstream file according to the determined codeword bit modification scheme, and obtain the encrypted AAC audio after modification.
[0019] Furthermore, the Huffman information in step S1 includes the starting bit storage position P of the Huffman codeword W corresponding to Q in the AAC code stream file, the codebook number B of the Huffman codeword W, the index I of the codeword in the codebook, and the Huffman codeword parsing sequence number Z corresponding to the QMDCT coefficient Q.
[0020] Further, in step S2, the condition for the coefficient q to contain the overflow word is: the absolute value of q is greater than or equal to 16, and all the sign bits of the coefficient containing the overflow word are used as embeddable messages; among them, the selection of the coefficient sign bits to embed the message needs to meet the following two conditions: (1) the absolute value of its coefficient q must be greater than 0 and less than or equal to the filtering threshold T, (2) the Huffman code character corresponding to the coefficient q can be reversed.
[0021] Furthermore, the condition for passing the sign reversal feasibility test is: after reversing the sign bit of the QMDCT coefficient q, the length of its corresponding Huffman codeword is not changed.
[0022] The methods for determining the code words in different codebooks include:
[0023] 1) For the QMDCT coefficient q of the corresponding codeword in number books 1, 2, 5, and 6, the corresponding Huffman codeword substitution should satisfy the following conditions:
[0024]
[0025] Where q = (q1, q2, ... q n Let S be a vector consisting of the n coefficients of the parsed source codeword w to which coefficient q belongs, and S = {-1, +1}. n Q is the coefficient matrix, W is the codeword matrix, Ψ represents the mapping from QMDCT coefficients q to codewords based on the G codebook, and |-| represents the bit length of the Huffman codeword; w t It is a row vector in W, representing a Huffman codeword;
[0026] 2) The QMDCT coefficient q corresponding to the remaining codebook has its sign bit represented as a single independent bit in the Huffman codeword.
[0027] Furthermore, the mapping of the overflow word LSB of the QMDCT coefficient q in step S2 is as follows:
[0028] Where f1 represents the mapping relationship, and c is the binary bit mapped to the overflow word LSB.
[0029] Furthermore, the mapping of the sign bit of the QMDCT coefficient q in step S2 is as follows:
[0030]
[0031] Where f2 represents the mapping relationship, and c is the binary bit mapped to the sign bit.
[0032] Furthermore, the modification of the distortion cost ρ in step S3 includes statistical distortion cost and perceptual distortion cost, wherein,
[0033] The cost of statistical distortion is calculated using the following statistical distortion function:
[0034]
[0035] Among them, Q i,j Q i,j+k Let K be the j-th and (j+k)-th adjacent coefficients in the i-th frame, K = {-1, 1}, and α be the adjustment parameter;
[0036] The cost of perceptual distortion is based on the human hearing threshold T in the psychoacoustic model. f The calculation yields qsthr(Q) i,j );
[0037] The distortion cost ρ is then calculated based on the following comprehensive distortion function:
[0038] D(Q i,j )=log2||ρ(Q i,j )||-log2||qsthr(Q i,j )||+β
[0039] Where β is the adjustment parameter.
[0040] Furthermore, the scrambling algorithm in step S5 is the Logistic chaotic mapping algorithm.
[0041] A steganography embedding system for AAC bitstreams based on semi-decoding, comprising:
[0042] The audio parsing module is used to perform partial decoding on AAC bitstream files, extracting QMDCT coefficients Q and Huffman information;
[0043] The embedding domain construction module is used to traverse the QMDCT coefficients Q of each audio frame, extract coefficients q that meet the filtering conditions from Q, map and concatenate the overflow word LSB and sign bit in coefficient q to obtain the carrier bit string C, and record the source frame number A and QMDCT coefficient index number K of each bit.
[0044] The joint distortion function calculation module is used to calculate the modification distortion cost ρ corresponding to the carrier bit string C for each frame, based on the QMDCT coefficients Q and their index numbers K.
[0045] The scrambling module is used to scramble the carrier bit string C and the distortion cost ρ using the same scrambling algorithm and parameter configuration to obtain the carrier C′ and ρ′.
[0046] The STC encoding module is used to embed secret information based on the modified distortion cost ρ and secret information M, using the check trellis encoding STC embedding carrier C′ to obtain the out-of-order secret bit string S′.
[0047] The reverse scrambling module is used to reverse scramble the bit string S′ using the reverse scrambling algorithm and parameter configuration to obtain an ordered dense bit string S;
[0048] The bit modification scheme formulation module is used to determine the codeword bit modification scheme based on the ordered encrypted bit string S, the source frame number A, and the QMDCT coefficient index number K, and based on the recorded QMDCT coefficients Q and Huffman information. The bit modification scheme consists of two parts: bit reversal and bit substitution.
[0049] The AAC bitstream modification module is used to modify some bits in the original AAC bitstream file according to a determined codeword bit modification scheme, and after modification, a encrypted AAC audio is obtained.
[0050] A method for extracting secret information from AAC bitstreams based on semi-decoding is provided, for extracting secret information from the secret-carrying AAC audio obtained by the steganography embedding method. The method includes the following steps:
[0051] P1: Use a decoder to decode the AAC audio file, extract the QMDCT coefficients Q, and record the codebook number B corresponding to the parsed codeword and the index I of the codeword in the codebook;
[0052] P2: For the QMDCT coefficients of a single frame, sequentially traverse each QMDCT coefficient q; for a single QMDCT coefficient q, determine whether the overflow word LSB embeds secret information based on the amplitude; if the overflow word LSB is invalid, determine whether the sign bit embeds secret information based on the amplitude, the filtering threshold T, the codebook number b of the corresponding codeword, and the codeword index i; perform bit mapping on the filtered coefficients q, and sequentially concatenate the sign bit of the embedded information and the overflow word LSB to obtain the binary secret bit string S of the single frame. i ;
[0053] P3: Concatenate the data carrier bit strings of all frames sequentially to obtain the final data carrier bit string S;
[0054] P4: Use the scrambling algorithm and parameter configuration to scramble the dense bit string S to obtain the extractable dense bit string S′;
[0055] P5: Use STC decoding to extract the secret information M from S′;
[0056] The filtering threshold T and bit mapping method in step P2 are the same as those in step S2 of the steganography embedding method; the scrambling algorithm and parameter configuration in step P4 are the same as those in step S4 of the steganography embedding method.
[0057] Furthermore, in step P2, for a single QMDCT coefficient q, the judgment function Φ(·) for determining whether secret information is embedded is:
[0058]
[0059] Among them, U b,i =1 indicates that the i-th code character in the b-th codebook can be reversed, and Φ(q) =1 indicates that the coefficient is embedded with secret information.
[0060] A system for extracting steganographic information from AAC bitstreams based on semi-decoding, comprising:
[0061] The audio parsing module is used to decode AAC audio files using a decoder, extract QMDCT coefficients Q, and record the codebook number B corresponding to the parsed codewords and the index I of the codewords in the codebook.
[0062] The embedding domain construction module is used to sequentially traverse each QMDCT coefficient q for a single frame of QMDCT coefficients. For a single QMDCT coefficient q, the magnitude of the overflow word LSB determines whether secret information is embedded. If the overflow word LSB is invalid, the magnitude of the overflow word LSB, the selection threshold T, the codebook number b of the corresponding codeword, and the codeword index i determine whether the sign bit is embedded with secret information. For the selected coefficients q, bit mapping is performed, and the sign bit of the embedded information and the overflow word LSB are sequentially concatenated to obtain the binary secret bit string S of the single frame. i ; Concatenate the data carrier bit strings of all frames sequentially to obtain the final data carrier bit string S;
[0063] The scrambling module is used to scramble the dense bit string S using a scrambling algorithm and parameter configuration to obtain an extractable dense bit string S′.
[0064] The STC decoding module is used to extract secret information M from S′ using STC decoding.
[0065] The advantages of this invention compared to the prior art are as follows:
[0066] (1) In this invention, the steganography algorithm has higher security. Compared with the traditional non-adaptive compressed audio steganography algorithm, this invention calculates the distortion cost and statistical distortion cost of the modified content at each embeddable point, and combines STC encoding to achieve adaptive steganography. Each embedding selects the best embedding path to minimize the total distortion of the modification, thereby improving the steganography method's resistance to steganography analysis.
[0067] (2) Compared with existing compressed audio steganography algorithms, this invention has a larger steganography capacity. Most compressed audio steganography methods have relatively small steganography capacities, which are difficult to meet the needs of practical applications. The method in this invention selects QMDCT coefficients with the sign bit or overflow word LSB meeting the requirements as the joint embedding field. The maximum steganography capacity can reach 15kbps for audio with a bitrate of 320kbps, and the maximum steganography capacity can also reach 7kbps for a bitrate of 128kbps.
[0068] (3) Compared with existing compressed audio steganography algorithms, the present invention has a higher embedding and extraction rate. Existing compressed audio steganography algorithms need to decode the compressed audio to the time domain and perform steganography embedding during the compression of the time domain signal; the steganography method in the present invention does not need to fully decode the AAC compressed audio, but adopts a semi-decoding method, only needs to decode the AAC audio to QMDCT coefficients, obtain and process the necessary information, and then directly embed it by modifying the AAC bitstream file. Attached Figure Description
[0069] Figure 1 This is a flowchart illustrating the implementation of the information steganography embedding method proposed in this invention.
[0070] Figure 2 This is a schematic diagram of the secret information steganography embedding process in an embodiment of the present invention;
[0071] Figure 3 This is the intention of constructing the joint embedded domain in this invention;
[0072] Figure 4 This is a block diagram of the information steganography embedding system proposed in this invention;
[0073] Figure 5 This is a schematic diagram of the secret information steganography extraction process in an embodiment of the present invention;
[0074] Figure 6 This is a block diagram of the information steganography extraction system proposed in this invention. Detailed Implementation
[0075] To make the various technical features, advantages, or effects of the present invention more apparent and understandable, a detailed description is provided below in conjunction with the accompanying drawings.
[0076] This invention proposes a method for embedding information steganography in AAC bitstreams based on semi-decoding. Figure 1 The flowchart below illustrates the implementation of this method, which includes the following specific steps:
[0077] S1: Use a decoder to partially decode the AAC bitstream file, extract the QMDCT coefficients Q, and record the Huffman information. The Huffman information includes the starting bit storage position P of the Huffman codeword W corresponding to Q in the AAC bitstream file, the codebook number B of the Huffman codeword W, and its index I in the codebook. Since a single Huffman codeword will decode into multiple QMDCT coefficients, it is also necessary to record the parsing sequence number Z of the Huffman codeword W corresponding to the QMDCT coefficient Q.
[0078] S2: QMDCT coefficients for the i-th audio frame i traverse Q in sequence i The process involves selecting QMDCT coefficients q that meet the selection criteria as embedding regions. For a single QMDCT coefficient q, the selection criteria are either that the Huffman codeword corresponding to q contains an overflow word, or that the magnitude of q meets the requirements of the set selection threshold T and that q does not affect the length of the corresponding Huffman codeword after sign inversion. Since the selection threshold T is set relatively small, a single coefficient will not simultaneously meet both of these conditions. When the selected embedding region contains a QMDCT coefficient with an overflow word, the bit mapping method for this coefficient is based on the least significant bit (LSB) of the overflow word. The mapping method for QMDCT coefficients selected by the threshold T is based on the sign of the coefficient value. After selection and mapping, the sign bit or overflow word LSB can be sequentially concatenated to obtain the binary carrier bit string C of a single frame. i Simultaneously record C i The corresponding source frame number A and QMDCT coefficient index number K for each bit i .
[0079] S3: Cost calculation will be performed frame by frame. For the sign bit, the modification method is to invert the sign of the QMDCT coefficient, and the modification magnitude is twice the QMDCT coefficient magnitude. The overflow word LSB is the inversion of its own bits, and its modification magnitude is always 1. For the carrier bit string C of a single frame... i Combined with QMDCT coefficient index number K i and QMDCT coefficient Q i Information, drawing on the concept of Uniform Embedding Distortion (UED), can be used to calculate C. i Corresponding statistical distortion cost Based on the human hearing threshold T in psychoacoustic models f C can be calculated i Corresponding perceptual distortion cost The Cost of Comprehensive Statistical Distortion and Get C i The corresponding distortion cost ρi .
[0080] S4: Repeat steps S2 and S3 until all audio frames have been traversed. Concatenate the binary carrier bit strings of all frames sequentially to obtain the binary carrier bit string C. Similarly, based on the correspondence, the modification cost ρ of all bits in C can be obtained. Record the source frame number A and QMDCT coefficient index number K corresponding to all bits in the binary carrier bit string C.
[0081] S5: Scramble the binary carrier bit string C and the distortion cost ρ using the same scrambling algorithm and parameter configuration to obtain carriers C′ and ρ′.
[0082] S6: Based on the modified distortion cost ρ and the secret information M, the secret information is embedded using the Syndrome trellis codes (STC) embedding carrier C′ to obtain the out-of-order secret binary bit string S′.
[0083] S7: Use the inverse scrambling algorithm and parameter configuration corresponding to step S5 to inverse scramble S′ to obtain an ordered binary bit string S.
[0084] S8: After the secret information STC embedding process is completed, based on the ordered secret binary bit string S obtained in step S7 and the source frame number A and QMDCT coefficient index number K obtained in step S4, and based on the QMDCT coefficient Q and Huffman information recorded in step S1, the codeword bit modification scheme is determined.
[0085] S9: Modify some bits in the original AAC bitstream file according to the bit modification scheme. After all bits are modified, the encrypted AAC audio is obtained.
[0086] The present invention provides an example as follows, which utilizes the information steganography embedding method proposed in this invention to achieve steganography of secret information. Figure 2 This is a schematic diagram of the secret information steganography process in this embodiment, showing the changes in the data stream during the information embedding process. After the embedding operation is completed, the final secret-carrying bit stream differs from the original bit stream by a difference of bits H. i+3 and H j The situation has changed. The specific implementation process of information steganography embedding is as follows:
[0087] S1: The AAC audio stream is partially decoded using a decoder, frame by frame. Decoding ends when the QMDCT coefficients of each frame are extracted. During decoding, Huffman codewords are parsed from the AAC stream as consecutive bits. At this point, the starting bit storage position P of all Huffman codewords in the AAC stream file can be recorded. Subsequently, based on the subband of the codeword and its codebook number B, the index I of the codeword in the codebook can be obtained by querying the codebook. Finally, based on the codebook parsing rules, each codeword will be parsed into 2 or 4 QMDCT coefficients. Since the QMDCT coefficients and codewords are not in a one-to-one correspondence, it is necessary to record which parsed value each QMDCT coefficient is from the source codeword, and denote the parsing sequence number as Z.
[0088] S2: Since not all QMDCT coefficients can be used for message embedding, they need to be filtered before using the overflow word LSB or sign bit for message embedding.
[0089] Overflow words can only occur when the QMDCT coefficients are large. For a single frame, the QMDCT coefficients Q... i The QMDCT coefficients of the overflow word exist in its Huffman codeword. It is expressed as follows:
[0090]
[0091] That is, only QMDCT coefficients with an absolute value greater than or equal to 16 will have an overflow bit.
[0092] For the overflow word, the QMDCT coefficient q overflow word is mapped to binary bits c. The mapping process f1 can be expressed as:
[0093]
[0094] Furthermore, since only the codewords in book number 11 have overflow words, and the overflow word part is a naive binary code, the overflow word LSB is equal to c. When modifying, you can simply reverse the overflow word LSB of the AAC code stream file.
[0095] Regarding the sign bit of the coefficients, since reversing the sign will result in a doubling of the coefficient amplitude, a threshold T needs to be set to filter QMDCT coefficients to avoid excessive steganographic perturbation. For a single frame of QMDCT coefficients Q... i QMDCT coefficients after threshold T filtering The expression is as follows:
[0096]
[0097] Unlike overflow words (LSBs) which directly correspond to individual bits in the codestream, reversing the sign of a single QMDCT coefficient does not necessarily reverse a specific bit in the codestream; it depends on the Huffman codeword from which the coefficient is resolved. Number books 3, 4, 7, 8, 9, and 10 contain bits that directly represent the sign bit, allowing for direct reversal of the corresponding sign bit to achieve the QMDCT coefficient sign reversal. However, for the remaining number books 1, 2, 5, and 6, Huffman codeword replacement is required to achieve the QMDCT coefficient sign reversal.
[0098] Regarding the Huffman codeword replacement strategy, if the replacement causes a change in the Huffman codeword length, it will disrupt the entire codestream structure. Therefore, QMDCT coefficient sign inversion cannot cause a change in the Huffman codeword length, and sign feasibility needs to be determined. For the QDMCT coefficient q of its source codeword w in codebook G, under the condition of meeting the threshold screening, it should also satisfy the following condition:
[0099]
[0100] Where q = (q1, q2, ... q n Let S be a vector consisting of the n coefficients of the parsed source codeword w to which coefficient q belongs, and S = {-1, +1}. n , Let Ψ be the coefficient matrix, W be the codeword matrix, Ψ represent the mapping from QMDCT coefficients q to codewords based on the G codebook, |·| represent the bit length of the Huffman codeword, and w t A row vector in W can represent a Huffman codeword.
[0101] Based on the above rules, all codewords in numbers 1, 2, 5, and 6 can be traversed in advance to determine whether each codeword meets the sign reversal feasibility criteria, thus forming a sign reversal feasibility table U. b,i =1 indicates that modifying the i-th code character of the b-th number book is feasible, U b,i =0 indicates that the codeword cannot undergo sign inversion of the QMDCT coefficients. The codewords in number books 3, 4, 7, 8, 9, and 10 are all set to 1 in the sign inversion feasibility table U.
[0102] Based on the sign reversal feasibility table U, in the... After performing symbol feasibility screening, we can obtain For a single QMDCT coefficient q, its sign bit mapping process f2 can be expressed as:
[0103]
[0104] Where c represents the mapped binary carrier bits.
[0105] The sign bit filtering threshold T is set to a small value, typically 2, to maintain minimal perturbation. Therefore, QMDCT coefficients filtered by this method will only be embedded in the message in either the sign bit or the overflow word LSB. For QMDCT coefficients filtered by the sign bit... and QMDCT coefficients filtered by overflow bits The following conditions must be met between the two:
[0106]
[0107] Will and After merging, the joint embedding domain of the i-th frame can be obtained. Figure 3 This is a schematic diagram of the joint embedding domain construction of the present invention.
[0108] When the sign bit threshold T is set to 2, the final bit mapping process for a single QMDCT coefficient q in the joint embedding region can be expressed as:
[0109]
[0110] Where c represents the binary bits mapped from coefficient q, and the sign bit and overflow word LSB are sequentially concatenated to obtain the binary carrier bit string C of a single frame. i Simultaneously record C i The corresponding QMDCT coefficient index number K i .
[0111] S3: The distortion function designed in this invention includes two parts: a statistical distortion function and a perceptual distortion function. The final distortion function is a combination of the two functions.
[0112] Statistical Distortion Function: To preserve the statistical properties of QMDCT coefficients, this invention references the design of distortion functions in image steganography. Borrowing the concept of Uniform Embedding Distortion (UED), this invention unifies the embedding modification to all possible amplitudes of quantized Discrete Cosine Transform (DCT) coefficients, aiming to reduce the average variation in DCT coefficient statistics. This invention also attempts to uniformly propagate the embedding modification to all possible amplitudes of QMDCT coefficients. In the proposed scheme, the secret information is embedded in a single channel of each frame. For the j-th QMDCT coefficient Q in the i-th frame... i,j The statistical distortion function is defined as:
[0113]
[0114] In this invention, only points Q within the same frame are considered. i,jTwo adjacent points Q i,j and Q i,j+k That is, K = {-1, 1}. α is an adjustment parameter, which was determined to be 1.3 through experiments.
[0115] Perceptual Distortion Function: To reduce auditory perception distortion, this invention proposes a perceptual distortion function using a psychoacoustic model. The psychoacoustic model analyzes the audio signal and calculates the available noise masking as a function of frequency. The absolute hearing threshold is the lowest sound pressure level of a pure tone that a person with normal hearing can hear in an absolutely quiet environment. The absolute hearing threshold can be approximated by the following formula:
[0116]
[0117] Where f represents the frequency value, T f The smaller the value, the more sensitive the human ear is to the current frequency.
[0118] According to ISO / IEC 13818-7:1997, psychoacoustic parameter thresholds for quiet conditions under different sampling rates and windows are provided. A psychoacoustic parameter threshold of 44.1 kHz with either a long or short Fast Fourier Transform (FFT) is selected. A smaller psychoacoustic parameter threshold implies higher sensitivity of the human ear to audio signals. The j-th QMDCT coefficient Q in the i-th frame is... i,j Perceptual distortion is represented as qsthr(Q i,j ), and calculate according to Tables C.1 to C.24 in ISO / IEC 13818-7:2006.
[0119] Combining the statistical and perceptual distortion of each QMDCT coefficient, for a single QDMCT coefficient Q i,j Synthetic distortion function D(Q) i,j ) is defined as:
[0120] D(Q i,j )=log2||ρ(Q i,j )||-log2||qsthr(Q i,j )||+β
[0121] Where β is an adjustment parameter, which can be set to 1 by default.
[0122] In practice, calculations are performed frame-by-frame. Statistical distortion is based on the cost of a single coefficient, which is calculated using both the coefficient itself and the coefficients of its neighbors. Perceptual distortion is based on indexed encoding of QMDCT coefficients. Therefore, according to K... i and Q i The binary carrier bit string C can then be calculated. i The corresponding distortion cost ρi .
[0123] S4: Repeat steps S2 and S3 until all audio frames have been traversed. Concatenate the binary carrier bit strings of all frames sequentially to obtain the binary carrier bit string C. Similarly, based on the correspondence, the modification cost ρ for all bits in C can be obtained. Record the source frame number A and QMDCT coefficient index number K corresponding to all bits in the binary carrier bit string C. This information is used to assist in subsequent bit modifications.
[0124] S5: Scramble the binary carrier bit string C and the modified distortion cost array ρ using the same scrambling algorithm and parameter configuration to obtain carriers C′ and ρ′. This invention uses the Logistic chaotic mapping algorithm to scramble C and ρ, using the same parameter configuration. The purpose of scrambling is to ensure a uniform distribution of the cost at the embedding points, achieving a better embedding path selection.
[0125] S6: Using STC encoding, the secret information M is embedded into the carrier C′, resulting in the secret bit string S′. The embedding process can be represented as follows:
[0126] S′=STC(C′,M,ρ,height,width)
[0127] Here, height and width are the height and width of the submatrix of the STC encoder, respectively. STC encoding can select the optimal embedding path based on the modification cost, minimizing the total distortion for each embedding.
[0128] S7: Using the corresponding chaotic inverse scrambling algorithm in step S5, and with the same parameter configuration, the encrypted string S′ is inversely scrambled to obtain the encrypted binary bit string S, so that its order is the same as the original order.
[0129] S8: Based on the difference between the encrypted binary bit string S and the carrier binary bit string C, a bit modification scheme for the AAC bitstream file can be constructed. The bit modification scheme consists of two parts: bit reversal and bit substitution.
[0130] Since the overflow word only exists in number book 11, if the carrier bit C i With the carrier bit S i The codes are inconsistent, and the corresponding codebook serial numbers are also inconsistent. Belongs to set B t If the sequence is {3, 4, 7, 8, 9, 10, 11}, then a bit reversal occurs. For number book 11, the bits to be reversed are the LSBs of the Huffman codeword overflow word; for other number books, they are the sign bits of the Huffman codeword. The specific reversal position can be based on the starting bit storage position. S iThe analytical sequence number corresponding to the QMDCT coefficients and the The number book The structure of each codeword is determined.
[0131] If the carrier bit C i With the carrier bit S i The two are inconsistent, and their corresponding codebook serial numbers do not belong to set B. t This embedding will result in bit substitution; however, not every pair of bits C... i With the carrier bit S i Inconsistent and not belonging to set B t In any of these cases, a bit substitution will occur. Assume C... i The corresponding QMDCT coefficients are Derived from codeword w, the set of QMDCT coefficients q = (q1, q2, q3, q4) obtained from the parsing of codeword w, and the parsing sequence number. If only q3 in q needs to be modified, then the modified coefficient q′=(q1, q2, -q3, q4) requires replacing the codeword w with w′=Ψ(q′). Correspondingly, in the AAC bitstream, the starting bit storage position needs to be... Replace the Huffman codeword w with codeword w′. If multiple QMDCT coefficients in a set of QMDCT coefficients q = (q1, q2, q3, q4) parsed from the same codeword w require inversion, then modify q once and map it, followed by a subsequent codeword replacement. Record the starting bit storage position and the target replacement codeword for all bit replacement operations.
[0132] S9: Based on the bit modification scheme obtained in step S8, modify the original AAC code stream H... c The corresponding bits are reversed or replaced. After all bits are modified, the encrypted AAC audio file is obtained. s .
[0133] This invention also proposes a semi-decoding-based information steganography embedding system for AAC bitstreams, used to implement the semi-decoding-based information steganography embedding method for AAC bitstreams proposed in this invention. Figure 4 This is the module structure diagram of the system, which mainly includes eight modules: audio parsing module, embedding domain construction module, joint distortion function calculation module, scrambling module, STC encoding module, inverse scrambling module, bit modification scheme formulation module, and AAC bitstream modification module.
[0134] The audio parsing module is used to execute step S1, and is responsible for processing the AAC encoded bitstream H of the carrier. cPerform semi-decoding to extract all QMDCT coefficients Q and their corresponding codeword start bit position P, codebook number B, codebook index I, and parsing sequence number Z of the codeword corresponding to the QMDCT coefficient in the file, stored in Huffman codeword format.
[0135] The embedding field construction module is used to execute step S2, which is responsible for concatenating the selected available QMDCT coefficient sign bits or overflow word LSBs to obtain the carrier bit string C and recording the source frame number A and QMDCT coefficient index number K for each bit.
[0136] The joint distortion function calculation module is used to execute step S3, which is responsible for calculating the modification cost of each embeddable position frame by frame based on the psychoacoustic model and the statistical distortion model, and obtaining the modification cost ρ corresponding to the carrier bit string C.
[0137] The scrambling module is used to execute step S5, which is responsible for scrambling C and ρ using the same scrambling algorithm and parameter configuration to obtain the scrambled bit string C′ and the scrambling modification cost ρ′.
[0138] The STC encoding module is used to execute step S6. It is responsible for using STC encoding combined with the modification cost of the embeddable position to embed the secret information M to be embedded into the scrambled bit string C′, so as to obtain the scrambled secret bit string S′.
[0139] The reverse scrambling module is used to execute step S7, and is responsible for reverse scrambling using the reverse scrambling algorithm and parameter configuration corresponding to the scrambling module to obtain the original sequence of the data-carrying bit string S.
[0140] The bit modification scheme formulation module is used to execute step S8, and is responsible for formulating a bit modification scheme based on the differences between C and S, combined with the mapped codeword information.
[0141] The AAC bitstream modification module is used to execute step S9 and is responsible for modifying the carrier AAC encoded bitstream H according to the bit modification scheme. c Modified to carrier AAC encoded bitstream H s .
[0142] The mathematical formula for the embedding algorithm can be expressed as:
[0143]
[0144] This invention also provides a method for extracting information from AAC bitstreams based on semi-decoding. Information extraction is the reverse process of information embedding, and includes the following steps:
[0145] P1: Use a decoder to decode the AAC bitstream file, extract the QMDCT coefficients Q, and record the codebook number B corresponding to the parsed codeword and the index I of the codeword in the codebook.
[0146] P2: For each QMDCT coefficient in a single frame, sequentially traverse each QMDCT coefficient q. For a single QMDCT coefficient q, firstly determine whether the overflow word LSB is embedded with information directly based on the amplitude. If the overflow word LSB is invalid, then determine whether the sign bit is embedded with information based on the amplitude, the same filtering threshold T as in step S2, and the codebook number b and codeword index i of the corresponding codeword. For the filtered ones, perform bit mapping in the same way as in step S2, and sequentially concatenate the sign bit with the embedded information and the overflow word LSB to obtain the binary data-carrying bit string S of the single frame. i .
[0147] P3: Concatenate the data carrier bit strings of all frames sequentially to obtain the final data carrier bit string S.
[0148] P4: Use the same scrambling algorithm and parameter configuration as in step S4 to scramble the dense bit string S, and obtain the extractable dense bit string S′.
[0149] P5: Use STC decoding to extract secret information M from S′. The parameters used in the extraction process, such as the width and height of the generation matrix, are consistent with the parameters used in the embedding process.
[0150] The present invention provides an embodiment as follows, which utilizes the information steganography extraction method proposed in this invention to realize the secret information extraction process. Figure 5 This is a schematic diagram illustrating the secret information extraction process of this embodiment. The specific implementation process of information steganography extraction is as follows:
[0151] P1: Use a decoder to decode the AAC bitstream file, extract the QMDCT coefficients Q, and record the codebook number B corresponding to the parsed codeword and the index I of the codeword in the codebook.
[0152] P2: The method for constructing the embedding region in the extraction process is completely consistent with the embedding process, and the extraction process is performed frame by frame. For the QMDCT coefficients of a single frame, each QMDCT coefficient q is sequentially traversed. For a single QMDCT coefficient q, the overflow word LSB is first determined directly by the amplitude size to determine whether information is embedded. If the overflow word LSB is invalid, the symbol bit is determined to carry bit information based on the amplitude size, the same filtering threshold T as in the embedding process, and the codebook number b and codeword index i of the corresponding codeword.
[0153] For a single QMDCT coefficient q, if its parsed source codeword is the i-th codeword in the b-th codebook, then the function Φ(·) to determine whether it carries bit information (i.e., secret information) can be expressed as:
[0154]
[0155] Among them, Ub,i =1 indicates that reversing the i-th code character in the b-th codebook is feasible, and Φ(q) =1 indicates that the coefficient carries bit information. For the QMDCT coefficient Q of a single frame... i After filtering, the embedded fields can be obtained.
[0156] Similar to the embedding process, when the sign bit threshold T is set to 2, for a single QMDCT coefficient q in the embedding region, the final bit mapping process can be expressed as:
[0157]
[0158] Where s represents the binary data carrier bits mapped from coefficient q, and sequentially concatenating all the data carrier bits of a single frame yields the binary data carrier bit string S. i ,
[0159] P3: Concatenate the data carrier bit strings of all frames sequentially to obtain the final data carrier bit string S.
[0160] P4: Use the same scrambling algorithm and parameter configuration as the embedding process to scramble the data-carrying bit string S, and obtain the extractable data-carrying bit string S′.
[0161] P5: Use the STC decoder to extract information from the encrypted string S′. The parameters used in the extraction process, such as the width and height of the generator matrix, are consistent with the parameters used in the embedding process. The STC decoder expression is as follows:
[0162] M = STC -1 (S′, height, width)
[0163] Where height is the height of the submatrix in the STC decoder, and width is the width of the submatrix in the STC decoder.
[0164] This invention also provides a system for extracting steganographic information from AAC bitstreams that have undergone partial audio decoding, used to implement the method for extracting steganographic information from AAC bitstreams that has undergone partial audio decoding proposed in this invention. Figure 6 This is a module structure diagram of the information steganography extraction system, which includes four modules: audio parsing module, embedding domain construction module, scrambling module, and STC decoding module.
[0165] The audio parsing module is used to execute step P1, which is responsible for processing the encrypted AAC encoded bitstream H. s Perform semi-decoding to extract all QMDCT coefficients Q and their corresponding codebook numbers B and codeword codebook index I.
[0166] The embedded field building module is used to execute steps P2 and P3, and is responsible for concatenating the selected QMDCT coefficient sign bits or overflow word LSBs to obtain the data-carrying bit string S.
[0167] The scrambling module is used to execute step P4, which is responsible for scrambling S using the same scrambling algorithm and parameter configuration as the embedding process, to obtain the scrambled carrier bit string S′.
[0168] The STC decoding module is used to execute step P5, which is responsible for extracting message M from the carrier bit string S′.
[0169] In this invention, the scrambling algorithm and cost function can be replaced by other different algorithms without affecting the system structure and process. Furthermore, the method of this invention is also applicable to MP3 (MPEG-1 standard) compressed audio.
[0170] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Those skilled in the art can modify or make equivalent substitutions to the technical solutions of the present invention without departing from the spirit and scope of the present invention. The scope of protection of the present invention should be determined by the claims.
Claims
1. A method for embedding information steganography in AAC bitstreams based on semi-decoding, characterized in that, Includes the following steps: S1: Perform semi-decoding on the AAC bitstream file to extract the QMDCT coefficients Q and Huffman information; S2: Traverse the QMDCT coefficients Q of each audio frame, extract coefficients q that meet the filtering conditions from Q. The filtering conditions include that the Huffman codeword corresponding to q contains an overflow word, the magnitude of q is not greater than the filtering threshold T, and q does not change the length of the corresponding Huffman codeword after sign inversion; map and concatenate the overflow word LSB and sign bit in coefficient q to obtain the carrier bit string C, and record the source frame number A and QMDCT coefficient index number K of each bit; S3: For each frame's carrier bit string C, calculate the modification distortion cost ρ corresponding to the carrier bit string C based on the QMDCT coefficients Q and their index numbers K. S4: Repeat steps S2 and S3 to perform frame-by-frame calculations until all audio frames have been traversed, obtain the carrier bit string C of all frames, the modification distortion cost ρ of all bits, and record the source frame number A and QMDCT coefficient index number K corresponding to all bits in the carrier bit string C. S5: Scramble the carrier bit string C and the distortion cost ρ using the same scrambling algorithm and parameter configuration to obtain the carrier C′ and ρ′; S6: Based on the modified distortion cost ρ and the secret information M, the secret information is embedded using the check trellis code STC embedding carrier C′ to obtain the out-of-order secret bit string S′. S7: Use the reverse scrambling algorithm and parameter configuration corresponding to step S5 to reverse scramble the bit string S′ to obtain the ordered dense bit string S; S8: Based on the ordered encrypted bit string S obtained in step S7 and the source frame number A and QMDCT coefficient index number K obtained in step S4, and based on the QMDCT coefficient Q and Huffman information recorded in step S1, determine the codeword bit modification scheme, which consists of two parts: bit reversal and bit substitution. S9: Modify some bits in the original AAC bitstream file according to the determined codeword bit modification scheme, and obtain the encrypted AAC audio after modification.
2. The information steganography embedding method as described in claim 1, characterized in that, In step S1, the Huffman information includes the starting bit storage position P of the Huffman codeword W corresponding to Q in the AAC code stream file, the codebook number B of the Huffman codeword W, the index I of the codeword in the codebook, and the Huffman codeword parsing sequence number Z corresponding to the QMDCT coefficient Q.
3. The information steganography embedding method as described in claim 1, characterized in that, The condition for coefficient q to contain overflow words in step S2 is: the absolute value of q is greater than or equal to 16, and all the sign bits of the coefficients containing overflow words are used as embeddable messages; among them, the following two conditions must be met when selecting the sign bits of the coefficients to embed messages: (1) the absolute value of the coefficient q must be greater than 0 and less than or equal to the filtering threshold T, (2) the Huffman code character corresponding to the coefficient q can be reversed.
4. The information steganography embedding method as described in claim 3, characterized in that, The condition for passing the sign reversal feasibility test is: after reversing the sign bit of the QMDCT coefficient q, the length of its corresponding Huffman codeword is not changed. The methods for determining the codewords of different codebooks include: 1) For the QMDCT coefficient q of the corresponding codeword in number books 1, 2, 5, and 6, the corresponding Huffman codeword substitution should satisfy the following conditions: Where q = (q1, q2, ... q n Let S be a vector consisting of the n coefficients of the parsed source codeword w to which coefficient q belongs, and S = {-1, +1}. n , Let Ψ be the coefficient matrix, W be the codeword matrix, Ψ represent the mapping from QMDCT coefficients q to codewords based on the G codebook, and |·| represent the bit length of the Huffman codeword; w t It is a row vector in W, representing a Huffman codeword; 2) The QMDCT coefficient q corresponding to the remaining codebook has its sign bit represented as a single independent bit in the Huffman codeword.
5. The information steganography embedding method as described in claim 1, characterized in that, The mapping of the overflow word LSB of the QMDCT coefficient q in step S2 is as follows: Where f1 represents the mapping relationship, and c is the binary bit that the overflow word LSB is mapped to; The mapping of the sign bit of the QMDCT coefficient q is as follows: Where f2 represents the mapping relationship, and c is the binary bit mapped to the sign bit.
6. The information steganography embedding method as described in claim 1, characterized in that, Step S3 modifies the distortion cost ρ, which includes statistical distortion cost and perceptual distortion cost. The cost of statistical distortion is calculated using the following statistical distortion function: Among them, Q i,j Q i,j+k Let K be the j-th and (j+k)-th adjacent coefficients in the i-th frame, K = {-1, 1}, and α be the adjustment parameter; The cost of perceptual distortion is based on the human hearing threshold T in the psychoacoustic model. f The calculation yields qsthr(Q) i,j ); The distortion cost ρ is then calculated based on the following comprehensive distortion function: D(Q i,j )=log2‖ρ(Q i,j )‖-log2‖qsthr(Q i,j )‖+β Where β is the adjustment parameter.
7. A steganography embedding system for AAC bitstreams based on semi-decoding, used to implement the steganography embedding method according to any one of claims 1-6, characterized in that, include: The audio parsing module is used to perform partial decoding on AAC bitstream files, extracting QMDCT coefficients Q and Huffman information; The embedding domain construction module is used to traverse the QMDCT coefficients Q of each audio frame, extract coefficients q that meet the filtering conditions from Q, map and concatenate the overflow word LSB and sign bit in coefficient q to obtain the carrier bit string C, and record the source frame number A and QMDCT coefficient index number K of each bit. The joint distortion function calculation module is used to calculate the modification distortion cost ρ corresponding to the carrier bit string C for each frame, based on the QMDCT coefficients Q and their index numbers K. The scrambling module is used to scramble the carrier bit string C and the distortion cost ρ using the same scrambling algorithm and parameter configuration to obtain the carrier C′ and ρ′. The STC encoding module is used to embed secret information based on the modified distortion cost ρ and secret information M, using the check trellis encoding STC embedding carrier C′ to obtain the out-of-order secret bit string S′. The reverse scrambling module is used to reverse scramble the bit string S′ using the reverse scrambling algorithm and parameter configuration to obtain an ordered dense bit string S; The bit modification scheme formulation module is used to determine the codeword bit modification scheme based on the ordered encrypted bit string S, the source frame number A, and the QMDCT coefficient index number K, and based on the recorded QMDCT coefficients Q and Huffman information. The bit modification scheme consists of two parts: bit reversal and bit substitution. The AAC bitstream modification module is used to modify some bits in the original AAC bitstream file according to a determined codeword bit modification scheme, and after modification, a encrypted AAC audio is obtained.
8. A method for extracting secret information from an AAC bitstream based on semi-decoding, used to extract secret information from the secret-carrying AAC audio obtained by the steganography embedding method according to any one of claims 1-6, characterized in that, The information steganography extraction method includes the following steps: P1: Use a decoder to decode the AAC audio file, extract the QMDCT coefficients Q, and record the codebook number B corresponding to the parsed codeword and the index I of the codeword in the codebook; P2: For the QMDCT coefficients of a single frame, sequentially traverse each QMDCT coefficient q; for a single QMDCT coefficient q, determine whether the overflow word LSB embeds secret information based on the amplitude; if the overflow word LSB is invalid, determine whether the sign bit embeds secret information based on the amplitude, the filtering threshold T, the codebook number b of the corresponding codeword, and the codeword index i; perform bit mapping on the filtered coefficients q, and sequentially concatenate the sign bit of the embedded information and the overflow word LSB to obtain the binary secret bit string S of the single frame. i ; P3: Concatenate the data carrier bit strings of all frames sequentially to obtain the final data carrier bit string S; P4: Use the scrambling algorithm and parameter configuration to scramble the dense bit string S to obtain the extractable dense bit string S′; P5: Use STC decoding to extract secret information M from S′; The filtering threshold T and bit mapping method in step P2 are the same as those in step S2 of the steganography embedding method; the scrambling algorithm and parameter configuration in step P4 are the same as those in step S4 of the steganography embedding method.
9. The information steganography extraction method as described in claim 8, characterized in that, In step P2, for a single QMDCT coefficient q, the judgment function Φ(·) for determining whether secret information is embedded is: Among them, U b,i =1 indicates that the i-th code character in the b-th codebook can be reversed, and Φ(q) =1 indicates that the coefficient is embedded with secret information.
10. A system for extracting steganographic information from AAC code streams based on semi-decoding, used to implement the steganographic information extraction method of claim 8 or 9, characterized in that, include: The audio parsing module is used to decode AAC audio files using a decoder, extract QMDCT coefficients Q, and record the codebook number B corresponding to the parsed codewords and the index I of the codewords in the codebook. The embedding domain construction module is used to sequentially traverse each QMDCT coefficient q for a single frame of QMDCT coefficients. For a single QMDCT coefficient q, the magnitude of the overflow word LSB determines whether secret information is embedded. If the overflow word LSB is invalid, the magnitude of the overflow word LSB, the selection threshold T, the codebook number b of the corresponding codeword, and the codeword index i determine whether the sign bit is embedded with secret information. For the selected coefficients q, bit mapping is performed, and the sign bit of the embedded information and the overflow word LSB are sequentially concatenated to obtain the binary secret bit string S of the single frame. i ; Concatenate the data carrier bit strings of all frames sequentially to obtain the final data carrier bit string S; The scrambling module is used to scramble the dense bit string S using a scrambling algorithm and parameter configuration to obtain an extractable dense bit string S′. The STC decoding module is used to extract secret information M from S′ using STC decoding.
Citation Information
Patent Citations
MDCT quantization coefficient small value region-based advanced audio coding (AAC) audio steganography and extraction method
CN104867496A
Compressed audio adaptive steganography method and system based on equal-length entropy codeword replacement
CN109346091A
