CN110689897A - Information hiding and hidden information extraction method based on linear prediction speech coding - Google Patents

Information hiding and hidden information extraction method based on linear prediction speech coding Download PDF

Info

Publication number
CN110689897A
CN110689897A CN201910953485.9A CN201910953485A CN110689897A CN 110689897 A CN110689897 A CN 110689897A CN 201910953485 A CN201910953485 A CN 201910953485A CN 110689897 A CN110689897 A CN 110689897A
Authority
CN
China
Prior art keywords
index point
embedded
index
frame
point
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201910953485.9A
Other languages
Chinese (zh)
Inventor
刘鹏
李松斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Research Station Of South China Sea Institute Of Acoustics Chinese Academy Of Sciences
Original Assignee
Research Station Of South China Sea Institute Of Acoustics Chinese Academy Of Sciences
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Research Station Of South China Sea Institute Of Acoustics Chinese Academy Of Sciences filed Critical Research Station Of South China Sea Institute Of Acoustics Chinese Academy Of Sciences
Priority to CN201910953485.9A priority Critical patent/CN110689897A/en
Publication of CN110689897A publication Critical patent/CN110689897A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
    • G10L19/018—Audio watermarking, i.e. embedding inaudible data in the audio signal
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
    • G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques

Landscapes

  • Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Signal Processing (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Compression, Expansion, Code Conversion, And Decoders (AREA)

Abstract

The invention discloses an information hiding and hidden information extracting method based on linear predictive speech coding, which comprises the following steps: determining the position of a frame to be embedded in the low-rate compressed voice stream according to the embedding rate, the Key and a random position selection algorithm; acquiring a frame to be embedded according to the position of the frame to be embedded, encoding the frame to be embedded to obtain an index point M of the frame to be embedded, and determining a group to which the index point M belongs according to an index point grouping result in a pre-established index point set I; reading the secret information according to a preset length to obtain a unit U to be embedded; judging whether the index point M belongs to the group corresponding to the unit U, if so, not replacing the index point M, otherwise, acquiring the nearest projection point of the index point M in the group corresponding to the unit U, replacing the index point M, and completing the embedding of the unit U; and coding the low-rate compressed voice stream embedded with the secret information to obtain a target compressed voice stream.

Description

Information hiding and hidden information extraction method based on linear prediction speech coding
Technical Field
The invention relates to the field of information security, in particular to an information hiding and hidden information extracting method based on linear prediction speech coding.
Background
The continuous development of information technology brings convenience to the life of people, but also brings a plurality of problems of information authentication, copyright protection, secret communication and the like. Information hiding techniques may well address these issues. The method is a technology for embedding secret information into a common carrier and achieving the purpose of transmitting the secret information by hiding secret communication.
Digital watermarking technology is an important branch of the field of information hiding, and the technology utilizes an information hiding algorithm (also called steganography algorithm) to embed authentication information into digital carriers such as audio and the like. The information embedding positions are very hidden and difficult to be detected or modified by lawless persons, but a producer or a distributor can judge whether the carrier data is replaced or tampered according to the watermark information so as to ensure the safety of the carrier data. Embedding watermarks in low-rate compressed speech streams is a very challenging task because there is little redundant information in the code stream after low-rate compressed speech encoding. The existing watermark embedding method in the low-rate compressed voice stream can be divided into three categories according to steganographic positions: the first type is that information embedding is realized by directly modifying some code elements in a compressed voice stream, and the information embedding and the voice coding process are independent in the method; the second and third categories are information embedding during speech coding. The first type carries out information embedding in the short-time predictor prediction step, and the second type and the third type select to carry out information embedding in the long-time predictor prediction step. For example, the Quantization Index Modulation (QIM) steganography algorithm embeds information during Vector Quantization (VQ) of linear prediction coefficients, and the pitch Modulation steganography embeds information during pitch prediction of speech subframes.
The core issue of concern in the QIM steganography algorithm is how to partition the original codebook. Currently, researchers have proposed many codebook partitioning methods. Such as a completely random partition, the random partition method introduces a large additional quantization distortion and degrades the quality of the decoded speech. To solve this problem, researchers have proposed some more sophisticated approaches. For example, Chiang et al implement codebook grouping based on code sub-clustering, Lu et al expand codeword search range based on shared codeword grouping to achieve the purpose of reducing additional quantization distortion, and related technologies also propose a codebook partitioning method based on graph theory, which is called Complementary Neighbor Vertex (CNV) algorithm, and the like. Wherein, the related CNV-QIM algorithm embeds the secret information by modifying the LPC quantization index, and the method inevitably influences the distribution characteristic of the quantization index. When the number of changes introduced in the embedding process is too large, the voice quality is greatly influenced and can be easily detected by a steganography detection method. In recent years, Tian et al have proposed an embedding algorithm called Sec-QIM that reduces the number of changes introduced by the embedding process, resulting in improved embedding efficiency but reduced embedding capacity.
Disclosure of Invention
The invention aims to overcome the technical defects and provides an information hiding and hidden information extracting method based on linear predictive speech coding, which can ensure the embedding capacity and improve the embedding efficiency in the process of embedding secret information into a low-rate compressed speech stream.
To achieve the above object, embodiment 1 of the present invention proposes an information hiding method based on linear predictive speech coding, including:
determining the position of a frame to be embedded in the low-rate compressed voice stream according to the embedding rate, the Key and a random position selection algorithm;
acquiring a frame to be embedded according to the position of the frame to be embedded, encoding the frame to be embedded to obtain an index point M of the frame to be embedded, and determining a group to which the index point M belongs according to an index point grouping result in a pre-established index point set I;
reading the secret information according to a preset length to obtain a unit U to be embedded;
judging whether the index point M belongs to the group corresponding to the unit U, if so, not replacing the index point M, otherwise, acquiring the nearest projection point of the index point M in the group corresponding to the unit U, replacing the index point M, and completing the embedding of the unit U;
and coding the low-rate compressed voice stream embedded with the secret information to obtain a target compressed voice stream.
As an improvement of the above method, the method further comprises:
acquiring a corresponding index point set I based on a linear predictive coding LPC analysis filter and a fixed codebook of an encoder;
the quantization index set is C ═ C i1,2, …, n, where c isiPoint to the corresponding codeword in the ith codebook; if n LPC quantization code books in the encoder are respectively L1,L2…LnThe number of code words in the code book is respectively | L1|,|L2|…|LnIf the index point set I corresponding to the encoder is expressed as:
Figure BDA0002226493980000021
and grouping the index points in the index point set I based on a quantum particle swarm algorithm.
As an improvement of the above method, the grouping index points in the index point set I based on the quantum-behaved particle swarm optimization specifically includes:
randomly generating a plurality of digit strings according to the number of the index points in the index point set I and a preset value range; each digital string corresponds to one particle in the quantum particle swarm algorithm, and the value of the number at each position in the digital string represents the grouping result of the index point corresponding to the position;
initializing each particle, and optimizing all the particles obtained after initialization by adopting a quantum particle swarm algorithm to obtain a target grouping digit string.
As an improvement of the above method, initializing each particle, and optimizing all particles obtained after initialization by using a quantum particle swarm algorithm to obtain a target grouped digit string specifically includes:
randomly generating an integer belonging to [1, N ] as a starting position number, wherein N is the length of a digit string;
traversing all the positions in the digital string by taking the position as a starting point, and judging whether K projection points closest to the index point corresponding to the current position are grouped or not when traversing each position;
if n projection points in the K projection points are divided into the same group, n-1 conflict occurs in the grouping, and n is larger than or equal to 2; at the moment, whether 7+ K-1 projection points with the nearest range to be determined are grouped or not is judged; randomly grouping the projection points which are not grouped;
determining the quantum search space dimension of a quantum particle swarm algorithm according to the length of the particles;
determining the population number of a quantum particle swarm algorithm according to the number of all particles obtained after initialization;
determining iteration times, and acquiring the global optimal position of the particles according to the quantum search space dimension, the population particle number and the iteration times;
obtaining a target digit string according to the global optimal position of the particles; the value of the number at each position in the target grouping digit string represents the grouping result of the corresponding index point.
As an improvement of the above method, the acquiring a nearest projection point of the index point M in the group corresponding to the unit U specifically includes:
calculating q-dimensional quantization residual error coefficient r corresponding to index point MMAccording to rMCalculating the distance E (M, M ') between any index point M' in the index point set I and the index point M;
for collections
Figure BDA0002226493980000031
The index point M satisfies M ∈ I and
Figure BDA0002226493980000032
the set of M projected points in I 'is I'MIf the point M ' satisfies M ' e I 'MAnd for any point M ∈ I'MIf E (M, M ') ≦ E (M, M ″), then M ' is the nearest projection point of M in the set I '.
Embodiment 2 of the present invention provides a hidden information extraction method, which is used for extracting information hidden by the method, and includes:
determining the position of the carrier frame according to the Key;
acquiring a group to which an index point corresponding to the carrier frame at the position of the carrier frame belongs;
and extracting the secret information according to the grouping to which the index point belongs.
Embodiment 3 of the present invention provides an information hiding system based on linear predictive speech coding, including:
the position acquisition module of the frame to be embedded is used for determining the position of the frame to be embedded in the low-rate compressed voice stream according to the embedding rate, the Key and the random position selection algorithm;
the index point grouping determination module is used for acquiring a frame to be embedded according to the position of the frame to be embedded, encoding the frame to be embedded to obtain an index point M of the frame to be embedded, and determining a group to which the index point M belongs according to an index point grouping result in a pre-established index point set I;
the embedded unit acquisition module is used for reading the secret information according to the preset length to obtain a unit U to be embedded;
the embedding module is used for judging whether the index point M belongs to the group corresponding to the unit U or not, if so, the index point M is not replaced, otherwise, the nearest projection point of the index point M in the group corresponding to the unit U is obtained, the index point M is replaced, and the embedding of the unit U is completed;
and the coding module is used for coding the low-rate compressed voice stream embedded with the secret information to obtain a target compressed voice stream.
Embodiment 4 of the present invention provides a hidden information extraction system, including:
the carrier frame position acquisition module is used for determining the position of the carrier frame according to the Key;
the index point grouping determination module is used for acquiring a grouping to which the index point corresponding to the carrier frame at the position of the carrier frame belongs;
and the secret information extraction module is used for extracting the secret information according to the grouping to which the index point belongs.
The invention has the advantages that:
the information hiding method of the linear prediction speech coding provided by the invention is characterized in that index points in the index point set are grouped based on a quantum particle swarm algorithm, and the grouping of index points M corresponding to a frame to be embedded is determined according to the grouping condition; and then, reading the secret information to be hidden according to the unit U, replacing the index point M by using the nearest projection point of the index point M in the group corresponding to the unit U when the index point M does not belong to the group corresponding to the unit U, realizing the embedding of the secret information, continuously encoding to obtain a target compressed voice stream embedded with the secret information, and improving the embedding efficiency and the information safety under the condition of ensuring the embedding capacity.
Drawings
Fig. 1 is a flowchart illustrating an information hiding method for linear predictive speech coding according to embodiment 1 of the present invention;
fig. 2 is a schematic diagram of a quantization index space provided in embodiment 1 of the present invention.
Detailed Description
The embodiments of the present invention will be described in further detail with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of and not restrictive on the broad invention. It should be further noted that, for convenience of description, only some structures, not all structures, relating to the embodiments of the present invention are shown in the drawings.
It should also be noted that, for the convenience of description, only some but not all of the pertinent contents of the embodiments of the present invention are shown in the drawings. Some example embodiments are described as processes or methods depicted as flow diagrams, which describe operations (or steps) as a sequential process, many of the operations can be performed in parallel, concurrently or simultaneously, and the order of the operations can be re-arranged. The process may be terminated when its operations are completed, but may have additional steps not included in the figure.
The basic processing of linear prediction coding typically includes: predicting the current sample value by using an LPC analysis filter, wherein the output of the LPC analysis filter is called residual error coefficient; quantizing the output residual error coefficient into a code word set based on a fixed codebook of an encoder, wherein the quantization process is different according to different encoders; then, in order to improve compression efficiency, the codeword set is expressed as an LPC quantization index set C ═ CiI |, 1,2, …, n }. Wherein, ciPointing to the corresponding codeword in the ith codebook. In decoding, the quantization index set C and the fixed codebook may be used to obtain residual coefficients. Finally, the LPC coefficients may be obtained by an LPC synthesis filter. In addition, the LPC coefficients, the line spectrum pair coefficients and the line spectrum frequency coefficients can be inter-converted according to the quantization and interpolation requirements. In a common low-rate speech coder, the LPC residual coefficients are quantized separately, usually divided into three sub-numbers during the quantization process. Therefore, the quantization index set C usually includes three quantization indexes, which correspond to three codebooks in the encoder.
The associated CNV-QIM algorithm sets the quantization index set C ═ C1,c2,c3Each index in the index is treated as a separate embedded unit. If a certain index is selected as the embedding carrier, 1-bit secret information can be embedded therein. Therefore, the CNV-QIM method can embed 3-bit secret information by modifying up to three indexes in each LPC vector quantization process. The Sec-QIM method improves the embedding efficiency by introducing matrix coding, which puts the entire quantization index set C ═ C1,c2,c3And the method is regarded as a QIM embedding unit, and according to a matrix coding rule, only one index is needed to be selected at most in the set C for modification so as to embed the 2-bit secret information.
The invention provides a more efficient compressed voice information embedding method based on a quantum particle swarm algorithm, which can ensure that the embedding capacity is the same as that of a CNV-QIM algorithm and has higher embedding efficiency than that of a Sec-QIM algorithm. Where the embedding capacity can be understood as the number of bits that can be embedded in a unit of embedding carrier. Embedding Efficiency (EE) represents the number of information bits that can be embedded per modified codeword on average. When embedding k-bit secret information, if k' code words are changed, the embedding efficiency can be obtained by the following equation (1):
Figure BDA0002226493980000051
fig. 1 is a flowchart of an information hiding method based on linear predictive speech coding according to embodiment 1 of the present invention, which is suitable for hiding secret information, such as embedding watermark information into a low-rate compressed speech stream. The method comprises the following steps:
and 110, acquiring a corresponding index point set I based on a linear predictive coding LPC analysis filter and a fixed codebook of an encoder, and grouping index points in the index point set I based on a quantum particle swarm algorithm.
The quantization index set C ═ { C } may be obtained according to the process of linear prediction coding described aboveiI is 1,2, …, n, and the LPC residual coefficients are usually divided into three sub-coefficients for quantization, and the case where n is 3 will be described as an example. At this time, one index set may be represented as C ═ { C ═ C1,c2,c3}. To describe the quantization index space, a cartesian three-dimensional rectangular coordinate system is established with (0,0,0) as the origin, and in this case, any one index set C ═ C1,c2,c3All can be regarded as one point (c) in the three-dimensional LPC quantization index space1,c2,c3) I.e. the index point. Suppose that three LPC quantization codebooks in the encoder are L respectively1、L2And L3The number of code words in the code book is respectively | L1|、|L2L and L3If the index point set I corresponding to the encoder is:
Figure BDA0002226493980000061
in Cartesian three-dimensional rectangular coordinate systemThree coordinate axes form three planes together, and the three planes are marked as P ═ rhoiI ═ 1,2,3 }. Set S ═ λ j1,2, …, ∞ is referred to as a set of spatially orthogonal planes, where λjRepresenting a plane in three-dimensional space. The plane satisfies the following constraints:
λj∈Por(λj//ρiandD(λj,ρi)∈N),1≤i≤3 (3)
wherein, D (λ)j,ρi) The euclidean distance between two planes is shown, and S represents a set of a plane formed by arbitrary two coordinate axes and a plane parallel to the formed plane and having a distance of a natural number, as can be seen from the above constraint. Since planes formed by any two coordinate axes in the spatial rectangular coordinate system are orthogonal to each other, any plane in the orthogonal plane set is orthogonal to planes in other directions. At this point, the index point corresponds to the intersection of the planes in the set S.
For index point M ═ M1,m2,m3) Assume that the index point after steganography becomes M '═ M'1,m′2,m′3) And M ≠ M'. If M 'is the projection point of M on the orthogonal plane set, M' is called the projection point of M, and has the following properties:
then
Figure BDA0002226493980000063
Wherein
Figure BDA0002226493980000064
And
Figure BDA0002226493980000065
respectively being a component plane rhoiUnit vector on two coordinate axes. Order to
Figure BDA0002226493980000066
Figure BDA0002226493980000071
Then, the following formula (5) can be obtained:
(m′1-m1)(x1-x′1)+(m′2-m2)(x2-x′2)+(m′3-m3)(x3-x′3)＝0 (6)
due to the fact that
Figure BDA0002226493980000072
And
Figure BDA0002226493980000073
is a unit vector located on two coordinate axes, thus (x)1-x′1)、(x2-x′2) And (x)3-x′3) Two of them are not 0, then (m'1-m1)、(m′2-m2) And (m'3-m3) Two items in the index list are 0, that is, only one index is needed to be modified at most when the original index point is replaced by the projection point.
Fig. 2 is a schematic diagram of a quantization index space provided in embodiment 1 of the present invention, and a description is given to a nearest projection point of the index point M with reference to fig. 2. As shown in fig. 2, three planes are formed between two coordinate axes, which are respectively denoted as α, β, and γ. In fig. 2, α 1, α 2, and α 3 are parallel to α and have a distance of a natural number, β 1, β 2, and β 3 are parallel to β and have a distance of a natural number, and γ 1, γ 2, and γ 3 are parallel to γ and have a distance of a natural number. Then the 16 planes marked in fig. 2 all belong to the set of orthogonal planes S. It should be noted that the orthogonal plane cluster includes an infinite number of planes, and only 16 planes are used herein for exemplary illustration, and all points shown by black dots "·" in fig. 2 are projected points of M.
For any index point M epsilon I, corresponding q-dimensional quantized residual error coefficient rMCan be obtained from the function f (M) defined by the respective encoder. In the embodiment of the present invention, a distance E (M, M ') between any two index points M and M' in the set I is defined as:
Figure BDA0002226493980000074
wherein r isi MAnd ri M′Respectively representing quantized residual coefficients rMAnd rM′The ith dimension coefficient of (1).
Set of assumptionsThe index point M satisfies M ∈ I and
Figure BDA0002226493980000076
the set of M projected points in I 'is I'MIf the point M ' satisfies M ' e I 'MAnd for any point M ∈ I'MIf the following expression (8) is satisfied, M 'is the nearest projection point of M in the set I'.
E(M,M′)≤E(M,M″) (8)
Before classification is carried out by using a quantum particle swarm algorithm, a plurality of digit strings are randomly generated according to the number of index points in the index point set I and a preset value range. Each digital string corresponds to one particle in the quantum particle swarm algorithm, and the value of the number of each position in the digital string represents the grouping result of the index point corresponding to the position.
Suppose that index points are divided into 8 groups, each group taking the value of [0,7 ]]Is an integer of (1). The size of the index point set I is | L1|·|L2|·|L3L, the acquisition length is | L1|·|L2|·|L3And | each position in the numeric string corresponds to an index point, and the numerical value of the position represents the grouping result of the index point. The value of each position in the digit string is [0,7 ]]Is an integer of (1). A plurality of strings of numbers are generated, each string of numbers representing a particle. Initializing each particle, and optimizing all the particles obtained after initialization by adopting a quantum particle swarm algorithm to obtain a final grouping digital string.
In order to evaluate the quality of the grouping result, the invention provides a concept of average replacement distance. Suppose index point M is atGroup Ii(1. ltoreq. i.ltoreq.7) is Mi(i is more than or equal to 1 and less than or equal to 7), and the distance between the projection points is calculated by utilizing the corresponding quantized Euclidean distance. Suppose index point MiIn set I is k-th from the point M, except for itselfi(i is more than or equal to 1 and less than or equal to 7) near projection points. At this time, the average replacement distance a at the time of replacing the point MMThe following equation (9) can be obtained:
Figure BDA0002226493980000081
the purpose of initializing each particle is to distribute the index point corresponding to each position in the number string corresponding to each particle and the K nearest projection points in different groups as much as possible. The specific process is as follows: an integer belonging to [1, N ] is randomly generated as a start position number, where N is the length of the digit string. All the positions in the digit string are traversed starting from this position. And when each position is traversed, judging whether K projection points closest to the index point corresponding to the current position are grouped or not. If n (n > -2) projection points in the K projection points are grouped into the same group, which indicates that n-1 collision occurs in the grouping, the range to be determined at this time is whether the nearest 7+ K-1 projection points are grouped or not. And randomly grouping the projection points which are not grouped.
After the initialization of the particles is completed, the optimal grouping can be found by using a quantum particle swarm algorithm. And adopting an average replacement distance function as a fitness function for calculating the fitness value of the particle, wherein the smaller the fitness value is, the better the position of the particle is. Determining the maximum iteration time T, and when the iteration time is T, the position of the ith particle in the quantum search space is as follows:
Xi(t)＝(xi1(t),xi2(t),....xin(t)),i∈[1,M](10)
the local optimum position of the particle is noted as:
Pi(t)＝(pi1(t),pi2(t),....pin(t)),i∈[1,M](11)
the global optimal position is noted as:
G(t)＝(g1(t),g2(t),....gn(t)), and g (t) Pg(t) (12)
The local optimal position mean value mbest for all particles is noted as:
Figure BDA0002226493980000082
in the iteration process, updating the positions of the particles, and updating the local optimal position of the particle by using the current position of the particle when the fitness value corresponding to the current position of each particle is smaller than the fitness value corresponding to the local optimal position of the particle; when the fitness value corresponding to the local optimal position of the particle is smaller than the fitness value corresponding to the global optimal position, updating the global optimal position by using the local optimal position of the particle, namely obtaining the global optimal position by comparing the fitness values of all the particles at different positions; calculating the average optimal position of the particles according to the local optimal positions of all the particles; and when the maximum iteration time T is reached or the adaptive value meets the preset condition, stopping updating the positions of the particles and acquiring the global optimal position.
And the numeric string corresponding to the global optimal position is a target grouping numeric string, and the value of the number at each position in the target grouping numeric string represents the grouping result of the corresponding index point. Grouping optimization of the index points in the index point set is carried out based on a quantum particle swarm algorithm, and the average replacement distance of the index points can be reduced as much as possible, so that the voice embedded with the secret information keeps high quality.
It should be noted that the process of grouping the index points in the index point set I only needs to be performed once, the grouping result can be reused, and the index points in the index point set I do not need to be grouped during each steganography.
And step 120, obtaining the position of the frame to be embedded in the low-rate compressed voice stream according to the embedding rate, the Key and the random position selection algorithm.
The embedding rate can be determined according to the required security level, and the Key can be generated by adopting the existing Key generation mode. Suppose a steganographic method bThe average accuracy when detected by the steganographic detection method d is c, then the security level L can be defined as Lb,d1-c. It is usually not necessary to embed in every carrier frame when embedding the watermark information, and therefore the embedding location needs to be chosen, and different embedding locations will correspond to different security levels. When i frames are selected for embedding among n carrier frames, the embedding rate can be expressed as R ═ i/n. The security level and the embedding rate are generally in the interval of R ^ 1/Lb,dAnd thus the corresponding security level can be determined by adjusting the embedding rate.
In order to ensure that the embedding rate is expected, all speech frames can be divided into a plurality of embedding units, each embedding unit comprises m carrier frames, wherein m is a preset value. And calculating the embedding rate according to the required security level, obtaining the number of the frames to be embedded according to the embedding rate, and then selecting the positions of all the frames to be embedded by using the Key and the strong random number generator.
Step 130, acquiring a frame to be embedded according to the position of the frame to be embedded, encoding the frame to be embedded to obtain an index point M of the frame to be embedded, and determining a group to which the index point M belongs according to an index point grouping result in the index point set I.
Step 140, reading the secret information according to a preset length, and acquiring the unit U to be embedded.
For example, the watermark information is converted into a binary sequence, the preset length may be 3 bits, each 3 bits is taken as a unit U, and when steganography is performed, one unit U is read each time for embedding.
And 150, judging whether the index point M belongs to the group corresponding to the unit U, if so, not replacing the index point M, and if not, replacing the index point M by using the nearest projection point of the index point M in the group corresponding to the unit U to complete the embedding of the unit U.
For example, when the length of the unit U is 3 bits, the unit U may have the same value as the 8 packet classes. And judging whether the group to which the index point of each frame to be embedded belongs is consistent with the corresponding embedding unit U, if so, not changing the index point M, and if the group to which the index point of a certain frame to be embedded belongs is 001 and the corresponding embedding unit U is also 001, not changing the index point M. And if the index points are inconsistent, all the index points belonging to the group corresponding to the unit U are obtained, and the nearest projection point of the current index point M is found from the obtained index points to replace the index point M, so that the unit U is embedded into the frame to be embedded corresponding to the index point M.
And step 160, continuously encoding the low-rate compressed voice stream embedded with the secret information to obtain a target compressed voice stream.
Embodiment 2 of the present invention provides a method for extracting information hidden by the above information hiding method, including:
step 201) determining the position of a carrier frame according to a Key;
step 202) obtaining a group to which an index point corresponding to the carrier frame at the position of the carrier frame belongs;
step 203) extracting the secret information according to the grouping to which the index point belongs.
Embodiment 3 of the present invention provides an information hiding system based on linear predictive speech coding, including:
the position acquisition module of the frame to be embedded is used for determining the position of the frame to be embedded in the low-rate compressed voice stream according to the embedding rate, the Key and the random position selection algorithm;
the index point grouping determination module is used for acquiring a frame to be embedded according to the position of the frame to be embedded, encoding the frame to be embedded to obtain an index point M of the frame to be embedded, and determining a group to which the index point M belongs according to an index point grouping result in a pre-established index point set I;
the embedded unit acquisition module is used for reading the secret information according to the preset length to obtain a unit U to be embedded;
the embedding module is used for judging whether the index point M belongs to the group corresponding to the unit U or not, if so, the index point M is not replaced, otherwise, the nearest projection point of the index point M in the group corresponding to the unit U is obtained, the index point M is replaced, and the embedding of the unit U is completed;
and the coding module is used for coding the low-rate compressed voice stream embedded with the secret information to obtain a target compressed voice stream.
Embodiment 4 of the present invention provides a hidden information extraction system, including:
the carrier frame position acquisition module is used for determining the position of the carrier frame according to the Key;
the index point grouping determination module is used for acquiring a grouping to which the index point corresponding to the carrier frame at the position of the carrier frame belongs;
and the secret information extraction module is used for extracting the secret information according to the grouping to which the index point belongs.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and are not limited. Although the present invention has been described in detail with reference to the embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (8)

1. A method for information concealment based on linear predictive speech coding, the method comprising:
determining the position of a frame to be embedded in the low-rate compressed voice stream according to the embedding rate, the Key and a random position selection algorithm;
acquiring a frame to be embedded according to the position of the frame to be embedded, encoding the frame to be embedded to obtain an index point M of the frame to be embedded, and determining a group to which the index point M belongs according to an index point grouping result in a pre-established index point set I;
reading the secret information according to a preset length to obtain a unit U to be embedded;
judging whether the index point M belongs to the group corresponding to the unit U, if so, not replacing the index point M, otherwise, acquiring the nearest projection point of the index point M in the group corresponding to the unit U, replacing the index point M, and completing the embedding of the unit U;
and coding the low-rate compressed voice stream embedded with the secret information to obtain a target compressed voice stream.
2. The method of claim 1, further comprising:
acquiring a corresponding index point set I based on a linear predictive coding LPC analysis filter and a fixed codebook of an encoder;
the quantization index set is C ═ Ci1,2, …, n, where c isiPoint to the corresponding codeword in the ith codebook; if n LPC quantization code books in the encoder are respectively L1,L2…LnThe number of code words in the code book is respectively | L1|,|L2|…|LnIf the index point set I corresponding to the encoder is expressed as:
I＝{(m1,m2,…mn)|m1＝0,1,2,…,|L1|-1；m2＝0,1,2,…,|L2|-1…mn＝0,1,2,…,|Ln|-1}
and grouping the index points in the index point set I based on a quantum particle swarm algorithm.
3. The method according to claim 2, wherein the grouping of the index points in the index point set I based on the quantum-behaved particle swarm optimization specifically comprises:
randomly generating a plurality of digit strings according to the number of the index points in the index point set I and a preset value range; each digital string corresponds to one particle in the quantum particle swarm algorithm, and the value of the number at each position in the digital string represents the grouping result of the index point corresponding to the position;
initializing each particle, and optimizing all the particles obtained after initialization by adopting a quantum particle swarm algorithm to obtain a target grouping digit string.
4. The method according to claim 3, wherein initializing each particle, optimizing all particles obtained after initialization by using a quantum particle swarm optimization algorithm, and obtaining a target grouping digit string specifically comprises:
randomly generating an integer belonging to [1, N ] as a starting position number, wherein N is the length of a digit string;
traversing all the positions in the digital string by taking the position as a starting point, and judging whether K projection points closest to the index point corresponding to the current position are grouped or not when traversing each position;
if n projection points in the K projection points are divided into the same group, n-1 conflict occurs in the grouping, and n is larger than or equal to 2; at the moment, whether 7+ K-1 projection points with the nearest range to be determined are grouped or not is judged; randomly grouping the projection points which are not grouped;
determining the quantum search space dimension of a quantum particle swarm algorithm according to the length of the particles;
determining the population number of a quantum particle swarm algorithm according to the number of all particles obtained after initialization;
determining iteration times, and acquiring the global optimal position of the particles according to the quantum search space dimension, the population particle number and the iteration times;
obtaining a target digit string according to the global optimal position of the particles; the value of the number at each position in the target grouping digit string represents the grouping result of the corresponding index point.
5. The method according to claim 4, wherein the obtaining a nearest projection point of the index point M in the group corresponding to the unit U specifically includes:
calculating q-dimensional quantization residual error coefficient r corresponding to index point MMAccording to rMCalculating the distance E (M, M ') between any index point M' in the index point set I and the index point M;
for collections
Figure FDA0002226493970000021
The index point M satisfies M ∈ I and
Figure FDA0002226493970000022
the set of M projected points in I 'is I'MIf the point M ' satisfies M ' e I 'MAnd for any point M ∈ I'MIf E (M, M ') ≦ E (M, M ″), then M ' is the nearest projection point of M in the set I '.
6. A hidden information extraction method for extracting information hidden by the method of any one of claims 1 to 5, comprising:
determining the position of the carrier frame according to the Key;
acquiring a group to which an index point corresponding to the carrier frame at the position of the carrier frame belongs;
and extracting the secret information according to the grouping to which the index point belongs.
7. An information hiding system based on linear predictive speech coding, the system comprising:
the position acquisition module of the frame to be embedded is used for determining the position of the frame to be embedded in the low-rate compressed voice stream according to the embedding rate, the Key and the random position selection algorithm;
the index point grouping determination module is used for acquiring a frame to be embedded according to the position of the frame to be embedded, encoding the frame to be embedded to obtain an index point M of the frame to be embedded, and determining a group to which the index point M belongs according to an index point grouping result in a pre-established index point set I;
the embedded unit acquisition module is used for reading the secret information according to the preset length to obtain a unit U to be embedded;
the embedding module is used for judging whether the index point M belongs to the group corresponding to the unit U or not, if so, the index point M is not replaced, otherwise, the nearest projection point of the index point M in the group corresponding to the unit U is obtained, the index point M is replaced, and the embedding of the unit U is completed;
and the coding module is used for coding the low-rate compressed voice stream embedded with the secret information to obtain a target compressed voice stream.
8. A hidden information extraction system, characterized in that the system comprises:
the carrier frame position acquisition module is used for determining the position of the carrier frame according to the Key;
the index point grouping determination module is used for acquiring a grouping to which the index point corresponding to the carrier frame at the position of the carrier frame belongs;
and the secret information extraction module is used for extracting the secret information according to the grouping to which the index point belongs.
CN201910953485.9A 2019-10-09 2019-10-09 Information hiding and hidden information extraction method based on linear prediction speech coding Pending CN110689897A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910953485.9A CN110689897A (en) 2019-10-09 2019-10-09 Information hiding and hidden information extraction method based on linear prediction speech coding

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910953485.9A CN110689897A (en) 2019-10-09 2019-10-09 Information hiding and hidden information extraction method based on linear prediction speech coding

Publications (1)

Publication Number Publication Date
CN110689897A true CN110689897A (en) 2020-01-14

Family

ID=69111672

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910953485.9A Pending CN110689897A (en) 2019-10-09 2019-10-09 Information hiding and hidden information extraction method based on linear prediction speech coding

Country Status (1)

Country Link
CN (1) CN110689897A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111291395A (en) * 2020-02-05 2020-06-16 北卡科技有限公司 Information hiding method based on intelligent calculation
CN116825117A (en) * 2023-04-06 2023-09-29 浙江大学 Microphone with privacy protection function and privacy protection method thereof

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104064193A (en) * 2013-03-21 2014-09-24 中国科学院声学研究所 Information hiding method and extraction method for linear prediction voice coding
US20150325246A1 (en) * 2014-05-06 2015-11-12 University Of Macau Reversible audio data hiding
CN107689226A (en) * 2017-08-29 2018-02-13 中国民航大学 A kind of low capacity Methods of Speech Information Hiding based on iLBC codings
CN108877818A (en) * 2018-05-16 2018-11-23 陕西师范大学 A kind of the audio camouflage and restoration methods of combination key and random units orthogonal basis
CN109192217A (en) * 2018-08-06 2019-01-11 中国科学院声学研究所 General information towards multiclass low rate compression voice steganography hides detection method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104064193A (en) * 2013-03-21 2014-09-24 中国科学院声学研究所 Information hiding method and extraction method for linear prediction voice coding
US20150325246A1 (en) * 2014-05-06 2015-11-12 University Of Macau Reversible audio data hiding
CN107689226A (en) * 2017-08-29 2018-02-13 中国民航大学 A kind of low capacity Methods of Speech Information Hiding based on iLBC codings
CN108877818A (en) * 2018-05-16 2018-11-23 陕西师范大学 A kind of the audio camouflage and restoration methods of combination key and random units orthogonal basis
CN109192217A (en) * 2018-08-06 2019-01-11 中国科学院声学研究所 General information towards multiclass low rate compression voice steganography hides detection method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
PENG LIU: "Steganography in vector quantizat ion processof linear predictive coding for low bit rate speech codec", 《MULTIMEDIA SYSTEMS》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111291395A (en) * 2020-02-05 2020-06-16 北卡科技有限公司 Information hiding method based on intelligent calculation
CN111291395B (en) * 2020-02-05 2022-07-19 北卡科技有限公司 Information hiding method based on intelligent calculation
CN116825117A (en) * 2023-04-06 2023-09-29 浙江大学 Microphone with privacy protection function and privacy protection method thereof

Similar Documents

Publication Publication Date Title
Jo et al. A digital image watermarking scheme based on vector quantisation
KR101654777B1 (en) Apparatus and method for scalable encoding 3d mesh, and apparatus and method for scalable decoding 3d mesh
Du et al. Adaptive data hiding based on VQ compressed images
He et al. Reversible data hiding in JPEG images based on negative influence models
CN110689897A (en) Information hiding and hidden information extraction method based on linear prediction speech coding
CN102857831B (en) H.264 video integrality authentication method
CN116939226A (en) Low-code-rate image compression-oriented generated residual error repairing method and device
Pan et al. Novel reversible data hiding scheme for two-stage VQ compressed images based on search-order coding
Zhang et al. Adaptive encoding based lossless data hiding method for VQ compressed images using tabu search
Li et al. Shortening the cover for fast JPEG steganography
Liu et al. Adversarial steganography embedding via stego generation and selection
Fan et al. Multiple histogram based adaptive pairwise prediction-error modification for efficient reversible image watermarking
CN101467459B (en) Generation method of vector quantization dictionary, encoder and decoder, and encoding and decoding method
Li et al. Steganography of steganographic networks
CN112714230A (en) Robust video steganography method and device based on audio side channel
Hong et al. Joint image coding and lossless data hiding in VQ indices using adaptive coding techniques
Zhang et al. Adaptive robust blind watermarking scheme improved by entropy-based SVM and optimized quantum genetic algorithm
Wu Efficient reversible data hiding simultaneously exploiting adjacent pixels
Ren et al. Silk steganography scheme based on the distribution of lsf parameter
CN114900701A (en) Video digital watermark embedding and extracting method and system based on deep learning
Hwang et al. Genetic entropy-constrained vector quantizer design algorithm
CN108615217B (en) Quantization-based JPEG compression resistant robust reversible watermarking method
Kubrick et al. Classified vector quantisation of images: codebook design algorithm
JP3404380B2 (en) Adaptive predictive encoding / decoding method and apparatus, and recording medium storing adaptive predictive encoding / decoding program
Wang et al. Efficient reversible data hiding scheme based on improved SMVQ index for color image

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20200114