CN111464717A - Reversible information hiding framework with contrast pull-up using histogram translation - Google Patents

Reversible information hiding framework with contrast pull-up using histogram translation Download PDF

Info

Publication number
CN111464717A
CN111464717A CN202010243872.6A CN202010243872A CN111464717A CN 111464717 A CN111464717 A CN 111464717A CN 202010243872 A CN202010243872 A CN 202010243872A CN 111464717 A CN111464717 A CN 111464717A
Authority
CN
China
Prior art keywords
information
histogram
image
embedding
gray value
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.)
Granted
Application number
CN202010243872.6A
Other languages
Chinese (zh)
Other versions
CN111464717B (en
Inventor
章劲松
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.)
Individual
Original Assignee
Individual
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 Individual filed Critical Individual
Priority to CN202010243872.6A priority Critical patent/CN111464717B/en
Publication of CN111464717A publication Critical patent/CN111464717A/en
Application granted granted Critical
Publication of CN111464717B publication Critical patent/CN111464717B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
    • H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
    • H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
    • H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
    • H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
    • H04N1/32347—Reversible embedding, i.e. lossless, invertible, erasable, removable or distorsion-free embedding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
    • G06F21/60—Protecting data
    • G06F21/602—Providing cryptographic facilities or services
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/40—Image enhancement or restoration by the use of histogram techniques
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
    • H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
    • H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
    • H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
    • H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
    • H04N1/32288—Multiple embedding, e.g. cocktail embedding, or redundant embedding, e.g. repeating the additional information at a plurality of locations in the image

Abstract

The invention discloses a reversible information hiding frame with contrast ratio pull-up by utilizing histogram translation, designs an improved optimal coding RDH (remote data link) based on a prediction error histogram, and expands a main algorithm into an efficient hiding frame with double embedding capability. Compared with many previous contrast-based pull-up RDH technologies, the framework provided by the invention not only has higher information embedding rate, but also can generate a high-quality confidential image, and in addition, the security test also proves that the framework can effectively resist many famous steganalysis technologies (such as an Ensemble classifier and the like). Moreover, the frame has a dual embedding function, and copyright information such as image identification, serial numbers, characters and the like is embedded into the carrier on the premise of ensuring the original appearance of the digital carrier, so that the purposes of copyright protection, integrity check and the like of the carrier are realized.

Description

Reversible information hiding framework with contrast pull-up using histogram translation
Technical Field
The invention relates to a reversible information hiding frame, in particular to a reversible information hiding frame with contrast ratio pull-up by utilizing histogram translation, and belongs to the technical field of reversible information hiding.
Background
Reversible information hiding (RDH) is proposed as a special information hiding technique. The use of reversible concealment not only allows the extraction of the embedded information from the dense carrier, but also allows the lossless recovery of the original carrier from the dense carrier. Unlike two traditional information hiding technologies, namely digital watermarking and steganography, reversible hiding mainly focuses on the reversibility of an embedding algorithm and a distortion measure between a secret carrier and an original carrier.
Reversible information hiding can be extended to many new fields of applications, such as reversible storage and reversible image processing. Reversible storage differs from traditional storage in that this technique hides the secret document in a natural document, such as a natural image, so that the secret document is imperceptible, these natural carrier documents being referred to as covert storage space. And the application of covert storage naturally requires that information hiding has reversibility, so that the covert storage space can be recycled, and the expansion and maintenance are easy. In addition, most of the current image processing algorithms are irreversible, i.e. the processed image cannot be restored to its original state. The irreversible image processing brings great inconvenience to users, users and enterprises need to realize reversible recovery of the original image by a method of storing the whole original image, so that the cost is almost doubled by storage resources, and the reversible image processing can greatly save the storage space of the original image.
At present, the embedding performance of the RDH algorithm is well improved, but most reversible hiding technology algorithms seek to ensure that the secret-carrying image has the peak signal-to-noise ratio as high as possible. Furthermore, in recent years, academic research has gradually applied RDH technology to more fields, such as designing reversible information hiding (RDH-CE) with contrast pull-up to realize reversible image processing, and efficient reversible information hiding (RDH-EI) of Encrypted Images. The reversible image processing technology researched by the invention is developed mainly for further improving the embedding performance of the RDH algorithm and expanding the application scene of the RDH.
Disclosure of Invention
To overcome the disadvantages of the prior art, an object of the present invention is to provide a reversible information hiding (RDH) framework with contrast pull-up, which can improve the contrast of a carrier image while embedding information, and also has a dual embedding function. The technology mainly acts on gray level images and can be simply expanded to the application of more carriers such as index images, RGB images and the like.
In order to achieve the above object, the present invention adopts the following technical solutions:
a reversible information hiding framework with contrast pull-up using histogram shifting, comprising:
(1) the optimal coding of the histogram constructs a transfer matrix: preprocessing a carrier image, reserving a certain amount of frequency vacancy in a carrier image histogram, and calculating a transfer matrix by using an OPTM algorithm to obtain a transfer matrix T finally used for information embedding;
in the present invention, the process of the OPTM algorithm can be referred to as: zhang W, Chen B, Yu N.Capacity-adaptive codes for reversible data linking [ C ]// International Workshop on information linking.
(2) Recursive information embedding extraction using arithmetic coding: adopting a block embedding method, converting the secret information into a gray value sequence according to the transfer matrix by using arithmetic decoding for each small block, thereby realizing the embedding of the secret information;
(3) dual information embedding of the framework: the first embedding is based on optimal coding of a prediction error histogram, the second embedding is a contrast pull-up RDH algorithm combined with histogram translation, and secret information to be hidden is embedded into a carrier image in a binary code stream mode;
(4) frame information extraction and image restoration: the process is the reverse process of information embedding, wherein the first stage is to embed the intermediate image I after the first round1And the second stage is to restore the original carrier image I.
Preferably, in (1), the specific algorithm process is as follows: given an input histogram of X ═ X0,x1,…,x255And let the final output histogram be Y ═ Y0,y1,…,y255}；
Constructing a transfer matrix T, and expressing the mapping relation between the carrier image and the image pixels after the first embedding:
Figure BDA0002433442340000031
wherein, ti，jRepresenting the number of pixels in the carrier image with a gray value i shifted to j, and in the initial state, T is the diagonal matrix, i.e.:
Figure BDA0002433442340000032
the distortion D caused to the image pixel values when information is embedded is defined as mean square Error distortion (MSE), i.e.:
Figure BDA0002433442340000033
wherein d isi，jRepresenting the distortion degree generated by converting the gray value i into j;
according to the transition matrix T, D, the following formula is calculated:
Figure BDA0002433442340000034
constructing a function L of the Lagrangian multiplier term based on the histogram information entropy according to the information theory knowledge:
L＝H(Y)-H(X)-λD
wherein the content of the first and second substances,
Figure BDA0002433442340000041
Figure BDA0002433442340000042
and: λ is the lagrange multiplier factor. Because H (Y), H (X), D are all ti，jI is more than or equal to 0 and less than or equal to 255, j is more than or equal to 0 and less than or equal to 255, and t is obtained from Li，jAnd making the partial derivative zero, we can obtain:
Figure BDA0002433442340000043
for all ternary combinations (i, j, k) in T, i is greater than or equal to 0 and less than or equal to 255, j is greater than or equal to 0 and less than or equal to 255, and k is greater than or equal to 0 and less than or equal to 255, different Lagrangian multiplier factors can be obtained:
Figure BDA0002433442340000044
using OPTM algorithm, and making each iteration process use the above-mentioned lambdai，j，kFor the minimum to be obtained, the transition matrix is calculated and Δ is used as the iteration termination condition, i.e. when max (λ)i，j，k) The OPTM calculation is terminated when the value is less than delta, and a transfer matrix T finally used for information embedding is obtained.
Preferably, in (2), the specific algorithmic processes are an embedding process and an extraction process, wherein the embedding process is as follows:
dividing the carrier information into X ═ X (X)1，x2，...，xL) The secret information is correspondingly classified as M ═ M (M)1，m2，...，mL) And divides the image into L small blocks correspondingly, the information hider recursively embeds the information into the image starting with the first small block, for small block k, assuming a small block BkOf which there are s different gray values, denoted as { b }k,1,bk,2,…,bk,sTheir pixel numbers are respectively marked as n (b)k,1),n(bk,2),…,n(bk,s)}；
Firstly, the pixels belonging to the same gray value in the small blocks are classified into a sequence,
bk，u＝[bk，u，bk，u，...，bk，u]T，|bk，u|＝n(bk，u)，1≤u≤s
according to the obtained transition matrix, the probability of transitioning an original pixel from value i to value j is:
Figure BDA0002433442340000051
thus, each bk,u(1. ltoreq. u. ltoreq.s) can be determined according to the corresponding
Figure BDA0002433442340000052
Shifting to other values, obtaining b from the shift matrixk,uAll transition probabilities corresponding to gray values
Figure BDA0002433442340000053
And uses arithmetic decoding to decode a section of binary secret information mk，uConverting into a sequence of gray values;
obtaining a secret-containing sequence
Figure BDA0002433442340000054
Figure BDA0002433442340000055
Wherein, Ari _ Dec shorthand arithmetic decoding;
then, corresponding n (b)k，u) Individual gray value of bk,uThe pixels of (a) are replaced according to the result of the arithmetic decoding, thereby completing the information embedding of the small block; then, it is necessary to further generate side information of the small block, so that the receiver can successfully recover the small block, and the side information will be
Figure BDA0002433442340000056
Is restored to bk,uThe small block B after the information is embedded is briefk' there are t different gray values, abbreviated as
Figure BDA0002433442340000057
The numbers of pixels are respectively abbreviated as n (c)k,1),n(ck,2),…,n(ck,t) Are generated by a similar method
Figure BDA0002433442340000058
Sequences for classifying pixels having the same gray value, from the transition matrix, the probability of a dense pixel being transformed from the value i to j can also be obtained
Figure BDA0002433442340000059
And because the embedder knows
Figure BDA00024334423400000510
The original gray value of each pixel in the sequence, abbreviated as ck，uUsing arithmetic coding, c is determined by transition probabilityk，uEncoding into binary sequences
Figure BDA0002433442340000061
Wherein Ari _ Enc corresponds to arithmetic coding;
b is obtained after the hidden information is arithmetically decoded by utilizing the gray values in all Bkk' will be the small block of information received by the receiver, hence, further to Bk' all gray values are arithmetically encoded, generating side information of a binary stream,
Figure BDA0002433442340000062
for the last small block of the intermediate image, it is necessary to compare
Figure BDA0002433442340000063
Directly hidden in the image to ensure that the information extractor can directly obtain the information, and simultaneously, the extractor also needs to obtain AMTo start information extraction, so the extractor needs to directly obtain
Figure BDA0002433442340000064
This part of the information is done by replacing the least significant bits (L SB) of the first tile and correspondingly, the L SB information of the first tile is also needed as part of the side information, included in the information to be embedded, to be embedded in the image in advance.
More preferably, in (2), the extraction process is as follows:
the receiver divides the image into L small blocks in the same blocking mode, and obtains L SB of the first small block
Figure BDA0002433442340000065
Then, using AMConstructing a transfer matrix of an initial state, obtaining the transfer matrix for information embedding by using the same iteration rule as an embedder, and
Figure BDA0002433442340000066
to start block extraction and image restoration for gray values therein
Figure BDA0002433442340000067
The transition probabilities between the various gray values are first obtained from the transition matrix:
P′L，u＝[pL，u(0)，...，pL，u(255)]
then for each
Figure BDA0002433442340000068
Using arithmetic decoding
Figure BDA0002433442340000069
Splitting into n (c)k,u) An integer of:
Figure BDA00024334423400000610
AL，uis dependent on n (c)k,u) And all possible codewords at decoding;
then, the extract will
Figure BDA00024334423400000611
With CL，uThe replacement is performed such that the grey value of the pixel before the information is embedded is obtained, whereby, using arithmetic decoding, BLCan be recovered without loss;
after the current tile is restored, the information extractor classifies the pixels in the tile as b by gray valueL，1，bL，2，...，bL，tFor each b }L，iThe extractor observes the pixels in the dense imageGradation value { bL，1，bL，2，...，bL，tThus all n (b) are coded arithmeticallyL，u) The code words corresponding to the pixels are encoded into embedded binary information:
Figure BDA0002433442340000071
through a pair of BLAfter extraction, the secret information m can be accurately recoveredL＝[mL，1，mL，2，...，mL，2]In the header of the extracted secret information, a hidden one can be obtained
Figure BDA0002433442340000072
By using
Figure BDA0002433442340000073
The same method can be used to iteratively extract the concealment at BL-1The secret information in (1) and so on;
finally, before the recovery of the first small block is executed, the extractor recovers the L SB information of the extracted first small block to the original position, and finally utilizes the information
Figure BDA0002433442340000074
Realize to B1Namely, the secret information M can be accurately recovered to be M1,m2…,mL}。
Further preferably, in (3), the first embedding process is as follows: dividing pixels of the carrier image into two groups which are not overlapped with each other according to a chessboard-shaped format, and setting the groups as Set A and setB; for each pixel p in SetAu，vUsing 4 pixels p adjacent thereto in SetBu+1，v、pu-1，v、pu，v+1、pu，v-1Predicting the data, and setting a prediction parameter used for prediction as w+1，0、w-1，0、w0，+1、w0，-1And initialized to 0.25;
predicted pixel gray scale value is set to
Figure BDA0002433442340000075
Figure BDA0002433442340000076
The prediction error is expressed as:
Figure BDA0002433442340000077
using traditional machine learning algorithm to obtain the optimal prediction coefficient w+1，0、w-1，0、w0，+1、w0，-1So that the mean square sum of the prediction errors is minimal:
Figure BDA0002433442340000078
next, SetA is obtained for each pixel pu，vPrediction error e ofu，vAnd e is combinedu，vThe statistics is in the form of histogram to obtain prediction error histogram G of SetASetA＝{...，g-1，g0，g1，...}
Using the same method, a prediction error histogram G of SetB is further obtainedSet B，
Generating a transfer matrix: for GSet A、GSet BRespectively carrying out simple quantization processing, setting a quantization interval to be r c/2048, carrying out truncation at two ends, discarding predicted values smaller than r c/4096 in the histogram, and obtaining a processed histogram G'Set A、G′Set BRespectively applying the optimal histogram coding transfer matrix generation method to G'Set A、G′Set BTo obtain a transfer matrix TSetA、TSet B；
The first round of embedding process: firstly, T isSetASecret information m1Is embedded in the prediction error of SetA using an iterative algorithm, and then m is added1Is left overPart, and side information from SetA, are combined, again embedded into the prediction error of SetA using an iterative embedding algorithm, and finally SetB generates side information ASetBWith the second part of secret information m2Combining, hiding the image in the second round of embedding, and making the generated dense image I after the embedding process1。
Specifically, the conventional machine learning method is a random gradient descent method.
Still further preferably, in (3), the second embedding process is as follows:
merging histogram frequency: obtaining a dense image I through first embedding1Histogram, let h ═ h0，h1，...，h255H isiThe histogram gray value which is equal to 0 is named as histogram vacancy, M original histogram vacancies in the carrier image are assumed, N histogram vacancies are assumed to be expected after the histogram vacancy reservation algorithm, so the residual N-M vacancies need to be additionally generated, N-M frequency numbers with the minimum pixel number in h are selected, the nearby and non-empty frequency numbers are combined according to the minimum modification degree principle, the corresponding frequency number mapping is carried out on the image, and the combined image is obtained
Figure BDA0002433442340000081
Reserving a histogram vacancy: all h areiThe gray value of 0 (histogram bin) is first removed from the h-sequence, resulting in a new sequence
Figure BDA0002433442340000082
Where K is 256-N, the slot is placed at hseqOn both sides of the element in (1), with ω ═ ω0，ω1，...，ωKThe abbreviation is placed at each hseqThe number of the vacant sites on the two sides of the element is used for carrying out iterative insertion on the reserved histogram vacant sites, the gray value with the largest number of pixels in the histogram is selected in the first round of vacant site insertion, and vacant sites are arranged on the sides of the gray value;
after the iteration is finished, inserting the vacant position into h according to omegaseqIn (1),thereby obtaining a new histogram he. Mapping pixels in the image after histogram merging processing according to the mapping relation corresponding to the gray value to obtain a preprocessed intermediate image, and finally, if the number of the original histogram vacancies is excessive and the number of the required vacancies is excessive, the excessive vacancies are added to two ends of the histogram, and the inserted rule is that the gravity center of the histogram is as close to one half of the maximum gray value as possible, thereby obtaining the image after vacancy reservation
Figure BDA0002433442340000091
Then, side information is generated for image restoration: first, interpolation E before and after histogram mergingmI-I' needs to be saved to distinguish which pixels are merged for the same merged gray value, the interpolation is saved as a binary position pattern and compressed into a binary sequence a using a compression algorithm (e.g., JBIG2 compression technique)E(ii) a In addition, h' and heThe difference also needs to be recorded, since the receiver gets
Figure BDA0002433442340000092
It can be observed how many gray values in the histogram are non-null, so only a binary sequence a of length 256 is neededOTo store heThat is, the way is to identify a gray value by "1" when it is not empty, otherwise, to identify it by "0"; finally, the two parts of information are connected to obtain the additional information A which needs to be embedded togetherC＝[AO，AE]；
Generating a transfer matrix: applying the optimal histogram coding transfer matrix generation method to heTo obtain a transfer matrix T2And using an iterative algorithm to convert m2Is embedded into
Figure BDA0002433442340000093
And the generated side information is hidden to L SB of the first small block of the image, which is directly obtained by the abstractor,the final generated dense image is I2。
Still further preferably, the aforementioned step of iteratively inserting reserved histogram bins is as follows:
first, the gray value with the largest number of pixels in the histogram is selected in the first round of null insertion, and the null is placed on its side. Order to
Figure BDA0002433442340000094
Representing the selected gray value in each round, let ω be in the first roundp，1 Update corresponding e 1p,rIn each subsequent round of fitting room, the potential entropy gain of the selected gray value is
Figure BDA0002433442340000101
And in the 2 nd to N th embedding, e is selected respectivelyp,rMaximum gray value and in ωp，rModified to omegap，r＝ωp，r+1 denotes the null placement and the latest entropy gain e for the current gray value is recalculatedp,r。
Still further preferably, the process of the aforementioned first stage is as follows: information extractors obtain I2Partitioning the image, obtaining side information for image restoration from L SB of the first small block, and extracting information1And the extraction of M hidden therein, which contains secret information M2And side information A for further restoring the original carrier imageSetB。
Still further preferably, in (4), the process of the second stage is as follows: information extractor is realized by1Recovery to I. Information extractors obtain I1Partitioning the image, obtaining side information for image restoration from L SB of the first small block, and extracting information1And the extraction of M hidden therein, which contains secret information M2And side information [ A ] for further restoring the original carrier imageO，AE](by A)SetBObtained), and A isO,AERespectively resolving into position pattern and merged histogram position information, and converting I into1Mapping the non-empty gray value to the position of the merged image, performing corresponding mapping on the pixels in the image to obtain an image I 'after histogram merging, and finally adding the I' and the position pattern to obtain a carrier image without loss recovery, namely I-Em+I′。
The invention develops a reversible information hiding algorithm (namely a main algorithm) with contrast pull-up by utilizing histogram translation, designs an improved RDH based on optimal coding of a prediction error histogram, and expands the main algorithm into an efficient hiding framework with double embedding capability. In the specific implementation process, the gray image is used as an RDH action object, so that the secret information can be reversibly hidden in the carrier image, the contrast of the carrier image is effectively improved, and the visual effect of the carrier image is improved. The secret information to be hidden is embedded into the carrier image in the form of a binary code stream, thus supporting any format document as a hidden object. A reversible information hiding method with contrast pull-up effect (RDHCE-HS) combined with histogram transfer is proposed, which allows for a double embedding of secret information, i.e. the information hiding can be divided into two stages. Firstly, the carrier image is preprocessed by the algorithm, a certain amount of frequency vacancy is reserved in a carrier image histogram, and the histogram vacancy can well help an information embedder to promote the entropy of the image in the subsequent process and ensure that the introduced mean square error is as small as possible. And a transfer matrix generation algorithm is designed, the entropy of the image histogram is expanded as much as possible, and the transfer probability among the frequency numbers is calculated. And adopting a block embedding method, and converting the secret information into a gray value sequence according to the transfer matrix by using arithmetic decoding for each small block so as to realize secret information embedding. In order to realize lossless recovery and information extraction, arithmetic coding is further adopted, and the original gray value corresponding to each frequency number containing the dense small blocks is transferred into binary side information from the arithmetic coding. The method is further combined with a conventional RDH method based on prediction error histogram transfer. Experiments prove that compared with many previous RDH technologies based on contrast ratio lifting, the framework provided by the invention not only has higher information embedding rate, but also can generate a confidential image with higher quality, and in addition, security tests also prove that the framework provided by the invention can effectively resist many famous steganalysis technologies (such as an Ensemble classifier and the like).
In particular, in a main algorithm, a "vacancy reservation method" is designed to preprocess a carrier image, the algorithm reserves some vacancies in a histogram by a histogram combination method, thereby generating an intermediate image (intermediate image), and then, a transition matrix generation algorithm is proposed, wherein the histogram entropy of the image is increased by using a Mean Squared Error (MSE) as a constraint term, so as to obtain transition probabilities between histograms, then, the intermediate image is divided into blocks according to the obtained transition probabilities, and secret information is iteratively embedded into the blocks.
Has the advantages that: the algorithm main part of the invention has reversibility, and the information extraction and image recovery process is the inverse process of the above process. As an extension, an improved prediction error histogram-based RDH-PEH method is proposed, so that the method is more suitable for the application scenario. Therefore, the information Embedding of the carrier image has Dual (Dual Embedding), one round of Embedding is carried out through RDHOVT, and the second round of Embedding is carried out through the RDHCE-HS main body part, so that the Embedding rate of most images is higher than that of some RDHCE methods which show the highest level. The contrast of the carrier image can be improved while information is embedded effectively. Compared with the traditional reversible information hiding method, the framework of the invention has double embedding functions, and copyright information such as image identification, serial numbers, characters and the like is embedded into the carrier on the premise of ensuring the original appearance of the digital carrier, so as to realize the purposes of copyright protection, integrity check and the like of the carrier.
Drawings
FIG. 1 is a body algorithm and overall framework schematic diagram of the reversible information hiding framework with contrast pull-up using histogram shifting of the present invention; wherein, fig. 1(a) is a schematic diagram of a main algorithm, fig. 1(b) is a schematic diagram of an overall framework, HS represents histogram translation, SI represents auxiliary information, and AD represents additional data;
FIG. 2 is a flow chart of a recursive information embedding process using arithmetic coding;
FIGS. 3(a) and 3(b) are schematic diagrams of a prediction error grid classification structure diagram and a prediction error histogram, respectively;
FIG. 4 is an information extraction and image restoration flow diagram; fig. 4(a) is a flowchart of the first stage recovery and information extraction, and fig. 4(b) is a flowchart of the second stage recovery and information extraction.
Detailed Description
The invention is described in detail below with reference to the figures and the embodiments.
In this embodiment, a grayscale image with 8-bit precision is taken as a carrier, and the length and width of the carrier image are r and c, respectively. Secret information m is embedded in an image in the form of a binary code stream and is divided into two parts for dual embedding, m ═ m { (m })1，m2}。
Referring to fig. 1, the technique of the present invention includes several aspects as follows: the histogram optimal coding transfer matrix calculation technology, the recursive information embedding and extraction by using arithmetic coding and decoding, the dual information embedding process of the frame, the frame information extraction and the image restoration process are described one by combining the following drawings:
first, histogram optimum coding transfer matrix calculation technology
Given an input histogram of X ═ X0,x1,…,x255And let the final output histogram be Y ═ Y0,y1,…,y255}. Constructing a transfer matrix T, and expressing the mapping relation between the carrier image and the image pixels after the first embedding:
Figure BDA0002433442340000131
wherein, ti，jRepresenting the number of pixels in the carrier image with a gray value i shifted to j, and in the initial state, T is the diagonal matrix, i.e.:
Figure BDA0002433442340000132
the distortion D caused to the image pixel values when information is embedded is defined as mean square Error distortion (MSE), i.e.:
Figure BDA0002433442340000133
then, according to the above transition matrix T, D, it is calculated by the following formula:
Figure BDA0002433442340000134
constructing a function L of the Lagrangian multiplier term based on the histogram information entropy according to the information theory knowledge:
L＝H(Y)-H(X)-λD
wherein the content of the first and second substances,
Figure BDA0002433442340000135
Figure BDA0002433442340000136
and: λ is the lagrange multiplier factor. Because H (Y), H (X), D are all ti，jI is more than or equal to 0 and less than or equal to 255, j is more than or equal to 0 and less than or equal to 255, and t is obtained from Li，jAnd making the partial derivative zero, we can obtain:
Figure BDA0002433442340000141
for all ternary combinations (i, j, k) in T, i is greater than or equal to 0 and less than or equal to 255, j is greater than or equal to 0 and less than or equal to 255, and k is greater than or equal to 0 and less than or equal to 255, different Lagrangian multiplier factors can be obtained:
Figure BDA0002433442340000142
using OPTM algorithm, and making each iteration process use the above-mentioned lambdai，j，kFor minimum values to be obtained, countingCalculating a transition matrix and using delta as an iteration termination condition, namely when max (lambda)i，j，k) The OPTM calculation is terminated when the value is less than delta, and a transfer matrix T finally used for information embedding is obtained.
Recursive information embedding extraction using arithmetic coding and decoding
A flowchart of a recursive information embedding process using arithmetic coding is shown in fig. 2. Given an input transfer matrix T, the carrier information is first divided into X ═ X (X)1，x2，...，xL) The secret information is correspondingly classified as M ═ M (M)1，m2，...，mL) The information concealer recursively embeds the information into the image starting with the first small block, for small block k, assuming a small block BkOf which there are s different gray values, denoted as { b }k,1,bk,2,…,bk,sTheir pixel numbers are respectively marked as n (b)k,1),n(bk,2),…,n(bk,s)}。
Firstly, the pixels belonging to the same gray value in the small blocks are classified into a sequence,
bk，u＝[bk，u，bk，u，...，bk，u]T，|bk，u|＝n(bk，u)，1≤u≤s。
according to the obtained transition matrix, the probability of transitioning an original pixel from value i to value j is:
Figure BDA0002433442340000143
thus, each bk,u(1. ltoreq. u. ltoreq.s) can be determined according to the corresponding
Figure BDA0002433442340000151
Branch to other values, thus, b is derived from the branch matrixk,uAll transition probabilities corresponding to gray values
Figure BDA0002433442340000152
And using arithmetic decoding to decode a segment of binary secret informationmk，uConverted into a sequence of gray values. The gray-scale value with low transition probability has low occurrence probability, so that the gray-scale value corresponds to a long secret information.
Thereby obtaining a secret-containing sequence
Figure BDA0002433442340000153
Figure BDA0002433442340000154
Wherein Ari _ Dec is abbreviated arithmetic decoding.
Then, corresponding n (b)k，u) Individual gray value of bk,uThe pixels of (a) are replaced according to the result of the arithmetic decoding, thereby completing the information embedding of the small block. Then, it is necessary to further generate side information of the small block so that the receiver can successfully recover the small block. The side information will
Figure BDA0002433442340000155
Is restored to bk,u. Small block B after information embeddingk' there are t different gray values, abbreviated as
Figure BDA0002433442340000156
The numbers of pixels are respectively abbreviated as n (c)k,1),n(ck,2),…,n(ck,t) Are generated by a similar method
Figure BDA0002433442340000157
And the sequence is used for classifying the pixels with the same gray value. From the transition matrix, the probability of transitioning a dense pixel from value i to j can also be obtained as
Figure BDA0002433442340000158
And because the embedder knows
Figure BDA0002433442340000159
The original gray value of each pixel in the sequence, abbreviated as ck，uThus, arithmetic coding can be usedCode, by transition probability ck，uEncoded into a binary sequence.
Figure BDA00024334423400001510
Where Ari _ Enc corresponds to arithmetic coding. B obtained by arithmetically decoding the gray values in all Bk to hide informationk' will be the small block of information received by the receiver, hence, further to Bk' all gray values are arithmetically encoded, generating side information of a binary stream,
Figure BDA0002433442340000161
for the last small block of the intermediate image, it is necessary to compare
Figure BDA0002433442340000162
Directly hidden in the image to ensure that the information extractor can directly obtain the information, and simultaneously, the extractor also needs to obtain AMTo start information extraction, so the extractor needs to directly obtain
Figure BDA0002433442340000163
This part of the information is done by replacing the least significant bits (L SB) of the first tile and correspondingly, the L SB information of the first tile is also needed as part of the side information, included in the information to be embedded, to be embedded in the image in advance.
The recipient divides the image into L patches in the same blocking manner and obtains L SB from the first patch
Figure BDA0002433442340000164
Then he utilizes AMAnd constructing a transition matrix of an initial state, and obtaining the transition matrix for information embedding by using the same iteration rule as that of an embedder. He then takes again
Figure BDA0002433442340000165
To turn on block fetchingWith image restoration, for gray values therein
Figure BDA0002433442340000166
He first obtains the transition probabilities between the various gray values from the transition matrix:
P′L，u＝[pL，u(0)，...，pL，u(255)]
then for each
Figure BDA0002433442340000167
Using arithmetic decoding
Figure BDA0002433442340000168
Splitting into n (c)k,u) An integer of:
Figure BDA0002433442340000169
AL，uis dependent on n (c)k,u) And all possible codewords at decoding time. Then, the extract will
Figure BDA00024334423400001610
With CL，uThe replacement is performed such that the grey value of the pixel before the information is embedded is obtained, whereby, using arithmetic decoding, BLCan be recovered without loss.
After the current tile is restored, the information extractor classifies the pixels in the tile as b by gray valueL，1，bL，2，...，bL，tFor each b }L，iThe extractor observes the gray values of these pixels in the dense image { b }L，1，bL，2，...，bL，tThus all n (b) are coded arithmeticallyL，u) The code words corresponding to the pixels are encoded as embedded binary information.
Figure BDA00024334423400001611
Through a pair of BLAfter extraction, can be recovered accuratelyReissue secret information, mL＝[mL，1，mL，2，...，mL，2]. In the header of the extracted secret information, the hidden one can be obtained
Figure BDA0002433442340000171
By using
Figure BDA0002433442340000172
The same method can be used to iteratively extract the concealment at BL-1And so on.
Finally, before the recovery of the first small block is executed, the extractor recovers the L SB information of the extracted first small block to the original position, and finally utilizes the information
Figure BDA0002433442340000173
Realize to B1The recovery of (1). By the method, secret information M can be accurately recovered by extracting1,m2…,mL}。
Triple, frame dual information embedding process
The framework provided by the invention has double information embedding functions, and an information extractor can restore an original image without damage and extract secret information.
1. Embedding for the first time: optimal coding technique based on prediction error histogram
1) Calculating a prediction error histogram for the carrier image according to the following rule:
the pixels of the carrier image are divided into two groups that do not overlap each other in a checkerboard-like format shown in fig. 2, and Set a and SetB are Set. For each pixel p in SetAu，vUsing 4 pixels p adjacent thereto in SetBu+1，v、pu-1，v、pu，v+1、pu，v-1Predicting the data, and setting a prediction parameter used for prediction as w+1，0、w-1，0、w0，+1、w0，-1And is initialized to 0.25.
Predicted pixel gray scale value is set to
Figure BDA0002433442340000174
Figure BDA0002433442340000175
The prediction error is expressed as
Figure BDA0002433442340000176
The optimal prediction coefficient w is found using a conventional machine learning algorithm, here using a stochastic gradient descent (sgd)+1，0、w-1，0、w0，+1、w0，-1So that the mean square sum of the prediction errors is minimal:
Figure BDA0002433442340000181
next, SetA is obtained for each pixel pu，vPrediction error e ofu，vAnd e is combinedu，vThe statistics is in the form of histogram to obtain prediction error histogram G of SetASetA＝{...，g-1，g0，g1，...}。
Using the same method, a prediction error histogram G of SetB can be further obtainedSet BSee fig. 3.
2) Generating a transfer matrix: for GSetA、GSetBRespectively carrying out simple quantization processing, setting a quantization interval to be r c/2048, carrying out truncation at two ends, discarding predicted values smaller than r c/4096 in the histogram, and obtaining a processed histogram G'Set A、G′Set B. Respectively applying the optimal histogram coding transfer matrix generation method to G'Set A、G′Set BTo obtain a transfer matrix TSetA、TSet B。
3) The first round of embedding process: firstly, T isSetASecret information m1Is embedded in the prediction of SetA using an iterative algorithmIn error, then m is1The remainder of (d), and the side information combination from SetA, are again embedded into the prediction error of SetA using an iterative embedding algorithm. Final SetB generated side information aSetBWith the second part of secret information m2The combination is made to hide the image in the second round of embedding. After the embedding process, the generated dense image is made to be I1。
2. And (3) embedding for the second time: contrast pull-up RDH algorithm (i.e., subject algorithm in FIG. 1) combined with histogram translation
1) Merging histogram frequency: obtaining a dense image I through first embedding1Histogram, let h ═ h0，h1，...，h255H isiThe histogram gray value which is equal to 0 is named as histogram vacancy, M original histogram vacancies in the carrier image are assumed, N histogram vacancies are assumed to be expected after the histogram vacancy reservation algorithm, so the residual N-M vacancies need to be additionally generated, N-M frequency numbers with the minimum pixel number in h are selected, the nearby and non-empty frequency numbers are combined according to the minimum modification degree principle, the corresponding frequency number mapping is carried out on the image, and the combined image is obtained
Figure BDA0002433442340000182
2) Reserving a histogram vacancy: all h areiThe gray value of 0 (histogram bin) is first removed from the h-sequence, resulting in a new sequence
Figure BDA0002433442340000191
Wherein K is 256-N. Placing a vacancy at hseqOn both sides of the element in (1), and therefore with ω ═ ω0，ω1，...，ωKThe abbreviation is placed at each hseqNumber of vacancies on both sides of the element. And performing iterative insertion on the reserved histogram blank, selecting the gray value with the maximum number of pixels in the histogram in the first round of blank insertion, and arranging the blank on the side of the gray value.
As known from the information theory knowledge, if a gray value (containing g)iIndividual pixels) are mapped to (1+ ω) by information hidingi) Different gray values, the gray values together can then represent g0log2(1+ωi) Bit information, therefore, the histogram entropy gain can be expressed as g0log2(1+ωi)-giThis is also the bit capacity that this grey value can theoretically be hidden from this. With e ═ e0，e1，...，eKRepresents the initial state of the quantity that could theoretically contribute to the histogram entropy gain after each insertion of a null,
Figure BDA0002433442340000192
the step of iteratively inserting the reserved histogram slots is as follows:
first, the gray value with the largest number of pixels in the histogram is selected in the first round of null insertion, and the null is placed on its side. Order to
Figure BDA0002433442340000193
Representing the selected gray value in each round, let ω be in the first roundp，1 Update corresponding e 1p,rIn each subsequent round of fitting room, the potential entropy gain of the selected gray value is
Figure BDA0002433442340000194
And in the 2 nd to N th embedding, e is selected respectivelyp,rMaximum gray value and in ωp，rModified to omegap，r＝ωp，r+1 denotes the null placement and the latest entropy gain e for the current gray value is recalculatedp,r。
In this embodiment, the upper limit ω of the number of vacancies is setb5. After the iteration is finished, inserting the vacant position into h according to omegaseqThus obtaining a new histogram he. And mapping the pixels in the image after the histogram merging processing according to the mapping relation corresponding to the gray value, so as to obtain a preprocessed intermediate image. Finally, if it isThe number of starting histogram bins is more than the number of bins needed, then more than one bin will be added at both ends of the histogram, with the rule of interpolation being such that the center of gravity of the histogram is as close as possible to one-half of the maximum value of the gray scale. Thus obtaining an image with reserved vacant positions
Figure BDA0002433442340000201
3) Generating side information for image restoration: first, interpolation E before and after histogram mergingmI-I' needs to be saved to distinguish which pixels are merged for the same merged gray value, the interpolation is saved as a binary position pattern and compressed into a binary sequence a using a compression algorithm (e.g., JBIG2 compression technique)E. In addition, h' and heThe difference also needs to be recorded, since the receiver gets
Figure BDA0002433442340000202
It can be observed how many gray values in the histogram are non-null, so only a binary sequence a of length 256 is neededOTo store heThat is, by identifying a gray value as non-empty by "1" or otherwise by "0". Finally, the two parts of information are connected to obtain the additional information A which needs to be embedded togetherC＝[AO，AE]。
4) Generating a transfer matrix: applying the optimal histogram coding transfer matrix generation method to heTo obtain a transfer matrix T2And using an iterative algorithm to convert m2Is embedded into
Figure BDA0002433442340000203
Hiding the generated side information to L SB of the first small block of the image, and obtaining the side information directly by the extracter2。
Fourthly, frame information extraction and image recovery process
The information extraction process of the framework is information embeddingAn inverse process, see fig. 4, can be specifically divided into two stages: the first stage is to embed the intermediate image I after the first round1And the second stage is to restore the original carrier image I.
1) And (3) information extraction and image restoration corresponding to the second round of information embedding: information extractors obtain I2Partitioning the image, obtaining side information for image restoration from L SB of the first small block, and extracting information1And the extraction of M hidden therein, which contains secret information M2And side information A for further restoring the original carrier imageSetB。
2) Information extraction and image restoration corresponding to the first round of information embedding: information extractor is realized by1Recovery to I. Information extractors obtain I1Partitioning the image, obtaining side information for image restoration from L SB of the first small block, and extracting information1And the extraction of M hidden therein, which contains secret information M2And side information [ A ] for further restoring the original carrier imageO，AE](by A)SetBObtained), and A isO,AEAnd respectively resolving into a position pattern and merged histogram position information. Will I1Mapping the non-empty gray value to the position of the merged image, performing corresponding mapping on the pixels in the image to obtain an image I 'after histogram merging, and finally adding the I' and the position pattern to obtain a carrier image without loss recovery, namely I-Em+I′。
In summary, the frame with contrast pull-up invertible information hiding and histogram translation of the invention firstly preprocesses the carrier image, reserves a certain amount of frequency vacancy in the carrier image histogram, and the histogram vacancy can well help the information embedder to ensure that the introduced mean square error is as small as possible while improving the image entropy in the subsequent process. And a transfer matrix generation algorithm is designed, the entropy of the image histogram is expanded as much as possible, and the transfer probability among the frequency numbers is calculated. And then, adopting a block embedding method, and converting the secret information into a gray value sequence according to the transfer matrix by using arithmetic decoding for each small block so as to realize secret information embedding. And finally, in order to realize lossless recovery and information extraction, further adopting arithmetic coding to transfer the original gray value corresponding to each frequency number containing the dense small blocks into binary side information from the arithmetic coding. The method is further combined with a traditional RDH method based on prediction error histogram transfer, compared with a plurality of previous RDH technologies based on contrast ratio pull-up, the method not only has higher information embedding rate, but also can generate a confidential image with higher quality, and in addition, the security test also proves that the method can effectively resist a plurality of famous steganalysis technologies (such as an Ensemble classifier and the like).
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It should be understood by those skilled in the art that the above embodiments do not limit the present invention in any way, and all technical solutions obtained by using equivalent alternatives or equivalent variations fall within the scope of the present invention.

Claims (10)

1. A reversible information hiding framework with contrast pull-up using histogram shifting, comprising:
(1) the optimal coding of the histogram constructs a transfer matrix: preprocessing a carrier image, reserving a certain amount of frequency vacancy in a carrier image histogram, and calculating a transfer matrix by using an OPTM algorithm to obtain a transfer matrix T finally used for information embedding;
(2) recursive information embedding extraction using arithmetic coding: adopting a block embedding method, converting the secret information into a gray value sequence according to the transfer matrix by using arithmetic decoding for each small block, thereby realizing the embedding of the secret information;
(3) dual information embedding of the framework: the first embedding is based on optimal coding of a prediction error histogram, the second embedding is a contrast pull-up RDH algorithm combined with histogram translation, and secret information to be hidden is embedded into a carrier image in a binary code stream mode;
(4) frame information elevatorTaking and restoring images: the process is the reverse process of information embedding, wherein the first stage is to embed the intermediate image I after the first round1And the second stage is to restore the original carrier image I.
2. The reversible information hiding framework with contrast pulling using histogram shifting as claimed in claim 1, wherein in (1), the specific algorithm procedure is: given an input histogram of X ═ X0，x1，...，x255And let the final output histogram be Y ═ Y0，y1，...，y255}；
Constructing a transfer matrix T, and expressing the mapping relation between the carrier image and the image pixels after the first embedding:
Figure FDA0002433442330000011
wherein, ti，jRepresenting the number of pixels in the carrier image with a gray value i shifted to j, and in the initial state, T is the diagonal matrix, i.e.:
Figure FDA0002433442330000012
the distortion D caused to the image pixel values when information is embedded is defined as Mean square error distortion (MSE), i.e.:
Figure FDA0002433442330000021
wherein d isi，jIndicating the degree of distortion resulting from the gray value i converted to j,
according to the transition matrix T, D, the following formula is calculated:
Figure FDA0002433442330000022
constructing a function L of the Lagrangian multiplier term based on the histogram information entropy according to the information theory knowledge:
L＝H(Y)-H(X)-λD
wherein the content of the first and second substances,
Figure FDA0002433442330000023
Figure FDA0002433442330000024
and: λ is the lagrange multiplier factor. Because H (Y), H (X), D are all ti，jI is more than or equal to 0 and less than or equal to 255, j is more than or equal to 0 and less than or equal to 255, and t is obtained from Li，jAnd making the partial derivative zero, we can obtain:
Figure FDA0002433442330000025
for all ternary combinations (i, j, k) in T, i is greater than or equal to 0 and less than or equal to 255, j is greater than or equal to 0 and less than or equal to 255, and k is greater than or equal to 0 and less than or equal to 255, different Lagrangian multiplier factors can be obtained:
Figure FDA0002433442330000026
using OPTM algorithm, and making each iteration process use the above-mentioned lambdai，j，kFor the minimum to be obtained, the transition matrix is calculated and Δ is used as the iteration termination condition, i.e. when max (λ)i，j，k) The OPTM calculation is terminated when the value is less than delta, and a transfer matrix T finally used for information embedding is obtained.
3. The reversible information hiding framework with contrast pulling using histogram shifting as claimed in claim 1, wherein in (2), the specific algorithm process is embedding process and extracting process, wherein the embedding process is as follows:
dividing the carrier information into X ═ X (X)1，x2，...，xL) The secret information is correspondingly classified as M ═ M (M)1，m2，...，mL) And divides the image into L small blocks correspondingly, the information hider recursively embeds the information into the image starting with the first small block, for small block k, assuming a small block BkOf which there are s different gray values, denoted as { b }k，1，bk，2，...，bk，sTheir pixel numbers are respectively marked as n (b)k，1)，n(bk，2)，...，n(bk，s)}；
Firstly, the pixels belonging to the same gray value in the small blocks are classified into a sequence,
bk，u＝[bk，u，bk，u，...，bk，u]T，|bk，u|＝n(bk，u)，1≤u≤s
according to the obtained transition matrix, the probability of transitioning an original pixel from value i to value j is:
Figure FDA0002433442330000031
thus, each bk，u(1. ltoreq. u. ltoreq.s) can be determined according to the corresponding
Figure FDA0002433442330000032
j is 0, 255, and the other values are transferred, and b is obtained according to the transfer matrixk，uAll transition probabilities corresponding to gray values
Figure FDA0002433442330000033
And uses arithmetic decoding to decode a section of binary secret information mk，uConverting into a sequence of gray values;
obtaining a secret-containing sequence
Figure FDA0002433442330000034
Figure FDA0002433442330000035
Wherein, Ari _ Dec shorthand arithmetic decoding;
then, corresponding n (b)k，u) Individual gray value of bk，uThe pixels of (a) are replaced according to the result of the arithmetic decoding, thereby completing the information embedding of the small block; then, it is necessary to further generate side information of the small block, so that the receiver can successfully recover the small block, and the side information will be
Figure FDA0002433442330000041
Is restored to bk，uThe small block B after the information is embedded is briefk' there are t different gray values, abbreviated as
Figure FDA0002433442330000042
The numbers of pixels are respectively abbreviated as n (c)k，1)，n(ck，2)，...，n(ck，t) Are generated by a similar method
Figure FDA0002433442330000043
Sequences for classifying pixels having the same gray value, from the transition matrix, the probability of a dense pixel being transformed from the value i to j can also be obtained
Figure FDA0002433442330000044
And because the embedder knows
Figure FDA0002433442330000045
The original gray value of each pixel in the sequence, abbreviated as Ck，uUsing arithmetic coding, c is determined by transition probabilityk，uEncoding into binary sequences
Figure FDA0002433442330000046
Wherein Ari _ Enc corresponds to arithmetic coding;
b obtained by arithmetically decoding the gray values in all Bk to hide informationkWill be received by the recipientSmall pieces of information, therefore, further for Bk' all gray values are arithmetically encoded, generating side information of a binary stream,
Figure FDA0002433442330000047
for the last small block of the intermediate image, it is necessary to compare
Figure FDA0002433442330000048
Directly hidden in the image to ensure that the information extractor can directly obtain the information, and simultaneously, the extractor also needs to obtain AMTo start information extraction, so the extractor needs to directly obtain
Figure FDA0002433442330000049
This part of the information is done by replacing the least significant bits (L SB) of the first tile and correspondingly, the L SB information of the first tile is also needed as part of the side information, included in the information to be embedded, to be embedded in the image in advance.
4. The reversible information hiding framework with contrast pulling using histogram shifting as claimed in claim 3, wherein in (2), the extraction process is as follows:
the receiver divides the image into L small blocks in the same blocking mode, and obtains L SB of the first small block
Figure FDA0002433442330000051
Then, using AMConstructing a transfer matrix of an initial state, obtaining the transfer matrix for information embedding by using the same iteration rule as an embedder, and
Figure FDA0002433442330000052
to start block extraction and image restoration for gray values therein
Figure FDA0002433442330000053
The transition probabilities between the various gray values are first obtained from the transition matrix:
P′L，u＝[pL，u(0)，...，pL，u(255)]
then for each
Figure FDA0002433442330000054
Using arithmetic decoding
Figure FDA0002433442330000055
Splitting into n (c)k，u) An integer of:
Figure FDA0002433442330000056
AL，uis dependent on n (c)k，u) And all possible codewords at decoding;
then, the extract will
Figure FDA0002433442330000057
With CL，uThe replacement is performed such that the grey value of the pixel before the information is embedded is obtained, whereby, using arithmetic decoding, BLCan be recovered without loss;
after the current tile is restored, the information extractor classifies the pixels in the tile as b by gray valueL，1，bL，2，...，bL，tFor each b }L，iThe extractor observes the gray values of these pixels in the dense image { b }L，1，bL，2，...，bL，tThus all n (b) are coded arithmeticallyL，u) The code words corresponding to the pixels are encoded into embedded binary information:
Figure FDA0002433442330000058
through a pair of BLAfter extraction, the secret information m can be accurately recoveredL＝[mL，1，mL，2，...，mL，2]In the header of the extracted secret information, a hidden one can be obtained
Figure FDA0002433442330000059
By using
Figure FDA00024334423300000510
The same method can be used to iteratively extract the concealment at BL-1The secret information in (1) and so on;
finally, before the recovery of the first small block is executed, the extractor recovers the L SB information of the extracted first small block to the original position, and finally utilizes the information
Figure FDA00024334423300000511
Realize to B1Namely, the secret information M can be accurately recovered to be M1，m2...，mL}。
5. The reversible information hiding framework with contrast pull-up using histogram shifting as claimed in claim 1, wherein in (3), the first embedding process is as follows: dividing pixels of the carrier image into two groups which are not overlapped with each other according to a chessboard-shaped format, and setting the groups as SetA and SetB; for each pixel p in SetAu，vUsing 4 pixels p adjacent thereto in SetBu+1，v、pu-1，v、pu，v+1、pu，v-1Predicting the data, and setting a prediction parameter used for prediction as w+1，0、w-1，0、w0，+1、w0，-1And initialized to 0.25;
predicted pixel gray scale value is set to
Figure FDA0002433442330000061
Figure FDA0002433442330000062
The prediction error is expressed as:
Figure FDA0002433442330000063
using traditional machine learning algorithm to obtain the optimal prediction coefficient w+1，0、w-1，0、w0，+1、w0，-1So that the mean square sum of the prediction errors is minimal:
Figure FDA0002433442330000064
next, SetA is obtained for each pixel pu，vPrediction error e ofu，vAnd e is combinedu，vThe statistics is in the form of histogram to obtain prediction error histogram G of SetASeta＝{...，g-1，g0，g1，...}
Using the same method, a prediction error histogram G of SetB is further obtainedSet B，
Generating a transfer matrix: for GSetA、GSet BRespectively carrying out simple quantization processing, setting a quantization interval to be r c/2048, carrying out truncation at two ends, discarding predicted values smaller than r c/4096 in the histogram, and obtaining a processed histogram G'Set A、G′Set BRespectively applying the optimal histogram coding transfer matrix generation method to G'Set A、G′Set BTo obtain a transfer matrix TSetA、Tset B；
The first round of embedding process: firstly, T isSetASecret information m1Is embedded in the prediction error of SetA using an iterative algorithm, and then m is added1And the side information from SetA, again using an iterative embedding algorithm to embed into the prediction error of SetA, the resulting force information a from SetBSetBWith the second part of secret information m2Combining, hiding the image in the second round of embedding, and making the generated dense image I after the embedding process1。
6. The reversible information hiding framework with contrast pulling-up using histogram shifting according to claim 5, characterized in that said traditional machine learning method is a stochastic gradient descent method.
7. The reversible information hiding framework with contrast pull-up using histogram shifting as claimed in claim 1, wherein in (3), the second embedding process is as follows:
merging histogram frequency: obtaining a dense image I through first embedding1Histogram, let h ═ h0，h1，...，h255H isiThe histogram gray value which is equal to 0 is named as histogram vacancy, M original histogram vacancies in the carrier image are assumed, N histogram vacancies are assumed to be expected after the histogram vacancy reservation algorithm, so the residual N-M vacancies need to be additionally generated, N-M frequency numbers with the minimum pixel number in h are selected, the nearby and non-empty frequency numbers are combined according to the minimum modification degree principle, the corresponding frequency number mapping is carried out on the image, and the combined image is obtained
Figure FDA0002433442330000071
Reserving a histogram vacancy: all h areiThe gray value of 0 (histogram bin) is first removed from the h-sequence, resulting in a new sequence
Figure FDA0002433442330000072
Where K is 256-N, the slot is placed at hseqOn both sides of the element in (1), with ω ═ ω0，ω1，...，ωKThe abbreviation is placed at each hseqThe number of the vacant sites on the two sides of the element is used for carrying out iterative insertion on the reserved histogram vacant sites, the gray value with the largest number of pixels in the histogram is selected in the first round of vacant site insertion, and vacant sites are arranged on the sides of the gray value;
after the iteration is finished, inserting the vacant position into h according to omegaseqThus obtaining a new histogram he. Combining the histogramsAnd mapping the pixels in the processed image according to the mapping relation corresponding to the gray value to obtain a preprocessed intermediate image, and finally, if the number of the original histogram vacancies is excessive and the number of the required vacancies is excessive, the excessive vacancies are added to the two ends of the histogram, and the rule of insertion is to make the gravity center of the histogram as close to one half of the maximum value of the gray value as possible, thereby obtaining the image with reserved vacancies
Figure FDA0002433442330000073
Then, side information is generated for image restoration: first, interpolation E before and after histogram mergingmI-I' needs to be saved to distinguish which pixels are merged for the same merged gray value, the interpolation is saved as a binary position pattern and compressed into a binary sequence a using a compression algorithm (e.g., JBIG2 compression technique)E(ii) a In addition, h' and heThe difference also needs to be recorded, since the receiver gets
Figure FDA0002433442330000081
It can be observed how many gray values in the histogram are non-null, so only a binary sequence a of length 256 is neededOTo store heThat is, the way is to identify a gray value by "1" when it is not empty, otherwise, to identify it by "0"; finally, the two parts of information are connected to obtain the additional information A which needs to be embedded togetherC＝[AO，AE]；
Generating a transfer matrix: applying the optimal histogram coding transfer matrix generation method to heTo obtain a transfer matrix T2And using an iterative algorithm to convert m2Is embedded into
Figure FDA0002433442330000082
Hiding the generated side information to L SB of the first small block of the image, directly obtaining the extracted side information, and finally generating the dense side informationThe image is I2。
8. The reversible information hiding frame with contrast pull-up using histogram shifting as claimed in claim 7 wherein said step of iteratively inserting reserved histogram bins is as follows:
first, the gray value with the largest number of pixels in the histogram is selected in the first round of null insertion, and the null is placed on its side. Order to
Figure FDA0002433442330000083
Representing the selected gray value in each round, let ω be in the first roundp，1Update corresponding e 1p，rIn each subsequent round of fitting room, the potential entropy gain of the selected gray value is
Figure FDA0002433442330000084
And in the 2 nd to N th embedding, e is selected respectivelyp，rMaximum gray value and in ωp，rModified to omegap，r＝ωp，r+1 denotes the null placement and the latest entropy gain e for the current gray value is recalculatedp，r。
9. The reversible information hiding framework with contrast pull-up using histogram shifting according to claim 1, wherein in (4), the procedure of the first stage is as follows: information extractors obtain I2Partitioning the image, obtaining side information for image restoration from L SB of the first small block, and extracting information1And the extraction of M hidden therein, which contains secret information M2And side information A for further restoring the original carrier imageSetB。
10. The reversible information hiding framework with contrast pull-up with histogram shifting as claimed in claim 1,characterized in that, in (4), the process of the second stage is as follows: information extractor is realized by1Recovery to I. Information extractors obtain I1Partitioning the image, obtaining side information for image restoration from L SB of the first small block, and extracting information1And the extraction of M hidden therein, which contains secret information M2And side information [ A ] for further restoring the original carrier imageO，AE](by A)SetBObtained), and A isO，AERespectively resolving into position pattern and merged histogram position information, and converting I into1Mapping the non-empty gray value to the position of the merged image, performing corresponding mapping on the pixels in the image to obtain an image I 'after histogram merging, and finally adding the I' and the position pattern to obtain a carrier image without loss recovery, namely I-Em+I′。
CN202010243872.6A 2020-03-31 2020-03-31 Reversible information hiding method with contrast ratio pull-up by utilizing histogram translation Active CN111464717B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010243872.6A CN111464717B (en) 2020-03-31 2020-03-31 Reversible information hiding method with contrast ratio pull-up by utilizing histogram translation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010243872.6A CN111464717B (en) 2020-03-31 2020-03-31 Reversible information hiding method with contrast ratio pull-up by utilizing histogram translation

Publications (2)

Publication Number Publication Date
CN111464717A true CN111464717A (en) 2020-07-28
CN111464717B CN111464717B (en) 2022-03-08

Family

ID=71680926

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010243872.6A Active CN111464717B (en) 2020-03-31 2020-03-31 Reversible information hiding method with contrast ratio pull-up by utilizing histogram translation

Country Status (1)

Country Link
CN (1) CN111464717B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112052854A (en) * 2020-09-14 2020-12-08 南京信息工程大学 Medical image reversible information hiding method for realizing self-adaptive contrast enhancement
CN113378881A (en) * 2021-05-11 2021-09-10 广西电网有限责任公司电力科学研究院 Instruction set identification method and device based on information entropy gain SVM model
CN115134142A (en) * 2022-06-28 2022-09-30 南京信息工程大学 Information hiding method and system based on file segmentation

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070085916A1 (en) * 2005-09-30 2007-04-19 Seiko Epson Corporation Image processing apparatus, image processing method and image processing program
CN107689026A (en) * 2017-08-24 2018-02-13 中国科学技术大学 Reversible steganography method based on optimum code
CN109617682A (en) * 2018-12-12 2019-04-12 北京邮电大学 A kind of ciphertext domain reversible information hidden method based on histogram left and right sidesing shifting
CN109671010A (en) * 2018-11-02 2019-04-23 中山大学 It is a kind of based on multiple two-dimensional histograms modification reversible information hide and extracting method

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070085916A1 (en) * 2005-09-30 2007-04-19 Seiko Epson Corporation Image processing apparatus, image processing method and image processing program
CN107689026A (en) * 2017-08-24 2018-02-13 中国科学技术大学 Reversible steganography method based on optimum code
CN109671010A (en) * 2018-11-02 2019-04-23 中山大学 It is a kind of based on multiple two-dimensional histograms modification reversible information hide and extracting method
CN109617682A (en) * 2018-12-12 2019-04-12 北京邮电大学 A kind of ciphertext domain reversible information hidden method based on histogram left and right sidesing shifting

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ZHANG,XINPENG: "Reversible Data Hiding With Optimal Value Transfer", 《TRANSACTIONS ON MULTIMEDIA》 *
应祺超: "基于直方图处理的图像可逆信息隐藏", 《硕士学位论文》 *
陆兴等: "基于二维直方图修改的JPEG图像可逆信息隐藏", 《网络空间安全》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112052854A (en) * 2020-09-14 2020-12-08 南京信息工程大学 Medical image reversible information hiding method for realizing self-adaptive contrast enhancement
CN112052854B (en) * 2020-09-14 2023-09-12 南京信息工程大学 Medical image reversible information hiding method for realizing self-adaptive contrast enhancement
CN113378881A (en) * 2021-05-11 2021-09-10 广西电网有限责任公司电力科学研究院 Instruction set identification method and device based on information entropy gain SVM model
CN115134142A (en) * 2022-06-28 2022-09-30 南京信息工程大学 Information hiding method and system based on file segmentation
CN115134142B (en) * 2022-06-28 2023-09-22 南京信息工程大学 Information hiding method and system based on file segmentation

Also Published As

Publication number Publication date
CN111464717B (en) 2022-03-08

Similar Documents

Publication Publication Date Title
Wang et al. Efficient generalized integer transform for reversible watermarking
Jo et al. A digital image watermarking scheme based on vector quantisation
CN111464717B (en) Reversible information hiding method with contrast ratio pull-up by utilizing histogram translation
Qin et al. Reversible data hiding in VQ index table with lossless coding and adaptive switching mechanism
Chang et al. Reversible steganographic method using SMVQ approach based on declustering
CN111898136A (en) High-fidelity reversible information hiding method and device based on prediction error value sequencing
Kumar et al. A reversible data hiding scheme using pixel location.
CN111277831B (en) Image compression method and system based on classified coding
Pan et al. Novel reversible data hiding scheme for two-stage VQ compressed images based on search-order coding
Fan et al. Multiple histogram based adaptive pairwise prediction-error modification for efficient reversible image watermarking
Chu et al. A digital image watermarking method based on labeled bisecting clustering algorithm
Su et al. Reversible data hiding using the dynamic block-partition strategy and pixel-value-ordering
Li et al. AMBTC based high payload data hiding with modulo-2 operation and Hamming code
Sachnev et al. Reversible watermarking algorithm with distortion compensation
Lee et al. A data hiding scheme based on turtle-shell for AMBTC compressed images
Mshir et al. A New Model for Creating Layer Planes Using Steganography for Text Hiding
CN108009974B (en) Robust reversible watermarking method resisting JPEG compression and digital television broadcasting system
CN113378186A (en) Method and device for encrypting hidden information based on coded image
Ding et al. A Robust Blind Image Watermarking Scheme Based on Classified Vector Quantization.
Ma et al. An improved reversible watermarking scheme
Shie et al. An improved steganographic scheme implemented on the compression domain of image using BTC and Histogram modification
Wu et al. Reversible image watermarking by rhombus prediction and histogram modification
Haijun et al. A Reversible Data Hiding Method by Compressing Overlapping Pixels
Renji et al. A Reversible Data Hiding Technique for Secure Image Transmission
Wang et al. A high performance multi-layer reversible data hiding scheme using two-step embedding

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
GR01 Patent grant
GR01 Patent grant