CN106683031A - Feature extraction method and extraction system for digital image steganalysis - Google Patents
Feature extraction method and extraction system for digital image steganalysis Download PDFInfo
- Publication number
- CN106683031A CN106683031A CN201611264561.8A CN201611264561A CN106683031A CN 106683031 A CN106683031 A CN 106683031A CN 201611264561 A CN201611264561 A CN 201611264561A CN 106683031 A CN106683031 A CN 106683031A
- Authority
- CN
- China
- Prior art keywords
- occurrence matrix
- lbp
- directions
- wave filters
- steganalysis
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2201/00—General purpose image data processing
- G06T2201/005—Image watermarking
- G06T2201/0065—Extraction of an embedded watermark; Reliable detection
Abstract
The invention discloses a feature extraction method and extraction system for digital image steganalysis. The method comprises the steps of filtering a steganalysis feature image to be extracted by ten filters to obtain residual images, the ten filters including four non-directional filters and six directional filters; encoding the residual images according to an LBPriu2 model and/or a DLBP model to obtain coding graphs, obtaining second-order co-occurrence matrices in four different directions according to the coding graphs, and for the six directional filters, adding the co-occurrence matrices calculated in different directions; for the calculated co-occurrence matrices, combining the co-occurrence matrices in different directions and decreasing the dimensions of the co-occurrence matrices, and then performing LOG function mapping to obtain co-occurrence matrix features. According to the invention, similar performance to the existing mainstream general steganalysis method SRM is gained, and the classification effect has higher accuracy than that of the SRM.
Description
Technical field
It is special the present invention relates to digital picture Steganalysis field, more particularly to a kind of digital picture steganalysis
Levy extracting method and extraction system.
Background technology
Digital picture steganalysis have Active Steganalysis and passive steganalysis two types, wherein passive steganalysis
It is main to judge the whether hiding secret information of carrier, and Active Steganalysis also to lead to while carrier has secret information being determined
Cross and estimate that the parameter of steganographic algorithm further extracts secret information.According to usage scenario, steganalysis are divided into special steganography again
Analysis and general steganalysis.General steganography method needs to detect various steganographic algorithms that its operating process is broadly divided into
Two parts of feature extraction and grader classification, wherein characteristic Design part are that the main contents of research are also this patent design
Algorithm acts on link, and the method for the present main flow of sort operation is to use integrated classifier.
In the general steganalysis feature of image spatial domain, most representative is spatial domain richness model (SRM), and it is first by being permitted
It is irregular that polyteny and nonlinear filter obtain the different surplus of image, carries out quantization to surplus afterwards and blocks, then by will not
With quantifying to block filtering surplus figure generation quadravalence co-occurrence matrix, finally feature is generated with different merging mode dimensionality reductions.Rich model
Feature achieves good effect on steganalysis to content-adaptive steganography, and SRM is more fully examined in the selection of wave filter
Consider but the worth improvement of its wave filter rule deficiency this respect.
With the appearance of content-adaptive steganography method HUGO, the steganographic algorithm based on minimum distortion framework goes out successively
Existing, these algorithms reflect the cost that insertion brings by defining a kind of distortion function, by minimize overall cost value by
Information is embedded in texture and fringe region, is more difficult to be detected with this.Secret information is embedded in carrier image mistake for steganographic algorithm
Cheng Zhong, can cause the change of image texture, select the detection more sensitive to texture region to have very great help to analysis steganography.
Local binary model (LBP) is a kind of operator for being combined structure and statistics, and its histogram feature is classified in image texture
In applied well, have been achieved with certain effect on the steganalysis of spatial domain using local binary model histogram feature,
But its performance is still inferior to SRM, histogram feature is not enough for being expressed in correlation between image pixel, it is difficult to catch steganography band
Carry out the change of image statistics, the definition of LBP and steganalysis problem are not that directly matching is also its weak point in addition.
Therefore, prior art has yet to be improved and developed.
The content of the invention
In view of above-mentioned the deficiencies in the prior art, carry it is an object of the invention to provide a kind of digital picture steganalysis feature
Take method and extraction system, it is intended to solve in the prior art using local binary model histogram feature on the steganalysis of spatial domain
Histogram feature is not enough for being expressed in correlation between image pixel, it is difficult to catch the change that steganography brings image statistics,
Not the problems such as definition of LBP and steganalysis problem are not the defects of direct matching.
Technical scheme is as follows:
A kind of digital picture steganalysis feature extracting method, wherein, the described method comprises the following steps:
A, steganalysis characteristic image to be extracted is filtered by 10 wave filters, obtains residual image;Wherein, 10
Individual wave filter includes 4 non-directional wave filters and 6 directional wave filters;
B, residual image is encoded according to LBPriu2 models and/or DLBP models after obtain code pattern, and according to volume
Code figure obtains 4 second order co-occurrence matrixs in direction, to wherein 6 directional wave filters, is added merging different directions and calculates
The co-occurrence matrix for arriving;
C, the co-occurrence matrix to being calculated, carry out LOG Function Mappings and obtain after merging different directions co-occurrence matrix and dimensionality reduction
To co-occurrence matrix feature.
The digital picture steganalysis feature extracting method, wherein, in the step A included by 10 wave filters 4
Individual primary filter is designated as D12, D13, D14 and D15 respectively, wherein:
[1,0]-[0,1]=[1, -1] (D12)
[1, -1,0]-[0,1, -1]=[1, -2,1] (D13)
[1, -2,1,0]-[0,1, -2,1]=[1, -3,3, -1] (D14)
[1, -3,3, -1,0]-[0,1, -3,3, -1]=[1, -4,6, -4,1] (D15);
6 wave filters derived by primary filter included by 10 wave filters be designated as respectively D22, D33, D44,
D55, D23 and D35, wherein:
Wherein, D22, D33, D44, D55 are 4 non-directional wave filters, and D12, D13, D14, D15, D23, D35 are 6
Directional wave filter, to directional wave filter, calculates the surplus in all directions and calculates wherein all directions respectively
The maximum surplus and minimum value surplus of surplus.
The digital picture steganalysis feature extracting method, wherein, filtered by 10 wave filters in the step A
The calculating formula that ripple obtains residual image is:
Wherein, I is steganalysis characteristic image to be extracted, and f is wave filter,It is convolution symbol, Row (f) and Colum
F () is respectively the number of lines and columns for seeking f, M, N are respectively the length and width of code pattern.
The digital picture steganalysis feature extracting method, wherein, the calculating formula of LBPriu2 models in the step B
For:
Wherein
Wherein
Wherein, gpIt is adjacent pixel values, gcIt is median pixel value, P, R are the numbers of the adjacent pixel defined in LBP codings
And position, take P=8 and R=1.
The digital picture steganalysis feature extracting method, wherein, the calculating formula of DLBP models is in the step B:
Wherein
Wherein, wherein gpIt is adjacent pixel values, gcIt is median pixel value, P, R are the adjacent pixels defined in LBP codings
Number and selection radius, take P=8 and R=1, α and are rule of thumb arranged to 0.3 and 0.5, fi,jIt is filter coefficient.
The digital picture steganalysis feature extracting method, wherein, 4 second order symbiosis squares in direction in the step B
Battle array production be:
C0(i, j)=# (x1, y1), (x1, y1+1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1, y1+1)=j }
C45(i, j)=# (x1, y1), (x1-1, y1+1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1-1, y1+1)=j }
C90(i, j)=# (x1, y1), (x1-1, y1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1-1, y1)=j }
C135(i, j)=# (x1, y1), (x1-1, y1+1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1-1, y1+1)=
j};
Wherein, C0It is second order co-occurrence matrix, C on 0 ° of direction45It is second order co-occurrence matrix, C on 45 ° of directions90For on 90 ° of directions
Second order co-occurrence matrix, C135It is second order co-occurrence matrix on 135 ° of directions, M, N are the length and width of code pattern, RLBPIt is LBP or DLBP
Code pattern.
The digital picture steganalysis feature extracting method, wherein, specifically included in the step C:
C1,6 directional wave filters are filtered with the residual image for obtaining different directions, calculate co-occurrence matrix and phase
The co-occurrence matrix of adduction and same wave filter different directions, to 6 directional wave filter maximum residual images and minimum
The co-occurrence matrix nonjoinder that value residual image and 4 non-directional wave filter residual images are obtained;
C2, co-occurrence matrix is added along diagonal doubling left to bottom right, and stretched as the one-dimensional vector on 0 ° of direction
Cf0, one-dimensional vector Cf on 45 ° of directions45, one-dimensional vector Cf on 90 ° of directions90And the one-dimensional vector Cf on 135 ° of directions135;
C3, by the one-dimensional vector Cf on 0 ° of direction0With the one-dimensional vector Cf on 90 ° of directions90Addition obtains Cf0+90, by 45 °
One-dimensional vector Cf on direction45With the one-dimensional vector Cf on 135 ° of directions135Addition obtains Cf45+135;
C4, f is calculated respectively1=log10 (Cf0+90+ 1) and f2=log10 (Cf45+135+ 1), by f1And f2It is total to after series connection
Raw matrix character.
A kind of digital picture steganalysis Feature Extraction System, wherein, including:
Filtration module, for steganalysis characteristic image to be extracted to be filtered by 10 wave filters, obtains residual error
Image;Wherein, 10 wave filters include 4 directional wave filters and 6 non-directional wave filters;
Coding module, for being encoded to residual image according to LBPriu2 models and/or DLBP models after encoded
Figure, and 4 second order co-occurrence matrixs in direction are obtained according to code pattern, to wherein 6 directional wave filters, it is added and merges different
The co-occurrence matrix that direction calculating is obtained;
Co-occurrence matrix characteristic extracting module, for the co-occurrence matrix to being calculated, merges different directions co-occurrence matrix simultaneously
LOG Function Mappings are carried out after dimensionality reduction and obtains co-occurrence matrix feature.
The digital picture steganalysis feature extracting method, wherein, the calculating of LBPriu2 models in the coding module
Formula is:
Wherein
Wherein
Wherein, gpIt is adjacent pixel values, gcIt is median pixel value, P, R are the numbers of the adjacent pixel defined in LBP codings
And position, take P=8 and R=1.
The digital picture steganalysis feature extracting method, wherein, the calculating formula of DLBP models in the coding module
For:
Wherein
Wherein, wherein gpIt is adjacent pixel values, gcIt is median pixel value, P, R are the adjacent pixels defined in LBP codings
Number and selection radius, take P=8 and R=1, α and are rule of thumb arranged to 0.3 and 0.5, fi,jIt is filter coefficient.
Digital picture steganalysis feature extracting method provided by the present invention and extraction system, method include:To wait to carry
Take steganalysis characteristic image to be filtered by 10 wave filters, obtain residual image;Wherein, 10 wave filters include 4
Non-directional wave filter and 6 non-directional wave filters;Residual image is entered according to LBPriu2 models and/or DLBP models
Code pattern is obtained after row coding, and 4 second order co-occurrence matrixs in direction are obtained according to code pattern;To the symbiosis square being calculated
Battle array, carries out LOG Function Mappings and obtains co-occurrence matrix feature after merging different directions co-occurrence matrix and dimensionality reduction.It is of the invention with it is existing
Main flow general steganalysis method SRM has close performance, and classifying quality has accuracy rate higher compared with SRM.
Brief description of the drawings
Fig. 1 is the flow chart of digital picture steganalysis feature extracting method preferred embodiment of the present invention.
Fig. 2 is 6 schematic diagrames of consecutive points selection range yardstick.
Fig. 3 is the functional block diagram of digital picture steganalysis Feature Extraction System preferred embodiment of the present invention.
Specific embodiment
The present invention provides a kind of digital picture steganalysis feature extracting method and extraction system, to make mesh of the invention
, technical scheme and effect it is clearer, clear and definite, the present invention is described in more detail below.It should be appreciated that described herein
Specific embodiment be only used to explain the present invention, be not intended to limit the present invention.
As shown in figure 1, be a kind of flow chart of digital picture steganalysis feature extracting method embodiment of the present invention,
The described method comprises the following steps:
Step S100, steganalysis characteristic image to be extracted is filtered by 10 wave filters, obtains residual image;
Wherein, 10 wave filters include 4 non-directional wave filters and 6 directional wave filters.
For 4 non-directional wave filters, each wave filter can obtain a residual image, directional for 6
Wave filter, each wave filter can obtain 4 direction residual images, 1 minimum value residual image and 1 maximum residual error respectively
Image.
Step S200, residual image is encoded according to LBPriu2 models and/or DLBP models after obtain code pattern,
And 4 second order co-occurrence matrixs in direction are obtained according to code pattern, to wherein 6 directional wave filters, it is added and merges not Tongfang
To the co-occurrence matrix being calculated.
Step S300, the co-occurrence matrix to being calculated, LOG functions are carried out after merging different directions co-occurrence matrix and dimensionality reduction
Mapping obtains co-occurrence matrix feature.
In embodiments of the invention, for steganalysis characteristic image to be extracted (the secondary gray level image of i.e. given one
I), it is necessary first to the surplus R of many is obtained by different wave filters.For in the selection of wave filter, here by derivative
Mode obtains four primary filters, and more rule is understandable and diversity is moderate for four designs of primary filter.
SRM features have taken into full account correlation between image pixel, but hidden for the more rapid self adaptation of development now
Analysis method emphasis steganography zone-texture noise region is write without special strategy.And tradition LBP models for image texture compared with
It is sensitivity, can fully recognizes image texture characteristic, but change of the steganography to image is very trickle, and original definition is using relative
The model that relation draws is not very applicable to steganography, therefore the DLBP models according to relative distance are disclosed in this patent,
Amplitude of variation is taken into account, increases the sensitivity to steganography.
For co-occurrence matrix, because it is the statistics relation between image assigned direction adjacent pixel, it has certain sparse
Property and maximin differ the several orders of magnitude, this is unfavorable for integrated classifier training classification, therefore by symbiosis in the present invention
The common LOG functions of matrix character are mapped, and enable characteristic value in the same order of magnitude.
Preferably, in the digital picture steganalysis feature extracting method, 10 wave filters in the step S100
4 included primary filters are designated as D12, D13, D14 and D15 respectively, wherein:
[1,0]-[0,1]=[1, -1] (D12)
[1, -1,0]-[0,1, -1]=[1, -2,1] (D13)
[1, -2,1,0]-[0,1, -2,1]=[1, -3,3, -1] (D14)
[1, -3,3, -1,0]-[0,1, -3,3, -1]=[1, -4,6, -4,1] (D15);
6 wave filters derived by primary filter included by 10 wave filters be designated as respectively D22, D33, D44,
D55, D23 and D35, wherein:
For two wave filters of D23 and D35, it is necessary to be used zero padding into 3*3 and 5*5 square matrices, in D23 bottoms
The null matrix of 1*3 is filled, in the null matrix of the under-filled 2*5 of D35.
4 non-directional wave filters that D22, D33, D44, D55 are, D12, D13, D14, D15, D23, D35 are 6 sides of having
Tropism wave filter, to directional wave filter, calculates the surplus in their all directions and calculates wherein all directions respectively
The maximum and minimum value surplus of surplus.
The digital picture steganalysis feature extracting method, wherein, entered by 10 wave filters in the step S100
Row filtering obtains the calculating formula of residual image and is:
Wherein, I is steganalysis characteristic image to be extracted, and f is wave filter,It is convolution symbol, Row (f) and Colum
F () is respectively the number of lines and columns for seeking f.
10 wave filters for more than, wherein because D22, D33, D44, D55 have had centre symmetry, it is not necessary to revolve
Turn, so can each obtain a residual image;For six wave filters of D12, D13, D14, D15, D23, D35, by them
Being rotated 4 times with 90 degree and obtain 4 residual images, then take maximum and minimum value by the residual image in respective 4 directions to obtain
To two non-linear residual images (being named as min and max), therefore 6 residual images can be obtained respectively.Therefore more than passing through
10 wave filters altogether can be with 40 residual images.
Preferably, in the digital picture steganalysis feature extracting method, LBPriu2 models in the step S200
Calculating formula be:
Wherein
Wherein
Wherein, gpIt is adjacent pixel values, gcIt is median pixel value, P, R are the numbers of the adjacent pixel defined in LBP codings
And position, take P=8 and R=1.
This model encodes one number of transitions of zero-sum, two principles by original LBP models according to invariable rotary characteristic and LBP
10 kinds are reduced to from 256 kinds, this is conducive to follow-up co-occurrence matrix conversion dimension to control.Several operations more than, first by residual error
Image quantization is blocked, and LBPriu2 model based codings are carried out afterwards can obtain having 10 LBPriu2 code patterns of span
RLBP。
Preferably, in the digital picture steganalysis feature extracting method, DLBP models in the step S200
Calculating formula is:
Wherein
Wherein, wherein gpIt is adjacent pixel values, gcIt is median pixel value, P, R are the adjacent pixels defined in LBP codings
Number and selection radius, take P=8 and R=1, α and are rule of thumb arranged to 0.3 and 0.5, fi,jIt is filter coefficient.
Adjacent 32 kinds of patterns are linearly classified as a kind of pattern by DLBP afterwards here by having 256 kinds of patterns after coding,
With this 8 kinds are reduced to by 256 kinds.Select to choose for DLBP model based codings in the selection for choosing neighbor pixel, in the present invention
The point of adjacent eight pixels takes an existing various range dimensions to take the side of consecutive points in addition as LBP coded references point
Method, i.e., equally take neighbouring eight consecutive points of center pixel as model based coding reference point, but is more than choosing adjacent pixel,
As shown in Fig. 2 it is hereby achieved that 6 selection range yardsticks, same operation is carried out with LBP encoding models.Operated according to more than,
Carrying out DLBP model based codings can obtain having 8 DLBP code patterns R of spanDLBP。
Preferably, in the digital picture steganalysis feature extracting method, 4 the two of direction in the step S200
The production of rank co-occurrence matrix is:
C0(i, j)=# (x1, y1), (x1, y1+1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1, y1+1)=j }
C45(i, j)=# (x1, y1), (x1-1, y1+1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1-1, y1+1)=j }
C90(i, j)=# (x1, y1), (x1-1, y1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1-1, y1)=j }
C135(i, j)=# (x1, y1), (x1-1, y1+1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1-1, y1+1)=
j};
Wherein, C0It is second order co-occurrence matrix, C on 0 ° of direction45It is second order co-occurrence matrix, C on 45 ° of directions90For on 90 ° of directions
Second order co-occurrence matrix, C135It is second order co-occurrence matrix on 135 ° of directions, M, N are the length and width of code pattern, RLBPIt is LBP or DLBP
Code pattern.
Specifically, in the digital picture steganalysis feature extracting method, being specifically included in the step S300:
Step S301, the residual image for obtaining is filtered to wave filter, calculates co-occurrence matrix;
Step S302, co-occurrence matrix is added along diagonal doubling left to bottom right, and stretched as on 0 ° of direction
Dimensional vector Cf0, one-dimensional vector Cf on 45 ° of directions45, one-dimensional vector Cf on 90 ° of directions90And on 135 ° of directions it is one-dimensional to
Amount Cf135;
Step S303, by the one-dimensional vector Cf on 0 ° of direction0With the one-dimensional vector Cf on 90 ° of directions90Addition is obtained
Cf0+90, by the one-dimensional vector Cf on 45 ° of directions45With the one-dimensional vector Cf on 135 ° of directions135Addition obtains Cf45+135;
Step S304, f is calculated respectively1=log10 (Cf0+90+ 1) and f2=log10 (Cf45+135+ 1), by f1And f2After series connection
Obtain co-occurrence matrix feature.
For the co-occurrence matrix Cx on x angle directions, along diagonal doubling left to bottom right be added for it by this patent, diagonally
Pixel is not processed on line, and one-dimensional vector Cfx is drawn into afterwards.To 4 direction one-dimensional vectors of surplus figure generation, will
Cf0And Cf90Addition, Cf45And Cf135Addition obtains Cf0+90And Cf45+135, seek f1=log10 (Cf respectively afterwards0+90+ 1) and f2
=log10 (Cf45+135+ 1), f1 and f2 are together in series the symbiosis square that can obtain the surplus figure under certain range dimension
Battle array becomes transform characteristics.By one residual image of above step, available LBP co-occurrence matrixs intrinsic dimensionality is 660*1 dimensions, is obtained
To DLBP co-occurrence matrix intrinsic dimensionalities be 432*1 dimension.
For this six directive wave filters of tool of D12, D13, D14, D15, D23, D35, four can be respectively obtained
Direction and minimum value and maximum totally six residual images, are respectively obtaining in this programme to four direction surplus figure therein
Co-occurrence matrix after, carry out addition and be merged into a co-occurrence matrix, remaining operations are carried out with this again, carry out by this method
Dimensionality reduction merges, and minimum value and maximum surplus figure then respectively obtain respective co-occurrence matrix feature and be not other merging behaviour
Make.The DLBP co-occurrence matrix characteristic performances of minimum value and maximum surplus are undesirable, therefore only use LBP co-occurrence matrix features.Cause
A total of (660*2+432*2) * 6+660*2*2*6=28944 of this this six filter D 12, D13, D14, D15, D23, D35
Dimension;For this four wave filters of D22, D33, D44, D55, a total of (660*2+432*2) * 4=8736 dimensions.Therefore a width is digital
Image can obtain 37680 dimensional features altogether.
Based on above method embodiment, the present invention also provides a kind of digital picture steganalysis Feature Extraction System.Such as Fig. 3
Shown, the digital picture steganalysis Feature Extraction System includes:
Filtration module 100, for steganalysis characteristic image to be extracted to be filtered by 10 wave filters, obtains residual
Difference image;Wherein, 10 wave filters include 4 non-directional wave filters and 6 directional wave filters;
Coding module 200, for being encoded to residual image according to LBPriu2 models and/or DLBP models after obtain
Code pattern, and 4 second order co-occurrence matrixs in direction are obtained according to code pattern, to wherein 6 directional wave filters, it is added and merges
The co-occurrence matrix that different directions are calculated;
Co-occurrence matrix characteristic extracting module 300, for the co-occurrence matrix to being calculated, merges different directions co-occurrence matrix
And carry out LOG Function Mappings after dimensionality reduction and obtain co-occurrence matrix feature.
Preferably, in the digital picture steganalysis feature extracting method, LBPriu2 in the coding module 200
The calculating formula of model is:
Wherein
Wherein
Wherein, gpIt is adjacent pixel values, gcIt is median pixel value, P, R are the numbers of the adjacent pixel defined in LBP codings
And position, take P=8 and R=1.
Preferably, in the digital picture steganalysis feature extracting method, DLBP models in the coding module 200
Calculating formula be:
Wherein
Wherein, wherein gpIt is adjacent pixel values, gcIt is median pixel value, P, R are the adjacent pixels defined in LBP codings
Number and selection radius, take P=8 and R=1, α and are rule of thumb arranged to 0.3 and 0.5, fi,jIt is filter coefficient.
In sum, digital picture steganalysis feature extracting method provided by the present invention and extraction system, method bag
Include:Steganalysis characteristic image to be extracted is filtered by 10 wave filters, residual image is obtained;Wherein, 10 filtering
Device includes 4 non-directional wave filters and 6 directional wave filters;According to LBPriu2 models and/or DLBP models to residual error
Image obtains code pattern after being encoded, and obtains 4 second order co-occurrence matrixs in direction according to code pattern;What filtering was obtained is residual
Difference image, calculates co-occurrence matrix, LOG Function Mappings is carried out after merging and obtains co-occurrence matrix feature.It is of the invention with existing main flow
General steganalysis method SRM has close performance, and classifying quality has accuracy rate higher compared with SRM.
It should be appreciated that application of the invention is not limited to above-mentioned citing, and for those of ordinary skills, can
To be improved according to the above description or converted, all these modifications and variations should all belong to the guarantor of appended claims of the present invention
Shield scope.
Claims (10)
1. a kind of digital picture steganalysis feature extracting method, it is characterised in that the described method comprises the following steps:
A, steganalysis characteristic image to be extracted is filtered by 10 wave filters, obtains residual image;Wherein, 10 filters
Ripple device includes 4 non-directional wave filters and 6 directional wave filters;
B, residual image is encoded according to LBPriu2 models and/or DLBP models after obtain code pattern, and according to code pattern
4 second order co-occurrence matrixs in direction are obtained, to wherein 6 directional wave filters, is added and is merged what different directions were calculated
Co-occurrence matrix;
C, the co-occurrence matrix to being calculated, carry out LOG Function Mappings and are total to after merging different directions co-occurrence matrix and dimensionality reduction
Raw matrix character.
2. digital picture steganalysis feature extracting method according to claim 1, it is characterised in that 10 in the step A
4 primary filters included by individual wave filter are designated as D12, D13, D14 and D15 respectively, wherein:
[1,0]-[0,1]=[1, -1] (D12)
[1, -1,0]-[0,1, -1]=[1, -2,1] (D13)
[1, -2,1,0]-[0,1, -2,1]=[1, -3,3, -1] (D14)
[1, -3,3, -1,0]-[0,1, -3,3, -1]=[1, -4,6, -4,1] (D15);
6 wave filters derived by primary filter included by 10 wave filters be designated as respectively D22, D33, D44, D55,
D23 and D35, wherein:
Wherein, D22, D33, D44, D55 are 4 non-directional wave filters, and D12, D13, D14, D15, D23, D35 are 6 sides of having
Tropism wave filter, to directional wave filter, calculates the surplus in all directions and calculates wherein all directions surplus respectively
Maximum surplus and minimum value surplus.
3. digital picture steganalysis feature extracting method according to claim 2, it is characterised in that lead in the step A
Cross 10 wave filters and be filtered and obtain the calculating formula of residual image and be:
Wherein, I is steganalysis characteristic image to be extracted, and f is wave filter,It is convolution symbol, Row (f) and Colum (f) points
The number of lines and columns of f Wei not be sought, M, N are respectively the length and width of code pattern.
4. digital picture steganalysis feature extracting method according to claim 1, it is characterised in that in the step B
The calculating formula of LBPriu2 models is:
Wherein, gpIt is adjacent pixel values, gcIt is median pixel value, P, R are number and the position of the adjacent pixel defined in LBP codings
Put, take P=8 and R=1.
5. digital picture steganalysis feature extracting method according to claim 1, it is characterised in that in the step B
The calculating formula of DLBP models is:
Wherein, wherein gpIt is adjacent pixel values, gcIt is median pixel value, P, R are the numbers of the adjacent pixel defined in LBP codings
With selection radius, take P=8 and R=1, α and be rule of thumb arranged to 0.3 and 0.5, fi,jIt is filter coefficient.
6. digital picture steganalysis feature extracting method according to claim 1, it is characterised in that 4 in the step B
The production of the second order co-occurrence matrix in direction is:
C0(i, j)=# (x1, y1), (x1, y1+1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1, y1+1)=j }
C45(i, j)=# (x1, y1), (x1-1, y1+1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1-1, y1+1)=j }
C90(i, j)=# (x1, y1), (x1-1, y1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1-1, y1)=j }
C135(i, j)=# (x1, y1), (x1-1, y1+1) ∈ M × N | RLBP(x1, y1)=i, RLBP(x1-1, y1+1)=j };
Wherein, C0It is second order co-occurrence matrix, C on 0 ° of direction45It is second order co-occurrence matrix, C on 45 ° of directions90It is second order on 90 ° of directions
Co-occurrence matrix, C135It is second order co-occurrence matrix on 135 ° of directions, M, N are respectively the length and width of code pattern, RLBPIt is LBP or DLBP
Code pattern.
7. digital picture steganalysis feature extracting method according to claim 1, it is characterised in that have in the step C
Body includes:
C1,6 directional wave filters are filtered with the residual image for obtaining different directions, calculate co-occurrence matrix and phase adduction
And the co-occurrence matrix of same wave filter different directions, it is residual to 6 directional wave filter maximum residual images and minimum value
The co-occurrence matrix nonjoinder that difference image and 4 non-directional wave filter residual images are obtained;
C2, co-occurrence matrix is added along diagonal doubling left to bottom right, and stretched as the one-dimensional vector Cf on 0 ° of direction0、
One-dimensional vector Cf on 45 ° of directions45, one-dimensional vector Cf on 90 ° of directions90And the one-dimensional vector Cf on 135 ° of directions135;
C3, by the one-dimensional vector Cf on 0 ° of direction0With the one-dimensional vector Cf on 90 ° of directions90Addition obtains Cf0+90, by 45 ° of directions
On one-dimensional vector Cf45With the one-dimensional vector Cf on 135 ° of directions135Addition obtains Cf45+135;
C4, f is calculated respectively1=log10 (Cf0+90+ 1) and f2=log10 (Cf45+135+ 1), by f1And f2Symbiosis square is obtained after series connection
Battle array feature.
8. a kind of digital picture steganalysis Feature Extraction System, it is characterised in that including:
Filtration module, for steganalysis characteristic image to be extracted to be filtered by 10 wave filters, obtains residual image;
Wherein, 10 wave filters include 4 non-directional wave filters and 6 directional wave filters;
Coding module, for being encoded to residual image according to LBPriu2 models and/or DLBP models after obtain code pattern,
And 4 second order co-occurrence matrixs in direction are obtained according to code pattern, to wherein 6 directional wave filters, it is added and merges not Tongfang
To the co-occurrence matrix being calculated;
Co-occurrence matrix characteristic extracting module, for the co-occurrence matrix to being calculated, merges different directions co-occurrence matrix and dimensionality reduction
After carry out LOG Function Mappings and obtain co-occurrence matrix feature.
9. digital picture steganalysis feature extracting method according to claim 8, it is characterised in that in the coding module
The calculating formula of LBPriu2 models is:
Wherein, gpIt is adjacent pixel values, gcIt is median pixel value, P, R are number and the position of the adjacent pixel defined in LBP codings
Put, take P=8 and R=1.
10. digital picture steganalysis feature extracting method according to claim 8, it is characterised in that the coding module
The calculating formula of middle DLBP models is:
Wherein, wherein gpIt is adjacent pixel values, gcIt is median pixel value, P, R are the numbers of the adjacent pixel defined in LBP codings
With selection radius, take P=8 and R=1, α and be rule of thumb arranged to 0.3 and 0.5, fi,jIt is filter coefficient.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611264561.8A CN106683031B (en) | 2016-12-30 | 2016-12-30 | Digital image steganalysis feature extraction method and extraction system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611264561.8A CN106683031B (en) | 2016-12-30 | 2016-12-30 | Digital image steganalysis feature extraction method and extraction system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106683031A true CN106683031A (en) | 2017-05-17 |
CN106683031B CN106683031B (en) | 2020-02-18 |
Family
ID=58850641
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611264561.8A Active CN106683031B (en) | 2016-12-30 | 2016-12-30 | Digital image steganalysis feature extraction method and extraction system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106683031B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107330845A (en) * | 2017-06-26 | 2017-11-07 | 中山大学 | A kind of digital picture steganalysis method based on convolutional neural networks |
CN107819568A (en) * | 2017-11-15 | 2018-03-20 | 深圳大学 | Information detecting method, device and receiving device based on constellation point distance |
CN107979711A (en) * | 2017-11-28 | 2018-05-01 | 上海大学 | Based on the method for optimizing distortion modification prevention hided transmission |
CN108171689A (en) * | 2017-12-21 | 2018-06-15 | 深圳大学 | A kind of identification method, device and the storage medium of the reproduction of indicator screen image |
CN108764270A (en) * | 2018-04-03 | 2018-11-06 | 上海大学 | A kind of Information Hiding & Detecting method integrated using convolutional neural networks |
CN109934760A (en) * | 2019-01-29 | 2019-06-25 | 深圳大学 | A kind of adaptive steganography method of spatial domain picture, system and device |
CN110490214A (en) * | 2018-05-14 | 2019-11-22 | 阿里巴巴集团控股有限公司 | The recognition methods and system of image, storage medium and processor |
CN110889456A (en) * | 2019-12-02 | 2020-03-17 | 深圳大学 | Neural network-based co-occurrence matrix feature extraction method, storage medium and terminal |
CN111402200A (en) * | 2020-02-18 | 2020-07-10 | 江苏大学 | Fried food detection system based on symbiotic double-current convolution network and digital image |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103077512A (en) * | 2012-10-18 | 2013-05-01 | 北京工业大学 | Feature extraction and matching method and device for digital image based on PCA (principal component analysis) |
US20150063625A1 (en) * | 2013-08-28 | 2015-03-05 | Morpho Trust USA Inc. | Dynamic digital watermark |
CN105117729A (en) * | 2015-05-11 | 2015-12-02 | 杭州金培科技有限公司 | Method and device for recognizing copied image |
CN106203492A (en) * | 2016-06-30 | 2016-12-07 | 中国科学院计算技术研究所 | The system and method that a kind of image latent writing is analyzed |
-
2016
- 2016-12-30 CN CN201611264561.8A patent/CN106683031B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103077512A (en) * | 2012-10-18 | 2013-05-01 | 北京工业大学 | Feature extraction and matching method and device for digital image based on PCA (principal component analysis) |
US20150063625A1 (en) * | 2013-08-28 | 2015-03-05 | Morpho Trust USA Inc. | Dynamic digital watermark |
CN105117729A (en) * | 2015-05-11 | 2015-12-02 | 杭州金培科技有限公司 | Method and device for recognizing copied image |
CN106203492A (en) * | 2016-06-30 | 2016-12-07 | 中国科学院计算技术研究所 | The system and method that a kind of image latent writing is analyzed |
Non-Patent Citations (3)
Title |
---|
TIMO OJALA等: "Multiresolution Gray-Scale and Rotation Invariant Texture Classification with Local Binary Patterns", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
李燕等: "基于LBP和共生矩阵的图像拼接篡改检测", 《武汉大学学报(理学版)》 * |
陈世媛等: "基于权值分配的隐写分析算法", 《计算机应用研究》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107330845A (en) * | 2017-06-26 | 2017-11-07 | 中山大学 | A kind of digital picture steganalysis method based on convolutional neural networks |
CN107819568B (en) * | 2017-11-15 | 2020-11-03 | 深圳大学 | Information detection method and device based on constellation point distance and receiving equipment |
CN107819568A (en) * | 2017-11-15 | 2018-03-20 | 深圳大学 | Information detecting method, device and receiving device based on constellation point distance |
CN107979711A (en) * | 2017-11-28 | 2018-05-01 | 上海大学 | Based on the method for optimizing distortion modification prevention hided transmission |
CN108171689A (en) * | 2017-12-21 | 2018-06-15 | 深圳大学 | A kind of identification method, device and the storage medium of the reproduction of indicator screen image |
CN108171689B (en) * | 2017-12-21 | 2021-12-14 | 深圳大学 | Identification method and device for copying display screen image and storage medium |
CN108764270A (en) * | 2018-04-03 | 2018-11-06 | 上海大学 | A kind of Information Hiding & Detecting method integrated using convolutional neural networks |
CN108764270B (en) * | 2018-04-03 | 2022-06-14 | 上海大学 | Information hiding detection method integrated by using convolutional neural network |
CN110490214A (en) * | 2018-05-14 | 2019-11-22 | 阿里巴巴集团控股有限公司 | The recognition methods and system of image, storage medium and processor |
CN110490214B (en) * | 2018-05-14 | 2023-05-02 | 阿里巴巴集团控股有限公司 | Image recognition method and system, storage medium and processor |
CN109934760A (en) * | 2019-01-29 | 2019-06-25 | 深圳大学 | A kind of adaptive steganography method of spatial domain picture, system and device |
CN110889456A (en) * | 2019-12-02 | 2020-03-17 | 深圳大学 | Neural network-based co-occurrence matrix feature extraction method, storage medium and terminal |
CN111402200A (en) * | 2020-02-18 | 2020-07-10 | 江苏大学 | Fried food detection system based on symbiotic double-current convolution network and digital image |
CN111402200B (en) * | 2020-02-18 | 2021-12-21 | 江苏大学 | Fried food detection system based on symbiotic double-current convolution network and digital image |
Also Published As
Publication number | Publication date |
---|---|
CN106683031B (en) | 2020-02-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106683031A (en) | Feature extraction method and extraction system for digital image steganalysis | |
Zhao et al. | Passive forensics for copy-move image forgery using a method based on DCT and SVD | |
CN109102469B (en) | Remote sensing image panchromatic sharpening method based on convolutional neural network | |
Dabov et al. | Image denoising by sparse 3-D transform-domain collaborative filtering | |
Li et al. | Complex contourlet-CNN for polarimetric SAR image classification | |
Mardanpour et al. | Robust transparent image watermarking with Shearlet transform and bidiagonal singular value decomposition | |
CN109859091B (en) | Image steganography detection method based on Gabor filtering and convolutional neural network | |
WO2010084714A1 (en) | Image identifier extracting apparatus | |
CN105096280A (en) | Method and device for processing image noise | |
CN104285239A (en) | Image processing device, image processing method, program, print medium, and recording medium | |
CN110490265A (en) | A kind of image latent writing analysis method based on two-way convolution sum Fusion Features | |
CN103544488B (en) | A kind of face identification method and device | |
CN107066959A (en) | A kind of hyperspectral image classification method based on Steerable filter and linear space correlation information | |
Cao et al. | Resampling detection of recompressed images via dual-stream convolutional neural network | |
Peng et al. | Revealing traces of image resampling and resampling antiforensics | |
WO2007105890A1 (en) | Recognizing the denomination of a note using wavelet transform | |
CN112785480A (en) | Image splicing tampering detection method based on frequency domain transformation and residual error feedback module | |
CN106530360B (en) | Complementary color wavelet color image processing method | |
CN105631860B (en) | Image point extracting method of the same name based on partial ordering's direction histogram description | |
Zou et al. | Joint wavelet sub-bands guided network for single image super-resolution | |
CN106709915A (en) | Image resampling operation detection method | |
CN107040790A (en) | A kind of video content certification and tampering location method based on many granularity Hash | |
CN108090460A (en) | Expression recognition feature extraction algorithm based on multi-direction description of weber | |
CN115272093B (en) | Hyperspectral image unmixing method based on spatial structure information constraint | |
CN107481196B (en) | Feature transformation face super-resolution reconstruction method based on nearest feature line |
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 |