CN108009434B - Rich model steganography detection feature selection method based on rough set α -positive domain reduction - Google Patents

Rich model steganography detection feature selection method based on rough set α -positive domain reduction Download PDF

Info

Publication number
CN108009434B
CN108009434B CN201711329640.7A CN201711329640A CN108009434B CN 108009434 B CN108009434 B CN 108009434B CN 201711329640 A CN201711329640 A CN 201711329640A CN 108009434 B CN108009434 B CN 108009434B
Authority
CN
China
Prior art keywords
feature
value
calculating
detection
components
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.)
Active
Application number
CN201711329640.7A
Other languages
Chinese (zh)
Other versions
CN108009434A (en
Inventor
罗向阳
马媛媛
李晓龙
包震坤
张祎
王道顺
杨春芳
刘粉林
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.)
Information Engineering University of PLA Strategic Support Force
Original Assignee
Information Engineering University of PLA Strategic Support Force
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 Information Engineering University of PLA Strategic Support Force filed Critical Information Engineering University of PLA Strategic Support Force
Priority to CN201711329640.7A priority Critical patent/CN108009434B/en
Publication of CN108009434A publication Critical patent/CN108009434A/en
Application granted granted Critical
Publication of CN108009434B publication Critical patent/CN108009434B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Hardware Design (AREA)
  • Computer Security & Cryptography (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Bioethics (AREA)
  • Image Analysis (AREA)
  • Other Investigation Or Analysis Of Materials By Electrical Means (AREA)

Abstract

The invention belongs to the technical field of information steganography analysis, and particularly relates to a rich model steganography detection feature selection method based on a rough set α -positive domain reduction.

Description

Rich model steganography detection feature selection method based on rough set α -positive domain reduction
Technical Field
The invention belongs to the technical field of information steganography analysis, and particularly relates to a rich model steganography detection feature selection method based on a rough set α -positive domain reduction.
Background
Information steganography techniques enable steganographic communication by embedding secret information into digital media, such as images, video, audio, text, and the like. In contrast, steganalysis techniques are directed to detecting and extracting secret information transmitted over a public channel. After about 20 years of development, the image steganography technology has made great progress, and the development has been from the early simple replacement of steganography by lsb (leastsignifican bit) to the current adaptive steganography. The image self-adaptive steganography technology places embedded changes in complex texture or noise regions which are difficult to model in an image, and excellent anti-detection performance is achieved. With the rapid development of image steganography, steganography analysis technology has also made great progress. In recent years, a detector constructs high-dimensional features by extracting multiple types of statistical attributes of an image and detects self-adaptive steganography by combining an integrated classifier, so that a good detection effect is achieved. However, although these high-dimensional features show good effect in detecting adaptive steganography, these features also bring huge computational overhead to extraction and corresponding classifier training, limiting practical application of this kind of method. Therefore, the invention aims to solve the problem of how to greatly reduce the dimension of the high-dimensional Rich Model under the condition of keeping the feature detection performance of the high-dimensional Rich Model to be equivalent.
The invention relates to a method for reducing dimension of a Rich Model, which aims at solving the problem of selection and dimension reduction of high-dimensional steganography detection features by introducing a rough set α -a positive domain reduction theory.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention aims to provide a rich model steganography detection feature selection method based on rough set α -positive domain reduction, so that the dimension of steganography detection features is reduced, and the efficiency of steganography detection is improved.
In order to achieve the purpose, the invention adopts the following technical scheme:
a rich model steganography detection feature selection method based on rough set α -positive domain reduction comprises the following steps:
step1, constructing a rich model steganography detection characteristic into a corresponding decision table;
step2, adopting attribute separability of each characteristic component in the ASM measurement decision table, and taking an ASM value as a basis for sorting and dividing subsets;
and 3, when the hidden-writing detection features of the rich model are reduced based on the rough set α -positive domain reduction, sequentially adding a plurality of continuous candidate feature components with larger ASM values into the reduction subset, and simultaneously eliminating redundant and conflicting feature components by using the attribute independence principle of the rough set α -positive domain reduction.
Further, the specific implementation process of step1 is as follows:
step1.1, the extracted rich model high-dimensional characteristics are normalized,
Figure BDA0001506359540000021
Figure BDA0001506359540000022
wherein the content of the first and second substances,
Figure BDA0001506359540000023
and
Figure BDA0001506359540000024
is the value of the ith feature component of the jth carrier and the secret image; after processing, the data in the decision table are all discrete and have the value of [ -1, 1 [ -1 [ ]]Numerical data in between;
step1.2, determining a decision attribute value, wherein a decision attribute set is represented by 0 or 1, 0 represents a carrier image, and 1 represents a secret image;
step1.3, a two-dimensional decision table T is constructed, each line represents the value of the characteristic component of a certain image, each column represents the value of the same-dimensional characteristic component of each image,
Figure BDA0001506359540000031
a value corresponding to the ith feature component representing the jth carrier image,
Figure BDA0001506359540000032
representing the value corresponding to the ith characteristic component of the jth secret-carrying image, wherein j is more than or equal to 1 and less than or equal to M, i is more than or equal to 1 and less than or equal to N, and the last column represents the decision attribute of the image;
the decision table form of the steganography detection feature is as follows:
Figure BDA0001506359540000033
further, the specific implementation process of step2 is as follows:
step2.1, calculating the average value of the characteristic components of the steganography detection; calculating the steganographic detection characteristic component f according to the value in the decision table TiMean value μ in the support and in the secret image+(fi) And mu-(fi)：
Figure BDA0001506359540000034
Figure BDA0001506359540000035
Step2.2, calculating the average value difference of the steganographic detection characteristic components; calculating the steganographic detection characteristic component f according to the result of the step2.1iAbsolute value μ of mean difference in carrier and secret images:
μ＝|μ+(fi)-μ-(fi)；
step2.3, calculating the standard deviation of the characteristic component of the steganography detection; calculating the characteristic component f of steganography detection according to the values in the decision table T and the result of the step2.1iStandard deviation delta in carrier and secret images+(fi) And delta-(fi)：
Figure BDA0001506359540000036
Figure BDA0001506359540000041
Step2.4, calculating the sum of standard deviations of the steganographic detection characteristic components; calculating the steganographic detection characteristic component f according to the result of the step2.3iSum of standard deviations δ in the support and dense images:
δ＝δ+(fi)+δ-(fi)；
step2.5, calculating attribute separability; calculating the steganographic detection characteristic component f according to the calculation results of the step2.2 and the step2.4iValue of attribute separability of (1):
Figure BDA0001506359540000042
step2.6, output ASM (f)i)。
Further, the specific implementation process of step3 is as follows:
step 3.1, initializing; reduction subset of high-dimensional steganography detection features
Figure BDA0001506359540000043
Step 3.2, deleting low separability characteristic components; setting a lower limit ASM of an ASM valueminIf ASM (f)i)<ASMminThen the steganographic detection feature component fiDeleting the low-separability feature components which are regarded as irrelevant feature components to obtain a set H' from which all the low-separability feature components are deleted;
step 3.3, calculating step length; dividing step size
Figure BDA0001506359540000044
Wherein λ is the step size, and m is the number of the expected feature subsets;
step 3.4, dividing subsets; according to the ASM value, the characteristic component in H' is reducedSorting the sequence, and dividing the sorted characteristic components into m characteristic subsets H ″ { H ″, according to the value of the step length lambda1,h2,…,hmWherein the ith feature subset
Figure BDA0001506359540000046
t is the number of features in the subset;
step 3.5, selecting characteristics;
Figure BDA0001506359540000045
Figure BDA0001506359540000051
and 3.6, selecting the characteristic subset, and selecting a reduction subset with good classification effect and low dimension.
Compared with the prior art, the invention has the following advantages:
after the steganography detection features are selected by the method, the potential (the number of set elements) of the feature subset is usually obviously lower than the dimension of the original features, so that the time for extracting the features is reduced, and compared with the high-dimensional features, the low-dimensional features can obviously reduce the pressure of a classifier and shorten the processing time of the classifier, so that the steganography detection based on the reduced features can obviously improve the detection efficiency. The method does not need to depend on a specific extraction algorithm, and has the advantages of simplicity in implementation, low time complexity and the like, so that the method is suitable for selecting high-dimensional steganography detection characteristics.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed in the prior art and the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
FIG. 1 is a schematic flow chart of a rich model steganography detection feature selection method based on rough set α -positive domain reduction according to the present invention.
Detailed Description
The core of the invention is to provide a rich model steganography detection feature selection method based on rough set α -positive domain reduction, so that the dimension of the steganography detection feature is reduced, and the efficiency of steganography detection is improved.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All embodiments obtained by a person of ordinary skill in the art based on the embodiments of the present invention without any creative efforts shall fall within the protection scope of the present invention.
The invention provides a rich model steganography detection feature selection method based on rough set α -normal domain reduction, which comprises the following steps:
step1, constructing a rich model steganography detection characteristic into a corresponding decision table;
step2, measuring the attribute Separability of each characteristic component in the decision table by adopting an ASM (attribute separation measure), and taking an ASM value as a basis for sequencing and dividing subsets;
and 3, when the hidden-writing detection features of the rich model are reduced based on the rough set α -positive domain reduction, sequentially adding a plurality of continuous candidate feature components with larger ASM values into the reduction subset, and simultaneously eliminating redundant and conflicting feature components by using the attribute independence principle of the rough set α -positive domain reduction.
For a more thorough and intuitive understanding of the rich model steganography detection feature selection method based on the rough set α -positive domain reduction, it is described in more detail below:
referring to fig. 1, fig. 1 is a flow chart of a rich model steganography detection feature selection method based on rough set α -normal domain reduction according to the present invention, the method includes the following steps of a carrier image set in an image library
Figure BDA0001506359540000071
Secret-carrying image collection
Figure BDA0001506359540000072
Extracted feature component set H ═ { f ═ f1,f2,...,fi,...fN}，fi＝{fc,i,fs,i}1≤i≤N。
Step 1: and (5) constructing a decision table. And normalizing the steganographic detection features of the rich model, so that the corresponding feature values are converted between [ -1, 1 ]. And constructing a decision table T, wherein each line represents the characteristic component of a certain image, each column represents the same-dimensional characteristic value of each image, the last column represents the decision attribute of the image, the decision value of the carrier image is 0, and the decision value of the secret-carrying image is 1.
Step1.1, normalizing the extracted rich model high-dimensional characteristics,
Figure BDA0001506359540000073
Figure BDA0001506359540000074
wherein the content of the first and second substances,
Figure BDA0001506359540000075
and
Figure BDA0001506359540000076
is the value of the ith feature component of the jth carrier and the secret image, i.e. the value of the feature component with the maximum absolute value divided by the value of all the feature components; after processing, the data in the decision table are all discrete and have the value of [ -1, 1 [ -1 [ ]]Numerical data in between.
Step1.2, defining decision attribute values, wherein a decision attribute set is represented by 0 or 1, 0 represents a carrier image, and 1 represents a secret image.
Step1.3, constructing a two-dimensional decision table T, wherein each line represents a characteristic component of a certain imageEach column representing the values of the same dimensional feature component of the respective image,
Figure BDA0001506359540000081
a value corresponding to the ith feature component representing the jth carrier image,
Figure BDA0001506359540000082
and j is more than or equal to 1 and less than or equal to M, i is more than or equal to 1 and less than or equal to N, and the last column represents the decision attribute of the image.
The decision table form of the steganography detection feature is as follows:
Figure BDA0001506359540000083
for ease of understanding, an example is given below to describe a specific process of constructing the decision table.
Example 1: assuming that 5 carrier images and corresponding 5 secret-carrying images are provided, 8-dimensional feature components are respectively extracted from the carrier and the secret-carrying images, and H ═ f1,f2,f3,f4,f5,f6,f7,f8The method concretely comprises the following steps:
f1＝{fc,1,fs,1}＝{(0.85,0.833,0.86,0.91,0.799),(0.646,0.969,0.844,0.96,0.969)}
f2＝{fc,2,fs,2}＝{(0.51,0.17,1.19,0.34,0.85),(0.85,0.85,0.119,1.7,1.02)}
f3＝{fc,3,fs,3}＝{(0.714,0.17,0.034,0.884,0.85),(0.394,0.85,1.19,0.544,1.224)}
f4＝{fc,4,fs,4}＝{(0.714,1.054,0.034,0.884,0.374),(0.85,0.85,1.19,1.7,1.02)}
f5＝{fc,5,fs,5}＝{(0.85,0.68,0.867,0.918,0.799),(0.663,0.986,0.901,0.969,1.02)}
f6＝{fc,6,fs,6}＝{(0.17,0.34,0.68,0.17,0.34),(1.53,0.34,1.19,0.17,1.19)}
f7＝{fc,7,fs,7}＝{(1.53,1.36,0.68,0.17,0.85),(0.17,0.34,0.17,0.17,1.19)}
f8＝{fc,8,fs,8}＝{(0.17,0.34,0.17,0.34,0.51),(0.51,0.68,0.85,0.51,0.34)}
the decision table is constructed as follows:
step1.1: and (6) normalizing. The maximum characteristic value of the carrier and the secret is 1.7, namely
Figure BDA0001506359540000084
Obtaining:
Figure BDA0001506359540000091
Figure BDA0001506359540000092
similarly, all the characteristic values are normalized.
Step1.2, the decision attribute value is determined, and the decision attribute set Q ═ 0,1, where 0 represents the carrier image and 1 represents the secret image.
Step1.3, generating a decision table. Constructing a decision table according to the normalized characteristic values in Step1.1, wherein the image set
Figure BDA0001506359540000093
Conditional attribute set H ═ f1,f2,f3,f4,f5,f6,f7,f8And (4) a decision attribute set Q is generated, and a decision table is shown in a table 1.
Table 1 steganographic feature decision table T
Figure BDA0001506359540000094
Step2, measure attribute separability. Measuring attribute separability of each feature component in the decision table T, wherein the measurement comprises three parts: calculating the mean value of each steganographic detection characteristic component, calculating the standard deviation of each steganographic detection characteristic component, and calculating the ASM value of each steganographic detection characteristic component.
Step2.1, calculating the average value of the steganography detection characteristic components; calculating the steganographic detection characteristic component f according to the value in the decision table TiMean value μ in the support and in the secret image+(fi) And mu-(fi)：
Figure BDA0001506359540000101
Figure BDA0001506359540000102
Step2.2, calculating the average value difference of the steganographic detection characteristic components; calculating the steganography detection characteristic component f according to the result of Step2.1iAbsolute value μ of mean difference in carrier and secret images:
μ＝|μ+(fi)-μ-(fi)。
step2.3, calculating the standard deviation of the characteristic component of steganography detection; calculating the steganography detection characteristic component f according to the values in the decision table T and the result of Step2.1iStandard deviation delta in carrier and secret images+(fi) And delta-(fi)：
Figure BDA0001506359540000103
Figure BDA0001506359540000104
Step2.4, calculating the sum of standard deviations of the steganographic detection characteristic components; calculating the steganography detection characteristic component f according to the result of Step2.3iSum of standard deviations δ in the support and dense images:
δ＝δ+(fi)+δ-(fi)。
step2.5, calculating attribute separability; according to Step2.2 and Step 2.4.4, calculating the characteristic component f of steganography detectioniValue of attribute separability of (1):
Figure BDA0001506359540000105
step2.6, output ASM (f)i)。
How to measure the attribute separability of the steganography detection feature component is described in detail by an example.
Example 2: the data are from example 1 (decision table construction results are shown in table 1), and the calculation process of the ASM values of the feature components of each dimension is as follows.
Step2.1, calculating the average value of the characteristic components in the carrier and the secret image:
Figure BDA0001506359540000111
Figure BDA0001506359540000112
Figure BDA0001506359540000113
Figure BDA0001506359540000114
Figure BDA0001506359540000115
Figure BDA0001506359540000116
Figure BDA0001506359540000117
Figure BDA0001506359540000118
Figure BDA0001506359540000119
Figure BDA00015063595400001110
Figure BDA00015063595400001111
Figure BDA00015063595400001112
Figure BDA00015063595400001113
Figure BDA00015063595400001114
Figure BDA00015063595400001115
Figure BDA00015063595400001116
step2.2, calculating the mean difference of the characteristic components in the carrier and the secret image:
|μ+(f1)-μ-(f1)|＝|0.502-0.524|＝0.022
|μ+(f2)-μ-(f2)|＝|0.36-0.66|＝0.3
|μ+(f3)-μ-(f3)|＝|0.312-0.612|＝0.3
|μ+(f4)-μ-(f4)|＝|0.36-0.66|＝0.3
|μ+(f5)-μ-(f5)|＝|0.484-0.534|＝0.05
|μ+(f6)-μ-(f6)＝|0.2-0.52|＝0.32
|μ+(f7)-μ-(f7)|＝|0.54-0.24|＝0.3
|μ+(f8)-μ-(f8)|＝|0.18-0.34|＝0.16
step2.3, calculating the standard deviation of the characteristic components in the carrier and the secret image:
Figure BDA0001506359540000121
Figure BDA0001506359540000122
Figure BDA0001506359540000123
Figure BDA0001506359540000124
Figure BDA0001506359540000125
Figure BDA0001506359540000126
Figure BDA0001506359540000127
Figure BDA0001506359540000128
Figure BDA0001506359540000129
Figure BDA00015063595400001210
Figure BDA00015063595400001211
Figure BDA00015063595400001212
Figure BDA00015063595400001213
Figure BDA00015063595400001214
Figure BDA00015063595400001215
Figure BDA00015063595400001216
step2.4, calculating the sum of the standard deviations of the feature components in the vector and the secret image:
δ+(f1)+δ－(f1)＝0.0232+0.0750＝0.0982
δ+(f2)+δ－(f2)＝0.2154+0.1855＝0.4009
δ+(f3)+δ－(f3)＝0.2100+0.1792＝0.3892
δ+(f4)+δ－(f4)＝0.2154+0.1855＝0.4009
δ+(f5)+δ－(f5)＝0.0476+0.0755＝0.1231
δ+(f6)+δ－(f6)＝0.1095+0.3124＝0.4219
δ+(f7)+δ－(f7)＝0.2871+0.2332＝0.5203
δ+(f8)+δ－(f8)＝0.0748+0.1020＝0.1768
step2.5, calculating characteristic component ASM value:
Figure BDA0001506359540000131
Figure BDA0001506359540000132
Figure BDA0001506359540000133
Figure BDA0001506359540000134
Figure BDA0001506359540000135
Figure BDA0001506359540000136
Figure BDA0001506359540000137
Figure BDA0001506359540000138
step3, reducing the characteristics, firstly calculating a division Step value lambda, secondly sorting the characteristic components in the set H' with all the low-separability characteristic components deleted in a descending order according to the ASM values of the characteristic components, secondly dividing the sorted characteristic components into a plurality of characteristic subsets according to the division Step value lambda, then selecting and testing the independence of the steganography detection characteristics by using a rough set α -positive domain reduction according to a threshold value α, outputting the reduction result, and finally finding a plurality of rough sets α -positive domain reduction subsets which accord with the positive domain non-reduction.
Step 3.1, initializing; reduction subset of high-dimensional steganography detection features
Figure BDA0001506359540000139
Step 3.2, deleting low separability characteristic components; setting a lower limit ASM of an ASM valueminIf ASM (f)i)<ASMminThen the steganographic detection feature component fiAnd deleting the feature components which are regarded as irrelevant feature components to obtain a set H' with all low separability feature components deleted.
Step 3.3, calculating step length; dividing step size
Figure BDA00015063595400001310
Where λ is the step size and m is the number of subsets of features desired.
Step 3.4, dividing subsets; sorting the feature components in H ' in descending order according to the ASM value, and dividing the sorted feature components into m feature subsets H ' { H ' (m is H) according to the value of the step length lambda1,h2,…,hmWherein the ith feature subset
Figure BDA0001506359540000141
And t is the number of features in the subset.
Step 3.5, selecting characteristics;
Figure BDA0001506359540000142
and 3.6, selecting the characteristic subset, and selecting a reduction subset with good classification effect and low dimension.
For the above reduction algorithm, a specific example is given below for explanation.
Example 3: the ASM values and their labels of the feature components are calculated in example 2, and the solution process of the reduced subset is as follows.
(1) And (5) initializing.
Figure BDA0001506359540000151
(2) The low separability feature is deleted. Setting ASMmin0.4, and f10.2242 < 0.4, so f1The irrelevant feature component is deleted, and a set H' ═ f obtained after all the low separability feature components are deleted is obtained2,f3,f4,f5,f6,f7,f8}。
(3) And calculating the dividing step length. Dividing step size
Figure BDA0001506359540000152
Where the number of expected feature subsets is 4.
(4) The subsets are divided. According to the lambda value, sorting the feature components in H' from large to small according to the ASM value and dividing the feature components into four feature subsets H ″ { H ″1,h2,h3,h4}＝{{f8}{f3,f6,f2,f4},{f7},{f5}}。
(5) Features are reduced.
Figure BDA0001506359540000153
For subset H in H ″iAnd sequentially executing reduction operations:
a)i＝1，h1＝{f8}，
a subset is added.
Then
Figure BDA0001506359540000154
CN' ═ {8 }; after deletion, H { { f { } { [ f { ] { [ H ] } { [ F ]3,f6,f2,f4},{f7},{f5}}；
Because P (X [ X ]]B)＝0.6＜α，
Figure BDA0001506359540000155
Thus, the
Figure BDA0001506359540000156
Non-compliant with positive domain non-subtractions;
meanwhile, independence test is not needed; }
The reduced subset is output. B ═ f8}，CN′＝{8}。
B＝{f8Is not a rough set α -a positive domain reduction subset of the steganography detection feature set H.
b)i＝2，h2＝{f3,f6,f2,f4}，
A subset is added.
Then B ═ f8}∪{f3,f6,f2,f4}＝{f8,f3,f6,f2,f4}; CN' {8,3,6,2,4 }; after deletion, H { { f { } { [ f { ] { [ H ] } { [ F ]7},{f5}}；
Because P (X | [ X ]]B)＝0.7≥α，
Figure BDA0001506359540000157
Thus, the
Figure BDA0001506359540000158
Conforming to positive domain nondecreasing;
independence tests were performed simultaneously. And t is 4, need to
Figure BDA0001506359540000159
Independence tests were performed in sequence.
Figure BDA00015063595400001510
Figure BDA0001506359540000161
Figure BDA0001506359540000162
Figure BDA0001506359540000163
CN′＝{8,3,6,2}；
B＝{f8,f3,f6,f2Is a rough set α -a positive domain reduction subset of the steganography detection feature set H
The reduced subset is output. B ═ f8,f3,f6,f2}，CN′＝{8,3,6,2}；
c)i＝3，h3＝{f7}，
A subset is added.
Then B ═ f8,f3,f6,f2}∪{f7}＝{f8,f3,f6,f2,f7}; CN' {8,3,6,2,7 }; after deletion, H { { f { } { [ f { ] { [ H ] } { [ F ]5}}；
Because P (X | [ X ]]B)＝0.8≥α，
Figure BDA0001506359540000164
Thus, the
Figure BDA0001506359540000165
Conforming to positive domain nondecreasing;
at the same time, the independence test is carried out,
Figure BDA0001506359540000166
because t is 1, no independence test is required;
B＝{f8,f3,f6,f2,f7is a rough set α -a positive domain reduction subset of the steganography detection feature set H
The reduced subset is output. B ═ f8,f3,f6,f2,f7}，CN′＝{8,3,6,2,7}；
d)i＝4，h4＝{f5}，
A subset is added.
Then B ═ f8,f3,f6,f2,f7}∪{f5}＝{f8,f3,f6,f2,f7,f5}; CN' {8,3,6,2,7,5 }; after the deletion, the user can delete the file,
Figure BDA0001506359540000169
because P (X [ X ]]B)＝0.6＜α，
Figure BDA0001506359540000167
Thus, the
Figure BDA0001506359540000168
Non-compliant with positive domain non-subtractions;
meanwhile, independence test is not needed;
B＝{f8,f3,f6,f2,f7,f5a rough set α -a positive domain reduction subset that is not a steganography detection feature set H
The reduced subset is output. B ═ f8,f3,f6,f2,f7,f5}，CN′＝{8,3,6,2,7,5}。
The feature subset B is obtained after reduction. The potential (the number of the set elements) of the feature subset B is usually obviously lower than the dimension of the original feature, so that the time for extracting the feature is reduced, the pressure of a classifier can be obviously reduced by the low-dimension feature compared with the high-dimension feature, and the processing time of the classifier is shortened, so that the detection efficiency can be obviously improved by performing steganography detection based on the reduced feature.
It should be noted that, in the present specification, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (1)

1. A Rich Model steganography detection feature selection method based on decision rough set α -positive domain reduction is characterized by comprising the following steps:
step1, constructing the Rich Model steganography detection characteristics into a corresponding decision table, and specifically realizing the following processes:
step1.1, the extracted Rich Model high-dimensional characteristics are normalized,
Figure FDA0002245919970000011
Figure FDA0002245919970000012
wherein the content of the first and second substances,
Figure FDA0002245919970000013
and
Figure FDA0002245919970000014
is the value of the ith feature component of the jth carrier and the secret image; after processing, the data in the decision table are all discrete and have the value of [ -1, 1 [ -1 [ ]]Numerical data in between;
step1.2, determining a decision attribute value, wherein a decision attribute set is represented by 0 or 1, 0 represents a carrier image, and 1 represents a secret image;
step1.3, construct a secondA dimension decision table T, each row representing the values of a feature component of an image, each column representing the values of the same dimension feature component of the respective image,
Figure FDA0002245919970000015
a value corresponding to the ith feature component representing the jth carrier image,
Figure FDA0002245919970000016
representing the value corresponding to the ith characteristic component of the jth secret-carrying image, wherein j is more than or equal to 1 and less than or equal to M, i is more than or equal to 1 and less than or equal to N, and the last column represents the decision attribute of the image;
the decision table form of the steganography detection feature is as follows:
Figure FDA0002245919970000017
step2, adopting attribute separability of each characteristic component in the ASM measurement decision table, and taking the ASM value as a basis for sorting and dividing subsets, wherein the specific implementation process is as follows:
step2.1, calculating the average value of the characteristic components of the steganography detection; calculating the steganographic detection characteristic component f according to the value in the decision table TiMean value μ in the support and in the secret image+(fi) And mu-(fi)：
Figure FDA0002245919970000021
Figure FDA0002245919970000022
Step2.2, calculating the average value difference of the steganographic detection characteristic components; calculating the steganographic detection characteristic component f according to the result of the step2.1iAbsolute value μ of mean difference in carrier and secret images:
μ＝|μ+(fi)-μ-(fi)|；
step2.3, calculating steganographyMeasuring the standard deviation of the characteristic components; calculating the characteristic component f of steganography detection according to the values in the decision table T and the result of the step2.1iStandard deviation delta in carrier and secret images+(fi) And delta-(fi)：
Figure FDA0002245919970000023
Figure FDA0002245919970000024
Step2.4, calculating the sum of standard deviations of the steganographic detection characteristic components; calculating the steganographic detection characteristic component f according to the result of the step2.3iSum of standard deviations δ in the support and dense images:
δ＝δ+(fi)+δ-(fi)；
step2.5, calculating attribute separability; calculating the steganographic detection characteristic component f according to the calculation results of the step2.2 and the step2.4iValue of attribute separability of (1):
Figure FDA0002245919970000025
step2.6, output ASM (f)i)；
Step3, when the Rich Model steganography detection features are reduced based on the decision rough set α -positive domain reduction, sequentially adding a plurality of continuous candidate feature components with larger ASM values into the reduction subset, and simultaneously eliminating redundant and conflicting feature components by using the attribute independence principle of the decision rough set α -positive domain reduction, wherein the specific implementation process is as follows:
step 3.1, initializing; reduction subset of high-dimensional steganography detection features
Figure FDA0002245919970000031
Step 3.2, deleting low separability characteristic components; setting a lower limit ASM of an ASM valueminSuch asFruit ASM (f)i)<ASMminThen the steganographic detection feature component fiDeleting the low-separability feature components which are regarded as irrelevant feature components to obtain a set H' from which all the low-separability feature components are deleted;
step 3.3, calculating step length; dividing step size
Figure FDA0002245919970000032
Wherein λ is the step size, and m is the number of the expected feature subsets;
step 3.4, dividing subsets; sorting the feature components in H ' in descending order according to the ASM value, and dividing the sorted feature components into m feature subsets H ' { H ' (m is H) according to the value of the step length lambda1,h2,…,hmWherein the ith feature subset
Figure FDA0002245919970000033
t is the number of features in the subset;
step 3.5, selecting characteristics;
Figure FDA0002245919970000034
Figure FDA0002245919970000041
and 3.6, selecting the characteristic subset, and selecting a reduction subset with good classification effect and low dimension.
CN201711329640.7A 2017-12-13 2017-12-13 Rich model steganography detection feature selection method based on rough set α -positive domain reduction Active CN108009434B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711329640.7A CN108009434B (en) 2017-12-13 2017-12-13 Rich model steganography detection feature selection method based on rough set α -positive domain reduction

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711329640.7A CN108009434B (en) 2017-12-13 2017-12-13 Rich model steganography detection feature selection method based on rough set α -positive domain reduction

Publications (2)

Publication Number Publication Date
CN108009434A CN108009434A (en) 2018-05-08
CN108009434B true CN108009434B (en) 2020-04-24

Family

ID=62058693

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711329640.7A Active CN108009434B (en) 2017-12-13 2017-12-13 Rich model steganography detection feature selection method based on rough set α -positive domain reduction

Country Status (1)

Country Link
CN (1) CN108009434B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112950445B (en) * 2021-03-11 2023-11-17 河南师范大学 Compensation-based detection feature selection method in image steganalysis
CN113556439A (en) * 2021-06-08 2021-10-26 中国人民解放军战略支援部队信息工程大学 Rich Model steganography detection feature selection method based on feature component correlation
CN114627340B (en) * 2022-03-19 2024-04-30 河南师范大学 Image steganography detection feature self-adaptive selection method based on triple measurement

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103914839A (en) * 2014-03-27 2014-07-09 中山大学 Image stitching and tampering detection method and device based on steganalysis
CN106530200A (en) * 2016-10-23 2017-03-22 深圳大学 Deep-learning-model-based steganography image detection method and system

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103914839A (en) * 2014-03-27 2014-07-09 中山大学 Image stitching and tampering detection method and device based on steganalysis
CN106530200A (en) * 2016-10-23 2017-03-22 深圳大学 Deep-learning-model-based steganography image detection method and system

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
基于信息量的悲观多粒度粗糙集粒度简约;孟慧丽;《南京大学学报（自然科学）》;20150408;第343-348页 *
基于噪声模型和特征联合的PS图像与隐写图像检测;罗向阳等;《计算机学报》;20100830;第1060-1072页 *
基于富模型的分层量化隐写分析算法;张聪等;《计算机应用研究》;20160316;第255-265页 *
面向隐写分析的图像富模型特征的改进;赵宝琴等;《科学技术与工程》;20161219;第56-65页 *

Also Published As

Publication number Publication date
CN108009434A (en) 2018-05-08

Similar Documents

Publication Publication Date Title
CN108573225B (en) Partial discharge signal pattern recognition method and system
CN109949317A (en) Based on the semi-supervised image instance dividing method for gradually fighting study
CN110633725B (en) Method and device for training classification model and classification method and device
CN104661037B (en) The detection method and system that compression image quantization table is distorted
CN108009434B (en) Rich model steganography detection feature selection method based on rough set α -positive domain reduction
CN108229550B (en) Cloud picture classification method based on multi-granularity cascade forest network
CN107392919B (en) Adaptive genetic algorithm-based gray threshold acquisition method and image segmentation method
EP2657884A2 (en) Identifying multimedia objects based on multimedia fingerprint
CN111243601B (en) Voiceprint clustering method and device, electronic equipment and computer-readable storage medium
CN112819020A (en) Method and device for training classification model and classification method
CN110457996B (en) Video moving object tampering evidence obtaining method based on VGG-11 convolutional neural network
CN112950445B (en) Compensation-based detection feature selection method in image steganalysis
CN115459782A (en) Industrial Internet of things high-frequency data compression method based on time sequence segmentation and clustering
CN104504406B (en) A kind of approximate multiimage matching process rapidly and efficiently
CN111104555A (en) Video hash retrieval method based on attention mechanism
CN110059288B (en) System and method for obtaining an optimal mother wavelet for facilitating a machine learning task
Ge et al. Structure-based low-rank model with graph nuclear norm regularization for noise removal
CN111935487B (en) Image compression method and system based on video stream detection
CN110968845A (en) Detection method for LSB steganography based on convolutional neural network generation
CN111901594B (en) Visual analysis task-oriented image coding method, electronic device and medium
CN110110120B (en) Image retrieval method and device based on deep learning
CN114581926A (en) Multi-line text recognition method, device, equipment and medium
CN105224954B (en) It is a kind of to remove the topic discovery method that small topic influences based on Single-pass
CN104867493B (en) Multifractal Dimension end-point detecting method based on wavelet transformation
CN105740916B (en) Characteristics of image coding method and device

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