CN111476702B - Image steganography detection method and system based on nonlinear mixed kernel feature mapping - Google Patents
Image steganography detection method and system based on nonlinear mixed kernel feature mapping Download PDFInfo
- Publication number
- CN111476702B CN111476702B CN202010263650.0A CN202010263650A CN111476702B CN 111476702 B CN111476702 B CN 111476702B CN 202010263650 A CN202010263650 A CN 202010263650A CN 111476702 B CN111476702 B CN 111476702B
- Authority
- CN
- China
- Prior art keywords
- image
- detected
- statistical
- feature
- mapping
- 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.)
- Expired - Fee Related
Links
- 238000013507 mapping Methods 0.000 title claims abstract description 106
- 238000001514 detection method Methods 0.000 title claims abstract description 48
- 238000000034 method Methods 0.000 claims abstract description 35
- 238000000605 extraction Methods 0.000 claims abstract description 22
- 230000002194 synthesizing effect Effects 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 107
- 238000012549 training Methods 0.000 claims description 83
- 239000013598 vector Substances 0.000 claims description 35
- 238000002156 mixing Methods 0.000 claims description 27
- 235000008331 Pinus X rigitaeda Nutrition 0.000 claims description 12
- 235000011613 Pinus brutia Nutrition 0.000 claims description 12
- 241000018646 Pinus brutia Species 0.000 claims description 12
- 238000005457 optimization Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 10
- 230000015572 biosynthetic process Effects 0.000 claims description 3
- 238000003786 synthesis reaction Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 description 11
- 238000004458 analytical method Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000002829 reductive effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
Images
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/259—Fusion by voting
-
- 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 relates to an image steganography detection method and system based on nonlinear mixed kernel feature mapping, which comprises the following steps: the method comprises the steps of obtaining an image to be detected, extracting features of the image to be detected by a rich model type feature extraction method, determining statistical features of the image to be detected, longitudinally and sequentially dividing the statistical features of the image to be detected to obtain a statistical feature block set of the image to be detected, mapping each statistical feature block of the image to be detected in the statistical feature block set of the image to be detected by using a feature mapping algorithm based on a nonlinear mixed kernel function to obtain a mapped statistical feature block set of the image to be detected, splicing each mapped statistical feature block of the image to be detected in the mapped statistical feature block set of the image to be detected, synthesizing high-dimensional rich model features of the image to be detected, and determining whether the image to be detected contains steganography information or not by using a trained integrated FLD classifier according to the high-dimensional rich model features of. The method of the invention improves the detection precision without increasing time complexity.
Description
Technical Field
The invention relates to the technical field of image steganography, in particular to an image steganography detection method and system based on nonlinear mixed kernel feature mapping.
Background
The image steganography technology is to embed secret information into an image by utilizing insensitivity of human sense and redundancy existing in a signal, and image steganography analysis is a reverse technology of image steganography and mainly aims to detect a stego image, extract stego messages and the like.
The existing steganography analysis is mainly divided into two types, one type is special steganography analysis, analysis and test are carried out aiming at a known steganography algorithm, and the existing steganography analysis has the priori knowledge of the steganography algorithm, so that the detection accuracy rate is high, but the application range is narrow. The other type is general steganalysis, the data to be detected is judged on the premise of not knowing a steganographic algorithm, and the specific steganographic algorithm is unknown, so that the application range is wider, the generalization capability is stronger, the method is more suitable for practical application, but the complexity and the accuracy are low.
Disclosure of Invention
The invention aims to provide an image steganography detection method and system based on nonlinear mixed kernel feature mapping, and solves the problems of complex general steganography analysis and low accuracy in the prior art.
In order to achieve the purpose, the invention provides the following scheme:
an image steganography detection method based on nonlinear mixed kernel feature mapping, the image steganography detection method comprising:
acquiring an image to be detected;
performing feature extraction on the image to be detected by adopting a rich model type feature extraction method, and determining the statistical features of the image to be detected;
longitudinally and sequentially dividing the statistical characteristics of the image to be detected to obtain a statistical characteristic block set of the image to be detected; the to-be-detected image statistical feature block set comprises a plurality of to-be-detected image statistical feature blocks;
mapping each image statistical feature block to be detected in the image statistical feature block set to be detected by using a feature mapping algorithm based on a nonlinear mixed kernel function to obtain a mapped image statistical feature block set to be detected; the mapped image statistical feature block set comprises a plurality of mapped image statistical feature blocks to be detected;
splicing each mapped image statistical feature block in the mapped image statistical feature block set to be detected, and synthesizing high-dimensional rich model features of the image to be detected;
and determining whether the image to be detected contains steganographic information or not by adopting a trained integrated FLD classifier according to the characteristic of the high-dimensional richness model of the image to be detected.
Optionally, the performing feature extraction on the image to be detected by using the rich model type feature extraction method further includes, after determining the statistical features of the image to be detected:
and determining the optimized statistical characteristics of the image to be detected by using a pine growth optimization method according to the statistical characteristics of the image to be detected.
Optionally, the mapping is performed on each statistical feature block of the to-be-detected image in the statistical feature block set of the to-be-detected image by using a feature mapping algorithm based on a nonlinear mixed kernel function, so as to obtain a statistical feature block set of the to-be-detected image after mapping, which specifically includes:
constructing a nonlinear mixed kernel function according to two different single kernel functions;
acquiring statistical characteristics of a training image;
determining a characteristic value and a characteristic vector of a nonlinear mixing kernel function according to the statistical characteristics of the training image and the nonlinear mixing kernel function;
determining a feature mapping algorithm based on a nonlinear mixing kernel function according to the eigenvalue, the eigenvector and the nonlinear mixing kernel function;
and mapping each to-be-detected image statistical feature block in the to-be-detected image statistical feature block set by using the feature mapping algorithm based on the nonlinear mixed kernel function to obtain a mapped to-be-detected image statistical feature block set.
Optionally, the nonlinear mixing kernel function is k (z, · k)1(z,·)·k2(z, ·); wherein k (z,. cndot.) represents a nonlinear mixing kernel function, k1(z,. and k)2(z, ·) represents two different mononuclear functions, z representing a feature vector.
Optionally, k is1(z,. cndot.) is a Linear kernel function, k2(z,. cndot.) is the Hellinger kernel function.
Optionally, the training process of the trained integrated FLD classifier specifically includes:
acquiring training sample data;
performing feature extraction on the training sample data by adopting a rich model type feature extraction method to determine the statistical features of the training images;
determining the statistical characteristics of the optimized training images by using a pine growth optimization method according to the statistical characteristics of the training images;
longitudinally and sequentially segmenting the optimized statistical characteristics of the training image to obtain a training image characteristic block set;
mapping each training image feature block in the training image feature block set by using the feature mapping algorithm based on the nonlinear mixed kernel function to obtain a mapped training image feature block set;
splicing each post-mapping training image feature block in the post-mapping training image feature block set to synthesize training high-dimensional rich model features;
and training the integrated FLD classifier according to the training high-dimensional model features to obtain the trained integrated FLD classifier.
Optionally, the acquiring training sample data specifically includes:
acquiring a plurality of images without steganography information as original images, and recording as an image set without steganography information;
embedding steganographic information into the original image according to a preset embedding rate by using a steganographic algorithm to obtain an image set containing the steganographic information; the training sample data comprises an image set without steganographic information and an image set with steganographic information.
Optionally, the training process of the feature values and the feature vectors specifically includes:
and determining the characteristic value and the characteristic vector of the nonlinear mixing kernel function according to the training image characteristic block set and the nonlinear mixing kernel function.
An image steganalysis system based on non-linear mixed kernel feature mapping, said image steganalysis system comprising:
the to-be-detected image acquisition module is used for acquiring an image to be detected;
the statistical characteristic determining module of the image to be detected is used for extracting the characteristic of the image to be detected by adopting a rich model type characteristic extracting method and determining the statistical characteristic of the image to be detected;
the device comprises a to-be-detected image statistical feature block set acquisition module, a to-be-detected image statistical feature block acquisition module and a to-be-detected image statistical feature block acquisition module, wherein the to-be-detected image statistical feature block set acquisition module is used for longitudinally and sequentially segmenting the statistical features of the to-be-detected image to obtain a to-; the to-be-detected image statistical feature block set comprises a plurality of to-be-detected image statistical feature blocks;
the mapped to-be-detected image statistical feature block set acquisition module is used for mapping each to-be-detected image statistical feature block in the to-be-detected image statistical feature block set by using a feature mapping algorithm based on a nonlinear mixed kernel function to acquire a mapped to-be-detected image statistical feature block set; the mapped image statistical feature block set comprises a plurality of mapped image statistical feature blocks to be detected;
the to-be-detected image high-dimensional rich model feature synthesis module is used for splicing each mapped to-be-detected image statistical feature block in the mapped to-be-detected image statistical feature block set to synthesize to-be-detected image high-dimensional rich model features;
and the steganography information detection module is used for determining whether the image to be detected contains steganography information or not by adopting a trained integrated FLD classifier according to the characteristics of the high-dimensionality model of the image to be detected.
Optionally, the image steganography detection system further includes:
and the optimized statistical characteristic determining module of the image to be detected is used for determining the optimized statistical characteristic of the image to be detected by using a pine growth optimization method according to the statistical characteristic of the image to be detected.
Optionally, the module for acquiring the statistical feature block set of the mapped image to be detected specifically includes:
the nonlinear mixed kernel function constructing unit is used for constructing a nonlinear mixed kernel function according to two different single kernel functions;
the statistical characteristic acquisition unit of the training image is used for acquiring the statistical characteristic of the training image;
a characteristic value and characteristic vector determining unit, configured to determine a characteristic value and a characteristic vector of a nonlinear mixing kernel function according to the statistical characteristics of the training image and the nonlinear mixing kernel function;
the characteristic mapping algorithm determining unit is used for determining a characteristic mapping algorithm based on a nonlinear mixed kernel function according to the characteristic value, the characteristic vector and the nonlinear mixed kernel function;
and the mapped to-be-detected image statistical feature block set obtaining unit is used for mapping each to-be-detected image statistical feature block in the to-be-detected image statistical feature block set by using the feature mapping algorithm based on the nonlinear mixed kernel function to obtain the mapped to-be-detected image statistical feature block set.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides an image steganography detection method and system based on nonlinear mixed kernel feature mapping, which comprises the following steps: the method comprises the steps of obtaining an image to be detected, extracting features of the image to be detected by a rich model type feature extraction method, determining statistical features of the image to be detected, longitudinally and sequentially dividing the statistical features of the image to be detected to obtain a statistical feature block set of the image to be detected, mapping each statistical feature block of the image to be detected in the statistical feature block set of the image to be detected by using a feature mapping algorithm based on a nonlinear mixed kernel function to obtain a mapped statistical feature block set of the image to be detected, splicing each mapped statistical feature block of the image to be detected in the mapped statistical feature block set of the image to be detected, synthesizing high-dimensional rich model features of the image to be detected, and determining whether the image to be detected contains steganography information or not by using a trained integrated FLD classifier according to the high-dimensional rich model features of. The method of the invention improves the detection precision without increasing time complexity.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in 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 inventive exercise.
FIG. 1 is a flowchart of an image steganography detection method based on nonlinear mixed kernel feature mapping according to an embodiment of the present invention;
FIG. 2 is a diagram illustrating an original feature PE according to an embodiment of the present inventionDirAnd PE after exp-Hellinger and multi-kernel projectionexp-H、PEMultiError comparison graph of (1);
FIG. 3 is a graph illustrating the impact of different kernel function projections on classification performance provided by an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an image steganography detection system based on nonlinear mixed kernel feature mapping according to an embodiment of the present invention.
Detailed Description
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 only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide an image steganography detection method and system based on nonlinear mixed kernel feature mapping, and solves the problems of complex general steganography analysis and low accuracy in the prior art.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Fig. 1 is a flowchart of an image steganography detection method based on nonlinear mixed kernel feature mapping according to an embodiment of the present invention, as shown in fig. 1, the image steganography detection method according to the present invention includes:
and S101, acquiring an image to be detected.
And S102, extracting the features of the image to be detected by adopting a rich model type feature extraction method, and determining the statistical features of the image to be detected.
The rich model (SRM) type feature extraction method comprises a plurality of different spatial domain high-pass filters, noise residuals of different forms are obtained by using linear and nonlinear high-pass filtering templates, a co-occurrence matrix of residual image is obtained, maximum values of embedded change rates of the four residuals are added to enhance the contribution of a possibly modified area to statistical features, the noise residuals are modeled into improved rich model features by the co-occurrence matrix, the improved rich model features are marked as maxSRM, all co-occurrence scanning directions are replaced by the inclination directions'd 2', and finally the rich model features maxSRMd2 are extracted to serve as the statistical features of steganalysis.
And S102, determining the optimized statistical characteristics of the image to be detected by using a pine growth optimization method according to the statistical characteristics of the image to be detected.
And removing the characteristics without influence, namely removing irrelevant characteristics by utilizing a Pine Growth Optimization (PGO) based characteristic selection algorithm, reducing the characteristic dimension, and avoiding a ill-conditioned matrix in the characteristic vector problem in the process of generating the kernel matrix.
S103, longitudinally and sequentially segmenting the statistical characteristics of the image to be detected to obtain a statistical characteristic block set of the image to be detected; the to-be-detected image statistical feature block set comprises a plurality of to-be-detected image statistical feature blocks.
Specifically, the statistical features of the image to be measured are sequentially and longitudinally decomposed into feature blocks with a set number from the first column according to the feature dimension, the set number is selected to be 8 in the embodiment of the invention, that is, the statistical features of the image to be measured are sequentially and longitudinally decomposed into 8 submodels from the first column according to the feature dimension.
S104, mapping each image statistical feature block to be detected in the image statistical feature block set to be detected by using a feature mapping algorithm based on a nonlinear mixed kernel function to obtain a mapped image statistical feature block set to be detected, wherein the mapped image statistical feature block set to be detected comprises a plurality of mapped image statistical feature blocks to be detected, and the method specifically comprises the following steps:
4-1) constructing a nonlinear mixed kernel function from two different single kernel functions.
The construction process of the nonlinear mixed kernel function specifically comprises the following steps: constructing a nonlinear mixed kernel function according to two different single kernel functions to obtain the nonlinear mixed kernel function; the non-linear mixing kernel functionIs k (z, ·) k1(z,·)·k2(z, ·); wherein k (z,. cndot.) represents a nonlinear mixing kernel function, k1(z,. and k)2(z, ·) represents two different mononuclear functions, z representing a feature vector. In the embodiment of the invention, k is1(z,. cndot.) is a Linear kernel function, k2(z,. cndot.) is the Hellinger kernel function.
4-2) acquiring the statistical characteristics of the training images.
4-3) determining the characteristic value and the characteristic vector of the nonlinear mixing kernel function according to the statistical characteristics of the training image and the nonlinear mixing kernel function.
4-4) determining a feature mapping algorithm based on a nonlinear mixing kernel function according to the eigenvalue, the eigenvector and the nonlinear mixing kernel function.
The characteristic mapping algorithm determination process based on the nonlinear mixed kernel function in the invention comprises the following steps:
in the embodiment of the invention, a new nonlinear mixed kernel function is proposed by using an existing single kernel function, and then each feature block is mapped based on the new nonlinear mixed kernel, and the nonlinear mixed kernel function formula (1) is named as a Multikernel kernel.
Wherein k is1And k2Two different single kernel functions are respectively represented, in the embodiment, a linear kernel function and a Hellinger kernel function are selected as basis functions, and vectors x and y respectively represent a carrier image feature cover and a stego of a stego image; x and y are component values of the vectors x and y respectively; d denotes the feature dimension.
The proposed nonlinear mixed kernel function is demonstrated:
proposition: suppose k1And k2Is that
Phi is X → RNProves that the formula (1) is a legal kernel function.
And (3) proving that: consider S as oneSet of limit points { x1,x2,...,xLAnd assume K1And K2To be limited to kernel function k1And k2The corresponding kernel matrix at these points. Consider an arbitrary vector α ∈ R+For all α, if α' K α ≧ 0, then K is semi-positive. Is provided with
α′k1α≥0，α′k2α≥0
Then there is alpha' (k)1·k2)α≥0
Thus, K1·K2Is semi-positive, k1·k2Is a kernel function.
Thus, this type of hybrid kernel function is significant when both of the single kernel functions are positive-valued, and the constructed nonlinear hybrid kernel space cannot be simply viewed as a direct product of each of the basic kernel spaces, which has a more complex kernel space structure.
The nonlinear mixing kernel first normalizes the features, i.e., divides each component of the feature vector by the sum of the vector components.
Finding a transformation task, and finding a transformation task to enable the dot product of two transformed vectors to coincide with the evaluation of the kernels of the two transformed vectors, wherein the specific expression is as follows: given that the number of images used for mapping training is greater than the feature dimension (M ≧ D), the carrier image feature vector is used
To train the mapping phi, find the vector phi (x)(i))∈RMPhi is a mapping function, namely for i, j epsilon { 1.. M }, the transformation task is as follows:
k(x(i),x(j))≈φ(x(i))·φ(x(j)) (2)
then, converting into an optimization problem solution:
in the first step, using phia(x) Denotes φ (x) e R M1 is more than or equal to a and less than or equal to M, so that the difference between the two in the formula (2) is minimum,
constraint conditions are as follows:
the constraint equation (4) makes the M-dimensional features of the new feature space non-redundant, i.e. the transformed feature vectors are uncorrelated.
The second step is that: obtain the kernel matrix K ═ Ki,j)∈R+ M×MFeature vector of
Wherein Ki,j＝k(x(i),x(j)) Equation (5) is established using lagrange multiplication:
Since the feature mapping relies only on a small number of carrier image features and not on a specific steganographic scheme or embedded payload. Therefore, the kernel matrices K and (are) calculated from the designed kernel function
As a partial matrix of the kernel matrix K) a finite number of eigenvalues
And the feature vector phiaThis only needs to be computed once in the training set, which reduces the mapping time.
The third step: mapping
R+ D→RE(E.ltoreq.M) named based on nonlinear mixed kernels
Approximate mapping, E is the feature dimension retained after mapping, for any feature vector z ∈ R+ DComputing the mapped feature vector
As shown in formula (6):
wherein K (z, ·) ═ (K (z, x)(1)),...,k(z,x(M))) (7)
When E is equal to D, the same feature dimension is reserved before and after mapping; in constructing the map
In the process, in order to obtain the result with optimal performance, the feature dimension is kept unchanged before and after the feature mapping is adopted, and the feature vector is mapped based on the nonlinear mixed kernel
As shown in formula (8):
Only the third step is then needed to complete when mapping the test features. Inputting a test feature pair, directly mapping the transmitted data and parameters through a training set, and calculating by formulas (6), (7) and (8) to obtain the mapped features. Note that the kernel matrix K (z,. cndot.) needs to be computed using equation (7) in both the training set and the test set. The mapping takes the form of a closed kernel, with feature projection being much less time-complex than classifier training.
The corresponding kernel matrix is:
x and y are respectively the characteristic cover extracted from the carrier image and the characteristic stego extracted from the stego image, wherein xi,yiFor the component value of the corresponding feature, i is 1 … M, where M is the number of images.
The specific mononuclear function used for the mapping is as follows, where D represents the feature dimension and M represents the number of images used to train the mapping.
Linear nucleus:
hellinger nucleus:
exp-Linear and exp-Hellinger nuclei:
Jensen-Shannon nucleus:
4-5) mapping each image statistical feature block to be detected in the image statistical feature block set to be detected by using the feature mapping algorithm based on the nonlinear mixed kernel function to obtain the mapped image statistical feature block set to be detected.
And S105, splicing each mapped image statistical feature block in the mapped image statistical feature block set to be detected, and synthesizing the high-dimensional rich model features of the image to be detected.
Specifically, each group of mapped features is spliced according to the original sequence to synthesize the high-dimensional model features.
And S106, determining whether the image to be detected contains steganographic information or not by adopting a trained integrated FLD classifier according to the characteristic of the high-dimensional richness model of the image to be detected.
The training process of the trained integrated FLD classifier specifically comprises the following steps:
6-1) obtaining training sample data, wherein the training sample data comprises an image set without steganography information and an image set with steganography information.
And acquiring a plurality of images without steganography information as original images, and recording as an image set without steganography information. Specifically, carrier images with different contents and textures are selected from a BOSSBase (v1.01) image library as original images, the set number is 10000, each image is an 8-bit uncompressed grayscale image, the size is 512 × 512, and the format is pgm.
And embedding steganographic information into the original image according to a preset embedding rate by using a steganographic algorithm to obtain an image set containing the steganographic information. Specifically, the selected original image is preprocessed, including operations such as image gray level conversion, scaling, clipping and the like and storage formats, and then, three mainstream steganographic algorithms maxSRMd2_ WOW, maxSRMd2_ MVG and S _ UNIWARD are respectively adopted for each preprocessed image to embed steganographic information according to embedding rates of 0.1, 0.2, 0.3, 0.4, 0.5 and the like, so that an image containing the steganographic information is generated. Namely, each steganographic algorithm is adopted to embed steganographic information according to embedding rates of 0.1, 0.2, 0.3, 0.4, 0.5 and the like, so that a steganographic image library is manufactured. The steganography detection error increases along with the reduction of the embedding rate, so various different embedding rates are adopted for steganography, and various steganography images are prepared.
And 6-2) performing feature extraction on the training sample data by adopting a rich model type feature extraction method, and determining the statistical features of the training images.
Specifically, feature extraction is respectively carried out on the carrier image and the corresponding stego image to obtain a statistical feature cover of the training image corresponding to the carrier image and a statistical feature stego of the training image corresponding to the stego image, so that a feature pair for stego detection is formed, the feature extraction method is the most advanced feature extraction method of a rich model (SRM) type, and a complete statistical feature of 34671 dimension can be obtained by aggregating 106 sub-templates.
6-3) determining the statistical characteristics of the optimized training images by using a pine growth optimization method according to the statistical characteristics of the training images.
The pine growth optimization method is developed based on the gradual growth of pine trees, the statistical characteristics of training images obtained in the previous stage are used as input, then each row of characteristics of each column is compared, and if the characteristics of each row of a certain column are the same, the characteristics cannot be considered and removed; otherwise, the feature is selected.
6-4) longitudinally and sequentially segmenting the optimized statistical characteristics of the training image to obtain a training image characteristic block set.
Specifically, according to the statistical features of the training images and the features of the feature mapping algorithm based on the nonlinear mixed kernel function (the ray algorithm must satisfy that the number of images used for mapping is larger than the feature dimension), the statistical features of the optimized training images are longitudinally decomposed into 8 feature blocks in sequence from the first column according to the feature dimension. In the embodiment of the present invention, the statistical feature size of the training image is 10000 × 34671, (i.e., 10000 images, the extracted feature dimension is 34671), the new feature size after removing the irrelevant features is 10000 × 32016 dimensions, and the 32016-dimensional features are longitudinally decomposed into 8 feature blocks, that is, the size of each feature block for training and testing is 5000 × 4002, i.e., 5000 images, and the feature dimension is 4002.
6-5) mapping each training image feature block in the training image feature block set by using the feature mapping algorithm based on the nonlinear mixed kernel function to obtain a mapped training image feature block set.
6-6) splicing each post-mapping training image feature block in the post-mapping training image feature block set to synthesize and train the high-dimensional rich model features.
6-7) training the integrated FLD classifier according to the training high-dimensional rich model characteristics to obtain the trained integrated FLD classifier.
Specifically, the integrated FLD (flash linear inverse classifier) classifier votes by using a plurality of FLD sub-classifiers to generate a final decision result, thereby determining whether an image contains steganographic information.
The method provided by the invention is verified as follows:
table 1 shows the detection errors of the three steganographic schemes under different payloads, and the extracted features are classified directly by the FLD integrated classifier, and then classified after single kernel and nonlinear mixed kernel projection.
TABLE 1
The image steganography detection error rate formula is as follows:
PFAfalse alarm rate, normal samples are detected as the percentage of the number of dense samples to the number of normal samples; pMDThe omission ratio is that the number of the dense samples is detected as the percentage of the number of the normal samples to the number of the dense samples; pETo detect error rates.
Table 1 shows that the reclassification effect of the rich model features corresponding to the three steganography schemes after projection is better than that of the direct classification, for the maxSRMd2_ WOW steganography scheme, when the load is 0.3, 0.4, and 0.5bpac, the feature projection is reduced by 2% compared with the PE of the direct classification, the performance of the projection algorithm based on the nonlinear mixed kernel is further reduced compared with the performance of the single kernel projection algorithm, and when the load is 0.3bpac, the detection error rate is reduced by 1%. Compared with the projection algorithm of an exp-Hellinger kernel, the projection algorithm based on the nonlinear mixed kernel reduces the detection error rate by 3% when the effective load is 0.2bpac and reduces the detection error rate by 1.35% when the effective load is 0.4 pac. The maxSRMd2_ S _ UNIWARD steganography scheme also has better classification results based on nonlinear mixed kernel feature mapping than the first two results.
Table 2 experiment software and hardware platform
FIG. 2 is a diagram illustrating an original feature PE according to an embodiment of the present inventionDirAnd PE after exp-Hellinger and multi-kernel projectionexp-H、PEMultiThe difference between the errors in percent (multiplied by 100) compares better with the improvement in detection, and the results show that consistent detection enhancement is obtained in all four embedding algorithms and the maximum improvement in WOW is obtained.
Fig. 3 is a diagram illustrating the effect of different kernel function projections on classification performance according to an embodiment of the present invention, as shown in fig. 3, after the new non-linear mixed kernel function projection is adopted, the detection error rate is the lowest.
The present invention further provides an image steganography detection system based on nonlinear mixed kernel feature mapping, as shown in fig. 4, the image steganography detection system includes:
and the to-be-detected image acquisition module 1 is used for acquiring the to-be-detected image.
And the statistical characteristic determining module 2 of the image to be detected is used for extracting the characteristics of the image to be detected by adopting a rich model type characteristic extraction method and determining the statistical characteristics of the image to be detected.
The to-be-detected image statistical feature block set acquisition module 3 is used for longitudinally and sequentially segmenting the statistical features of the to-be-detected image to obtain a to-be-detected image statistical feature block set; the to-be-detected image statistical feature block set comprises a plurality of to-be-detected image statistical feature blocks.
The mapped to-be-detected image statistical feature block set acquisition module 4 is used for mapping each to-be-detected image statistical feature block in the to-be-detected image statistical feature block set by using a feature mapping algorithm based on a nonlinear mixed kernel function to obtain a mapped to-be-detected image statistical feature block set; the mapped image statistical feature block set to be tested comprises a plurality of mapped image statistical feature blocks to be tested.
And the to-be-detected image high-dimensional model feature synthesis module 5 is used for splicing each mapped to-be-detected image statistical feature block in the mapped to-be-detected image statistical feature block set to synthesize the to-be-detected image high-dimensional model feature.
And the steganography information detection module 6 is used for determining whether the image to be detected contains steganography information or not by adopting a trained integrated FLD classifier according to the characteristics of the high-dimensionality model of the image to be detected.
Preferably, the image steganography detection system further comprises:
and the optimized statistical characteristic determining module of the image to be detected is used for determining the optimized statistical characteristic of the image to be detected by using a pine growth optimization method according to the statistical characteristic of the image to be detected.
Preferably, the module for acquiring the statistical feature block set of the mapped image to be detected specifically includes:
and the nonlinear mixed kernel function constructing unit is used for constructing a nonlinear mixed kernel function according to two different single kernel functions.
And the statistical characteristic acquisition unit of the training image is used for acquiring the statistical characteristics of the training image.
And the characteristic value and characteristic vector determining unit is used for determining the characteristic value and the characteristic vector of the nonlinear mixing kernel function according to the statistical characteristics of the training image and the nonlinear mixing kernel function.
And the characteristic mapping algorithm determining unit is used for determining a characteristic mapping algorithm based on the nonlinear mixing kernel function according to the characteristic value, the characteristic vector and the nonlinear mixing kernel function.
And the mapped to-be-detected image statistical feature block set obtaining unit is used for mapping each to-be-detected image statistical feature block in the to-be-detected image statistical feature block set by using the feature mapping algorithm based on the nonlinear mixed kernel function to obtain the mapped to-be-detected image statistical feature block set.
MehdiBorouman et al use a machine learning paradigm to project extracted features, combining explicit nonlinear feature mapping with a simple classifier, and improving the accuracy of steganalysis detectors currently constructed as binary classifiers. The Hellinger kernel in an exponential form realizes the optimal performance, and improves the spatial domain content self-adaptive steganography detection precision by 2-3%. In the feature projection process, kernel selection is the key to improve the classification performance after approximate transformation. Due to the fixed format of a single kernel function and the relatively narrow space of change, the normalization capability and robustness have limitations. When the sample data contains heterogeneous information or the sample data has a high-dimensional feature space with a non-planar distribution, the performance of a single kernel function is not ideal.
Therefore, the invention provides an image steganography detection method and system based on nonlinear mixed kernel feature mapping, and the ideas of high-dimensional rich model feature segmentation, mapping and splicing are firstly put forward; in the mapping process, a new nonlinear mixed kernel function is constructed to replace a single kernel function to map the high-dimensional features so as to overcome the phenomena that the sample features contain heterogeneous information, the sample scale is huge and the data is unevenly distributed in the high-dimensional feature space, the generalization capability is improved through the flexible design of the multi-kernel function, the integrated classification performance is further improved, the detection precision can be further improved, the time complexity is not increased, and meanwhile, the problem of high operating memory is also solved by adopting a packet projection method for the high-dimensional features.
Meanwhile, the mixed kernel function constructed by the invention is simple, the calculated amount is small, the performance of the classifier can be effectively improved, and the main contributions are as follows: firstly, grouping, projecting and reclassifying the characteristics of the high-dimensional model, solving the problem that tens of thousands of dimensional high-dimensional characteristics cannot be directly projected, and reducing the memory requirement; secondly, a new nonlinear mixed kernel function improved feature projection algorithm is proposed, so that the performance is further improved.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (10)
1. An image steganography detection method based on nonlinear mixed kernel feature mapping is characterized in that the image steganography detection method comprises the following steps:
acquiring an image to be detected;
performing feature extraction on the image to be detected by adopting a rich model type feature extraction method, and determining the statistical features of the image to be detected;
longitudinally and sequentially dividing the statistical characteristics of the image to be detected to obtain a statistical characteristic block set of the image to be detected; the to-be-detected image statistical feature block set comprises a plurality of to-be-detected image statistical feature blocks;
mapping each image statistical feature block to be detected in the image statistical feature block set to be detected by using a feature mapping algorithm based on a nonlinear mixed kernel function to obtain a mapped image statistical feature block set to be detected; the mapped image statistical feature block set comprises a plurality of mapped image statistical feature blocks to be detected;
splicing each mapped image statistical feature block in the mapped image statistical feature block set to be detected, and synthesizing high-dimensional rich model features of the image to be detected;
and determining whether the image to be detected contains steganographic information or not by adopting a trained integrated FLD classifier according to the characteristic of the high-dimensional richness model of the image to be detected.
2. The image steganography detection method based on nonlinear mixed kernel feature mapping according to claim 1, wherein the performing feature extraction on the image to be detected by using a rich model type feature extraction method further comprises, after determining the statistical features of the image to be detected:
and determining the optimized statistical characteristics of the image to be detected by using a pine growth optimization method according to the statistical characteristics of the image to be detected.
3. The image steganography detection method based on nonlinear mixed kernel feature mapping according to claim 1, wherein the mapping is performed on each to-be-detected image statistical feature block in the to-be-detected image statistical feature block set by using a feature mapping algorithm based on a nonlinear mixed kernel function to obtain a mapped to-be-detected image statistical feature block set, specifically comprising:
constructing a nonlinear mixed kernel function according to two different single kernel functions;
acquiring statistical characteristics of a training image;
determining a characteristic value and a characteristic vector of a nonlinear mixing kernel function according to the statistical characteristics of the training image and the nonlinear mixing kernel function;
determining a feature mapping algorithm based on a nonlinear mixing kernel function according to the eigenvalue, the eigenvector and the nonlinear mixing kernel function;
and mapping each to-be-detected image statistical feature block in the to-be-detected image statistical feature block set by using the feature mapping algorithm based on the nonlinear mixed kernel function to obtain a mapped to-be-detected image statistical feature block set.
4. The image steganography detection method based on nonlinear mixed kernel feature mapping of claim 3, wherein the nonlinear mixed kernel function is k (z,) k 1(z,) k2(z,); where k (z,) represents a nonlinear mixed kernel function, k 1(z,) and k2(z,) represent two different single kernel functions, and z represents a feature vector.
5. The image steganography detection method based on nonlinear mixed kernel feature mapping according to claim 4, wherein k 1(z, ·) is a Linear kernel function, and k2(z, ·) is a Hellinger kernel function.
6. The image steganography detection method based on nonlinear mixed kernel feature mapping according to claim 1, wherein the training process of the trained integrated FLD classifier specifically comprises:
acquiring training sample data;
performing feature extraction on the training sample data by adopting a rich model type feature extraction method to determine the statistical features of the training images;
determining the statistical characteristics of the optimized training images by using a pine growth optimization method according to the statistical characteristics of the training images;
longitudinally and sequentially segmenting the optimized statistical characteristics of the training image to obtain a training image characteristic block set;
mapping each training image feature block in the training image feature block set by using the feature mapping algorithm based on the nonlinear mixed kernel function to obtain a mapped training image feature block set;
splicing each post-mapping training image feature block in the post-mapping training image feature block set to synthesize training high-dimensional rich model features;
and training the integrated FLD classifier according to the training high-dimensional model features to obtain the trained integrated FLD classifier.
7. The image steganography detection method based on nonlinear mixed kernel feature mapping according to claim 6, wherein the acquiring training sample data specifically includes:
acquiring a plurality of images without steganography information as original images, and recording as an image set without steganography information;
embedding steganographic information into the original image according to a preset embedding rate by using a steganographic algorithm to obtain an image set containing the steganographic information; the training sample data comprises an image set without steganographic information and an image set with steganographic information.
8. An image steganography detection system based on nonlinear mixed kernel feature mapping, the image steganography detection system comprising:
the to-be-detected image acquisition module is used for acquiring an image to be detected;
the statistical characteristic determining module of the image to be detected is used for extracting the characteristic of the image to be detected by adopting a rich model type characteristic extracting method and determining the statistical characteristic of the image to be detected;
the device comprises a to-be-detected image statistical feature block set acquisition module, a to-be-detected image statistical feature block acquisition module and a to-be-detected image statistical feature block acquisition module, wherein the to-be-detected image statistical feature block set acquisition module is used for longitudinally and sequentially segmenting the statistical features of the to-be-detected image to obtain a to-; the to-be-detected image statistical feature block set comprises a plurality of to-be-detected image statistical feature blocks;
the mapped to-be-detected image statistical feature block set acquisition module is used for mapping each to-be-detected image statistical feature block in the to-be-detected image statistical feature block set by using a feature mapping algorithm based on a nonlinear mixed kernel function to acquire a mapped to-be-detected image statistical feature block set; the mapped image statistical feature block set comprises a plurality of mapped image statistical feature blocks to be detected;
the to-be-detected image high-dimensional rich model feature synthesis module is used for splicing each mapped to-be-detected image statistical feature block in the mapped to-be-detected image statistical feature block set to synthesize to-be-detected image high-dimensional rich model features;
and the steganography information detection module is used for determining whether the image to be detected contains steganography information or not by adopting a trained integrated FLD classifier according to the characteristics of the high-dimensionality model of the image to be detected.
9. The non-linear mixed-kernel feature mapping-based image steganography detection system of claim 8, further comprising:
and the optimized statistical characteristic determining module of the image to be detected is used for determining the optimized statistical characteristic of the image to be detected by using a pine growth optimization method according to the statistical characteristic of the image to be detected.
10. The image steganography detection system based on nonlinear mixed kernel feature mapping according to claim 8, wherein the module for acquiring the statistical feature block set of the image to be detected after mapping specifically comprises:
the nonlinear mixed kernel function constructing unit is used for constructing a nonlinear mixed kernel function according to two different single kernel functions;
the statistical characteristic acquisition unit of the training image is used for acquiring the statistical characteristic of the training image;
a characteristic value and characteristic vector determining unit, configured to determine a characteristic value and a characteristic vector of a nonlinear mixing kernel function according to the statistical characteristics of the training image and the nonlinear mixing kernel function;
the characteristic mapping algorithm determining unit is used for determining a characteristic mapping algorithm based on a nonlinear mixed kernel function according to the characteristic value, the characteristic vector and the nonlinear mixed kernel function;
and the mapped to-be-detected image statistical feature block set obtaining unit is used for mapping each to-be-detected image statistical feature block in the to-be-detected image statistical feature block set by using the feature mapping algorithm based on the nonlinear mixed kernel function to obtain the mapped to-be-detected image statistical feature block set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010263650.0A CN111476702B (en) | 2020-04-07 | 2020-04-07 | Image steganography detection method and system based on nonlinear mixed kernel feature mapping |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010263650.0A CN111476702B (en) | 2020-04-07 | 2020-04-07 | Image steganography detection method and system based on nonlinear mixed kernel feature mapping |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111476702A CN111476702A (en) | 2020-07-31 |
| CN111476702B true CN111476702B (en) | 2021-02-02 |
Family
ID=71749829
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010263650.0A Expired - Fee Related CN111476702B (en) | 2020-04-07 | 2020-04-07 | Image steganography detection method and system based on nonlinear mixed kernel feature mapping |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111476702B (en) |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060239501A1 (en) * | 2005-04-26 | 2006-10-26 | Verance Corporation | Security enhancements of digital watermarks for multi-media content |
| US7885470B2 (en) * | 2007-01-19 | 2011-02-08 | New Jersey Institute Of Technology | Method and apparatus for steganalysis for texture images |
| CN100440256C (en) * | 2007-03-29 | 2008-12-03 | 上海大学 | Digital image LSB substitution information hiding rapid detecting method |
| CN102411771B (en) * | 2011-08-03 | 2013-02-13 | 北京航空航天大学 | Reversible image steganalysis method based on histogram peak value fluctuation quantity |
| CN103914839B (en) * | 2014-03-27 | 2017-02-15 | 中山大学 | Image stitching and tampering detection method and device based on steganalysis |
| CN106250578A (en) * | 2016-06-05 | 2016-12-21 | 乌鲁木齐职业大学 | Coal mine gas detection method |
| CN106096649B (en) * | 2016-06-08 | 2019-08-06 | 北京科技大学 | Sense of taste inductive signal otherness feature extracting method based on core linear discriminant analysis |
| US10699358B2 (en) * | 2018-02-22 | 2020-06-30 | Mcafee, Llc | Image hidden information detector |
| CN110362683A (en) * | 2019-06-26 | 2019-10-22 | 五邑大学 | A kind of information steganography method based on recurrent neural network, device and storage medium |
-
2020
- 2020-04-07 CN CN202010263650.0A patent/CN111476702B/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN111476702A (en) | 2020-07-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Simmonds et al. | Is more data always better? A simulation study of benefits and limitations of integrated distribution models | |
| CN110706302B (en) | System and method for synthesizing images by text | |
| CN110119753B (en) | Lithology recognition method by reconstructed texture | |
| CN111027576B (en) | Cooperative significance detection method based on cooperative significance generation type countermeasure network | |
| Singh et al. | Steganalysis of digital images using deep fractal network | |
| CN113627482A (en) | Cross-mode image generation method and device based on audio-tactile signal fusion | |
| Li et al. | Detecting double JPEG compression and its related anti-forensic operations with CNN | |
| CN110619347A (en) | Image generation method based on machine learning and method thereof | |
| KR20210034462A (en) | Method for training generative adversarial networks to generate per-pixel annotation | |
| CN114783034A (en) | Facial expression recognition method based on fusion of local sensitive features and global features | |
| CN116363489A (en) | Copy-paste tampered image data detection method, device, computer and computer-readable storage medium | |
| JP2001326811A (en) | Discrimination for dividing digital signal to insert electronic watermark signal and insertion relating thereto | |
| Zhao et al. | Universal embedding strategy for batch adaptive steganography in both spatial and JPEG domain | |
| He et al. | One-way or two-way factor model for matrix sequences? | |
| CN115147601A (en) | Urban street point cloud semantic segmentation method based on self-attention global feature enhancement | |
| CN111476702B (en) | Image steganography detection method and system based on nonlinear mixed kernel feature mapping | |
| CN116228753B (en) | Tumor prognosis evaluation method, device, computer equipment and storage medium | |
| CN105046286A (en) | Supervision multi-view feature selection method based on automatic generation of view and unit with l1 and l2 norm minimization | |
| Zhong et al. | A novel steganalysis method with deep learning for different texture complexity images | |
| CN113111906B (en) | Method for generating confrontation network model based on condition of single pair image training | |
| CN116309465A (en) | Tongue image detection and positioning method based on improved YOLOv5 in natural environment | |
| CN109902720A (en) | The image classification recognition methods of depth characteristic estimation is carried out based on Subspace Decomposition | |
| CN114494387A (en) | Data set network generation model and fog map generation method | |
| Quan et al. | Unsupervised deep learning for phase retrieval via teacher-student distillation | |
| Wu et al. | A Steganalysis framework based on CNN using the filter subset selection method |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210202 |