CN102147913A - Steganalysis method based on image smoothness variation characteristics - Google Patents

Abstract

The invention discloses a steganalysis method based on image smoothness variation characteristics, which comprises a process for extracting image smoothness and a process for extracting image smoothness variation characteristics, wherein the process for extracting the image smoothness comprises a segment covering step and a smoothness calculating step; the segment covering step comprises: covering the segment in a zero-degree direction, covering the segment in a 45-degree direction, covering the segment in a 90-degree direction and covering the segment in a 135-degree direction, selecting parameters after the segments are covered and calculating the image smoothness; the process for extracting the image smoothness variation characteristics comprises the following four steps: calculating the image smoothness characteristics before writing hidden information, writing hidden information by using the steganography to be analyzed, calculating the image smoothness characteristics after writing hidden information and calculating the smoothness variation characteristics. In the method, the hidden information writing of a grey image based on least significant bit (LSB) and histogram modification can be analyzed and the accuracy of the method is high for the two steganographies.

Description

A kind of steganalysis method based on Image Smoothness variable quantity feature

Technical field

The invention belongs to the Information hiding detection range, particularly relate to a kind of steganalysis method based on Image Smoothness variable quantity feature.

Background technology

Since early 1990s, Information hiding has caused the attention of international academic community as the important topic in the information security.At first be the sharply intensification of digital watermarking research to protection media product copyright, the paper of publishing rises year by year, and the company that much develops the digital watermarking product arises at the historic moment.In more than ten years, the struggle of forward concealing technology and inverse detection technology grows in intensity, and becomes the research focus that information security field attracts people's attention in the past.

Because the present information concealing technology is a kind of new covert communications technology, thereby is with a wide range of applications in military affairs, safety, communication and business circles, information steganography technology (Steganography) correspondingly obtains paying attention to day by day.The information steganography technology is the important branch of Information Hiding Techniques, has occurred some simple steganography methods very early.Begun extensive exploration to this field to the turn of the century.Latent writing is with normal digital carrier on the surface, as shielding, embed secret information as image, Voice ﹠ Video etc. therein, hiding data neither change the audio visual effect of carrier signal, do not change the size and the form of computer documents yet, thereby can realize lost covert communications.Carry close medium and mix with a large amount of normal media articles usually, particularly the internet is propagated through various channels.What be different from conventional cipher communication is that " ongoing communication " this fact also has been hidden itself, thereby can be used for the safe transfer of important messages.In recent years, along with being extensive use of of Internet technology, latent writing technology has obtained flourish.This technology is adopted by the important department that military establishment, government department, financial institution etc. relate to national economy.Yet information steganography also is a double-edged sword, and hostile force, terrorist etc. also can utilize it to be engaged in the unlawful activities that destroy social stability, harm national security.Occurred thousands of kinds of Information hiding softwares on the internet at present, using these softwares does not need advanced professional knowledge, and related data shows: some terroristic organizations, illegal group and hostile force once utilized Information Hiding Techniques to carry out hidden transmission.This shows, carry out the research of Information hiding detection technique,,, have very important practical sense safeguarding country and army's information security to finding, follow the tracks of in the overt channel based on the lawbreaking activities of covert communications.

Till now, at the existing many steganography methods of the different bearer type of image, wherein LSB (least significant bit) is latent writes appearance the earliest, because its method is simple, data volume is widely used greatly.A large amount of research has been carried out in the latent analysis of writing at LSB simultaneously, many effective analytical approachs occurred.In order to improve security, the researcher improves at latent the writing constantly of LSB, has proposed the LSB secret writing method of multiple correction, can resist some steganalysis method.

In existing research, reversible graph mainly can be divided into spatial domain method and transform domain method as information steganography method according to the difference that embeds the territory.Wherein method application in spatial domain is comparatively extensive, and reversible information steganography method the earliest is to be proposed in 1997 by Barton.Along with going deep into of research, scholars propose a large amount of spatial domain reversible information steganography methods in recent years, these methods are divided into roughly that the reversible information that uses data compression is latent to be write, write based on the reversible information of histogram modification is latent, and (Histogram Modification HM) and based on the reversible letter of difference expansion method such as writes.By changing the histogram of image, with the secret information embedded images, these class methods have higher picture quality based on the reversible steganography method of histogram modification.

Write correspondingly with latent, the purpose of steganalysis is to disclose the hidden latent existence of writing in the medium, even just points out to exist in the medium dubiety of latent write information.Reversible graph is the new problem of information security field proposition over the past two years as the information steganography analysis.At present still be in the starting stage as the steganalysis of steganography method about reversible graph, the correlative study achievement is less, mainly comprise steganalysis based on characteristic statistics, as RCM (reversible contrast mapping) reversible graph as steganalysis, based on carrying the unusual steganalysis of close characteristics of image, as based on the reversible graph of histogram displacement as steganalysis etc., but from detecting on the meaning, the latent writing detection method of parts of traditional is suitable equally for reversible steganalysis.

At present, according to steganalysis at steganography method divide, can be divided into specific steganalysis method and universal blind checking method at ad hoc approach.A kind of specific steganalysis method is only analyzed a kind of specific steganography method, has higher accuracy rate, but after this kind steganography method improves, steganalysis method at this steganography method is just no longer suitable, after for example classical LSB steganography method proposes, chi-square analysis method at classical LSB steganography method has appearred, RS steganalysis method etc., after this, improved LSB steganography method is suggested to resist these at the latent steganalysis method of writing of classical LSB, histogram compensation secret writing for example, minimum histogram distortion (least histogram abnormality, LHA) secret writing etc.And the general steganalysis method is because its versatility, and the steganography method of analysis is more, but accuracy rate is also lower simultaneously, at different steganography methods, has different effects usually.

Summary of the invention

The technical problem to be solved in the present invention is: overcome the deficiencies in the prior art, a kind of steganalysis method based on Image Smoothness variable quantity feature is provided, combine the advantage of specific steganography method and blind checking method, have certain universality, have higher accuracy rate simultaneously.

The technical solution adopted for the present invention to solve the technical problems: a kind of steganalysis method based on Image Smoothness variable quantity feature comprises Image Smoothness extracting method and Image Smoothness variable quantity feature extraction two large divisions.The Image Smoothness extracting method comprises line segment covering method and smoothness calculating two parts.The line segment covering method comprises that 0 degree direction line segment covers, 45 degree direction line segments cover, 90 degree direction line segments cover and 135 degree direction line segments cover, and choose parameter and computed image smoothness after line segment covers.The feature extraction of smoothness variable quantity comprises the computed image smoothness, with steganography method to be analyzed image is concealed and writes, and calculates latent back Image Smoothness and four steps of calculating smoothness variable quantity feature write.After extracting smoothness variable quantity feature, utilize support vector machine that the feature of training sample is carried out model training and treat the test sample eigen classifying, the support vector machine technology belongs to known technology, the present invention mainly introduces smoothness variable quantity characteristic extraction procedure, and the smoothness extracting method is the used a kind of basic operation of smoothness variable quantity feature extraction.

0 degree direction line segment overwrite procedure is:

(1), obtains one-dimension array P0 by the line direction whole gray level image of lining by line scan;

(2) the line segment capacity C is set, each pixel among the array P0 is also rounded downwards divided by C;

(3) cover adjacent pixel among the one-dimension array P0 with line segment successively with same pixel value;

(4) frequency of statistics different length line segment obtains frequency array f0.

45 degree direction line segment overwrite procedures are:

(1) by the whole gray level image of 45 degree direction zigzag scans, obtains one-dimension array P45;

(2) the line segment capacity C is set, each pixel among the array P45 is also rounded downwards divided by C;

(3) cover adjacent pixel among the one-dimension array P45 with line segment successively with same pixel value;

(4) frequency of statistics different length line segment obtains frequency array f45.

90 degree direction line segment overwrite procedures are:

(1) presses column direction by the whole gray level image of column scan, obtain one-dimension array P90;

(2) the line segment capacity C is set, each pixel among the array P90 is also rounded downwards divided by C;

(3) cover adjacent pixel among the one-dimension array P90 with line segment successively with same pixel value;

(4) frequency of statistics different length line segment obtains frequency array f90.

135 degree direction line segment overwrite procedures are:

(1) by the whole gray level image of the 135 degree reverse zigzag scans of direction, obtains one-dimension array P135;

(2) the line segment capacity C is set, each pixel among the array P135 is also rounded downwards divided by C;

(3) cover adjacent pixel among the one-dimension array P135 with line segment successively with same pixel value;

(4) frequency of statistics different length line segment obtains frequency array f135.

Image Smoothness computation process is:

Line segment for 4 directions covers, difference computed image smoothness, and computation process is:

(1) selected threshold K, the line segment that expression length is not less than K is long line segment;

(2) calculate the shared ratio of long line segment, this ratio is Image Smoothness, and computing formula is:

$S_{\mathrm{dir}}\left(K\right)=\frac{{\mathrm{\Σ}}_{i=K}^{t_{\mathrm{dir}}}{f}_{\mathrm{dir}}\left(i\right)\×i}{{\mathrm{\Σ}}_{i=1}^{t_{\mathrm{dir}}}{f}_{\mathrm{dir}}\left(i\right)\×i},$ dir＝0，45，90，135，K＝3，4，5

Sdir (K) is the Image Smoothness of image under dir direction threshold k, and tdir is the length of nose section during line segment covers on the dir direction, and the dir direction is 0 degree direction, 45 degree directions, 90 degree directions and 135 degree directions.

Smoothness variable quantity characteristic extraction procedure is:

(1) calculates piece image on 0 degree direction, 45 degree directions, 90 degree directions, 135 degree directions, the smoothness S1 under the different threshold k;

(2) with steganography method to be analyzed image is concealed and write;

(3) calculate the latent back image of writing on 0 degree direction, 45 degree directions, 90 degree directions, 135 degree directions, the smoothness S ' under the different threshold k;

(4) calculate the latent variable quantity of writing each smoothness of back image, this variable quantity is as feature, and computing formula is:

$R_{\mathrm{dir}}\left(K\right)=\frac{S_{\mathrm{dir}}\left(K\right)-{S_{\mathrm{dir}}}^{,}\left(K\right)}{S_{\mathrm{dir}}\left(K\right)},$ dir＝0，45，90，135，

Rdir (K) be image on the dir direction, the Image Smoothness variable quantity under the threshold k, Sdir (K) for latent writing before the smoothness of image on the dir direction, Sdir ' (K) is the latent smoothness of back image on the dir direction of writing.

The advantage that the present invention is compared with prior art had is:

(1) the steganalysis method that proposes of the present invention is applicable to analysis based on the steganography method of LSB with based on the steganography method of histogram modification, rather than only at the steganalysis of certain single steganography method, adaptability is strong;

(2) the steganalysis method of the present invention's proposition is compared with universal blind checking method, has higher accuracy rate;

(3) the present invention covers by the line segment of four direction, can calculate and obtain the Image Smoothness of image on four direction, for some latent steganography method that has direction to select problem in the process of writing, has good analytical effect;

(4) the Image Smoothness computing method of the present invention's extraction are flexible and changeable, by adjusting line segment capacity C and threshold k, can obtain a plurality of Image Smoothness, to satisfy different application requirements.

Description of drawings

Fig. 1 is the input of the eigenwert among the present invention synoptic diagram;

Fig. 2 is the testing image file analysis synoptic diagram among the present invention;

Fig. 3 is the smoothness variable quantity feature extraction synoptic diagram among the present invention;

Fig. 4 among the present invention along the 0 degree synoptic diagram of lining by line scan;

Fig. 5 covers synoptic diagram for the one dimension image array line segment among the present invention;

Fig. 6 among the present invention along 45 degree zigzag scan synoptic diagram;

Fig. 7 is spending by the column scan synoptic diagram along 90 among the present invention;

Fig. 8 among the present invention along the reverse zigzag scan synoptic diagram of 135 degree.

Embodiment

Framework such as Fig. 1 and Fig. 2 of the steganalysis method that covers based on line segment of the present invention: extract the latent eigenwert of the smoothness variable quantity feature of sample image file and clean sample audio file of writing, extract the smoothness variable quantity eigenwert input support vector machine of band altimetric image file as support vector machine; Support vector machine generates training pattern according to eigenwert, and the smoothness variable quantity feature of training testing image file contains latent write information when judging the testing image file.Support vector machine is called for short SVM, is a kind of of existing sorter, and the present invention chooses support vector machine to carry out sort operation, and the proper vector that criteria for classification adopts the smoothness variable quantity feature of image file to constitute.The present invention notices, writes and based on the latent latent write information of writing of histogram modification if contain in the image file based on LSB is latent, and through after latent the writing once more, the smoothness quantitative changeization of image is little; LSB is latent to be write and the latent latent write information of writing of histogram if do not contain in the image file, and through after latent the writing once more, the smoothness variable quantity of image is bigger.Therefore, the present invention proposes, extract the latent eigenwert of the smoothness variable quantity feature of sample image file and clean sample image of writing as support vector machine, support vector machine is the training vector feature with the smoothness variable quantity, can obtain the conclusion whether the testing image file contains latent write information according to the smoothness variable quantity feature of testing image file.In order to guarantee accuracy rate, the latent quantity of writing sample image file and clean sample image file should have thousands of, with a plurality of latent smoothness variable quantity feature input support vector machine of writing sample image file and clean sample image file, can generate training pattern.

No matter be to write sample image file and clean sample image file to latent, or the testing image file, unified smoothness variable quantity feature extraction mode adopted, the judgement of can conveniently classifying.Wherein, utilize support vector machine that the feature and the latent feature of writing the sample image file of clean sample image file are generated training pattern and utilize the feature of training pattern and testing image file that the testing image file is classified, belong to existing known technology.Referring to Fig. 3, key of the present invention is smoothness variable quantity Feature Extraction mode, comprise Image Smoothness computing method and Image Smoothness variable quantity feature extraction two large divisions, the smoothness computing method are the used a kind of basic operations of smoothness variable quantity feature extracting method.

One, Image Smoothness computing method

The Image Smoothness computing method adopt the line segment covering method, at first with line segment overlay image on 0 degree direction, 45 degree directions, 90 degree directions, 135 degree direction four directions, add up the frequency of different length line segment then, with the shared Image Smoothness of ratio presentation video on 0 degree direction, 45 degree directions, 90 degree directions, 135 degree direction four directions of long line segment, by adjusting the parameter in the Image Smoothness computing formula, can obtain a plurality of Image Smoothness.

0 degree direction line segment overwrite procedure is:

Step 1. by the 0 degree direction entire image of lining by line scan, obtains one dimension array of pixels P0 as shown in Figure 4, array length is the image pixel number, and for example: the gray level image two-dimensional array is IMG[m] [n], m is a picture altitude, n is a picture traverse, image pixel number len=m * n, and P0 is:

P0[1……len]＝[IMG[1][1……n]，IMG[2][1……n]，……，IMG[m][1……n]]；

It is C that step 2. is set the line segment capacity, and all pixel values among the one dimension array of pixels P0 are also rounded downwards divided by C, as follows:

Step 3. covers the adjacent pixel with same pixel value among the P0 with line segment successively, and as shown in Figure 5, the P0 line segment of a certain image covers (part);

Step 4. is added up each bar line segment length (the length of line segment is the number of the pixel of this line segment covering) successively, obtain line segment length array L0, L0[i] length of expression i bar line segment, the length of array L0 is lenl0, the length t0=max (L0[1 of nose section ... lenl0]; The frequency of statistics different length line segment obtains frequency array f0, and the length of f0 is t0, and computing formula is:

$f_0\left(i\right)=\underset{j=1}{\overset{{\mathrm{lenl}}_0}{\mathrm{\Σ}}}\mathrm{sign}\left(L_0\right[j]-i),$ Wherein:

$\mathrm{sign}\left(x\right)=\left\{\begin{array}{cc}1,& x=0\\ 0,& x\≠0\end{array}\right.,$ Lenl0 is the length of array L0, f0 (i) be illustrated in 0 the degree direction on length be the number of the line segment of i;

For example, parts of images shown in Figure 5 can calculate L0=[2, and 1,3,1,1,5,1,1,1 ... ], f0=[6,1,1,0,1 ... ].

45 degree direction line segment overwrite procedures are:

Step 1. as shown in Figure 6, from the upper left corner to the lower right corner, image is carried out zigzag scan by 45 degree directions, obtain one dimension array of pixels P45, array length is the image pixel number, for example: the gray level image two-dimensional array is IMG[m] [n], m is a picture altitude, n is a picture traverse, image pixel number len=m * n, P45 is:

P ₄₅=zigzag (IMG), wherein:

Zigzag (x) is the zigzag scan function, zigzag scan two-dimensional array x;

It is C that step 2. is set the line segment capacity, and all pixel values among the one dimension array of pixels P45 are also rounded downwards divided by C, as follows:

Step 3. covers the adjacent pixel with same pixel value among the P45 with line segment successively, and the P45 overwrite procedure is identical with the P0 overwrite procedure;

Step 4. is added up each bar line segment length (the length of line segment is the number of the pixel of this line segment covering) successively, obtain line segment length array L45, L45[i] length of expression i bar line segment, the length of array L45 is lenl45, length t45=max (the L45[1 of nose section ... lenl45], the frequency of statistics different length line segment obtains frequency array f45, the length of f45 is t45, and computing formula is:

$f_{45}\left[i\right]=\underset{j=1}{\overset{{\mathrm{lenl}}_{45}}{\mathrm{\Σ}}}\mathrm{sign}\left(L_{45}\right[j]-i),$ Wherein:

$\mathrm{sign}\left(x\right)=\left\{\begin{array}{cc}1,& x=0\\ 0,& x\≠0\end{array}\right.,$ Lenl45 is the length of array L45, f45 (i) be illustrated in 45 the degree directions on length be the number of the line segment of i.

90 degree direction line segment overwrite procedures are:

Step 1. by the column scan entire image, obtains one dimension array of pixels P90 by 90 degree directions as shown in Figure 7, array length is the image pixel number, and for example: the gray level image two-dimensional array is IMG[m] [n], m is a picture altitude, n is a picture traverse, image pixel number len=m * n, and P90 is:

P90[1……len]＝[IMG[1……m][1]，IMG[1……m][2]，……，IMG[1……m][n]]；

It is C that step 2. is set the line segment capacity, and all pixel values among the one dimension array of pixels P90 are also rounded downwards divided by C, as follows:

Step 3. covers the adjacent pixel with same pixel value among the P90 with line segment successively, and the P90 overwrite procedure is identical with the P0 overwrite procedure;

Step 4. is added up each bar line segment length (the length of line segment is the number of the pixel of this line segment covering) successively, obtain line segment length array L90, L90[i] length of expression i bar line segment, the length of array L90 is lenl90, the length t90=max (L90[1 of nose section ... lenl90]; The frequency of statistics different length line segment obtains frequency array f90, and line segment length is the number of the pixel of this line segment covering, and the length of f45 is t45, and computing formula is:

$f_{90}\left[i\right]=\underset{j=1}{\overset{{\mathrm{lenl}}_{90}}{\mathrm{\Σ}}}\mathrm{sign}\left(L_{90}\right[j]-i),$ Wherein:

$\mathrm{sign}\left(x\right)=\left\{\begin{array}{cc}1,& x=0\\ 0,& x\≠0\end{array}\right.,$ Lenl90 is the length of array L90, and length is the number of the line segment of i on f90 (i) the expression 90 degree directions.

135 degree direction line segment overwrite procedures are:

Step 1. as shown in Figure 8, from the upper right corner to the lower left corner, image is carried out reverse zigzag scan by 135 degree directions, obtain one dimension array of pixels P135, array length is the image pixel number, for example: the gray level image two-dimensional array is IMG[m] [n], m is a picture altitude, n is a picture traverse, image pixel number len=m * n, P135 is:

P ₁₃₅＝zigzag2(IMG)，

Wherein zigzag2 (x) is reverse zigzag scan function, oppositely zigzag scan two-dimensional array x;

It is C that step 2. is set the line segment capacity, and all pixel values among the one dimension array of pixels P135 are also rounded downwards divided by C, as follows:

Step 3. covers the adjacent pixel with same pixel value among the P135 with line segment successively, and the P135 overwrite procedure is identical with the P0 overwrite procedure;

Step 4. is added up each bar line segment length (the length of line segment is the number of the pixel of this line segment covering) successively, obtain line segment length array L135, L135[i] length of expression i bar line segment, the length of array L135 is lenl135, the length t135=max (L135[1 of nose section ... lenl135]; The frequency of statistics different length line segment obtains frequency array f135, and line segment length is the number of the pixel of this line segment covering, and the length of f135 is t135, and computing formula is:

$f_{135}\left[i\right]=\underset{j=1}{\overset{{\mathrm{lenl}}_{135}}{\mathrm{\Σ}}}\mathrm{sign}\left(L_{135}\right[j]-i),$ Wherein:

$\mathrm{sign}\left(x\right)=\left\{\begin{array}{cc}1,& x=0\\ 0,& x\≠0\end{array}\right.,$ Lenl135 is the length of array L135, and length is the number of the line segment of i on f135 (i) the expression 135 degree directions.

The smoothness computing method are on 0 degree direction, 45 degree directions, 90 degree directions, the 135 degree directions:

Step 1. setting threshold K, expression length is long line segment more than or equal to the line segment of K, and the K value is different, and long line segment number difference, the smoothness that calculates are also different;

Step 2. is calculated the shared ratio of long line segment, and the length that adds line segment during calculating is as weight, and computing formula is:

$S_{\mathrm{dir}}\left(K\right)=\frac{{\mathrm{\Σ}}_{i=K}^{t_{\mathrm{dir}}}{f}_{\mathrm{dir}}\left(i\right)\×i}{{\mathrm{\Σ}}_{i=1}^{t_{\mathrm{dir}}}{f}_{\mathrm{dir}}\left(i\right)\×i},$ dir＝0，45，90，135，K＝3，4，5，

Sdir (K) be image on the dir direction, the smoothness under the threshold k, dir is the direction at the smoothness place of being calculated, tdir is the length of nose section during line segment covers on the dir direction;

The smoothness computing formula is respectively on the 0 degree direction:

K=3,4,5,

The smoothness computing formula is respectively on the 45 degree directions:

K=3,4,5,

The smoothness computing formula is respectively on the 90 degree directions: K=3,4,5,

The smoothness computing formula is respectively on the 135 degree directions: K=3,4,5;

During line segment in this example on 0 degree direction, 45 degree directions, 90 degree directions, the 135 degree directions covers, the line segment capacity C at first is set, the line segment capacity C is being controlled the quantity that line segment covers the long line segment in back, the duration line segment is more greatly as C, when long line segment more after a little while, the smoothness numerical value that calculates is less, is unfavorable for being used for analyzing; C is unified in this example is set at 1, and when being used for other steganalysis method, if C is that 1 duration line segment is less, smoothness numerical value hour can change the C value; Threshold k also influences the quantity of long line segment, the K value hour, long line segment is more, smoothness numerical value is bigger, the K value is 3,4,5 in this example.

Two, Image Smoothness variable quantity feature extraction

The feature extraction of Image Smoothness variable quantity utilizes the Image Smoothness computing method, and the variable quantity of computed image before and after latent writing is as feature, and leaching process is:

Step 1. is calculated piece image on 0 degree direction, 45 degree directions, 90 degree directions and 135 degree directions, the smoothness S under the different threshold k:

0 degree directional smoothing degree: K=3,4,5,

45 degree directional smoothing degree:

K=3,4,5,

90 degree directional smoothing degree:

K=3,4,5,

135 degree directional smoothing degree: K=3,4,5;

Step 2. is concealed image with steganography method to be analyzed and is write, and latent write information be the information that produces at random, and the latent capacity of writing adopts latent the writing of steganography method single to be analyzed can conceal the max cap. of writing, and for example, LSB conceals that to write capacity be 1bpp;

Step 3. is calculated the latent back image of writing on 0 degree direction, 45 degree directions, 90 degree directions and 135 degree directions, the smoothness S ' under the different threshold k:

0 degree directional smoothing degree:

K=3,4,5,

45 degree directional smoothing degree:

K=3,4,5,

90 degree directional smoothing degree:

K=3,4,5,

135 degree directional smoothing degree:

K=3,4,5;

F0 ' wherein, f45 ', f90 ', f135 ' spends 45 degree, 90 degree for the latent back image of writing 0, line segment covers the frequency of back different length line segment on the 135 degree directions, t0 ', t45 ', t90 ', t135 ' spends 0 for the latent back image of writing, 45 degree, 90 degree, the length of nose section during line segment covers on the 135 degree directions;

Step 4. is calculated the latent variable quantity of writing each smoothness of back image, and this variable quantity is as feature, and computing formula is:

$R_{\mathrm{dir}}\left(K\right)=\frac{S_{\mathrm{dir}}\left(K\right)-{S_{\mathrm{dir}}}^{,}\left(K\right)}{S_{\mathrm{dir}}\left(K\right)},$ dir＝0，45，90，135，

Rdir (K) be image on the dir direction, the Image Smoothness variable quantity under the threshold k, Sdir (K) for latent writing before the smoothness of image on the dir direction, Sdir ' (K) is the latent smoothness of back image on the dir direction of writing;

0 degree directional smoothing degree variable quantity: $R_0\left(K\right)=\frac{S_0\left(K\right)-{S_0}^{,}\left(K\right)}{S_0\left(K\right)},$

45 degree directional smoothing degree variable quantities: $R_{45}\left(K\right)=\frac{S_{45}\left(K\right)-{S_{45}}^{,}\left(K\right)}{S_{45}\left(K\right)},$

90 degree directional smoothing degree variable quantities: $R_{90}\left(K\right)=\frac{S_{90}\left(K\right)-{S_{90}}^{,}\left(K\right)}{S_{90}\left(K\right)},$

135 degree directional smoothing degree variable quantities: $R_{135}\left(K\right)=\frac{S_{135}\left(K\right)-{S_{135}}^{,}\left(K\right)}{S_{135}\left(K\right)};$

The dir value is 0,45,90,135 in this example, and the K value is 3,4,5, so for a sub-picture, get 12 Image Smoothness variable quantities in this example as eigenwert.

The part that the present invention does not elaborate belongs to techniques well known.

Claims (3)

1. the steganalysis method based on Image Smoothness variable quantity feature is characterized in that: comprise Image Smoothness extracting method and Image Smoothness variable quantity feature extraction two large divisions; Described Image Smoothness extracting method comprises line segment covering method and Image Smoothness calculating two parts, the line segment covering method comprises that 0 degree direction line segment covers, 45 degree direction line segments cover, 90 degree direction line segments cover and 135 degree direction line segments cover, line segment for described 4 directions that obtain covers, respectively the computed image smoothness; The feature extraction of Image Smoothness variable quantity comprises the computed image smoothness, with steganography method to be analyzed image is concealed and writes, and calculates latent back Image Smoothness and four steps of calculating smoothness variable quantity feature write;

Described 0 degree direction line segment overwrite procedure is:

(1), obtains one-dimension array P0 by the line direction whole gray level image of lining by line scan;

(2) the line segment capacity C is set, each pixel among the array P0 is also rounded downwards divided by C;

(3) cover adjacent pixel among the one-dimension array P0 with line segment successively with same pixel value;

(4) frequency of statistics different length line segment obtains frequency array f0;

Described 45 degree direction line segment overwrite procedures are:

(1) by the whole gray level image of 45 degree direction zigzag scans, obtains one-dimension array P45;

(2) the line segment capacity C is set, each pixel among the array P45 is also rounded downwards divided by C;

(3) cover adjacent pixel among the one-dimension array P45 with line segment successively with same pixel value;

(4) frequency of statistics different length line segment obtains frequency array f45;

Described 90 degree direction line segment overwrite procedures are:

(1) presses column direction by the whole gray level image of column scan, obtain one-dimension array P90;

(2) the line segment capacity C is set, each pixel among the array P90 is also rounded downwards divided by C;

(3) cover adjacent pixel among the one-dimension array P90 with line segment successively with same pixel value;

(4) frequency of statistics different length line segment obtains frequency array f90;

Described 135 degree direction line segment overwrite procedures are:

(1) by the whole gray level image of the 135 degree reverse zigzag scans of direction, obtains one-dimension array P135;

(2) the line segment capacity C is set, each pixel among the array P135 is also rounded downwards divided by C;

(3) cover adjacent pixel among the one-dimension array P135 with line segment successively with same pixel value;

(4) frequency of statistics different length line segment obtains frequency array f135;

Described Image Smoothness computation process is:

Cover for the line segment of above-mentioned 4 directions, promptly 0 degree direction line segment covers, 45 degree direction line segments cover, 90 degree direction line segments cover and 135 degree direction line segments cover, computed image smoothness respectively, and computation process is:

(1) selected threshold K, the line segment that expression length is not less than K is long line segment;

(2) calculate the shared ratio of long line segment, this ratio is Image Smoothness, and the Image Smoothness computing formula is:

$S_{\mathrm{dir}}\left(K\right)=\frac{{\mathrm{\Σ}}_{i=K}^{t_{\mathrm{dir}}}{f}_{\mathrm{dir}}\left(i\right)\×i}{{\mathrm{\Σ}}_{i=1}^{t_{\mathrm{dir}}}{f}_{\mathrm{dir}}\left(i\right)\×i},$ dir＝0，45，90，135，K＝3，4，5

Sdir (K) is that image is on the dir direction, Image Smoothness under the threshold k, tdir is the length of nose section during line segment covers on the dir direction, and the dir direction is 0 degree direction, 45 degree directions, 90 degree directional rays and 135 degree directions, and fdir is the frequency of different length line segment on the dir direction;

Described Image Smoothness variable quantity characteristic extraction procedure is:

(1) calculates piece image on 0 degree direction, 45 degree directions, 90 degree directions, 135 degree directions, the smoothness S under the different threshold k;

(2) with steganography method to be analyzed image is concealed and write;

(3) calculate the latent back image of writing on 0 degree direction, 45 degree directions, 90 degree directions, 135 degree directions, the smoothness S ' under the different threshold k;

(4) calculate the latent variable quantity of writing each Image Smoothness of back, this variable quantity is as feature, and computing formula is:

$R_{\mathrm{dir}}\left(K\right)=\frac{S_{\mathrm{dir}}\left(K\right)-{S_{\mathrm{dir}}}^{,}\left(K\right)}{S_{\mathrm{dir}}\left(K\right)},$ dir＝0，45，90，135，

Rdir (K) be image on the dir direction, the Image Smoothness variable quantity under the threshold k, Sdir (K) for latent writing before the smoothness of image on the dir direction, Sdir ' (K) is the latent smoothness of back image on the dir direction of writing.

2. the steganalysis method based on Image Smoothness variable quantity feature according to claim 1 is characterized in that: the covering of direction line segments is spent in described 0 degree direction line segment covering, 45,90 degree direction line segments cover and the 135 line segment capacity C values of spending in the covering of direction line segments are 1,2,3,4.

3. the steganalysis method based on Image Smoothness variable quantity feature according to claim 1 is characterized in that: the threshold k value in the described Image Smoothness computing formula is 2,3,4,5,6,7.

Images