CN102609891B - Texture-characteristic-based method for passively and blindly obtaining evidence of digital image - Google Patents

Abstract

The invention discloses a texture-characteristic-based method for passively and blindly obtaining the evidence of a digital image. The method is characterized in that the image content is expressed by selecting six texture characteristic quantities, namely the image texture mean value, the texture standard deviation, the texture smoothness, the texture third moment, the texture consistency and the texture entropy. Compared with the existing method for expressing the information of the image content by selecting the indirect characteristic quantity of an image transform domain, the method is used for expressing the image information by directly selecting the texture characteristic of the image, so that the information of the image can be more adequately expressed; and compared with the existing method, the method can be used for reducing the dimensionality of the characteristic quantity, and the texture characteristic selected by the method has the characteristics of layer property, dimensional property, translation invariance and the like, so that the image falsifying and evidence obtaining detection rate and the robustness can be improved, and the method is applicable to the fields of identification on the content truth of images and the like.

Description

The passive blind evidence collecting method of a kind of digital picture based on textural characteristics

Technical field

The invention belongs to Signal and Information Processing technical field, be specifically related to copy-paste in image forge to distort the passive blind evidence collecting method based on textural characteristics.

Background technology

The arriving of digital Age, bringing the leading visual effect of enjoyment to people when, also makes some lawless persons by some forgeries of the broadcasting medias such as network and the image of distorting, to individual and society, cause negative effect.So digital Age also becomes one " double-edged sword " when being convenient for people to.It is a kind of the most common image forge means that digital picture copy-paste is distorted, and generally can be divided into two large classes: a class is that the copy-paste of same piece image is distorted to forgery; Another kind of is that the copy-paste between different images is distorted to forgery.

The method that the existing copy-paste for same piece image is distorted evidence obtaining has:

In the U.S. in 2004, reach the passive blind evidence collecting method of a kind of digital picture copy-paste based on principal component analysis (PCA) (PCA) proposing in special thatch thatch university computer science and technology report meeting, adopt image transform domain characteristic quantity to carry out presentation video content, exist the defect of poor robustness.

The copy-paste based on discrete cosine transform (DCT) coefficient that electric electronic engineering association (IEEE) information processing international conference the 53 volume second phase of the collected works 758-767 page of holding for 2005 proposes is distorted passive blind evidence collecting method, although realized the evidence obtaining that copying image-stickup is distorted, there is the defect that calculated amount is large.

The passive blind evidence collecting method in region is forged in the copying image based on small echo and svd that electric electronic engineering association (IEEE) multimedia messages of holding for 2007 and forensic technologies international conference collected works third phase 1750-1753 page propose, although further reduced the calculated amount of evidence collecting method, still cannot collect evidence fast for image larger in reality.

A kind of digital picture copy-paste based on main transfer vector that China's " Chinese journal of computers " the 30 volume ten first phase 1998-2007 pages in 2007 are mentioned is distorted passive blind evidence collecting method and can only be collected evidence to the distorted image of single form.

A kind of image zone duplicating and altering evidence collecting method based on gray scale symbiosis square of China's " computer utility " the 31 the 6th phase of volume 1621-1630 page proposition in 2011, exists calculated amount greatly and the defect of poor robustness.

Summary of the invention

The object of the invention is to propose the passive blind evidence collecting method of a kind of image based on textural characteristics, to overcome the above-mentioned defect of prior art, raising, to distorted image evidence obtaining verification and measurement ratio and robustness, makes the method can be applied to the fields such as discriminating of the content authenticity of image.

The passive blind evidence collecting method of digital picture that the present invention is based on textural characteristics, is characterized in that:

First carry out texture feature extraction:

The first step, suspect image to be detected is carried out to gray scale judgement, non-gray level image, is converted into gray level image if;

Second step, the texture part that is converted to the image to be detected after gray level image is cut out out;

The 3rd step, by cutting out the gray level image texture part obtaining, carry out overlap partition; The 4th step, according to following formula, calculate respectively texture mean value, texture standard deviation, texture smoothness, third moment, using texture homogeneity and the texture entropy of each Streak block, as the characteristic statistic of each sub-block, represent each Streak block content information:

1) texture Mean Value Formulas

$\mathrm{ave}=\frac{1}{s\×s}\underset{x,y}{\mathrm{\Σ}}n(x,y)$

In formula, ave is texture mean value, the size that s * s is Streak block, and n (x, y) is the gray-scale value of (x, y) point;

2) texture standard deviation formula

$\mathrm{var}=\underset{x,y}{\mathrm{\Σ}}{[n(x,y)-\mathrm{ave}]}^2/(s\×s)$

In formula, var is texture standard deviation;

3) texture smoothness formula

$P=1-\frac{1}{1+\frac{{\mathrm{var}}^2}{{(L-1)}^2}}$

In formula, P is texture smoothness, and L is texture gray level;

4) third moment formula

$S=\frac{E{(n(x,y)-\mathrm{ave})}^3}{{\mathrm{var}}^3}$

In formula, S is texture third moment, and E represents to get the computing of average statistical;

5) using texture homogeneity formula

K=∑(p ²)

In formula, K is using texture homogeneity, and p is texture histogram average;

6) texture entropy formula

$E=-\underset{x,y}{\mathrm{\Σ}}{p}_{x,y}\mathrm{lo}{g}_2\left(p_{x,y}\right)$

In formula, E is texture entropy, p _x,ydistribution probability for (x, y) point;

Then carry out the similar coupling of textural characteristics, specifically comprise following steps:

The 5th step, the characteristic quantity of each sub-block obtaining is normalized;

The 6th step, the proper vector of establishing the textural characteristics sub-block after normalization are (ave _i, var _i, P _i, S _i, K _i, E _i), wherein, i=(M-b+1) * (N-b+1), textural characteristics sub-block vector is arranged in order and obtained N _w* 6 eigenmatrix, the eigenmatrix T after being sorted, T _ia line in representation feature matrix T, i=1,2 ... N _w, M, N represents the size of test pattern texture region, the size that b is detection window, N _wthe number of the sub-block comprising in expression test pattern;

The 7th step, traversal eigenmatrix T, the side-play amount of adjacent two row-coordinate values in calculated characteristics matrix T, obtains side-play amount matrix;

The 8th step, according to following similarity decision rule formula

Each row to the side-play amount matrix obtaining carries out similarity matching; In formula, D represents to calculate the Euclidean distance of two vectors, S _i, S _jfor the row vector in excursion matrix, δ is judgement threshold values;

The 9th step, by be less than judgement threshold values the corresponding image texture part Zhong of row vector position be labeled as 1, the position mark that does not meet texture part corresponding to the row vector of judgement threshold values is 0;

The tenth step, to the non-connected region producing in evidence obtaining process, first to its carry out morphology open operation and then it is carried out to morphology closed operation and eliminates.

These six texture statistics amounts of texture mean value, texture standard deviation, texture smoothness, third moment, using texture homogeneity and texture entropy of each Streak block in the inventive method, have been chosen as the proper vector of image subblock, and existing method is minimum, must choose seven characteristic quantities as the proper vector of image subblock, so the relatively existing method of the present invention has reduced the dimension of characteristic quantity.Six texture characteristic amounts directly choosing image due to the present invention carry out presentation video content information, compare existing method and all select the indirect characteristic quantity of image transform domain to carry out representative image content information, more the information of image can be showed fully.The textural characteristics of choosing due to the present invention has the features such as level, yardstick and translation invariance, and the feature of these textural characteristics can strengthen the robustness of the method, can be applicable to the fields such as discriminating of the content authenticity of image.

Accompanying drawing explanation

Fig. 1 is the flow process principle schematic that the present invention is based on the passive blind evidence collecting method of image of textural characteristics.

Embodiment

Embodiment 1:

The passive blind forensic technologies of the present embodiment Applied Digital image is collected evidence to the content authenticity of suspect image.Fig. 1 has provided the flow process principle schematic of the passive blind evidence collecting method of image that the present invention is based on textural characteristics.Referring now to Fig. 1, to the present embodiment, the specific operation process of the passive blind evidence collecting method of image based on textural characteristics is described below:

First carry out texture feature extraction:

The first step, decision steps A, carry out gray scale judgement to suspect image P to be detected, if image to be detected is non-gray level image, image P to be detected is converted to gray level image;

Second step, cut out step B, the texture part of the image to be detected after judgement is cut out out, cast out and after those are cut out, be less than half marginal portion of piecemeal window;

The 3rd step, piecemeal step C, by cutting out the gray level image texture part obtaining, carry out overlap partition, concrete operations are: with the window of b * b size from the pixel of sliding from left to right up and down at every turn, if image texture to be detected part size is M * N, the sub-block quantity obtaining be so (M-b+1) (N-b+1);

The 4th step, characteristic statistic calculation procedure D, according to following texture formula, calculate respectively texture mean value, texture standard deviation, texture smoothness, third moment, using texture homogeneity and the texture entropy of each Streak block, as the characteristic statistic of each sub-block, represent each Streak block content information:

1) texture Mean Value Formulas

$\mathrm{ave}=\frac{1}{s\×s}\underset{x,y}{\mathrm{\Σ}}n(x,y)$

In formula, ave is texture mean value, the size that s * s is Streak block, and n (x, y) is the gray-scale value of (x, y) point;

2) texture standard deviation formula

$\mathrm{var}=\underset{x,y}{\mathrm{\Σ}}{[n(x,y)-\mathrm{ave}]}^2/(s\×s)$

In formula, var is texture standard deviation;

3) texture smoothness formula

$P=1-\frac{1}{1+\frac{{\mathrm{var}}^2}{{(L-1)}^2}}$

In formula, P is texture smoothness, and L is texture gray level;

4) third moment formula

$S=\frac{E{(n(x,y)-\mathrm{ave})}^3}{{\mathrm{var}}^3}$

In formula, S is texture third moment, and E represents to get the computing of average statistical;

5) using texture homogeneity formula

K=∑(p ²)

In formula, K is using texture homogeneity, and p is texture histogram average;

6) texture entropy formula

$E=-\underset{x,y}{\mathrm{\Σ}}{p}_{x,y}\mathrm{lo}{g}_2\left(p_{x,y}\right)$

In formula, E is texture entropy, p _x,ydistribution probability for (x, y) point;

Then carry out the similar coupling of textural characteristics, specifically comprise following steps:

The 5th step, normalization step e, be normalized the characteristic quantity of each sub-block obtaining;

The 6th step, ordered steps F, the proper vector of establishing the textural characteristics sub-block after normalization is (ave _i, var _i, P _i, S _i, K _i, E _i), wherein, i=(M-b+1) * (N-b+1); By textural characteristics sub-block vector by obtain N along seeking arranging _w* 6 eigenmatrix, then by this eigenmatrix by the line ordering of advancing, obtain eigenmatrix T, T _ia line in representation feature matrix T, i=1,2 ... N _w, M, N represents the size of test pattern texture region, the size that b is detection window, N _wthe number of the sub-block comprising in expression test pattern; ;

The 7th step, side-play amount calculation procedure G, traversal eigenmatrix T, the side-play amount of adjacent two row-coordinate values in calculated characteristics matrix T, obtains side-play amount matrix;

The 8th step, similarity matching step H, carry out similarity matching according to following similarity decision rule formula to each row of the side-play amount matrix obtaining, and in this example, τ gets 0.02;

Similarity decision rule formula

In formula, D represents to calculate the Euclidean distance of two vectors, S _i, S _jfor the row vector in excursion matrix, τ is judgement threshold values;

The 9th step, mark can region step I, by be less than judgement threshold values the corresponding image texture part Zhong of row vector position be labeled as 1, the position mark that does not meet texture part corresponding to the row vector of judgement threshold values is 0;

The tenth step, process the region step G of non-UNICOM, to the non-connected region producing in evidence obtaining process, first it is carried out to morphology and open operation and then it is carried out to morphology closed operation elimination.

Comparing existing method all selects the indirect characteristic quantity of image transform domain to carry out representative image content information, the passive blind evidence collecting method of the digital picture based on textural characteristics that the present invention takes, directly choose image texture characteristic presentation video information, more the information of image can be showed fully.

These six texture statistics amounts of texture mean value, texture standard deviation, texture smoothness, third moment, using texture homogeneity and texture entropy of each Streak block in the inventive method, have been chosen as the proper vector of image subblock, and existing method is minimum, must choose seven characteristic quantities as the proper vector of image subblock, so the relatively existing method of the present invention has reduced the dimension of characteristic quantity.

The textural characteristics of choosing due to the present invention has the features such as level, yardstick and translation invariance, and the feature of these textural characteristics can strengthen the robustness of the method, can be applicable to the fields such as discriminating of the content authenticity of image.

Claims (1)

1. the passive blind evidence collecting method of the digital picture based on textural characteristics, is characterized in that:

First carry out texture feature extraction:

The first step, suspect image to be detected is carried out to gray scale judgement, non-gray level image, is converted into gray level image if;

Second step, the texture part that is converted to the image to be detected after gray level image is cut out out;

The 3rd step, by cutting out the gray level image texture part obtaining, carry out overlap partition; The 4th step, according to following formula, calculate respectively texture mean value, texture standard deviation, texture smoothness, third moment, using texture homogeneity and the texture entropy of each Streak block, as the characteristic statistic of each sub-block, represent each Streak block content information:

1) texture Mean Value Formulas

$\mathrm{ave}=\frac{1}{s\×s}\underset{x,y}{\mathrm{\Σ}}n(x,y)$

In formula, ave is texture mean value, the size that s * s is Streak block, and n (x, y) is the gray-scale value of (x, y) point;

2) texture standard deviation formula

$\mathrm{var}=\underset{x,y}{\mathrm{\Σ}}{[n(x,y)-\mathrm{ave}]}^2/(s\×s)$

In formula, var is texture standard deviation;

3) texture smoothness formula

$P=1-\frac{1}{1+\frac{{\mathrm{var}}^2}{{(L-1)}^2}}$

In formula, P is texture smoothness, and L is texture gray level;

4) third moment formula

$S=\frac{E{(n(x,y)-\mathrm{ave})}^3}{{\mathrm{var}}^3}$

In formula, S is texture third moment, and E represents to get the computing of average statistical;

5) using texture homogeneity formula

K=∑(p ²)

In formula, K is using texture homogeneity, and p is texture histogram average;

6) texture entropy formula

$E=-\underset{x,y}{\mathrm{\Σ}}{p}_{x,y}\mathrm{lo}{g}_2\left(p_{x,y}\right)$

In formula, E is texture entropy, p _x,ydistribution probability for (x, y) point;

Then carry out the similar coupling of textural characteristics, specifically comprise following steps:

The 5th step, the characteristic quantity of each sub-block obtaining is normalized;

The 6th step, the proper vector of establishing the textural characteristics sub-block after normalization are (ave _i, var _i, P _i, S _i, K _i, E _i), its

In, i=(M-b+1) * (N-b+1), textural characteristics sub-block vector is arranged in order and obtained N _w* 6 eigenmatrix, the eigenmatrix T after being sorted, T _ia line in representation feature matrix T, i=1,2 ... N _w, M, N represents the size of test pattern texture region, the size that b is detection window, N _wthe number of the sub-block comprising in expression test pattern;

The 7th step, traversal eigenmatrix T, the side-play amount of adjacent two row-coordinate values in calculated characteristics matrix T, obtains side-play amount matrix;

The 8th step, according to following similarity decision rule formula

Each row to the side-play amount matrix obtaining carries out similarity matching; In formula, D represents to calculate the Euclidean distance of two vectors, S _i, S _jfor the row vector in excursion matrix, δ is judgement threshold values;

The 9th step, by be less than judgement threshold values the corresponding image texture part Zhong of row vector position be labeled as 1, the position mark that does not meet texture part corresponding to the row vector of judgement threshold values is 0;

The tenth step, to the non-connected region producing in evidence obtaining process, first to its carry out morphology open operation and then it is carried out to morphology closed operation and eliminates.

Images