CN101493927B - Image reliability detecting method based on edge direction characteristic - Google Patents

Abstract

The invention discloses an image credibility testing method based on edge direction characteristics, which comprises: utilizing a vector gradient to solve a gradient direction angle Theta (x, y)=(1/2)*arctan(2gxy/(gxx-gyy)) of an image to be tested so as to solve a direction characteristic K of pixel dots in the image to be tested; finding out all doubtful dots and forgery dots in the image; respectively recording the numbers of the doubtful dots and the forgery dots as num1 and num 2; the image credibility=1-(num2/num1); respectively expressing the forgery dots and non-forgery dots with two different gray levels, thus obtaining an binary image; and performing a morphology operation to the forgery dots in the binary image to obtain a forgery area. The method expresses tested results in the form of credibility and overcomes the defect that the tested results of the prior algorithm are relatively arbitrary.

Description

Image reliability detecting method based on edge direction characteristic

Technical field

The present invention relates to the image content information security fields, particularly a kind of blind checking method of picture material authenticity.

Background technology

Current protection to digital image content mainly utilizes digital watermarking and signature technology to realize, the basic thought that these technology adopt all is by adding additional information image to be carried out the authenticity discriminating.But do not contain digital watermarking or signature in present most image.Along with popularizing of digital camera and image processing software with popular, increasing high-quality forgery image is propagated in a large number by network, " soon " that makes people be " for real " no longer, caused the negative effect of aspects such as politics, culture, news and science authenticity.Further, a large amount of existence of forging photo have influence on individual's consciousness of the public probably, finally allow people lose the trust of comparison film.Therefore, carrying out of this research has military significance, civilian meaning and scientific meaning to society and people's life.

The content reliability of digital picture evaluation is one of branch of information security technology, is the technology of under the situation of sign such as embed watermark the authenticity of digital image content being estimated in image in advance not.This technology is by analysis of image data and statistical property thereof, comes the forgery in the image and alters vestige and detect and locate, thereby the confidence level of picture material is estimated.

At present, the research of content reliability of digital picture evaluation still is in the starting stage,, has obtained some achievements in research along with domestic and international associated mechanisms improves day by day to its degree of concern.These researchs are primarily aimed at certain and specifically distort operation and image is carried out the true and false detect, comprising fuzzy operation, and dual JPEG squeeze operation, partial copy-paste operation, re-sampling operations, concatenation or the like.These technology all are based on such hypothesis: exist some statistical character in natural image, then can change its potential statistical law if view data made amendment.This hypothesis also is to judge whether piece image is modified and carries out the foundation of tampering location.

At present existing detection method at the forgery and the image of juggling the figures, mainly contain the Blind image splicing and photomontage detecting using higherorder statistics (utilizing the higher order statistical characteristic that image mosaic is carried out blind Detecting) that Haney Farid and Tian-Tsong propose, and the Detection of copy-move forgery in digital images that proposes by Fridrich J and Soukal D (paste to forge detect duplicating in the image).These methods are the digital image evidence collecting method at the simple image splicing.Yet forging and altering the modal operation of image is exactly synthesizing and retouching of image local.Therefore, except to image is synthetic detect, to the detection of retouching operations such as fuzzy, the emergence of forging image, the gradual change emphasis of digital image evidence collecting research especially.All there are some problems in present digital image evidence collecting algorithm, for example can't the image that carry out the retouching operational processes after the splicing be detected, and is subjected to bigger restriction in actual applications.

Aspect synthetic edge, occurred some and detected the technology of the artificial fuzzy operation vestige of synthetic edge.Patent publication No. is CN1862598 (open day: in image forge 2006.11.15) in the method for obtaining evidence of abnormal hue rate of fuzzy operation, utilize fuzzy operation to digital picture local color attribute cause unusual, by set of definition abnormal hue and abnormal hue rate, locally coherence and degree of relevancy to picture tone quantize, and then the digital picture of a width of cloth through forgery and artificial fuzzy operation detected.People such as Zhou Linna publish an article in the KES-AMSTA2007 meeting, utilize the method for homomorphic filtering, moving average filtering and mathematical morphology that the digital picture of crossing through artificial Fuzzy Processing is distorted detection.The two-value result who is "Yes" or " non-" that these already present detection techniques obtain, it is comparatively dogmatic therefore to detect conclusion.

Therefore, at present aspect synthetic edge, detect the method and the achievement in research of forging image although proposed some, but with respect to performance image editing software from strength to strength and the adulterator who is skilled in technique, these methods and achievement in research still strength a little less than, also can't fundamentally contain the paces of image forge.

Summary of the invention

In order to overcome the effectively deficiency of the detected image true and false of prior art, the invention provides a kind of image reliability detecting method based on edge direction characteristic, summed up and forged the weak point rule that may exist in the image, analyzed the vestige that the image forge process may be left over, authenticity to image content information is reliably estimated, remedy the deficiency of testing result arbitrary decision, made testing result more reasonable reliable.

Principle analysis of the present invention is as follows: Photoshop is an image processing software commonly used at present, owing to have easy, tractable characteristics, it is subjected to liking of vast image fan deeply, and therefore, most forgery image is distorted by Photoshop software and obtained.In image forge, a kind of pseudo-making method that is in daily use synthesizes two width of cloth images exactly, and the area-of-interest that is about in the piece image is set up the constituency, and copy-paste is in another width of cloth image.Most of images are through after simple synthetic, in order to eliminate vision or the statistical distortion that produces at synthetic edge, all can adopt operations such as fuzzy, emergence, gradual change to carry out post-processed.These retouching operations can make the color of new insertion portion and scenery around it or border etc. seamlessly transit, and eliminate regional uncontinuity.No matter adopt which kind of retouching method, its net result all can make gray level change too violent splicing edge and become and seamlessly transit, and makes new insertion portion and its color of scenery or transition on border on every side seem more natural.Therefore, these operations all inevitably make image become level and smooth and fuzzy.After various retouching operations, must increase " the perfect degree " of synthetic edge transition, promptly the direction character of edge pixel reaches unanimity; Near the synthetic edge gradient angle also can have stronger smoothness properties, and promptly smoothness properties is also compared more " perfection " with other zones.Therefore, the present invention is directed to the retouching operations such as emergence of forging image, by the image edge direction feature being detected the confidence level of process decision chart picture.

The technical solution adopted for the present invention to solve the technical problems may further comprise the steps:

By with the conceptual expansion of gradient in vector function, utilize vectorial gradient to ask for the gradient direction angle θ of image to be detected.

(x, y), gradient is defined as that (x y) locates to point to the vector (amplitude and direction) of the maximum rate of change direction of f, that is: at coordinate for scalar function f

$\▿f=\left[\begin{array}{c}{G}_x\\ G_y\end{array}\right]=\left[\begin{array}{c}\frac{\∂f}{\∂x}\\ \frac{\∂f}{\∂y}\end{array}\right]---\left(1\right)$

Below with the conceptual expansion of gradient in the scalar function in vector function, try to achieve the gradient (amplitude and direction) of vector, thereby can ask gradient coloured image at the place, arbitrfary point.Make r, g, b be in the RGB color space along the vector of unit length of three color coordinates axles, make u, v respectively the remarked pixel point (x, the color vector of y) locating (R, G, B) along continuous straight runs and gradient vertically, u then, v is expressed as follows:

$u=\frac{\∂R}{\∂x}r+\frac{\∂G}{\∂x}g+\frac{\∂B}{\∂x}b---\left(2\right)$

$v=\frac{\∂R}{\∂y}r+\frac{\∂G}{\∂y}g+\frac{\∂B}{\∂y}b---\left(3\right)$

Quantity g _Xx, g _Yy, g _XyBe defined as these vectorial dot products, as follows:

$g_{\mathrm{xx}}=u\·u=u^{T}u={\left|\frac{\∂R}{\∂x}\right|}^2+{\left|\frac{\∂G}{\∂x}\right|}^2+{\left|\frac{\∂B}{\∂x}\right|}^2---\left(4\right)$

$g_{\mathrm{yy}}=v\·v=v^{T}v={\left|\frac{\∂R}{\∂y}\right|}^2+{\left|\frac{\∂G}{\∂y}\right|}^2+{\left|\frac{\∂B}{\∂y}\right|}^2---\left(5\right)$

$g_{\mathrm{xy}}=u\·v=u^{T}v=\frac{\∂R}{\∂x}\frac{\∂R}{\∂y}+\frac{\∂G}{\∂x}\frac{\∂G}{\∂y}+\frac{\∂B}{\∂x}\frac{\∂B}{\∂y}---\left(6\right)$

Pixel (x, three the color component R, G, B and the resultant g that y) locate _Xx, g _Yy, g _XyBe the function of x and y, utilize this representation, vectorial c (x, y)=(R, G, maximum rate of change direction B), promptly gradient direction can be by providing with lower angle:

$\mathrm{\θ}(x,y)=\frac{1}{2}\arctan \left[\frac{{2g}_{\mathrm{xy}}}{g_{\mathrm{xx}}-g_{\mathrm{yy}}}\right]---\left(7\right)$

Point (promptly gradient magnitude is provided by following formula for x, the y) value of rate of change on the θ direction:

$F_{\mathrm{\&theta;}}(x,y)=\sqrt{1/2[(g_{\mathrm{xx}}+g_{\mathrm{yy}})+(g_{\mathrm{xx}}-g_{\mathrm{yy}})\mathrm{<figure-callout\; id="2"\; label="cos"\; filenames="H2009100213191E00011.png"\; state="\{\{state\}\}">cos</figure-callout>}2\mathrm{\&theta;}+{2g}_{\mathrm{xy}}\sin 2\mathrm{\&theta;}}---\left(8\right)$

In the present invention, it is as follows to ask for the detailed process of image gradient deflection:

A. with the value of each pixel in the image divided by 255, the span of pixel value is transformed in the interval [0,1];

B. three passages to the RGB coloured image use certain edge detection operator (spendable operator comprises the Sobel operator, Prewitt operator, Roberts operator) to carry out convolution respectively, obtain desired derivative in the above-mentioned equation (2) (3) $\frac{\&PartialD;R}{\&PartialD;x},\frac{\&PartialD;R}{\&PartialD;y},\frac{\&PartialD;G}{\&PartialD;x},\frac{\&PartialD;G}{\&PartialD;y},\frac{\&PartialD;B}{\&PartialD;x},\frac{\&PartialD;B}{\&PartialD;y};$

C. in the traversing graph picture each pixel (x y), utilizes formula (4) (5) (6) to calculate g respectively _Xx, g _Yy, g _XyThereby, calculate the gradient direction angle θ of each pixel, see formula (7).

2. obtain the direction character k of pixel in the image to be detected;

(being designated as (m, n)) is the center, defines 16 directions, D with pixel q _j=j π/8, j=0,1 ..., 15.With q is the center, along D _jGet the individual pixel of 2W+1 (1≤W≤5) on the direction, their gray-scale value is respectively y _-W, y _-W+1..., y _W-1, y _WSuppose that these pixel values can fit with following linear function:

y＝kx+b+ε (9)

Wherein, x ∈ W ,-W+1 ..., 0 ..., W-1, W}, y represents gray-scale value, and k and b represent the slope and the intercept of straight line respectively, and ε represents error.The estimation of slope k can be described with following formula:

$k=\frac{{\mathrm{\&Sigma;}}_{=-W}^W(x_i-\stackrel{\&OverBar;}{x})(y_i-\stackrel{\&OverBar;}{y})}{{\mathrm{\&Sigma;}}_{=-W}^W{(x_i-\stackrel{\&OverBar;}{x})}^2}---\left(10\right)$

Because x ∈ W ,-W+1 ..., 0 ..., W-1, W}, x=0, formula (10) can be reduced to:

$k=\frac{{\mathrm{\&Sigma;}}_{i=-W}^W{x}_i(y_i-\stackrel{\&OverBar;}{y})}{{\mathrm{\&Sigma;}}_{i=-W}^W{x}_i^2}---\left(11\right)$

Wherein,

$\stackrel{\&OverBar;}{y}=\frac{1}{2W+1}\underset{i=-W}{\overset{W}{\mathrm{\&Sigma;}}}{y}_i---\left(12\right)$

Further, the estimation of variance can be described with following formula:

${\mathrm{\&sigma;}}^2=\frac{1}{2W+1}[\underset{i=-W}{\overset{W}{\mathrm{\&Sigma;}}}{(y_i-\stackrel{\&OverBar;}{y})}^2-{\mathrm{<figure-callout\; id="2"\; label="k"\; filenames="H2009100213191E00011.png"\; state="\{\{state\}\}">k</figure-callout>}}^2\underset{i=-W}{\overset{W}{\mathrm{\&Sigma;}}}{x}_i^2]---\left(13\right)$

If σ=0 illustrates that this straight line can well fit the pixel of this direction, promptly have a few all to drop on to fit on the straight line, σ is big more, and it is poor more then to fit effect.Each direction D for pixel q _j, j=0,1 ..., 15, estimate the slope k and the variances sigma that fit straight line respectively.If variance along the Dj direction Minimum then will

Direction character as pixel q.

In the present invention, it is as follows to ask in the band detected image the directional characteristic concrete steps of pixel:

A) with image to be detected by rgb space or under itself color space conversion extract luminance component I to the HSI space;

B) to 16 direction D of pixel q definition among the luminance component I _j=j π/8, j=0,1 ..., 15;

C) with each direction D _jGoing up with q is that 2W+1 the pixel at center carried out linearity and fitted, and estimates slope

And variance

, see formula (11) (13);

D) with the slope of the minimum value institute line correspondence in 16 variances direction character k as pixel q;

E) each pixel in the traversal luminance component, execution in step b), c), d), thereby obtain the direction character of each pixel.

3. the zone is forged in the location, and image reliability is estimated.

By discovery that a large amount of forgery images are experimentized, normally one " complete scenery " is inserted in another image during image forge, the operation of promptly sprouting wings acts on scenery edge usually, therefore, in order to improve detection speed, can only detect the scenery edge.Find that simultaneously the direction character of the pixel in the zone of sprouting wings and gradient direction all have certain compatibility with surrounding pixel.

Definition 1: so-called compatible, be meant pixel q ₁With pixel q ₂The difference of attribute k less than a certain threshold value, that is:

Wherein, λ _k, λ _θFor the experiment in choose appropriate threshold (1.0≤λ _k≤ 10.0,0.1≤λ _θ≤ 0.5).

Definition 2: compatible degree, establish pixel q, the number of pixels compatible with it in its a certain appointment neighborhood Ω is called the compatible degree of pixel q.

Definition 3: compatibility, the ratio of the sum of all pixels among the compatible degree that is meant pixel q and the neighborhood Ω.

Based on above three definition, the concrete steps that zone and computed image confidence level are forged in the location are as follows:

A). (being designated as (m, n)) is the center, and definition length is the neighborhood Ω of 2W+1 (1≤W≤5), Ω={ (m with pixel q ₁, n ₁) || m ₁-m|≤W, | n ₁-n|≤W}.Calculate the direction character compatibility size s of pixel q respectively _kWith gradient direction angle compatibility size s _θ

$s_k=\frac{{\mathrm{num}}_k}{(2W+1)(2W+1)}---\left(14\right)$

$s_{\mathrm{\&theta;}}=\frac{{\mathrm{num}}_{\mathrm{\&theta;}}}{(2W+1)(2W+1)}---\left(15\right)$

Wherein, num _kThe direction character compatible degree of expression central pixel point q, num _θThe gradient direction angle compatible degree of expression central pixel point q.

B). the similarity size of definition pixel q and its neighborhood territory pixel is:

s＝αs _k+βs _θ (16)

Wherein, α, β is a weighting coefficient, and satisfies alpha+beta=1, α 〉=β.Choose appropriate threshold δ _sAnd μ (0.5≤δ _s≤ 0.9,0.03≤μ≤0.1), as s 〉=δ _sDuring-μ, q is labeled as suspicious points with pixel; As s 〉=δ _sThe time, pixel q is labeled as the forgery point.

C). each pixel in the traversing graph picture, difference execution in step a, b, thus must find out suspicious points all in the image and forge point, the number of record suspicious points and forgery point is respectively num ₁And num ₂, the following expression of the confidence level of image:

$\mathrm{credibility}=1-\frac{{\mathrm{<figure-callout\; id="2"\; label="num"\; filenames="H2009100213191E00011.png"\; state="\{\{state\}\}">num</figure-callout>}}_2}{{\mathrm{num}}_1}---\left(17\right)$

D). will forge a little and represent (such as 1 and 0) with two kinds of different gray scales respectively with non-forgery point, obtain a width of cloth bianry image, forgery in bianry image point is carried out morphology operations (burn into expands, is communicated with), make its zone that obtains a sealing or be communicated with, should the zone as forging the zone.

The invention has the beneficial effects as follows: owing to start with in the synthetic edge from forge image, the retouching of having analyzed in the forgery process is operated to the pixel orientation feature at synthetic edge and the influence that the gradient direction angle is brought, and detects the forgery image with this.At first, the gradient direction angle is adopted and is tried to achieve based on the method for vectorial gradient, is better than classic method; Secondly, defined the direction character of pixel, itself and gradient direction angle combined find out suspicious points and forge point, utilize morphologic corrosion and dilation operation to orient forgery vestige in the image then, and the form of testing result with confidence level showed, overcome the comparatively dogmatic deficiency of existing algorithm testing result.

The present invention is further described below in conjunction with drawings and Examples.

Description of drawings

Fig. 1 is 16 direction synoptic diagram of pixel.

Fig. 2 utilizes edge direction characteristic to detect sample image and the testing result synoptic diagram of forging image.

Fig. 3 is original image and the testing result contrast synoptic diagram of forging image.

Embodiment

Detect for ease of carrying out image forge, seminar has set up by the forgery image data base of specifically distorting the means classification, existing at present nearly 4000 width of cloth images.Simple operations such as existing direct splicing in the image data base also has and forges the complex operations that the back image is polished modification again to simple; The image that the single instrument of existing employing is handled, the image that also has comprehensive multiple instrument to handle, this image library can be carried out research by easy stages from simple to complexity, is beneficial to the feasibility analysis that improves algorithm.

Embodiments of the invention as shown in Figure 1, wherein Fig. 1 (a), Fig. 1 (b) are original image, Fig. 1 (c) is for forging image, it is the tower among former Fig. 2 (b) to be copied to the appropriate location obtains among Fig. 2 (a) figure, and has used post-processing operation such as edge fog.We utilize edge direction characteristic that it is forged image graph 2 (c) and detect.The detection step is as follows:

(1) by with the conceptual expansion of gradient in vector function, utilize vectorial gradient to ask for the gradient direction angle θ of image to be detected.

A) with the value of each pixel in the image divided by 255, the span of pixel value is transformed in the interval [0,1];

B) three passages to image (c) use the Sobel operator to carry out convolution respectively, obtain desired derivative in the above-mentioned equation (2) (3) $\frac{\&PartialD;R}{\&PartialD;x},\frac{\&PartialD;R}{\&PartialD;y},\frac{\&PartialD;G}{\&PartialD;x},\frac{\&PartialD;G}{\&PartialD;y},\frac{\&PartialD;B}{\&PartialD;x},\frac{\&PartialD;B}{\&PartialD;y};$

C) in the traversing graph picture each pixel (x y), utilizes formula (4) (5) (6) to calculate g respectively _Xx, g _Yy, g _XyThereby, calculate the gradient direction angle θ of each pixel, see formula (7).

(2) obtain the direction character k of pixel in the image to be detected;

A) image (c) is converted to the HSI space by rgb space, extracts luminance component I;

B) to 16 direction D of pixel q definition among the luminance component I _j=j π/8, j=0,1 ..., 15;

C) with each direction D _jGoing up with q is that 5 pixels at center are carried out linearity and fitted, and estimates slope

And variance , see formula (11) (13);

D) with the slope of the minimum value institute line correspondence in 16 variances direction character k as pixel q;

E) each pixel among the traversal luminance component I, execution in step b), c), d), thereby obtain the direction character of each pixel.

(3) zone is forged in the location, and image reliability is estimated.

A) to each pixel q of image (c), be center definition 5 * 5 neighborhoods with q, utilize formula (14) (15) to calculate the direction character compatible degree num of q successively _k, compatibility s _kWith gradient direction angle compatible degree num _θ, compatibility s _θ, wherein selecting threshold value is λ _k=3.5, λ _θ=0.2;

B) utilize the two weighted sum (α=0.7, β=0.3) to estimate the similarity size s of pixel q and its neighborhood territory pixel, see formula (16), similarity setting threshold δ _s=0.85, μ=0.05, whether marker image vegetarian refreshments q is suspicious points or forges point;

C) find out suspicious points all in the image (c) and forgery point, and utilize formula (17) that the confidence level of image is estimated, the confidence level that calculates image (c) is 89.93%;

D) all forgery points are carried out morphology and expand and erosion operation, the connected region that obtains is regional as forging, and wherein the zone of white line sign is detected forgery zone shown in Fig. 2 (d).

Validity for verification algorithm, we have carried out identical experiment to the original image in the accompanying drawing 2 (b) by above-mentioned steps, and the testing result of testing result and image (c) contrasted, comparing result as shown in Figure 3, for ease of observing and contrast and experiment, we intercept the experimental result that rectangle indicates the zone, and amplify demonstration.Fig. 3 (a) is depicted as the forgery image (c) in the accompanying drawing 2, has shown forgery image detection result among Fig. 3 (b), and detected forgery point can constitute a connected region, can think that this zone is for forging the zone.Fig. 3 (c) is depicted as the original image (b) in the accompanying drawing 2, has shown the testing result of original image among Fig. 3 (d), and detected forgery point can not constitute connected region, so can not judge that this image is for forging image.The confidence level of the original image that calculates is 89.93%, and the confidence level of forging image is 52.16%.

Claims (1)

1. based on the image reliability detecting method of edge direction characteristic, it is characterized in that comprising the steps: the first step, utilize vectorial gradient to ask for the gradient direction angle θ of image to be detected, detailed process is as follows:

A. with the value of each pixel in the image divided by 255, the span of pixel value is transformed in the interval [0,1];

B. three passages to the RGB coloured image use edge detection operator to carry out convolution respectively, obtain color vector (R, G, derivative B)

C. (x y), calculates for each pixel in the traversing graph picture

Thereby calculate the gradient direction angle of each pixel

Second goes on foot, and obtains the direction character k of pixel in the image to be detected, and concrete steps are as follows:

1) with image to be detected by rgb space or under itself color space conversion extract luminance component I to the HSI space;

2) to 16 direction D of pixel q definition among the luminance component I _j=j π/8, j=0,1 ..., 15;

3) with each direction D _jGoing up with q is that 2W+1 the pixel at center carried out linearity and fitted, and according to formula

With

Estimate slope

And variance 1≤W≤5 wherein;

4) with the slope of the minimum value institute line correspondence in 16 variances direction character k as pixel q;

5) each pixel in the traversal luminance component, execution in step 2), 3) and, 4), thereby obtain the direction character of each pixel;

In the 3rd step, the zone is forged in the location, and image reliability is estimated, and concrete steps are as follows:

I). (m n) is the center, and definition length is the neighborhood Ω of 2W+1, Ω={ (m with pixel q ₁, n ₁) || m ₁-m|≤W, | n ₁-n|≤W}; Calculate the direction character compatibility size s of pixel q respectively _kWith gradient direction angle compatibility size s _θ

$s_k=\frac{{\mathrm{num}}_k}{(2W+1)(2W+1)}$

$s_{\mathrm{\&theta;}}=\frac{{\mathrm{num}}_{\mathrm{\&theta;}}}{(2W+1)(2W+1)}$

Wherein, num _kThe direction character compatible degree of expression central pixel point q, num _θThe gradient direction angle compatible degree of expression central pixel point q;

II). the similarity size of definition pixel q and its neighborhood territory pixel is s=α s _k+ β s _θ, wherein, α, β is a weighting coefficient, and satisfies alpha+beta=1, α 〉=β; Choose appropriate threshold δ _sAnd μ, 0.5≤δ _s≤ 0.9,0.03≤μ≤0.1 is as s 〉=δ _sDuring-μ, q is labeled as suspicious points with pixel; As s 〉=δ _sThe time, pixel q is labeled as the forgery point;

III). each pixel in the traversing graph picture, execution in step I respectively) and II), thereby must find out suspicious points all in the image and forge point, the record suspicious points is respectively num with the number of forging point ₁And num ₂, the following expression of the confidence level of image:

$\mathrm{credibility}=1-\frac{{\mathrm{num}}_2}{{\mathrm{num}}_1}$

IV). will forge a little and represent with two kinds of different gray scales respectively, obtain a width of cloth bianry image, the point of the forgery in the bianry image will be carried out morphology operations, make its zone that obtains a sealing or be communicated with non-forgery point, should the zone as forging the zone.

Images