CN100452884C - Method for detecting GIF infomration hidden - Google Patents

Abstract

The present invention relates to a method for detecting information hiding of palette image files in the information technical field. The brightness value of each pixel of palette images corresponds to an index value in palettes. Firstly, image data are judged and positioned according to the format of GIF image files, index values of images are read in and grouped; secondly, a corresponding difference function and a corresponding replacement function are formed, and the image data are divided into three types; thirdly, the image data are carried out replacing operation, the change situation of normal sets and singular sets before and after randomization is calculated, and the present invention judges whether the images carry out information hiding or not. The present invention has the advantage that the present invention has high detection rate and higher accuracy rate to the effects of randomly embedded detection of information hiding of the GIF image files than those of similar detection methods.

Description

Method at the detection of palette image hiding file information

Technical field

What the present invention relates to is a kind of method of areas of information technology, particularly a kind of method that detects at the palette image hiding file information.

Background technology

Information disguising is a kind of confidential corespondence means, and purpose is in order to transmit information in confidence, obtains the safety of confidential corespondence by the existence of hiding secret information, and it is a kind of new means that ensure information safety except cryptographic technique for an information security.The visual observation that whether is concealed with secret information in a large amount of pictures of propagating only is on network by human eye is difficult to perceive, and must come the existence of detection of concealed information by certain detection technique.At present, BMP (bitmap file form), JPG (JPEG (joint photographic experts group)), three kinds of picture formats of GIF (BIIF) are the image file format usually seen of right and wrong, along with popularizing of computer, network, these image files are by more and more widely use.So with the image is the Information hiding of blindage, the picture format that is adopted mainly is above three kinds.Mainly be based on the time-space domain at the Information hiding that with BMP, GIF format-pattern is blindage, and be that the Information hiding of blindage mainly is based on frequency domain with the JPG format-pattern.The Information hiding detection technique develops along with information hiding technology, so the Information hiding detection technique also proposes at corresponding Information Hiding Techniques and realizes.Because the Information hiding based on the time-space domain mainly is that LSB (Least significant bit position) replaces, so the Information hiding detection also is primarily aimed at LSB and proposes.In the prior art, someone proposes the method for detection of concealed information in coloured image, be called method RQP (the original image color contrast), its basic principle is after embedding LSB information, it is right that new palette can produce a lot of close colors, and the appearance that number of colors is right shows to exist and hides Info.Be applicable to palette image, but be not suitable for more image of independent color number and gray level image;

Find that through the literature search to prior art the article " Detecting LSBsteganography in color and gray-scale images " that the people such as Jessica Fridrich deliver at " Magazine of IEEEMultimedia Special Issue on Security " (the safe monograph of international IEEE multimedia magazine) calendar year 2001 the 8th volume the 4th phase 22-28 page or leaf (detecting the Least significant bit position in colour and gray scale scalogram picture hides Info) has proposed a kind of LSB Information hiding detection method---RS method based on harmless embedding capacity. This method is thought to the harmless capacity of the randomization meeting reduction LSB bit plane of LSB; Is different but embed a plurality of bit planes to the influence that can't harm capacity.Basic principle is to utilize the responsive antithesis statistic that derives from be correlated with in the spatial domain of image to detect the existence of classified information.But this method also only is aimed at the image of BMP form, is not suitable for the application (mainly being meant the GIF image file format) in palette image.

Summary of the invention

The objective of the invention is to overcome deficiency of the prior art, a kind of method that detects at the palette image hiding file information is provided, make its image file at the GIF form, judging whether it contains hides Info, for information embedding rate reach 0.202% and the Information hiding of above GIF image can realize correct detection, the effect that detects for embedding at random of hiding of GIF image file information is than embedding the accuracy rate height that detects continuously.

The present invention is achieved by the following technical solutions, an index value in the corresponding palette of the brightness value of each pixel of palette image of the present invention, at first, form according to the GIF image file, judge and orient view data, read in the image index value, carry out image packets; Next constructs corresponding discernment function and permutation function, and view data is divided three classes; The 3rd, view data is carried out replacement operator, calculate the situation of change of conventional group before and after the randomization and unusual group, judge whether whether image has carried out Information hiding.

Described image packets, specifically: suppose that blindage image (not by the latent original image of writing) is the pixel of M * N, pixel value is included among the set P, promptly be the brightness value of GIF format-pattern here, all (palette of 256 looks sorts these brightness values corresponding to corresponding index value in the palette, sort according to brightness), the palette after the ordering is listed the index value of ordering, so the brightness value of each pixel is just corresponding to an index value in the palette.For example, an index value of representing with 8bit, then brightness set P={0...255}.Image is divided into n adjacent brightness (x arranged ₁... x _n) form discontinuous group, the grouping size is delegation or one 's 4 * 4 a piece, specifically according to the visual characteristic of image, chooses and will make it with respect to neighbor is level and smooth relatively on every side.For example, can get the continuous colourity of two row two row of n=4 as a group.

Described structure discernment function, specifically: in order to obtain flatness and the regularity of brightness group G, define discernment function f, its complexity is detecting searching compromise between accuracy rate and the computation complexity.Select brightness group (x ₁... x _n) variance be that formula (1) is as discernment function.

$f(x_1,x_2,...x_n)=\underset{i=1}{\overset{n-1}{\mathrm{\Σ}}}{|x_{i+1}-x_i|}^2---\left(1\right)$

Described transposition of structures function, specifically: definition substitution operation F.F (G) is meant brightness group G is adopted the operation of F function.F is defined as follows:

To F ₁: $0\↔\mathrm{1,2}\↔3,\·\·\·,254\↔255.$

To F _-1: $-1\↔\mathrm{0,1}\↔2,\·\·\·,255\↔256.$

For integrality, definition F ₀: F (x)=x

F ₁And F _-1Between have relation: F _-1(x)=F ₁(x+1)-1, definition F (G) is (F _{M (1)}(x ₁), F _{M (2)}(x ₂) ..., F _{M (n)}(x _n)). adopt different replacement operators for brightness values different among the G, definition mask M, the M value can be-1,0, or 1.

Described view data is divided three classes, specifically: brightness value is divided three classes-R according to defined f in front and F, S, U:

Conventional group: $G\∈R\⇔f\left(F\left(G\right)\right)>f\left(G\right)$

Unusual group: $G\&Element;S\&DoubleLeftRightArrow;f\left(F\left(G\right)\right)<f\left(G\right)$

Abandon group: $G\&Element;U\&DoubleLeftRightArrow;f\left(F\left(G\right)\right)=f\left(G\right)$

When in image, introducing an amount of noise, can cause the increase of discernment function, will make the R group more than the number of S group.

To typical initial carrier image, brightness group G is used the increase that permutation function can cause the discernment function value.Thereby after picture breakdown being become several brightness groups, the number of conventional brightness group will be much larger than the number of unusual brightness group.

Definition R _MBe the relative number (percentage in all group of pixels) of conventional group under non-negative mask operator M, S _MBe unusual group relative number, thereby R is arranged _M+ S _M≤ 1 and R _-M+ S _-M≤ 1.Studies show that randomization LSB is to R _-MAnd S _-MInfluence be opposite, promptly their difference becomes big along with the increase that embeds information m.

The embedding at random that the present invention adopts GIF image data packets method based on the palette technology, hide at the discernment function of GIF image and permutation function building method and GIF image file information with embed detection method continuously, Information hiding at the GIF format-pattern, has the higher detection rate, embed for random information simultaneously, accuracy rate is higher.The present invention can be applicable to corresponding products such as internet content safety filtering gateway, Email intelligently filters system, effectively detect hiding Info in the palette image, whether decision needs to be filtered and block, can unit such as radio and television, information office and in/be applied in primary school's campus network construction.

Description of drawings

Fig. 1 the inventive method flow chart

The pending image that Fig. 2 embodiment of the invention is used

Fig. 3 embodiment of the invention embeds the image of information

Embodiment

Further provide following specific embodiment in conjunction with content of the present invention:

As Fig. 1, testing process: at first, judge pending image file, be defined as entering this flow process behind the GIF form; Secondly,, orient view data position hereof, read in image index data, the line data of going forward side by side grouping according to the form of GIF image file; Based on the present invention, construct corresponding discernment function and permutation function then, view data is divided three classes: conventional group, and abandons group by unusual group; Then, view data is carried out replacement operator, the situation of change of routine group and unusual group of number before and after the calculating randomization, if the routine group is big with unusual group relative number change, the decidable image has carried out Information hiding; Otherwise if the relative number of conventional group and unusual group does not have very variation, the decidable image does not carry out Information hiding, no embedding information.

Lena with 512 * 512 is an example, constructs the lena-hide that a width of cloth has embedding information, and a former figure lena describes respectively.Below be lena-hide (Fig. 3) and lena (Fig. 2)

Embodiment 1: for lena-hide

At first, judge the image file format of pending lena, be defined as entering this flow process behind the GIF form;

Secondly,, orient view data position hereof, read in image index data according to the form of GIF image file, the line data of going forward side by side grouping, four pixels with per 2 * 2 are one group, the image for 512 * 512 can be divided into 6656 groups;

Then, based on the present invention, construct corresponding discernment function and permutation function, wherein discernment function is

$f(x_1,x_2,x_3,x_4)=\underset{i=1}{\overset{3}{\mathrm{\&Sigma;}}}|x_{i+1}-x_i|,$

Permutation function is: F _-1(x)=F ₁(x+1)-1, mask operator M=[0 1 0],

View data is divided three classes: conventional group, unusual group, abandon group, classification foundation be formula as indicated above;

Then, view data is carried out replacement operator, calculates the situation of change of conventional group in randomization front and back and unusual group of number, obtain:

$R_{{-M}_1}-S_{{-M}_1}=14.6641,$ $R_{M_1}-S_{M_1}=6.0928,$ R _-M2-S _-M2＝0.1240， $R_{M_2}-S_{M_2}=19.4691$

Four parameters by above can fit out a curve, can obtain embedding rate p=0.4056, are 0.025 according to defined judgement fault value, and the decidable image has carried out Information hiding.

Embodiment 2: for lena

At first, judge the image file format of pending lena, be defined as entering this flow process behind the GIF form;

Secondly,, orient view data position hereof, read in image index data according to the form of GIF image file, the line data of going forward side by side grouping, four pixels with per 2 * 2 are one group, the image for 512 * 512 can be divided into 6656 groups;

Then, based on the present invention, construct corresponding discernment function and permutation function, wherein discernment function is

$f(x_1,x_2,x_3,x_4)=\underset{i=1}{\overset{3}{\mathrm{\&Sigma;}}}|x_{i+1}-x_i|,$

Permutation function is: F _-1(x)=F ₁(x+1)-1, mask operator M=[0 1 0],

View data is divided three classes: conventional group, unusual group, abandon group, classification foundation be formula as indicated above;

Then, view data is carried out replacement operator, calculates the situation of change of conventional group in randomization front and back and unusual group of number, obtain:

$R_{{-M}_1}-S_{{-M}_1}=11.6367,$ $R_{M_1}-S_{M_1}=11.7432,$ R _-M2-S _-M2＝4.9011， $R_{M_2}-S_{M_2}=16.8808$

Four parameters by above can fit out a curve, can obtain embedding rate p=0.0171, are 0.025 according to defined judgement fault value, and the decidable image does not carry out Information hiding, no embedding information.

By embodiment 1 and 2 resultant R _-M-S _-MThe discovery of comparing, for image after the embedding information, its value obviously can be bigger than the relative value of the information that does not embed.The present invention is directed to the Information hiding of GIF format-pattern, have the higher detection rate, embed for random information simultaneously, accuracy rate is higher.

Claims (2)

1, a kind of method that detects at the palette image hiding file information, it is characterized in that, an index value in the corresponding palette of the brightness value of each pixel of palette image, at first, form according to the GIF image file, judge and orient view data, read in the image index value, carry out image packets; Secondly, construct corresponding discernment function and permutation function, view data is divided three classes; The 3rd, view data is carried out replacement operator, calculate the situation of change of conventional group before and after the randomization and unusual group, judge whether whether image has carried out Information hiding;

The described image packets of carrying out, specifically: suppose that the blindage image is the pixel of M * N, pixel value is included among the set P, promptly be the brightness value of GIF format-pattern here, these brightness values are all corresponding to corresponding index value in the palette, palette after the ordering is listed the index value of ordering, so the brightness value of each pixel is just corresponding to an index value in the palette, image is divided into n adjacent brightness (x arranged ₁... x _n) group formed, the grouping size is delegation or one 's 4 * 4 a piece, specifically according to the visual characteristic of image, chooses and will make its image index value with respect to neighbor is level and smooth relatively on every side;

Described structure discernment function, specifically: in order to obtain flatness and the regularity of brightness group G, define discernment function f, its complexity is selected brightness group (x detecting searching compromise between accuracy rate and the computation complexity ₁... x _n) variance as discernment function:

$f(x_1,x_2,...x_n)=\underset{i=1}{\overset{n-1}{\mathrm{\&Sigma;}}}{|x_{i+1}-x_i|}^2;$

Described transposition of structures function, specifically: definition substitution operation F, F (G) are meant brightness group G are adopted the operation of F function that F is defined as follows:

To F ₁: $0\&LeftRightArrow;\mathrm{1,2}\&LeftRightArrow;3,\&CenterDot;\&CenterDot;\&CenterDot;,254\&LeftRightArrow;255$

To F _-1: $-1\&LeftRightArrow;0,1\&LeftRightArrow;2,\&CenterDot;\&CenterDot;\&CenterDot;,255\&LeftRightArrow;256;$

Described brightness group G, each brightness value adopts corresponding replacement operator among the G, definition mask M, the M value is-1,0, or 1;

Described transposition of structures function, for integrality, definition F ₀: F (x)=x, F ₁And F _-1Between have relation: F _-1(x)=F ₁(x+1)-1, definition F (G) is (F _{M (1)}(x ₁), F _{M (2)}(x ₂) ..., F _{M (n)}(x _n)).

2, the method that detects at the palette image hiding file information according to claim 1 is characterized in that, described view data is divided three classes, and specifically: according to defined f and F brightness value being divided three classes is R, S, and U:

Conventional group: $G\&Element;R\&DoubleLeftRightArrow;f\left(F\left(G\right)\right)>f\left(G\right)$

Unusual group: $G\&Element;S\&DoubleLeftRightArrow;f\left(F\left(G\right)\right)<f\left(G\right)$

Abandon group: $G\&Element;U\&DoubleLeftRightArrow;f\left(F\left(G\right)\right)=f\left(G\right).$

Images