CN100534033C - Text numerical watermark method for resisting analog domain attack - Google Patents

Abstract

The invention relates to an anti-analog field attack text digital watermarking method, which comprises: a) pretreatment before embedding: dividing, combining and screening the letter block in original text image to form watermarking embedded block sequence; b) watermarking embedding: embedding the binary sequence into block sequence by turns to embed one bit watermarking information per horizontal placement; c) pretreatment before collecting watermarking: dividing the watermarking image with same letter block as in a), generating and locating the sequence in watermarking image and opposite location sequence by original image; d) collecting decision for watermarking: using block placement. This invention has strong robustness and well property.

Description

The text digital water mark method that a kind of anti-analog domain is attacked

Technical field

The present invention relates to a kind of Information Hiding Techniques, the text digital water mark method that especially a kind of anti-analog domain is attacked belongs to the secure authentication technology field.

Background technology

Digital watermarking is as a kind of Information Hiding Techniques of novelty, for solve copyright protection, the source authentication on the open network, a series of problems such as authentication, distribution on the net, usertracking and authentication of distorting provide brand-new solution thinking.But now people have focused on research at the digital watermark technology of digital picture, Voice ﹠ Video to attentiveness mostly, relate to text image as the research of the digital watermarking algorithm of carrier with use but seldom.Yet the application demand of text digital water mark is very urgent; especially for the mechanism of state administrative organs, large scale business tissue and so on; often need give the huge affiliated institutions of quantity with document distribution; because multiple factor, these to the documents that receive bear the protection responsibility affiliated institutions may with prints, scan, duplicate, reprint books in a reduce format, various ways leakage documents such as fax, photo.If in distributing documents, add unique watermark information, intercepting and capturing by behind the document of revealing, just can realize revealing the source investigation of document in order to identify each affiliated institutions' identity.Yet, as the digital watermarking algorithm of carrier following design difficulty is arranged with text image:

1) text image only can be represented entire image with simple several color relations, does not almost have what texture and details to say, what have only is different geometries, utilizes such carrier to be difficult to design and meets the algorithm that digital watermarking requires;

2) mostly be continuum piece in the text image with same color value or gray value, come embed watermark information if adopt the method for some selected pixel color of common modification or gray value, just may cause the pixel that perceptible different colours of naked eyes or brightness in same color value or gray value zone, occur, destroy the disguise of digital watermarking and the aesthetic property of text;

3) not only effective at the digital watermark method of text at numeric field, more need pass through the watermark information that can correctly extract embedding after a series of analog domains are attacked (include but not limited to print, scan, duplicate, fax, convergent-divergent, shooting etc. have the process of modulus/digital to analog conversion) at text.

Existing general Electronic Coding technology or digital watermarking algorithm all well do not solve for above-mentioned difficult point.

Summary of the invention

The present invention is for solving above-mentioned design difficulty, purpose is to provide a kind of text digital water mark method of anti-analog domain attack, adopt the watermark information of the embedding of this method to have very strong robustness, attack after the noise that causes, rotation, translation, convergent-divergent, word edge blurry etc. change at the various analog domains of experience, still can correctly extract watermark.This strong robustness of this digital watermark method is not only effective to the numeric field text image, also be effectively to the text hard copy (as paper document) of analog domain, and the text after watermarked has good aesthetic property and watermark disguise.

The present invention achieves the above object by the following technical programs: the text digital water mark method that a kind of anti-analog domain is attacked may further comprise the steps:

A) watermarked preceding preliminary treatment;

B) watermark embeds;

C) extract the preceding preliminary treatment of watermark;

D) watermark extracting judgement;

Detailed process is as follows:

A) watermarked preceding preliminary treatment;

The object of the digital watermark method among the present invention is for only comprising the text image of content of text, and text exists in the ranks septal area and word interval district, can utilize these characteristics to mark off the rectangle block.There is spacer region in the horizontal direction and the wrong situation that is divided into a plurality of part blocks of quilt for some block inside, the part block is merged.Carry out the screening of block then, check that successively whether each block satisfies the requirement that configure in advance wide to word, complexity and integrality are set simultaneously: 1) the wide E of word, the wide unit of word is a pixel, needs to satisfy $E>\frac{2}{3}\left(E_m\right)-3$ , and E＜(E _m+ 3), E _mFor the maximum word of complete block in text image that font, font size and print resolution are fixing wide; 2) complexity, the definition of complexity are the ratio of all pixel quantity in interior foreground pixel quantity of block and the block, and the requirement of complexity is greater than 1/15; 3) requirement of integrality is that block can not be incomplete Chinese character, letter or a symbol.As satisfy condition then add to can be watermarked the block set, be used for subsequent watermark and embed and watermark extracting.

Then carry out the generation that watermark embeds the block sequence: can be able to select watermark embedding block in the watermarked block set by key and pseudo random sequence, the watermark block must satisfy spacing with the most contiguous block of the left and right sides more than or equal to 1/50 inch, and this block left and right sides all exist belong to can watermarked block set block.These two blocks are for locating block, and its position can not be moved (just can not be chosen as watermark embedding block again in case certain block is chosen as the location block) in watermark embed process, are used to when extracting watermark judge that the relative position of determining watermark embedding block changes.The watermark of how much selecting sufficient amount successively according to embed watermark information embeds block formation watermark embedding block sequence.

B) watermark embeds:

Watermark information binaryzation sequence is pressed the watermarked embedding block of bit sequence, corresponding respectively watermark bit is 1 or 0 situation, watermark is embedded the distance that block is turned left in the horizontal direction or right integral body is mobile certain (displacement needs more than or equal to 1/100 inch), embed all bits successively and then finish the embedding of watermark information, watermark information is looked application need can comprise error correcting code and check code.

C) extract the preceding preliminary treatment of watermark:

The text image that is used to extract watermark may pass through the various operations that comprise that analog domain is attacked, and causes its foreground to be compared with original image with background colour variation has all taken place.For correct differentiation prospect district (text area) and background area, at first need watermarking images (watermarked text image) is carried out binaryzation, then according to a) watermarked before in the preliminary treatment same rule watermarking images carried out block divide, but do not carry out subsequent operations such as merging that block divides, screening.Then original figure territory text image (not watermarked text image) is carried out the division of identical block when watermarked, and generate same watermark and embed the block sequence, calculate in the watermarking images position that corresponding each watermark embeds block by coordinate transform, realize that watermark in the watermarking images embeds the coarse positioning of block sequence.In view of there are small difference in coarse positioning result and actual value, need to divide by the block in the search watermarking images, find out with coarse positioning as a result the difference minimum promptly obtain accurate position.Guaranteed that watermark embeds the block left and right sides and all has the location block when a) choosing watermark embedding block in the watermarked preceding preliminary treatment, watermark by original image embeds the location block sequence that the block sequence generates correspondence, the method that adds search according to similar coarse positioning, can obtain the accurate position of all location blocks in the watermarking images, generate with watermark and embed the corresponding location block sequence of block sequence.

Described watermark to the watermark text image embeds the method that block in block sequence and the location block sequence carries out coarse positioning: according to this block in original image the abscissa at two ends, the left and right sides of position coordinates, the place literal line of this block in original image, and the abscissa at this literal line two ends, the left and right sides of corresponding literal line in the watermark text image, calculate the rough position coordinate of corresponding block in the watermark text image by coordinate transform.

The described method of obtaining the accurate position of block by search is: in watermarking images, with block by coarse positioning obtain the left end abscissa and block place literal line in the left end abscissa calculation deviation amount of all blocks, the left end abscissa exact value that is the block of searching for of departure minimum; In watermarking images, with block by coarse positioning obtain the right-hand member abscissa and block place literal line in the right-hand member abscissa calculation deviation amount of all blocks, the right-hand member abscissa exact value that is the block of searching for of departure minimum.

Watermarking images is owing to experienced a series of analog domains attacks, variations such as noise, rotation, translation, convergent-divergent, word edge blurry have unavoidably been introduced, block width and block all have been subjected to influence at interval, merge, screen if adopt the method identical to carry out block, be easy to produce mistake with original image.And the operation in original image is fully accurate and reproducible, the watermark that the block that can utilize original image to divide, merge, filter out obtains in the watermarking images embeds block and location block, and the method that this coarse positioning adds search can make watermarking algorithm to comprising the various attack that analog domain is attacked stronger robustness be arranged.

D) watermark extracting judgement:

Embed the change in location relation of block by each watermark in the watermark text image with respect to original image, can judge that it is toward moving to left or past moving to right that watermark embeds block when watermarked bit, the corresponding rule that embeds can be extracted the watermark bit that wherein embeds, and extracts all watermarks successively and embeds the watermark information that the watermark bit in the block sequence obtains embedding.

Embed the barycenter of block and location block thereof by watermark, can calculate in watermarking images and original image watermark respectively and embed the ratio that distance between the block barycenter is located in block and both sides.From the original image to the watermarking images, this ratio is because watermark embeds that block is turned left or right translation and variation has taken place, can judge the translation direction that embeds block with respect to watermark according to its variation relation, contrast embedding rule can determine that watermarked bit is 1 or 0 again.Can carry out subsequent treatment such as error correction, check code verification according to actual conditions.

The anti-rotation of the relative position of barycenter in block, Pan and Zoom, watermark embeds block and the anti-translation of barycenter spacing and the rotation of locating block, and watermark embeds block and left positioner block barycenter spacing is anti-rotation, translation, convergent-divergent with the ratio that watermark embeds block and right positioner block barycenter spacing, and also can not change a lot after the various analog domains attacks being subjected to.Adopt the relative position variation relation of barycenter between block to carry out the extraction of watermark information, make that the digital watermark method among the present invention has very strong robustness, can effectively resist various analog domains and attack, correctly extract embedded watermark information.

Beneficial effect of the present invention:

1. the present invention has provided a kind of digital watermark method of attacking at the anti-analog domain of text image, utilized text image to have the characteristics of septal area and word interval district in the ranks dexterously, as embedding target, succinctly and effectively embedded watermark information with the block that marks off;

2. the horizontal level by mobile block comes embed watermark information, has good aesthetic property and disguise, has avoided producing the prospect color dot or produce the influence to visual effect that the background color dot causes in the background colour district in the foreground district;

3. adopt watermark to embed block and embed and extract watermark information with the variation relation that the relative position of the location block of the left and right sides is compared with original image, and ingeniously combine the method that the barycenter that uses block calculates the distance between the block, can effectively resist various analog domains and attack variations such as the noise that causes, rotation, translation, convergent-divergent, word edge blurry;

4. adopt key and pseudo random sequence to generate watermark and embed the block sequence, in scramble watermark information binary sequence, make the position that embeds block also change, improved the fail safe of watermark information;

5. when extracting watermark, with the block in the numeric field image serves as with reference to carrying out coarse positioning, in watermarking images, search for then and obtain embedding block of watermark accurately and location block thereof, avoided analog domain to attack the block that causes and divided mistake, effectively improved the accuracy rate of watermark extracting.

Description of drawings:

Fig. 1 is the flow chart that watermark of the present invention embeds;

Fig. 2 is a watermark extracting judgement flow chart of the present invention;

Fig. 3 is used for watermarked original figure territory text image among the embodiment 1;

Fig. 4 is to the numeric field watermark picture and text image behind Fig. 3 embed watermark information;

Fig. 5 scans the gray scale watermark text image that obtains again for after printing with Fig. 4;

Fig. 6 is for carrying out the schematic diagram that block is divided to delegation's literal among Fig. 3 before watermarked;

Fig. 7 is a digital watermarking block among Fig. 3 and the schematic diagram of locating block thereof.

Embodiment:

Embodiment 1: below in conjunction with the watermark embedding and the watermark extracting judging process of actual text image, the present invention is further elaborated by embodiment:

Fig. 1 is the flow chart that watermark of the present invention embeds, Fig. 2 is a watermark extracting of the present invention judgement flow chart, Fig. 3 be imitation Song-Dynasty-style typeface font, little No. three font sizes, the generation of A4 paper mold e-text 300dpi resolution, be of a size of the bmp form two-value urtext image I of 2481 x, 3509 pixels; Fig. 4 is for being the seed key of pseudo random sequence with integer 211, with the length of integer 94728 correspondences is that the binary sequence (error correction information that has added 12 bits before the embedding, final watermarking information are 32 bits) of 20 bits (not enough 20 bits are high-order mends 0) embeds the watermarking images I that Fig. 3 obtains _w(key here and watermark information can be specified arbitrarily, are not particular value); Fig. 5 for Fig. 4 by the common laser printer with the 300dpi resolution printing, the gray scale watermark text image I ' of the 256 grades of gray scales of bmp form that obtain with the 300dpi resolution scan through the plain scan instrument is of a size of 2550 x, 3509 pixels again.(the concrete parameter in the example is only established for explanation, and it is fixed to be come by actual conditions in the application, below describes identical):

The whole embedding of the digital watermark method among the present invention and testing process can be segmented as follows and be described: a) watermarked preceding preliminary treatment; B) watermark embeds; C) extract the preceding preliminary treatment of watermark; D) watermark extracting judgement.

A) watermarked preceding preliminary treatment:

I is an original figure territory text image 1, and background colour (as white) is expressed as W, and foreground image (word content is as black) is expressed as B.At first divide and determine in this style of writing word the position coordinates at two ends about all blocks, in the literal line that marks off, utilize the word interval district to determine the position coordinates at two ends, the block left and right sides then, thereby realize the division of all blocks by row.At first utilize the capable division 2 of literal line spacer region, pixel with the lower left corner of original image I is the origin of coordinates (0,0) sets up rectangular coordinate system, original image I in the vertical direction is carried out transverse projection, also promptly add up each pixel column (capable delegation's literal that is not meant here, and be meant one-row pixels) have a number of foreground pixel, obtain the one-dimension array ProY[of 3509 elements like this], ProY[i] the expression ordinate is the foreground number of pixels (the element numbering of i and all arrays of occurring in the back is all since 0) of the pixel column of i.Blank spaces district ProY[i between the edge of text image blank parts and row and row] be zero, and the regional ProY[i at the word place of respectively composing a piece of writing] non-vanishing.From ordinate is that 0 pixel column begins to check array ProY[successively] (because the initial point of rectangular coordinate system is in the lower left corner of original image I, the row order of original image I is for from bottom to up), as ProY[i]〉0 and ProY[i-1]=0 the time, show that capable i is the initial pixel column of delegation's literal; As ProY[i+1]〉0 and ProY[i]=0 the time, show that capable i is the end pixel row of delegation's literal.Scanned after 3509 row, obtain having among the I 22 style of writing words, and the initial pixel column and the end pixel row-coordinate of each literal line have been obtained, with two bit array RowPos[22] [2] expression, RowPos[j wherein] the initial pixel column ordinate of [0] expression literal line j, RowPos[j] the end pixel row ordinate of [1] expression literal line j.So far the row of finishing I is cut apart.

With delegation's literal is the process that block division 3 is carried out in the example explanation: to literal line j, taking-up is by RowPos[j] [0] and RowPos[j] [1] literal rectangle region of limiting, carry out upright projection in the horizontal direction, add up the foreground number of pixels of each pixel column, obtain the one-dimension array Prox[2481 of 2481 elements], Prox[k wherein] the expression abscissa is the foreground number of pixels of the pixel column of k.From abscissa is that 0 pixel column begins to check successively Prox[], as Prox[k] 0 and Prox[k-1]=0 the time, show that row k is the initial pixel column of a block; As Prox[k+1]〉0 and Prox[k]=0 the time, show that row k is the end pixel row of a block.With literal line 0 (promptly " series report ") is example, mark off 6 blocks, use array WordPos[22] [6] [2] expressions, WordPos[j] the initial row abscissa of block n in [n] [0] expression literal line j, WordPos[j] the end column abscissa of block n in [n] [1] expression literal line j.To WordPos[0] [] [] check as can be known, block 1 and block 2 be respectively " row " this word about part, it is carried out block merges 4,5 blocks to the end, the division result before and after merging is as shown in Figure 6.(left, top) (right bottom) determines each block with lower right corner coordinate Y by its upper left corner coordinate X.Here each block is specifically by WordPos[j] [n] [2] and the RowPos[j that is expert at] [2] four values limit, also be upper left corner coordinate X (left, top)=(WordPos[j] [n] [0], RowPos[j] [1]), lower right corner coordinate Y (right, bottom)=(WordPos[j] [n] [1], RowPos[j] [0]).Can finish the block division of other row and the merging of part block after the same method.

Whether the block that marks off by inspection satisfies simultaneously that word is wide, the requirement of complexity and integrality, can be with such as ', ', ' 1 ', ' one ' and so on too simple block, and some part block that does not obtain merging weeds out, and finishes block screening 4.Remaining block constitutes and can gather Q by watermarked block, is used for subsequent watermark and embeds and watermark extracting.Here can need satisfy following condition by the interior block of watermarked block set: 1) the wide E of word, the wide unit of word is a pixel, $E>\frac{2}{3}\left(E_m\right)-3,$ And E＜(E _m+ 3), E _mFor the maximum word of complete block in text image that font, font size and print resolution are fixing wide, in the present embodiment word wide be less than 64 pixels greater than 38 pixels; 2) complexity.The definition of complexity is the ratio of all pixel quantity in interior foreground pixel quantity of block and the block, requires greater than 1/15; 3) integrality.Require block can not be incomplete Chinese character, letter or a symbol.

Watermark embeds the generation 5 of block sequence: the row of selecting some, with belong in these row can be watermarked the block of block set Q connect into a sequence A, can select a block s by key K 6 and pseudo random sequence in sequence A, judge whether this block satisfies following condition simultaneously: this block is before not selected to be that watermark embeds block; Can the watermarked block s left and right sides in original image I close position all exist belong to can watermarked block set Q location block f ₁, f ₂(being used for when extracting watermark, judging that watermark embeds the translation direction of block), and these two blocks are not that watermark embeds block; S respectively with f ₁, f ₂Distance all to surpass certain threshold value t, to avoid after moving block to overlap or the block interval variation too obviously influences visual effect, t changes to some extent according to the different of font and font size, can specify in advance or self adaptation is calculated and obtained.Add s to watermark embedding block sequence as above-mentioned condition is satisfied simultaneously, otherwise seek next block, up to obtaining the watermark embedding block sequence S={s that length is M _j(X _j, Y _j), j=1,2 ..., till the M}, the searching times that surpasses some then represents to find enough blocks watermarked.In multiline text, can begin to generate S with the literal behavior radix of some, then increase the literal line retry as generating the S failure, successfully generate S and then begin to attempt generating another S to repeat embed watermark information from next line.According to the embedding ability difference of text, can generate and mostly be the individual S of p (P=0) most, be used to repeat to embed identical watermark information to improve information redundance, information redundance can effectively improve the accuracy rate and the robustness of watermark extracting.

The watermark information that embeds is 32 bit-binary sequences, begin to attempt generating S with 4 style of writing words as radix, with the sequence number is the literal behavior example of 0-3, the block that belongs to Q in this 4 row is ascending by the row sequence number, block is linked to be one dimension block sequence A by left-to-right order in the row, and the length of A here is 62 in this example.With the seed generation random number random of key 211 as pseudo random sequence, calculate the position number iPos of the block in the A with random%62 (asking modular arithmetic), the block of checking the iPos correspondence need meet the following conditions simultaneously: be not chosen as watermark and embed block, the location block of left and right sides arest neighbors all belongs to A and is not chosen as watermark embedding block, with the blank spaces of the most contiguous block of the left and right sides more than or equal to 1/50 inch, be 6 more than the pixel under the resolution of concrete 300dpi in the present embodiment.Embed block if satisfy then elect it as watermark, add among the S, otherwise seek the next block that embeds with pseudo random sequence.If continuous 5000 times are not all found new block to add among the S, expression can not be found enough blocks, then increases delegation and regenerates A and attempt.In this example, the A that 0-3 is capable has 62 blocks can only generate 23 watermarks embedding blocks that satisfy condition, and the capable A of 0-4 has 83 blocks successfully to generate the long S of 32 blocks ₀Attempt generating next S with capable beginning of 5-8, last result is that 5-9 is capable, and 10-15 is capable, and the 16-20 provisional capital has successfully generated different S respectively ₁, S ₂, S ₃, can repeat to embed 4 times watermark information so in this text.

B) watermark embeds:

Here with S ₀Be example explanation watermark information telescopiny 7, the important information binaryzation sequence C={ c of watermark information 8 for needing protection that embed _j, j=1,2 ..., M} ∈ 0, and 1}, this important information can comprise error correcting code and check code, decides on different application.The integer 94728 that will embed is here represented with 20 bit-binary, adds that the error correcting code of 12 bits constitutes the 32 bit watermark information that will embed.S ₀={ s _j(X _j, Y _j), j=1,2 ..., each the block s among the 32} _jBe WordPos[j] [n] [2] and the RowPos[j that is expert at] [2] four rectangular areas that value limited, j represents the row number of place literal line here, n is the block sequence number in literal line.The 32 bit-binary watermark informations that add error correcting code are C=" 00010111001000001000101100001100 ", watermark bit are embedded that (lowest order is c according to the order of little-endian ₀, highest order is c ₃₁), if watermark bit c _j=1, with the watermark embedding block s of correspondence _jWhole toward 3 pixels of left, if watermark bit c _j=0, with the watermark embedding block s of correspondence _jWhole toward 3 pixels of right translation.So far finish S ₀Carry out the process that watermark embeds, can be according to same step to S ₁, S ₂, S ₃Embed same watermark information, generate watermarking images 10---the Iw of numeric field.

C) extract the preceding preliminary treatment of watermark:

By the common laser printer with 300dpi resolution printing I _w, again through the plain scan instrument with 256 grades of gray-scale map I ' of bmp form that the 300dpi resolution scan obtains, be of a size of 2550 x, 3509 pixels.For ease of effectively distinguishing foreground picture and Background, I ' is carried out binaryzation, here the threshold value of Cai Yonging is the empirical value 150 of appointment, (threshold value also can be determined by adaptive approach) is promptly to each pixel in the image, if its gray value, is judged to be preceding scenic spot pixel less than 150, make its gray value equal 0 (stain), otherwise be judged to be the background area pixel, make that its gray value is 255 (white points).

Because print scanned process 9 certainly will be introduced noise easily, especially the noise in background area can cause block to merge, screen and make a mistake, need carry out a step denoising, if but big noise spot, the stain of non-content of text and the adhesion of block etc. that whether watermark extracting failure hand inspection has denoising not remove can manually be repaired subsequently.Experiment shows that the water mark method among the present invention has robustness to text image in (0.5 degree, 0.5 degree) interior rotation of scope, does not need to do any registration process.

Next in gray scale watermark text image I ', carry out the extraction 14 that watermark embeds the location block of block and correspondence.At first original image I is carried out when watermarked identical block and divide 3, and use same key 6 and pseudo random sequence to extract watermark embedding block sequence S5, gray scale watermark text image I ' is also carried out same row division 11 and block divide 12, but do not carry out the subsequent operations such as merging of part block.With original image I is template, and the watermark of extracting as follows among the gray scale watermark text image I ' embeds block sequence S ₀' 14: establish block s ∈ S who is embedded into watermark among the original image I, its abscissa of being expert at V high order end foreground pixel and the rightest section foreground pixel is respectively D and E, and the abscissa of middle high order end foreground pixel of the corresponding row V ' among the I ' and the rightest section foreground pixel is respectively D ' and E '.The left end of s and the abscissa of right-hand member are L and R, concern the left end abscissa L ' and right-hand member abscissa R ' of the block s ' that is calculated in the middle correspondence of V ' by coordinate transform according to Pan and Zoom, and computing formula is as follows:

L′＝(L-D+1)·(E′-D′+1)/(E-D+1)+D′-1；

(1)

R′＝(R-D+1)·(E′-D′+1)/(E-D+1)+D′-1。

L ' that above formula draws and R ' have certain difference with actual value.For obtaining the exact value of L ' and R ',, find out the actual value that is L ' and R ' of difference minimum with the left end abscissa and the right-hand member abscissa calculation deviation amount of the block that has marked off among L ' that calculates and the same respectively V ' of R '.All watermarks that can find out successively among the I ' embed block.Choose watermark and embed in the block in a), one of them condition is to have two location block f among the Q ₁And f ₂Lay respectively at the left and right sides,, can obtain the location block f of s ' left and right sides according to calculating the same method of s ' ₁' and f ₂'.The extraction of location block is very simple, and when successfully generating s, the most contiguous block of the s left and right sides in the corresponding at this moment block sequence A is location block f ₁, f ₂

Here use identical pseudo random sequence seed key integer 211 and pseudo random sequence to obtain identical watermark to I and embed block sequence S ₀, S ₁, S ₂, S ₃With S ₀Be example, the location block sequence F of generation ₀Be expressed as F ₀={ (f _1j, f _2j), produce 1,2 ..., 32}.Watermark in obtaining I ' embeds block sequence S ₀' and location block sequence F ₀', need utilize in the ranks septal area and word interval district that gray scale watermark text image I ' is carried out preliminary block and divide, but not carry out the subsequent operations such as merging of part block.

Here with s ₀∈ S ₀Be example, s in I ₀The word " gold " of being expert at and being gone out by frame in 2, its upper left corner coordinate is (1210,706), lower right corner coordinate is (1271,641), as shown in Figure 7, the prospect color dot abscissa of 2 high order ends of being expert at is 379, and low order end is 2097, and the width at the preceding scenic spot of this journey is 2097-379+1=1719.The high order end and the low order end prospect color dot abscissa of the row 2 among the corresponding I ' are respectively 411,2084, and its prospect sector width is 2084-411+1=1674.Row 2 after attacking through analog domain as can be seen among the I ' is compared with the row 2 among the I Pan and Zoom (rotation amount of generation is very little, does not influence watermark extracting) has been taken place.According to the Pan and Zoom relation, utilize formula (1) can calculate s ₀Watermark in the middle correspondence of I ' embeds block s ₀' the abscissa at two ends, the left and right sides be respectively L '=1220, R '=1279.There are certain deviation in L ' that calculates and R ' with actual value, with L ' and R ' respectively with the middle s of I ' ₀' the left end abscissa and the right-hand member abscissa calculation deviation amount of all blocks in 2 of being expert at, find the actual value that is L ' and R ' of departure minimum.Here the row 2 of I ' has 28 blocks, obtains actual value L '=1223 by search, and R '=1283 are respectively 3 and 4 with the departure of calculated value 1220 and 1279, adds its be expert at 2 ordinate up and down 687 and 751 and defines block s ₀', its upper left corner coordinate is (1223,751), lower right corner coordinate is (1283,687).Can obtain all watermarks embedding blocks and location, the left and right sides block, wherein s among I and the I ' after the same method ₀The location block as shown in Figure 7.

D) watermark extracting judgement:

According to the embedding order of watermark information binaryzation sequence C, the process of watermark extracting 15 is: to s ∈ S and corresponding s ' ∈ S ', f ₁, f ₂, f ₁', f ₂', calculate the barycenter Z of each block respectively _S, Z _{S '}, Z ₁, Z ₂,

Centroid calculation with s is an example, s by its upper left corner coordinate (left, top) and lower right corner coordinate (right bottom) limits, and then its barycenter is:

$Z_s=\frac{\underset{y=\mathrm{top}}{\overset{\mathrm{bottom}}{\mathrm{\Σ}}}\underset{x=\mathrm{left}}{\overset{\mathrm{right}}{\mathrm{\Σ}}}{P}_{x,y}B(x,y)}{\underset{y=\mathrm{top}}{\overset{\mathrm{bottom}}{\mathrm{\Σ}}}\underset{x=\mathrm{left}}{\overset{\mathrm{right}}{\mathrm{\Σ}}}B(x,y)}---\left(2\right)$

Here P _{X, y}For vector (x, y),

The selection standard that watermark embeds block s has guaranteed all to exist in its left and right sides location block f ₁And f ₂, and f ₁And f ₂Can not be chosen as watermark and embed block, also be f ₁And f ₂Horizontal level can not change.If s has been moved to left, then s is near left positioner block f ₁And away from right positioner block f ₂If s has been moved to right, then s is away from left positioner block f ₁And near right positioner block f ₂S, f ₁, f ₂Barycenter between relative position relation just can reflect that s is with respect to f ₁And f ₂The direction of translation also just can extract the watermark bit of embedding.The relative position of barycenter in block is anti-rotation, Pan and Zoom, s respectively with f ₁, f ₂The distance of barycenter be anti-translation, rotation, therefore, can obtain s and f ₁The distance and s and f of barycenter ₂The ratio of distance of barycenter be anti-rotation, translation, convergent-divergent.Can be by Z _S, Z _{S '}, Z ₁, Z ₂,

Judge the translation direction of block by following formula:

If $\mathrm{Ratio}=\frac{D(Z_{S\&prime;},Z_1^{\&prime;})}{D(Z_{S\&prime;},{\mathrm{<figure-callout\; id="2"\; label="Z"\; filenames="C200510060488E00131.png"\; state="\{\{state\}\}">Z</figure-callout>}}_2^{\&prime;})}/\frac{D(Z_S,Z_1)}{D(Z_S,Z_2)}>1,$ Then watermark embedding block s has been moved to left;

(3)

If $\mathrm{Ratio}=\frac{D(Z_{S\&prime;},Z_1^{\&prime;})}{D(Z_{S\&prime;},{\mathrm{<figure-callout\; id="2"\; label="Z"\; filenames="C200510060488E00131.png"\; state="\{\{state\}\}">Z</figure-callout>}}_2^{\&prime;})}/\frac{D(Z_S,Z_1)}{D(Z_S,Z_2)}\&le;1,$ Then watermark embedding block s has been moved to right.

Here D (Z ₁, Z ₂) expression Z ₁, Z ₂2 Euclidean distance.

Rule when the contrast watermark information embeds can know that s toward moving to left, is 1 or 0 toward the corresponding respectively watermark bit that moves to right, and then can extracts complete watermark information.Can carry out subsequent treatment such as error correction, check code verification according to actual conditions.

s ₀Its upper left corner coordinate of ' block is (1223,751), and lower right corner coordinate is (1283,687); Left positioner block upper left corner coordinate is (1092,751), and lower right corner coordinate is (1147,687); Right positioner block upper left corner coordinate is (1288,751), and lower right corner coordinate is (1339,687).S among the original image I ₀Its upper left corner coordinate of block is (1210,706), and lower right corner coordinate is (1271,641); Left positioner block upper left corner coordinate is (1078,706), and lower right corner coordinate is (1134,641); Right positioner block upper left corner coordinate is (1279,706), and lower right corner coordinate is (1332,641).Calculate the barycenter of above 6 blocks respectively according to formula (2), the result is s ₀' barycenter (1252.55,714.736), s ₀' left positioner block barycenter (1117.97,720.988), s ₀' right positioner block barycenter (1312.09,716.948), s ₀Barycenter (1240.31,669.198), s ₀Left positioner block barycenter (1106.21,675.004), s ₀Right positioner block barycenter (1304.22,671.273).

Can calculate Ratio=0.8615＜1 according to formula (3), be judged as watermark and embed block and moved to right, according to the watermarked rule of correspondence, the block watermarked bit of representative that moved to right is 0.Can extract all watermarks according to same step and embed the watermark bit that embeds in the block, it is linked to be binary sequence in order, carry out error correction and just can obtain watermark extracting result 16.S in the present embodiment ₀, S ₁, S ₂, S ₃In can both correctly extract the integer 94728 of embedding.

Claims (10)

1. the text digital water mark method that anti-analog domain is attacked is characterized in that, may further comprise the steps:

A) watermarked preceding preliminary treatment: generation is divided, merges, screened to the block in the urtext image can gather by watermarked block, from selecting watermark embedding block the watermarked block set, formation watermark embedding block sequence;

B) watermark embeds: the horizontal shift that embeds the block position by each watermark embeds the watermark information of 1 bit, and value is 0 or 1; Again the binaryzation sequence of watermark information is embedded successively the watermark that the urtext image selects and embed in the block sequence, obtain adding the watermark text image of watermark;

C) extract the preceding preliminary treatment of watermark: the watermark text image is distinguished preceding scenic spot and background area; The urtext image is carried out and a) identical operation, generate watermark and embed block sequence and location block sequence thereof; To the watermark text image carry out with a) in identical block divide, and utilize the watermark of urtext image embed block sequence and location block sequence thereof assist generate and location watermark text image in watermark embed the block sequence and locate the block sequence;

D) watermark extracting judgement: embed the moving direction extraction watermark information of block according to each watermark in the watermark text image, adjudicate with respect to the watermark embedding block of urtext image;

At d) described in the watermark extracting judgement stage, extract watermark embedding information with the relative position of location, left and right sides block with the variation relation that the watermark embedding block in the original image is compared with the relative position of location, left and right sides block by the watermark embedding block in the judgement watermark text image, wherein the position of watermark embedding block is represented with center-of-mass coordinate.

2. the text digital water mark method that anti-analog domain according to claim 1 is attacked, it is characterized in that, the division of described block is to rely on septal area in the ranks and word interval district in the text image to carry out the block division, there is spacer region in the horizontal direction in the merging of described block to block inside and is divided into a plurality of part blocks by mistake, carrying out block merges, the block that the screening of described block is meant that word is wide, complexity and integrality satisfy predefined requirement simultaneously, just can enter can watermarked block set.

3. the text digital water mark method that anti-analog domain according to claim 2 is attacked is characterized in that, the requirement that wide to word during the screening of described block, complexity and integrality are set is: 1) the wide E of word, the wide unit of word is a pixel, needs to satisfy $E>\frac{2}{3}\left(E_m\right)-3,$ And $E<(E_m+3),$ E _mWide for being the fixing maximum word of complete block in text image of font, font size and print resolution; 2) complexity, the definition of complexity are the ratio of all pixel quantity in interior foreground pixel quantity of block and the block, and the requirement of complexity is greater than 1/15; 3) requirement of integrality is that block can not be incomplete Chinese character, letter or a symbol.

4. the text digital water mark method that anti-analog domain according to claim 1 is attacked, it is characterized in that, generating watermark embeds when each watermark embeds block in the block sequence, guaranteeing that this block left and right sides exists simultaneously is not that watermark embeds the location block of block, and must with the spacing of the block of the left and right sides respectively all more than or equal to 1/50 inch.

5. the text digital water mark method that anti-analog domain according to claim 1 is attacked is characterized in that described watermark embeds the formation of block sequence and controlled by key and pseudo random sequence.

6. the text digital water mark method that anti-analog domain according to claim 1 is attacked, it is characterized in that, embed the definition of 1 bit watermark information: by watermark selected in the urtext image is embedded block toward moving left, embed watermark information is a bit information of " 1 ", toward the bit information of the corresponding watermark information that embeds for " 0 " that move to right; Vice versa; Displacement needs more than or equal to 1/100 inch, and maximum moving distance is the upper limit can not cause after moving block to overlap.

7. the text digital water mark method that anti-analog domain according to claim 1 is attacked is characterized in that, described location block is positioned at the watermark embedding block left and right sides and embeds two the most contiguous blocks of block with this watermark for embedding in the block set.

8. the text digital water mark method of attacking according to the described anti-analog domain of the arbitrary claim of claim 1-7, it is characterized in that, before watermark extracting, the position coordinates that embeds block sequence and location block sequence with the watermark of the urtext image being divided generation is reference, watermark embedding block sequence to the watermark text image is carried out coarse positioning with the block of locating in the block sequence, obtains watermark by search watermark text image again and embeds block sequence and the accurate position of locating each block in the block sequence.

9. the text digital water mark method that anti-analog domain according to claim 8 is attacked, it is characterized in that, described watermark to the watermark text image embeds the method that block in block sequence and the location block sequence carries out coarse positioning: according to the abscissa at the two ends, the left and right sides of the position coordinates of this block in the urtext image, the place literal line of this block in the urtext image, and the abscissa at this literal line two ends, the left and right sides of corresponding literal line in the watermark text image, calculate the rough position coordinate of corresponding block in the watermark text image by coordinate transform.

10. the text digital water mark method that anti-analog domain according to claim 8 is attacked, it is characterized in that, the described method of obtaining the accurate position of block by search is: in the watermark text image, the left end abscissa calculation deviation amount of all blocks in left end abscissa that obtains by coarse positioning with block and the block place literal line, the left end abscissa exact value that is the block of searching for of departure minimum; In the watermark text image, the right-hand member abscissa calculation deviation amount of all blocks in right-hand member abscissa that obtains by coarse positioning with block and the block place literal line, the right-hand member abscissa exact value that is the block of searching for of departure minimum.

Images