CN108711132A - Digital watermark method based on Harris angle point resist geometric attacks - Google Patents
Digital watermark method based on Harris angle point resist geometric attacks Download PDFInfo
- Publication number
- CN108711132A CN108711132A CN201810438065.2A CN201810438065A CN108711132A CN 108711132 A CN108711132 A CN 108711132A CN 201810438065 A CN201810438065 A CN 201810438065A CN 108711132 A CN108711132 A CN 108711132A
- Authority
- CN
- China
- Prior art keywords
- watermark
- characteristic
- image
- point
- harris
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
- G06T1/005—Robust watermarking, e.g. average attack or collusion attack resistant
- G06T1/0064—Geometric transfor invariant watermarking, e.g. affine transform invariant
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Editing Of Facsimile Originals (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of digital watermark methods based on Harris angle point resist geometric attacks, and using the base embedded extraction watermarking algorithm of Harris, step is：First, the region of watermark insertion is found in the sign of rotation area to carrier image insertion and detection rotation angle, watermark region logo area, by completing image preprocessing to having resisted rotation attack and translation attack since indicating the adjustment of zone position, the correction of angle；Harris operator extraction characteristic points are used after completing image preprocessing, screen stable characteristic point, and by being divided into different small rectangular areas, then the size and quantity of the characteristic area of watermark information size requirements screening are pressed, again to corresponding watermarking images piecemeal, embedded watermark after the characteristic area wavelet transformation of its watermark information of selection matching, enhances the robustness of the anti-shearing attack of algorithm, completes the insertion and extraction of watermark.
Description
Technical field
The present invention relates to a kind of digital image watermark embedding methods, especially a kind of to use Harris angle point resist geometric attacks
Digital watermark method.
Background technology
Nowadays, can the digital information epoch arrive already, and multimedia technology is quickly grown, effective protection internet
Information security and the copyright of the numerous products of digital printing have become one of the hot spot of researchers' research.Digital picture water
The robustness for printing resist geometric attacks ability is also to be hoisted, and geometric attack can be in the synchronism of watermark extracted, be embedded in
Certain influence is generated, even if small attack can all make watermark wreck, extraction is caused to fail.Therefore, it is possible to find one
The method that kind resists geometric attack water mark inlaying of good performance is particularly important.It is most commonly seen in numerous methods of watermark insertion to be
The embedded watermark in spatial domain and transform domain, the typically direct modification wavelet field of traditional method water mark inlaying in wavelet field are
Number, although such methods can guarantee that the invisibility of watermark is relatively good, but be vulnerable to the shadow of threshold value in the compressed transform of image
Sound wrecks so as to cause watermark information.It is that water mark inlaying is typically directly modification in wavelet field that watermark, which is embedded in traditional method,
Although the coefficient of wavelet field, such methods can guarantee that the invisibility of watermark is relatively good, but in the compressed transform of image easily by
Influence to threshold value wrecks so as to cause watermark information.
Carrier image is embedded in by a kind of new anti-rotation blind watermarking algorithm based on logo area that Zhou Guangzhou et al. is proposed to be revolved
The method for turning logo area and watermark symbol area, this algorithm have good Shandong to any rotation to attacking and translating attack
Stick, but the invisibility analysis of watermark insertion is not done, and is influenced can only to be embedded in a small amount of watermark letter by carrier image
Breath.
Existing watermark embedding method invisibility is poor, similarity degree is low compared with original watermark, and watermark information insertion is bright
It can only be embedded in a small amount of watermark information by significantly limitation.It is good using algorithm watermark invisibility herein that experiment shows,
PSNR values are all higher than 65dB, and watermark similarity is high, and NC values are all higher than 0.8, and the embedded quantity of watermark is big.It is embedded in water using this method
Print, the extraction process of watermark do not need original image, have stronger Shandong to attacks such as rotation, translation, the spiced salt, shearing, JPEG
Stick.
Invention content
The present invention using Harris operators and is combined to carrier image by carrying out analysis of experimental results to existing method
The algorithm of embedding logo area is improved, and proposes the more excellent blind watermarking algorithm of resist geometric attacks performance, the algorithm is to carrier figure
As embedded logo area is sign of rotation area and watermark symbol area, sign of rotation area detection rotation angle, watermark region mark respectively
The region of watermark insertion is found in will area, by having resisted rotation attack peace since indicating the adjustment of zone position, the correction of angle
Move attack；This method uses Harris operator extraction characteristic points after completing image preprocessing, screens stable characteristic point by step
It is divided into different small rectangular areas, then presses the size and quantity of the characteristic area of watermark information size requirements screening, then right
Corresponding watermarking images piecemeal selects embedded watermark after the characteristic area wavelet transformation of its watermark information of matching, enhances algorithm
Anti-shearing attack robustness, can more easily complete the insertion and extraction of watermark.
The technical scheme is that：A kind of digital watermark method based on Harris angle point resist geometric attacks, using base
With the embedded extraction watermarking algorithm of Harris, step is：First, the sign of rotation area and detection of carrier image insertion are rotated
Angle, watermark region logo area find the region of watermark insertion, by having been supported since indicating the adjustment of zone position, the correction of angle
Anti-rotation is attacked and image preprocessing is completed in translation attack；Harris operator extraction characteristic points are used after completing image preprocessing,
Stable characteristic point is screened, and by different small rectangular areas is divided into, then presses the feature of watermark information size requirements screening
The size and quantity in region, then to corresponding watermarking images piecemeal, the characteristic area small echo that selection matches its watermark information becomes
Embedded watermark, enhances the robustness of the anti-shearing attack of algorithm, completes the insertion and extraction of watermark after changing.
It is described to use Harris operator extraction characteristic points, screen stable characteristic point, and by being divided into different small rectangles
Region the specific steps are：
One, characteristic point is extracted
Choose a small rectangular window, it is assumed that target pixel points are rectangular window center, make the rectangular window along each side
To minute movement is made, grey scale change amount is indicated using analytical expression, it is assumed that the wicket centered on (x, y) is in X direction
Mobile u, moves v, grey scale change amount analytic expression is expressed as along Y-direction：
Wherein：E (x, y) is window gray scale variable quantity；DefinitionFor window function；I is gradation of image
Function；Abbreviation omits infinite event：
EX, y=∑ ωX, y[u2(Ix)2+v2(Iy)2+2uvIxIy]
=Λ u2+2Cuv+Bv2 (2)
Wherein
EX, yQuadratic form indicates：
Wherein：M is real symmetric matrix：
Wherein：IxFor the gradients of image I in the x-direction；IyFor the gradients of image I in the y-direction.Angle point receptance function is CRF,
According to real symmetric matrix M, definition is expressed as：
CRF=det (M)-ktrace2(M) (5)
Wherein：Det (M) is the determinant of matrix M；Trace (M) is the mark of matrix；K is constant, takes 0.04~0.06；Than
Compared with response CRF and threshold value ThSize, the position of characteristic point is detected according to result；
Two, characteristic area screens
(1) the center l (c of labeled vector imagex, cy), it is distributed in the upper lefts l, upper right, lower-left, bottom right four panel region
Characteristic point takes it next to the position in the rectangular characteristic region of characteristic point in the upper left of characteristic point, upper right, lower-left, bottom right；
(2) all characteristic points obtained by Harris detections are extracted successively, these characteristic points is denoted as gathering
Ω1, then again from set omega1In seriatim filter out the larger characteristic point of those responses, be denoted as P0；
(3) from set omega1In delete P0There are the spies of lap in the corresponding small rectangular characteristic region of characteristic point in addition
Point is levied, preserves and is denoted as set omega2；
(4) by P0From set omega1Removal；Judge set omega1In the corresponding rectangular characteristic region of remaining characteristic point whether also
There is mutually superimposed phenomenon, above step is repeated if having Superposition Characteristics region；
(5) set omega obtained in above-mentioned steps1In, divide carrier image, containing in all zonules after piecemeal
Characteristic point central point hiDistance to l is denoted as set omega2And it sorts；
(6) according to set omega in step (5)2Characteristic area where order from big to small chooses corresponding characteristic point is made
For the position of embedded watermark, characteristic area is close to the center of carrier image when screening can be effectively ensured.
3. the digital watermark method according to claim 1 based on Harris angle point resist geometric attacks, feature exist
In：The base includes watermarking algorithm and watermark extraction algorithm with the embedded extraction watermarking algorithm of Harris：
One, watermarking algorithm
(1) carrier image f (x, y) is pre-processed first, it is special to standard picture f (x/ α, y/ α) extractions Harris after processing
Sign point, whereinWherein δ is preset value, m00For the zeroth order square of image, stability height, rectangular area are screened one by one
Be independent of each other the characteristic points of non-overlapping copies, is denoted as set omega1, in set omega1Rectangular area in the central points of all characteristic points
hiDistance to original image center l is denoted as set omega2, to set omega2It sorts by size；
(2) characteristic area is selected, from set omega2In choose the central point insertion water of four corresponding characteristic points from top to bottom
Print, i.e. characteristic area of the optimum selecting four close to carrier image center；
(3) watermarking images piecemeal carries out piecemeal according to characteristic area number and size to 32 × 32 watermarking images, will test
Watermarking images are allocated as 4 fritters, and pixel value is 16 × 16；
(4) wavelet transformation, a wavelet decomposition is carried out by the rectangular block for the characteristic area selected out, and h is carried out to LL subbands
× h does not repeat piecemeal,
(5) mean value solves, and solves the Coefficient Mean of each details coefficients and the mean value of each LL sub-band coefficients；
The number of wherein h × h wavelet coefficient in the block is expressed as M, and the sub-block of h × h is expressed as SDi, (i, j) block
Wavelet coefficient is expressed as c(i, j)(a, b)；
(6) modulation factor is calculated, it is as follows to calculate modulation factor：
P (i, j)=A+B*T (i, j) (7)
Wherein, T (i, j) is defined as follows：
Wherein indicate that intensity factor, β are constant with A, B, T (i, j) can still be embedded in some strength in the state of going to zero
Watermark；
(7) embedded watermark when watermark information W (i, j)=1, adjusts wavelet coefficient：
c' (i, j)(a, b)=c(i, j)(a, b)-E (i, j) (9)
When watermark information W (i, j)=0, wavelet coefficient is adjusted：
c' (i, j)(a, b)=c(i, j)(a, b)-E (i, j)+P (i, j) (10)
(8) wavelet inverse transformation restores to artwork size the subband small echo inversion containing embedded watermark, obtains and believe containing watermark
The image of breath；
Two, watermark extraction algorithm
(1) normalized, unified to image with watermarked information to make pretreatment i.e. image normalization, normalized can
It excludes certain image scaling etc. and brings unnecessary influence；
(2) Harris feature point extractions, if containing watermarking images by geometric attack, to continuing Harris by attack image
Corner Detection, invariant feature point can still retain；
(3) characteristic area is screened, the image characteristic region screening technique to having been inserted into watermark selectes characteristic area；
(4) wavelet transformation, the characteristic area that screening is completed carries out level-one wavelet transformation, on this basis by its LL subband
H × h piecemeals；
(5) sum, ask every piece coefficient and, the coefficients of each h × h fritters is added, summation is calculated：
(6) watermark extracting, as embedded W (i, j)=1, the coefficient and sum of fritter*Be similar to zero, when embedded W (i,
When j)=0, the coefficient and sum of fritter*It is similar to M × P (i, j), the number of the wavelet block coefficient of h × h is indicated with M：
Wherein, ThFor threshold value.
The beneficial effects of the invention are as follows：
1, solve the problems, such as that existing watermarking algorithm invisibility after embedded watermark is poor.
2, after solving the watermark extracting after attacking the image of embedded watermark, watermark and the initial condition print of extraction
Compare the low problem of similarity degree.
3, solve the problems, such as that existing method insertion watermark information amount is small, the watermark information amount of this method insertion is big.
4, the extraction process of this method watermark does not need original image.
5, this method with stronger robustness and has good resist geometric attacks ability to common signal processing attack.
Description of the drawings
Fig. 1 is the result figure of Harris characteristic points detection；
Fig. 2 is that watermark is embedded in flow chart；
Fig. 3 is characterized region the selection result figure；
Fig. 4 is original watermark image block diagram；
Fig. 5 is the extraction flow chart of watermark；
Fig. 6 is original image and embedded watermarking images；
Wherein：A is original image, and b is free of watermarking images (PSNR=72.95)；
Fig. 7 is the watermark figure of original watermark and extraction；
Wherein：A is original watermark, the watermark (NC=1) of b extractions.
Specific implementation mode
The invention will be further described with embodiment below in conjunction with the accompanying drawings.
A kind of digital watermark method of base Harris angle point resist geometric attacks, using the base embedded extraction water of Harris
Algorithm is printed, first, sign of rotation area and detection rotation angle to carrier image insertion, the searching watermark of watermark region logo area are embedding
The region entered, by completing image to having resisted rotation attack and translation attack since indicating the adjustment of zone position, the correction of angle
Pretreatment；Harris operator extraction characteristic points are used after completing image preprocessing, screen stable characteristic point, and by being divided into not
Then the size and quantity of the characteristic area of watermark information size requirements screening are pressed in same small rectangular area, then to corresponding
Watermarking images piecemeal selects embedded watermark after the characteristic area wavelet transformation of its watermark information of matching, enhances the anti-shearing of algorithm
The robustness of attack completes the insertion and extraction of watermark.
One, the method for screening characteristic area
1.Harris operators as a kind of feature extraction operator, not only rationally effectively again be easily achieved, detect basic thought retouch
State for：Choose a small rectangular window, it is assumed that target pixel points are rectangular window center, so that the window is made along all directions small
It is mobile, indicate grey scale change amount using analytical expression.Assuming that the wicket centered on (x, y) moves u along the directions M, along Y
V is moved in direction, and grey scale change amount analytic expression is expressed as：
Wherein：E (x, y) is window gray scale variable quantity；General definitionFor window function；I is image
Gamma function；Abbreviation omits infinite event：
EX, y=∑ ωX, y[u2(Ix)2+v2(Iy)2+2uvIxIy
=Λ u2+2Cuv+Bv2 (2)
Wherein
EX, yQuadratic form indicates：
Wherein：M is real symmetric matrix：
Wherein：IxFor the gradients of image I in the x-direction；IyFor the gradients of image I in the y-direction.Angle point receptance function is CRF,
According to real symmetric matrix M, definition is expressed as：
CRF=det (M)-ktrace2(M) (5)
Wherein：Det (M) is the determinant of matrix M；Trace (M) is the mark of matrix；K is constant, takes 0.04~0.06；Than
Compared with response CRF and threshold value ThSize, the position of characteristic point is detected according to result.Experiment under normal circumstances,
Make the characteristic point quantity that Harris Corner Detections obtain few larger it is necessary to which set threshold value is arranged.In addition to this,
Response CRF is bigger to indicate that the robustness of corresponding characteristic point is better.
It is that standardized test chart Lena figures and Peppers scheme used in Harris Corner Detections confirmatory experiment in this programme,
The use of platform is Matlab2015b, laboratory test results figure is shown in Figure 1.Can intuitively it find out from extraction result in figure, feature
For the main integrated distribution of point in texture complex region and grey scale change visibility point, algorithm picks response CRF is larger
Characteristic point.
Rational screening characteristic area needs to consider various factors, as that may contain phase in different characteristic region
Same characteristic point is also easy to lead to the friendship of watermark information if screening technique is not proper to not only result in characteristic area overlapping repeatedly
Mistake influences each other, therefore also needs to further screen, so as to get characteristic area include not only that stable characteristic point does not weigh mutually again
It is folded.
2. characteristic area screens
Step1：Center l (the c of labeled vector imagex, cy), it is distributed in the upper lefts l, upper right, lower-left, four section of bottom right
The characteristic point in domain takes it next to the position in the rectangular characteristic region of characteristic point in the upper left of characteristic point, upper right, lower-left, bottom right.
Step2：All characteristic points obtained by Harris detections are extracted successively, these characteristic points are denoted as collecting
Close Ω1, then again from set omega1In seriatim filter out the larger characteristic point of those responses, be denoted as P0。
Step3：From set omega1In delete P0There are laps in the corresponding small rectangular characteristic region of characteristic point in addition
Characteristic point, preserve and be denoted as set omega2。
Step4：By P0From set omega1Removal；Judge set omega1In the corresponding rectangular characteristic region of remaining characteristic point be
It is no to also have mutually superimposed phenomenon, repeat above step if having Superposition Characteristics region.
Step5：In the set omega that above-mentioned steps have obtained1In, divide carrier image, all zonules are included after piecemeal
The central point h of some characteristic pointsiDistance to l is denoted as set omega2And it sorts.
Step6：The size and number of characteristic area depend on the size of embedded watermarking images, characteristic area selection
Size and number are required to be adapted with watermark information and be matched.According to Ω in Step52Order from big to small chooses corresponding feature
Position of the characteristic area as embedded watermark where point, characteristic area is close to carrier image when screening can be effectively ensured in this way
Center.
Two, embedded extraction watermarking project algorithm
1. the algorithm of watermark insertion
Watermark is embedded in flow referring to Fig. 2.
(1) it Harris feature point extractions and screens.Carrier image f (x, y) is pre-processed first, to standard picture after processing
F (x/ α, y/ α) extracts Harris characteristic points, whereinWherein δ is preset value, m00For the zeroth order square of image, one by one
The characteristic point that screening stability is high, rectangular area is independent of each other non-overlapping copies, is denoted as set omega1, in set omega1Rectangular area
The interior central point h all characteristic pointsiDistance to original image center l is denoted as set omega2, to set omega2It arranges by size
Sequence.
(2) characteristic area is selected.It is tested herein from Ω2In choose from top to bottom four corresponding characteristic points central point it is embedding
Enter watermark, i.e. characteristic area of the optimum selecting four close to carrier image center, as shown in Figure 3.
(3) watermarking images piecemeal.Piecemeal is carried out according to characteristic area number and size to 32 × 32 watermarking images.Literary scheme
Experiment watermarking images are allocated as 4 fritters, pixel value is 16 × 16, as shown in Figure 4.
(4) wavelet transformation, a wavelet decomposition is carried out by the rectangular block for the characteristic area selected out, and h is carried out to LL subbands
× h does not repeat piecemeal.
(5) mean value solves.Solve the Coefficient Mean of each details coefficients and the mean value of each LL sub-band coefficients.
The number of wherein h × h wavelet coefficient in the block is expressed as M, and the sub-block of h × h is expressed as SDi, (i, j) block
Wavelet coefficient is expressed as c(i, j) (a, b).
(6) modulation factor is calculated.It is as follows to calculate modulation factor：
P (i, j)=A+B*T (i, j) (7)
Wherein, T (i, j) is defined as follows：
Wherein A, B indicate intensity factor, and β is constant, and T (i, j) can still be embedded in the water of some strength in the state of going to zero
Print.
(7) embedded watermark.
When watermark information W (i, j)=1, wavelet coefficient is adjusted：
c' (i, j)(a, b)=c(i, j)(a, b)-E (i, j) (9)
When watermark information W (i, j)=0, wavelet coefficient is adjusted：
c' (i, j)(a, b)=c(i, j)(a, b)-E (i, j)+P (i, j) (10)
(8) wavelet inverse transformation.Subband small echo inversion containing embedded watermark is restored to artwork size, obtains and believes containing watermark
The image of breath.
2. watermark extraction algorithm
The extraction process of watermark is the inverse process of watermarking algorithm, and detailed process can be found in Fig. 5.
(1) normalized.Unified to image with watermarked information to make pretreatment i.e. image normalization, normalized can
It excludes certain image scaling etc. and brings unnecessary influence.
(2) Harris feature point extractions.If containing watermarking images by geometric attack, to continuing Harris by attack image
Corner Detection, invariant feature point can still retain.
(3) characteristic area is screened.The same procedure that image to having been inserted into watermark is introduced with 1.2 selectes characteristic area.
(4) wavelet transformation.The characteristic area that screening is completed carries out level-one wavelet transformation, on this basis by its LL subband
H × h piecemeals.
(5) it sums.Ask every piece coefficient and, the coefficients of each h × h fritters is added, summation is calculated：
(6) watermark extracting.As embedded W (i, j)=1, the coefficient and sum of fritter*Be similar to zero, when embedded W (i,
When j)=0, the coefficient and sum of fritter*It is similar to M × P (i, j), the number of the wavelet block coefficient of h × h is indicated with M：
Three, this programme experimental result and analysis
Usually the invisibility of watermark is weighed with Y-PSNR (PSNR)[15], PSNR is bigger, embedded watermark can not
Opinion property is better；It is weighed by the robustness after Attack Digital Watermarking with normalizated correlation coefficient (NC), NC is bigger, and watermark robustness is better.
I (i, j) wherein in formula (13) indicates original image；That I ' (i, j) is indicated is the image after embedded watermark, M × N
Indicate image size.W (i, j) is original watermark in formula (14)；W ' (i, j) is the watermark of extraction.
The Lena gray-scale maps that pixel value is 1024 × 1024 are used in testing herein as carrier image, pixel value 32
× 32 bianry image makees watermarking images, and such as Fig. 6 a, 6b, 7a, shown in 7b, experiment uses Matlab2015b as verification platform.
Watermark is embedded, extraction is using parameter：β=0.318, A=7, B=1, Th=13.8.
The experiment of literary scheme is divided into 4 features that the watermark information that 4 fritter pixel values are 16 × 16 is sequentially embedded screening
In region.The algorithm watermark capacity is considerable, and the maximum embedded quantity of watermark information depends on the number of characteristic area.Watermark is believed
Breath piecemeal, which requires nothing more than, is divided into the pixel that fritter pixel is less than rectangular characteristic region, and watermark piecemeal quantity is no more than the spy filtered out
Sign region quantity can select the small rectangle being equal with watermark number of blocks successively from top to bottom according to the quantity of watermark information piecemeal
Characteristic area block is matching.The quantity of characteristic area is by shadows such as the sizes, texture complexity degree, grey scale change of carrier image
It rings, so carrier image is selected as far as possible big in experiment, watermarking images select smaller, can ensure the complete embedding of watermark information
Enter.
The experiment extracted after 1 rotation attack of table
Other geometric attack experimental results of table 2
After the operation for completing watermark insertion, 8 kinds of conventional attacks in table 1,2 are carried out respectively to the image embedded with watermark, it will
These images are considered as target image, and the performance simulation that watermark extracting is carried out to target image is tested, and table 1,2 is watermark figure respectively
Picture, NC and PSNR parameter lists.Data are respectively worth in analytical table：Normal image signal is attacked, extraction watermark robustness detection is real
It tests and 0.8033 is more than for rotation, translation, spiced salt attack NC values, 0.9170 is more than for other normal signals attack NC values, not
Under fire, be rotated by 90 °, 180 ° when NC values be 1, it is seen that by it is different attack extraction watermarks it is still high-visible.Transparency inspection
The PSNR for testing and obtaining is surveyed more than 66.024dB, and PSNR values are big to indicate that the invisibility of algorithm watermark insertion is fabulous.From each
The NC value PSNR values of the lower extraction watermark of kind attack fully demonstrate the New Image digital blind watermark algorithm proposed in text and are directed to
The conventional digital images processing attack such as rotation, translation, shearing, salt-pepper noise and JPEG compression shows good robustness, no
Visibility.
This programme proposes the resist geometric attacks digital watermarking algorithm based on Harris angle points, and the algorithm is in spatial domain embedded rotating
Logo area calculates image rotation angle, is corrected to its reverse rotation, by cleverly screening in stable characteristic area, at it
Wavelet field is embedded in watermark.The decoding process of the algorithm does not need original image, and watermark capacity is big, anti-rotation robustness, to cutting
Cut attack, routine information processing attack robustness it is all fine.But in view of watermark information embedded quantity is by characteristic area number
Limit value, so as possible selecting the bigger watermark information images of carrier image choosing smaller.
Claims (3)
1. a kind of digital watermark method based on Harris angle point resist geometric attacks, using the base embedded extraction watermark of Harris
Algorithm, it is characterised in that：First, the sign of rotation area to carrier image insertion and detection rotation angle, watermark region logo area
The region for finding watermark insertion, by having been attacked to resistance rotation attack and translation since indicating the adjustment of zone position, the correction of angle
It hits and completes image preprocessing；Harris operator extraction characteristic points are used after completing image preprocessing, screen stable characteristic point, and
By different small rectangular areas is divided into, the size and quantity of the characteristic area of watermark information size requirements screening are then pressed, then
To corresponding watermarking images piecemeal, embedded watermark after the characteristic area wavelet transformation of its watermark information of selection matching, enhancing is calculated
The robustness of the anti-shearing attack of method completes the insertion and extraction of watermark.
2. the digital watermark method according to claim 1 based on Harris angle point resist geometric attacks, it is characterised in that：Institute
It states and uses Harris operator extraction characteristic points, screen stable characteristic point, and by being divided into the specific of different small rectangular areas
Step is：
One, characteristic point is extracted
Choose a small rectangular window, it is assumed that target pixel points are rectangular window center, and the rectangular window is made to make along all directions
Minute movement indicates grey scale change amount, it is assumed that the wicket centered on (x, y) moves in the x-direction using analytical expression
U moves v along Y-direction, and grey scale change amount analytic expression is expressed as：
Wherein：E (x, y) is window gray scale variable quantity；DefinitionFor window function；I is gradation of image function；
Abbreviation omits infinite event：
Wherein
EX, yQuadratic form indicates：
Wherein：M is real symmetric matrix：
Wherein：IxFor the gradients of image I in the x-direction；IyFor the gradients of image I in the y-direction.Angle point receptance function is CRF, according to
Real symmetric matrix M, definition are expressed as：
CRF=det (M)-ktrace2(M) (5)
Wherein：Det (M) is the determinant of matrix M；Trace (M) is the mark of matrix；K is constant, takes 0.04~0.06；Compare sound
It should value CRF and threshold value ThSize, the position of characteristic point is detected according to result；
Two, characteristic area screens
(1) the center l (c of labeled vector imagex, cy), be distributed in the upper lefts l, upper right, lower-left, four panel region of bottom right feature
Point takes it next to the position in the rectangular characteristic region of characteristic point in the upper left of characteristic point, upper right, lower-left, bottom right；
(2) all characteristic points obtained by Harris detections are extracted successively, these characteristic points is denoted as set omega1, so
Afterwards again from set omega1In seriatim filter out the larger characteristic point of those responses, be denoted as P0；
(3) from set omega1In delete P0There are the features of lap in the corresponding small rectangular characteristic region of characteristic point in addition
Point preserves and is denoted as set omega2；
(4) by P0From set omega1Removal；Judge set omega1In the corresponding rectangular characteristic region of remaining characteristic point whether also have it is mutual
Superimposed phenomenon repeats above step if having Superposition Characteristics region；
(5) set omega obtained in above-mentioned steps1In, divide carrier image, the feature contained in all zonules after piecemeal
The central point h of pointiDistance to l is denoted as set omega2And it sorts；
(6) according to set omega in step (5)2Order from big to small chooses the characteristic area where corresponding characteristic point as embedding
Enter the position of watermark, characteristic area is close to the center of carrier image when screening can be effectively ensured.
3. the digital watermark method according to claim 1 based on Harris angle point resist geometric attacks, it is characterised in that：Institute
The embedded extraction watermarking algorithm for stating base Harris includes watermarking algorithm and watermark extraction algorithm：
One, watermarking algorithm
(1) carrier image f (x, y) is pre-processed first, Harris characteristic points is extracted to standard picture f (x/ α, y/ α) after processing,
WhereinWherein δ is preset value, m00For the zeroth order square of image, screen that stability is high, the mutual not shadow in rectangular area one by one
The characteristic point for ringing non-overlapping copies, is denoted as set omega1, in set omega1Rectangular area in the central point h of all characteristic pointsiTo original
The beginning distance of image center location l is denoted as set omega2, to set omega2It sorts by size；
(2) characteristic area is selected, from set omega2In choose the central point insertion watermarks of four corresponding characteristic points from top to bottom, i.e.,
Characteristic area of the optimum selecting four close to carrier image center；
(3) watermarking images piecemeal carries out piecemeal according to characteristic area number and size to 32 × 32 watermarking images, will test watermark
Image is allocated as 4 fritters, and pixel value is 16 × 16；
(4) wavelet transformation, a wavelet decomposition is carried out by the rectangular block for the characteristic area selected out, and h × h is carried out not to LL subbands
Piecemeal is repeated,
(5) mean value solves, and solves the Coefficient Mean of each details coefficients and the mean value of each LL sub-band coefficients；
The number of wherein h × h wavelet coefficient in the block is expressed as M, and the sub-block of hxh is expressed as SDi, the wavelet systems of (i, j) block
Number is expressed as c(i, j)(a, b)；
(6) modulation factor is calculated, it is as follows to calculate modulation factor：
P (i, j)=A+B*T (i, j) (7)
Wherein, T (i, j) is defined as follows：
Wherein indicate that intensity factor, β are constant with A, B, T (i, j) can still be embedded in the watermark of some strength in the state of going to zero；
(7) embedded watermark when watermark information W (i, j)=1, adjusts wavelet coefficient：
c′(i, j)(a, b)=c(i, j)(a, b)-E (i, j) (9)
When watermark information W (i, j)=0, wavelet coefficient is adjusted：
c′(i, j)(a, b)=c(i, j)(a, b)-E (i, j)+P (i, j) (10)
(8) wavelet inverse transformation restores to artwork size the subband small echo inversion containing embedded watermark, obtains containing watermark information
Image；
Two, watermark extraction algorithm
(1) normalized, pretreatment i.e. image normalization, normalized of making unified to image with watermarked information can exclude
Certain image scaling etc. brings unnecessary influence；
(2) Harris feature point extractions, if containing watermarking images by geometric attack, to continuing Harris angle points by attack image
Detection, invariant feature point can still retain；
(3) characteristic area is screened, the image characteristic region screening technique to having been inserted into watermark selectes characteristic area；
(4) wavelet transformation, the characteristic area that screening is completed carries out level-one wavelet transformation, on this basis by its LL subband h × h
Piecemeal；
(5) sum, ask every piece coefficient and, the coefficients of each hxh fritters is added, summation is calculated：
(6) watermark extracting, as embedded W (i, j)=1, the coefficient and sum of fritter*It is similar to zero, as embedded W (i, j)=0
When, the coefficient and sum of fritter*It is similar to MxP (i, j), the number of the wavelet block coefficient of h × h is indicated with M：
Wherein, ThFor threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810438065.2A CN108711132A (en) | 2018-05-09 | 2018-05-09 | Digital watermark method based on Harris angle point resist geometric attacks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810438065.2A CN108711132A (en) | 2018-05-09 | 2018-05-09 | Digital watermark method based on Harris angle point resist geometric attacks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108711132A true CN108711132A (en) | 2018-10-26 |
Family
ID=63868874
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810438065.2A Pending CN108711132A (en) | 2018-05-09 | 2018-05-09 | Digital watermark method based on Harris angle point resist geometric attacks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108711132A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685711A (en) * | 2018-12-29 | 2019-04-26 | 中山大学 | A kind of anti-rotation water mark method in characteristic area insertion cyclic graph |
| CN110570343A (en) * | 2019-08-14 | 2019-12-13 | 袁小晨 | Image watermark embedding method and device based on self-adaptive feature point extraction |
| CN113469862A (en) * | 2021-05-28 | 2021-10-01 | 安徽大学 | Tamper-proof digital watermark embedding method and image restoration method for file scanning |
| CN116630125A (en) * | 2023-04-20 | 2023-08-22 | 淮阴工学院 | Robust reversible watermarking algorithm based on SIFT and Harris corner detection |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101702230A (en) * | 2009-11-10 | 2010-05-05 | 大连理工大学 | Stable digital watermark method based on feature points |
-
2018
- 2018-05-09 CN CN201810438065.2A patent/CN108711132A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101702230A (en) * | 2009-11-10 | 2010-05-05 | 大连理工大学 | Stable digital watermark method based on feature points |
Non-Patent Citations (4)
| Title |
|---|
| 周广州 等: "一种基于Harris特征点和DWT_SVD的图像盲水印算法", 《包装工程》 * |
| 周广州 等: "一种新的小波域自适应图像盲水印算法", 《数据通信》 * |
| 李健 等: "一种基于harris角点的抗几何攻击的数字水印算法", 《太原理工大学学报》 * |
| 陈青 等: "一种新的基于标志区的抗旋转盲水印算法", 《包装工程》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685711A (en) * | 2018-12-29 | 2019-04-26 | 中山大学 | A kind of anti-rotation water mark method in characteristic area insertion cyclic graph |
| CN109685711B (en) * | 2018-12-29 | 2022-09-30 | 中山大学 | Anti-rotation watermark method for embedding periodogram in characteristic region |
| CN110570343A (en) * | 2019-08-14 | 2019-12-13 | 袁小晨 | Image watermark embedding method and device based on self-adaptive feature point extraction |
| CN110570343B (en) * | 2019-08-14 | 2023-04-07 | 袁小晨 | Image watermark embedding method and device based on self-adaptive feature point extraction |
| CN113469862A (en) * | 2021-05-28 | 2021-10-01 | 安徽大学 | Tamper-proof digital watermark embedding method and image restoration method for file scanning |
| CN113469862B (en) * | 2021-05-28 | 2022-05-17 | 安徽大学 | Tamper-proof digital watermark embedding method and image restoration method for file scanning |
| CN116630125A (en) * | 2023-04-20 | 2023-08-22 | 淮阴工学院 | Robust reversible watermarking algorithm based on SIFT and Harris corner detection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108711132A (en) | Digital watermark method based on Harris angle point resist geometric attacks | |
| Walia et al. | Digital image forgery detection: a systematic scrutiny | |
| Li et al. | Passive detection of doctored JPEG image via block artifact grid extraction | |
| CN110033471B (en) | Frame line detection method based on connected domain analysis and morphological operation | |
| Tang et al. | A feature-based robust digital image watermarking scheme | |
| US6408082B1 (en) | Watermark detection using a fourier mellin transform | |
| Kutter et al. | Towards second generation watermarking schemes | |
| Guo et al. | Duplication forgery detection using improved DAISY descriptor | |
| Uliyan et al. | A novel forged blurred region detection system for image forensic applications | |
| Fitton et al. | Optimising the application of the Hough transform for automatic feature extraction from geoscientific images | |
| CN112241929B (en) | Watermark embedding method, watermark detecting method, watermark embedding device, watermark detecting device and electronic equipment | |
| Zhu et al. | Blind image splicing detection via noise level function | |
| CN110189290A (en) | Metal surface fine defects detection method and device based on deep learning | |
| Alshayeji et al. | Detection method for counterfeit currency based on bit-plane slicing technique | |
| EP2240887A1 (en) | Feature-based signatures for image identification | |
| CN107908969A (en) | A kind of adaptive steganography method of jpeg image based on Spatial characteristic | |
| Gul et al. | SVD based image manipulation detection | |
| Wang et al. | Digital image watermarking based on texture block and edge detection in the discrete wavelet domain | |
| Lin et al. | Histogram-oriented watermarking algorithm: colour image watermarking scheme robust against geometric attacks and signal processing | |
| Shih et al. | 16 A Comparison Study on Copy–Cover Image Forgery Detection | |
| Panda et al. | Filtering corrupted image and edge detection in restored grayscale image using derivative filters | |
| CN108648130A (en) | A kind of total blindness's digital watermark method with copyright protection and tampering location function | |
| Dai et al. | Recognition of tunnel lining cracks based on digital image processing | |
| Hu et al. | Image fusion based visible watermarking using dual-tree complex wavelet transform | |
| Leung et al. | A robust watermarking scheme using selective curvelet coefficients |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181026 |
|
| RJ01 | Rejection of invention patent application after publication |