CN102262773A - Dual-threshold image lossless data embedding method - Google Patents
Dual-threshold image lossless data embedding method Download PDFInfo
- Publication number
- CN102262773A CN102262773A CN2010102004282A CN201010200428A CN102262773A CN 102262773 A CN102262773 A CN 102262773A CN 2010102004282 A CN2010102004282 A CN 2010102004282A CN 201010200428 A CN201010200428 A CN 201010200428A CN 102262773 A CN102262773 A CN 102262773A
- Authority
- CN
- China
- Prior art keywords
- embedding
- data
- value
- image
- gray
- 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
Images
Abstract
The invention provides a dual-threshold image lossless data embedding method, comprising the following steps: 1) setting an embedding threshold T and a fluctuation threshold TF and setting a variable P to be equal to T; 2) using an image window to scan an embedded image, calculating a grey prediction error y of a central pixel in the image window and a fluctuation value F of an adjacent domain at each scanning position, if F is less than or equal to TF, comparing the grey prediction error y with P, constructing a histogram pair, embedding data, determining a grey value of the central pixel in the image window and moving the image window to the next scanning position, and otherwise, moving the image window to the next scanning position directly; and 3) judging whether the to-be-embedded data are embedded completely, if the to-be-embedded data are embedded completely, taking P as an embedded stop point S and ending, and otherwise, setting P again and executing the step 2). The method is large in embeddable information, low in change of the data-embedded image and high in PSNR (peak signal to noise ratio).
Description
Technical field
The present invention relates to the image lossless data embedding grammar of lossless data embedded technology, particularly a kind of dual threshold.
Background technology
Image lossless data embedded technology, promptly the lossless data hiding technology can be divided into wavelet method and predicted method, wavelet method, as:
[1] J.Tian, " Reversible data embedding using a difference expansion; " IEEE Transactionon Circuits and Systems for Video Technology, 2003,13 (8): 890-896. (small echo adds the image lossless data embedding grammar of embedded location figure method);
[3] L.Kamstra and H.J.A.M.Heijmans, " Reversible data embedding into images usingwavelet techniques and sorting; " IEEE Transactions on Image Processing, vol.14, no.12, pp.2082-2090, December 2005. (small echo add embedded location figure method use the image lossless data embedding grammar of waiting line approach again);
[4] D.Coltuc and J.M.Chassery, " Very fast watermarking by reversible contrastmapping; " IEEE Signal Processing Letters, Vol.14, No.4, pp.255-258, April 2007. (small echo add embedded location figure method use the image lossless data embedding grammar of waiting line approach again);
[5] S.Lee, C.D.Yoo and T.Kalker, " Reversible image watermarking based oninteger-to-integer wavelet transform; " IEEE Transactions on Information Forensics and Security, vol.2, issue3, part 1, pp.321-330, September 2007. (small echo add embedded location figure method use the image lossless data embedding grammar of waiting line approach again);
[6] G.Xuan, Y.Q.Shi, P.Chai, J.Teng, Z.Ni, X.Tong, " Optimum histogram pair basedimage lossless data embedding, " Transactions on Data Hiding and Multimedia Security IV, LNCS 5510, Springer, pp.84-102, March2009 (the image lossless data embedding grammar that the small echo histogram is right);
[7] G.Xuan, Y.Q.Shi, P.Chai, X Cui, Z.Ni, X.Tong, " Optimum histogram pair basedimage lossless data embedding; " Proceedings of International Workshop on DigitalWatermarking, LNCS 5041, Springer, pp.264-278, Guangzhou, China, December 2007. (the image lossless data embedding grammar that the small echo histogram is right).
Predicted method, as:
[2] M.Thodi and J.J.Rodriguez, " Reversible watermarking by prediction-errorexpansion; " Proceedings of 6th IEEE Southwest Symposium on Image Analysis andInterpretation, pp.21-25, Lake Tahoe, CA, USA, March 28-30,2004. (the image lossless data embedding grammars of window predicted method);
[8] V.Sachnev, H.J.Kim, J.Nam, S.Suresh, Y.Q.Shi, " Reversible watermarkingalgorithm using sorting and prediction; " IEEE Transactions on Circuits and Systems for VideoTechnology, vol.19, no.7, pp.989-999, July 2009. (predicated error is the image lossless data embedding grammar of method and waiting line approach one by one).
With the Lena image is example: PSNR (PeakSignal to Noise Ratio)=45 when [1] J.Tian paper embeds capacity such as 0.1bpp (Bit Per Pixel) embeds max cap.=0.5bpp.Characteristics: small echo adds that embedded location figure method (LocationMap) embeds.The Tian[1 of field army in 2003] difference expansion (difference expansion technique) image reversible data embedded technology of proposing, started the beginning of the image lossless data hiding technique of high-quality and big embedding capacity.Field army is follow-up to improve one's methods: PSNR=50 when [3] L.Kamstra paper embeds capacity such as 0.1bpp, embed more than max cap.=0.7bpp, and [4] [5] are nearest development, PSNR=51 when wherein [5] S.Lee paper embeds capacity such as 0.1bpp.Characteristics: small echo adds that embedded location figure method (Location Map) embeds, and adds queuing (sorting) technology.Follow-up have an a series of improvement method, makes embedded performance that bigger raising be arranged.Histogram is proposed to method [7] (magazine [6] is delivered) in our the IWDW2007 meeting in 2007.Characteristics: small echo adds that histogram is to method.This method comprises than field army and follow-up improving one's methods thereof that than field army method higher performance is arranged.Our small echo histogram is to method, and PSNR=52 when embedding capacity such as 0.1bpp embeds more than max cap.=2.4bpp.Histogram further expands again method and is applied to jpeg compressed image in addition, and bianry image etc. become general insertion tool.
Be example with the Lena image equally, PSNR=52 when [2] M.Thodi paper in 2004 embeds capacity such as 0.1bpp.Characteristics: 3*3 window predicted method embeds data.Embedded performance and field army method are comparable, but it is cumbersome to solve harmless recovery problem.Because solve the improvement of harmless recovery problem in recent years, predicted method has the trend above the small echo embedding grammar.PSNR=54 when [8] V.Sachnev paper in 2009 embeds capacity such as 0.1bpp, the embedding max cap. is smaller.Characteristics: PSNR is very high, reaches at that time highest level in the world.Predicated error is the image lossless data embedding grammar of method and waiting line approach one by one, and the odevity row and column is separately done, but can only be with 4 neighborhoods of 3 * 3 windows.
Summary of the invention
It is big that the image lossless data embedding grammar that the purpose of this invention is to provide a kind of dual threshold, this method can embed information, embed data after image modification little, the PSNR height.
The image lossless data embedding grammar of dual threshold of the present invention, it is to be feature with the gray scale predicated error, in conjunction with the embedding threshold value T and the fluctuating threshold value T that set
F, adopt the right principle of histogram to embed data, this method may further comprise the steps:
1) sets embedding threshold value T and fluctuating threshold value T according to the embedding capacity
F, will embed threshold value T as embedding starting point, variable P=T is set;
2) sequential scanning is by the embedding image from left to right, from top to bottom with image window, and the gray scale predicated error y of center pixel and neighborhood fluctuating value F in each scanning position computed image window are if neighborhood fluctuating value F is less than or equal to fluctuating threshold value T
F, then relatively gray scale predicated error y and P, the structure histogram is right, embeds data, determines the gray-scale value of center pixel in the image window according to embedding gray scale predicated error y ' after the data, and image window moves to next scanning position then; Otherwise image window directly moves to next scanning position;
3) judge to treat whether embedding finishes the embedding data,, P at this moment as embedding halt S, is finished if embedding is intact; Otherwise, by positive and negative alternately and the principle that progressively reduces of absolute value reset P, go to step 2) continue to embed and treat the embedding data.
Preferably, gray scale predicated error y is actual grey value and prediction gray-scale value poor of center pixel in the image window, and wherein, described prediction gray-scale value is rounded downwards after weighted mean by the gray-scale value of each neighborhood territory pixel of described center pixel and obtains.Neighborhood fluctuating value F is on average obtained with the gray value differences square rear weight of its each neighborhood territory pixel respectively by the prediction gray-scale value of described center pixel.Preferred 3 * 3 image windows of image window preferably adopt 8 neighborhoods to calculate gray scale predicated error y and neighborhood fluctuating value F.Described fluctuating threshold value T
FWith the positive correlation of the capacity of embedding, embed the Y-PSNR and the fluctuating threshold value T of back image
FNegative correlation.
In step 2) in adopt one by one method to calculate gray scale predicated error y, promptly after the center pixel of this scanning position embeds data, with the gray-scale value after the embedding data, as the neighborhood territory pixel value of the center pixel of next scanning position.
Step 2) in, preferably adopt displacement method structure histogram right.
Step 2) method that embeds data in is: when P 〉=0, the gray scale predicated error y ' that embeds after the data equals gray scale predicated error y and the current embedding data sum for the treatment of; When P＜0, embed gray scale predicated error y ' after the data and equal gray scale predicated error y and currently treat the poor of embedding data; The described current embedding data for the treatment of are a bit.
In step 2) in, determine that according to the gray scale predicated error y ' after the embedding data method of the gray-scale value of center pixel in the image window is: the gray-scale value that embeds rear center's pixel equals described prediction gray-scale value
With the gray scale predicated error y ' sum after the embedding data.
Further also can comprise the step that the histogram of gray scale predicated error is adjusted, to overcome overflow and underflow problem, wherein, corresponding histogram adjustment information is with the described embedding data embedded images for the treatment of.
The present invention is feature with the predicated error, adopts to embed threshold value T and fluctuating threshold value T
F, the principle right by histogram embeds data, not only can realize the harmless recovery of image and embedding data, and can reach higher PSNR.Through test, adopt the Lena.bmp512*512 image, PSNR was higher than 55dB when the embedding capacity was 0.1bpp, and PSNR was higher than 67dB when the embedding capacity was 0.01bpp, and performance is better than existing other method.
Description of drawings
The process flow diagram that Fig. 1 embeds and extracts for data of the present invention, wherein (a) is for embedding the process flow diagram of data, (b) for extracting the process flow diagram of data;
Fig. 2 is the PSNR curve contrast figure of the inventive method and other method;
Fig. 3 is the grey level histogram of experimental test with figure and correspondence.
Embodiment
The present invention will be further described below in conjunction with drawings and Examples.
The principle of this dual threshold image lossless data embedding grammar:
Prediction error methods: the image lossless data embed and can carry out in the spatial domain, because natural image continuity in the spatial domain is obvious, correlativity is too strong, information redundancy is little, and it is more flat to be embodied in information distribution, and the information that therefore can embed is little, image modification is big after embedding data, and PSNR is low.By the predicated error conversion of image, continuity is not obvious in the predicated error, and correlativity dies down, and information redundancy is big, is embodied in relatively point of information distribution, and the information that therefore can embed is big, and image modification is little after the embedding data, the PSNR height.
The present invention adopts the predicated error embedding grammar, add and embed threshold value (also crying histogram right optimal threshold) and two threshold values of fluctuating threshold value (also cry and instruct threshold value), realize that by the right embedding principle of histogram the image lossless data embed, can access more performance, the common use of two kinds of threshold values is called " the image lossless data of dual threshold embed ".
The present invention is on the basis of document [6] [7], and a kind of new lossless data of proposition embeds scheme, and comparing the present invention with document [6] [7] has two important difference: 1, data are embedded in the gray scale predicated error of pixel, do not use wavelet transformation; 2, increased a threshold value, the threshold value that promptly rises and falls T
F, be used for the embedding of vectoring information.Its essence is that a kind of predicated error histogram that rises and falls under instructing at neighborhood embeds scheme to lossless data.
With reference to Fig. 1 (a), the inventive method specifically may further comprise the steps:
1) sets embedding threshold value T and fluctuating threshold value T according to the embedding capacity
F, will embed threshold value T as embedding starting point, variable P=T is set;
2) sequential scanning is by the embedding image from left to right, from top to bottom with image window, and the gray scale predicated error y of center pixel and neighborhood fluctuating value F in each scanning position computed image window are if neighborhood fluctuating value F is less than or equal to fluctuating threshold value T
F, then relatively gray scale predicated error y and P, the structure histogram is right, embeds data, determines the gray-scale value of center pixel in the image window according to embedding gray scale predicated error y ' after the data, and image window moves to next scanning position then; Otherwise image window directly moves to next scanning position;
3) judge to treat whether embedding finishes the embedding data,, P at this moment as embedding halt S, is finished if embedding is intact; Otherwise, by positive and negative alternately and the principle that progressively reduces of absolute value reset P and (promptly, then make P=-P-1 when P≤0; When P＞0, then make P=-P), go to step 2) continue to embed and treat the embedding data.
Gray scale predicated error y is actual grey value and prediction gray-scale value poor of center pixel in the image window, and wherein, described prediction gray-scale value is rounded downwards after weighted mean by the gray-scale value of each neighborhood territory pixel of described center pixel and obtains.Neighborhood fluctuating value F is on average obtained with the gray value differences square rear weight of its each neighborhood territory pixel respectively by the prediction gray-scale value of described center pixel.Preferred 3 * 3 image windows of image window preferably adopt 8 neighborhoods to calculate gray scale predicated error y and neighborhood fluctuating value F.Now be the calculating of example explanation gray scale predicated error y and neighborhood fluctuating value F with 3 * 3 image windows:
Definition 3 * 3 image windows are suc as formula (1)
X5 is the gray-scale value of center pixel in these 3 * 3 image windows in the formula (1), and x1, x4, x7, x2, x8, x3, x6, x9 are the gray-scale value of its 8 neighborhood territory pixels, just can obtain the prediction gray-scale value of center pixel by the gray-scale value of 8 neighborhood territory pixels
Here adopt formula (2) to obtain the prediction gray-scale value of center pixel
As seen, the prediction gray-scale value of center pixel
Round downwards after being actually the gray-scale value weighted mean of its 8 neighborhood territory pixels, be equivalent to:
The gray scale predicated error y of center pixel obtains by formula (3)
Neighborhood fluctuating value F represents intensity of variation, and the little expression of neighborhood fluctuating value F is smooth, and data are embedded among the flat gray scale predicated error y as far as possible among the present invention, make vision be difficult to perceive, and neighborhood fluctuating value F obtains by formula (4)
Certainly, image window of the present invention is not limited to 3 * 3 image windows.
In step 2) in adopt one by one method to calculate gray scale predicated error y, promptly after the center pixel of this scanning position embeds data, with the gray-scale value after the embedding data, as the neighborhood territory pixel value of the center pixel of next scanning position.Adopt one by one method can obtain the harmless recovery of data hidden at an easy rate.
Treat among the present invention that the embedding data are necessary for binary number, establish and treat that the embedding data are D that among the embedding data D is treated in each embedding here, with Di ∈ { 0,1} represents the current embedding data for the treatment of, wherein i represents the current embedding data i position in treating the embedding binary sequence for the treatment of, i is the integer more than or equal to zero.
The right structure of histogram belongs to prior art, all uses in the document [6] [7] (document [6] [7] is the inventor's paper) that background technology is quoted, and is clearer for making the present invention, and the right structure of histogram is done following explanation:
The definition that histogram is right: two adjacent feature x ∈ a, the histogram of b} is to being defined as follows: if a 〉=0 remembers that its histogram is to being h=[m, 0], or h=[h (a)=m, [h (b)=0], wherein a is an original position, the b extension bits.If a＜0 remembers that its histogram is to being h=[0, m], or h=[h (a)=0, [h (b)=m], wherein b is an original position, a extension bits.
By top definition as can be seen, for a pair of histogram, the number of times that must have an eigenwert to occur in sample is 0.The structure histogram to be exactly the number of times how in a pair of adjacent feature value, one of them eigenwert to be occurred in sample be 0, the simplest method is exactly a displacement method.The particular content of displacement method is: { if a 〉=0, then its histogram is to being [m, 0] for a, b}, and m is the number of times that a occurs in sample, if h (b) is 0, then itself is exactly that a pair of histogram is right, does not need to construct again to establish sample; If h (b) is not 0, then in the sample all more than or equal to the feature of b toward moving to right one (being that eigenwert adds 1), eigenwert is that the number of samples of b has been 0 just in the sample afterwards, histogram is to constructing thus.If b＜0, then its histogram is to being [0, m], if h (a) is 0, then itself is that a pair of histogram is right just, does not need to construct again; If h (a) is not 0, then in the sample all smaller or equal to the feature of a toward moving to left one (being that eigenwert subtracts 1), eigenwert is that the number of samples of a has been 0 just in the sample afterwards, histogram is to constructing thus.
Above-mentioned steps 2) in, adopt displacement method structure histogram right.Above-mentioned steps 2) in, compare gray scale predicated error y and P, the structure histogram is right, and the method that embeds data is:
When P 〉=0, if y=P then embeds data Di, y '=y+Di;
If y＞P does not then embed data, just move right one, y '=y+1;
Otherwise y does not embed data, is not shifted yet.
When P＜0, if y=P then embeds data Di, y '=y-Di;
If y＜P does not then embed data, just be moved to the left one, y '=y-1;
Otherwise y does not embed data, is not shifted yet.
Above-mentioned steps 2) method that embeds data in is: when P 〉=0, the gray scale predicated error y ' that embeds after the data equals gray scale predicated error y and current embedding data sum, i.e. the y '=y+Di for the treatment of; When P＜0, the gray scale predicated error y ' that embeds after the data equals gray scale predicated error y and current the poor of embedding data, i.e. the y '=y-Di for the treatment of; The described current embedding data for the treatment of are a bit Di.
Above-mentioned steps 2) in, determine that according to the gray scale predicated error y ' after the embedding data method that embeds the gray-scale value x5 ' of center pixel in the image window of back is: the gray-scale value x5 ' of embedding rear center pixel equals described prediction gray-scale value
With the gray scale predicated error y ' sum after the embedding data, promptly
The performance that information embeds, not only relevant with predicated error histogram taper degree, and also relevant with the pixel fluctuating around the window center.Therefore the present invention adopts and embeds threshold value T and fluctuating threshold value T
FTwo threshold value adjustment satisfy two conditions simultaneously and just embed data, can obtain higher PSNR like this.It is too little to embed threshold value T, can cause the embedding capacity not enough, and T is too big for the embedding threshold value, can cause PSNR to descend.Equally, fluctuating threshold value T
FWith the positive correlation of the capacity of embedding, embed the Y-PSNR (being PSNR) and fluctuating threshold value T of back image
FNegative correlation.
In this method, embedding starting point can be to embed threshold value T, also can be-T just can bear.Embedding halt S can be positive number, negative or zero, and satisfy 0≤| S|≤T.Will from embed starting point to the sequence T that embeds halt S ,-T, T-1 ,-(T-1) ..., S is called the optimum sequence that embeds, and represents with t, when the embedding capacity is given, determines to embed threshold value T and fluctuating threshold value T
F, promptly obtain optimum sequence t, the fluctuating threshold value T of embedding by T and S
FWith the selection of optimum embedding sequence t, the Y-PSNR that lossless data is embedded is the highest, promptly
Need repeatedly to attempt, select optimum fluctuating threshold value T
FEmbed sequence t with optimum, specifically can realize by above-mentioned principle coding.
Above-mentioned embedding grammar further also can comprise the step that the histogram of gray scale predicated error is adjusted, to overcome overflow and underflow problem.Just when detecting to overflow the time, image spatial domain histogram is adjusted in advance, that is, make histogram from the left side and/or the right to the third side to narrowing down, represent from the right the progression that shrinks to the center with parameter GR, represent from the left side to the progression of center contraction with parameter GL.Corresponding histogram adjustment information is recorded in " additional capacity " parameter, and data will be with the described embedding data D embedded images for the treatment of in " additional capacity " parameter.The histogram adjustment technology is a prior art, and disclosed histogram method of adjustment in the preferred employing document [6] [7] be not described in detail in detail here.
Data extract just in time is the anti-process that data embed.With reference to Fig. 1 (b), data extraction method of the present invention may further comprise the steps:
5) make variable P=S;
6) be embedded with the image of data with image window reverse scan (promptly from right to left, from top to bottom), at each scanning position, the gray scale predicated error y and the neighborhood fluctuating value F of center pixel in the computed image window are if neighborhood fluctuating value F is less than or equal to fluctuating threshold value T
F, then compare gray scale predicated error y and P, extract data, it is right to remove histogram, and the gray-scale value of the interior center pixel of image window moves to next scanning position then after definite extraction data; Otherwise image window directly moves to next scanning position;
7) whether judgment data has been extracted, if extracted, then finishes; Otherwise, by positive and negative alternately and the principle that progressively increases of absolute value reset P and (promptly, then make P=-P when P≤0; When P＞0, then make P=-P-1), go to step 6) and continue to extract data.
In the step 6), relatively gray scale predicated error y and P extract data, remove the right method of histogram and are:
When P 〉=0, if y=P or y=P+1 then extract data Di, y "=y-Di;
If y＞P+1 does not then extract data, just retract one left, y "=y-1;
Otherwise y does not extract data, is not shifted yet.
When P＜0, if y=P or y=P-1 then extract data Di, y "=y+Di;
If y＜P-1 does not then extract data, just move right one, y "=y+1;
Otherwise y does not extract data, is not shifted yet.
In the step 6), determine to extract after the data gray-scale value x5 of center pixel in the image window " method be: the gray-scale value x5 that extracts rear center's pixel " equal described prediction gray-scale value
With the gray scale predicated error y after the extraction data " sum, promptly
The present invention illustrate below in order can more to be expressly understood.The present invention preferably adopts 8 neighborhood methods to determine the gray scale predicted value of center pixel in the image window
Predicated error y and neighborhood fluctuating value F, but in order to simplify, be predicted as example with 4 neighborhoods below, illustrate:
The partial pixel value of supposing image is as follows:
154 158 160 158 160 162 162 162 167 168
158 158 159 160 158 163 162 162 166 169
153 158 157 158 161 162 162 163 164 157 definition, 3 * 3 image windows,
X5 is the gray-scale value of center pixel in the image window, and with respect to the prediction of 4 neighborhoods, the prediction gray-scale value x5 of center pixel calculates with formula (5)
The predicated error of center pixel is calculated with formula (6)
Neighborhood fluctuating value F calculates with formula (7)
(1) embedding of data
If fluctuating threshold value T
FBe 4, embedding threshold value T is 0, treats that embedding data D is { 1,0} is two bits altogether, treats that embedding data D is necessary for binary sequence, from left to right scans with 3 * 3 image windows, first scanning position following (in the wherein big solid box presentation video window, little solid box is center pixel)
The actual grey value x5=158 of center pixel in this moment window calculates the gray scale predicated error y=0 of center pixel, neighborhood fluctuating value F=1＜T with formula (5)-(7)
F, this point is selected, embeds first " 1 " for the treatment of embedding data D, so y '=y+1=1, the gray-scale value of this window center pixel
Shown in seeing in second scanning position in the frame of broken lines.Continue scanning, second scanning position is
The actual grey value x5=159 of center pixel by calculating, tries to achieve gray scale predicated error y=0, field fluctuating value F=6, F＞T at this moment
FSo this point does not have selected, do not embed data and do not change current point yet.Window continues to move, and the 3rd scanning position is
The actual grey value x5=160 of center pixel by calculating, tries to achieve gray scale predicated error y=2＞0, field fluctuating value F=1＜T at this moment
F, carry out the histogram expansion, y '=y+1=3, therefore, the gray-scale value of window center pixel
Shown in seeing in the 4th scanning position in the frame of broken lines.Window continues to move, and the 4th scanning position is
The actual grey value x5=158 of center pixel by calculating, tries to achieve gray scale predicated error y=-3＜0, field fluctuating value F=5＞T at this moment
FSo this point does not have selected, do not embed data and do not change current point yet.Window continues to move, and the 5th scanning position is
The actual grey value x5=163 of center pixel by calculating, tries to achieve gray scale predicated error y=2＞0, field fluctuating value F=12＞T at this moment
FSo this point does not have selected, do not embed data and do not change current point yet.Window continues to move, and the 6th scanning position is
The actual grey value x5=162 of center pixel by calculating, tries to achieve gray scale predicated error y=0, field fluctuating value F=1＜T at this moment
F, satisfy the embedding condition, embed second " 0 " for the treatment of embedding data D, so y '=y+0=0, the gray-scale value of this window center pixel
Promptly remain unchanged.
So far, data all embed and finish, and algorithm stops, and this point is for embedding cut off, S=0.
(2) extraction of data
The extraction of data is opposite with the data telescopiny, be from embedding cut off, and the scanning sequency of image window is opposite during with embedding.Make P=S=0, scan the image that has embedded data with 3 * 3 image windows, first scanning position is
This moment center pixel actual grey value x5=162, calculate the gray scale predicated error y=0=P of window center, field fluctuating value F=1＜T with formula (5)-(7)
F, satisfy extraction conditions, extract data " 0 ", therefore, y "=y-0=0,
Recover the gray-scale value 162 → 162 of window center pixel, see in second scanning position in the frame of broken lines shown in.Continue scanning, second scanning position is
Calculate the gray scale predicated error y=2 of center pixel, field fluctuating value F=12＞T
F, do not do any change, continue scanning, the 3rd scanning position is
Calculate the gray scale predicated error y=-3 of center pixel, field fluctuating value F=5＞T
F, do not do any change, continue scanning, the 4th scanning position is
Calculate the gray scale predicated error y=3＞P+1 of center pixel, field fluctuating value F=1＜T
F, represent that this point is selected, y "=y-1=2, therefore, the gray-scale value of window center pixel
Rehabilitation center's pixel, promptly 161 → 160, see in the 5th scanning position in the frame of broken lines shown in, do not extract data, continue scanning, the 5th scanning position is
Calculate the gray scale predicated error y=0 of center pixel, field fluctuating value F=6＞T
F, do not do change, continue scanning, the 6th scanning position is
Calculate the gray scale predicated error y=1=P+1 of center pixel, field fluctuating value F=1＜T
F, satisfy extraction conditions, extract data " 1 ", recover the gray-scale value of window center simultaneously, promptly 159 → 158, it is as follows to recover the back
As can be seen, the view data after the recovery with embed before the same, the data of extraction are arranged according to the order of sequence, obtain lossless data 1,0}.
Contrast test:
Respectively Lena.bmp 512*512 image is embedded data with the inventive method, [8] method, IWDW2007 (5,3) wavelet method, and carry out multiple embedding capacity experiment, experimental result is as shown in table 1.
Table 1 Lena.bmp 512*512 dual threshold histogram compares lossless data embedding grammar and additive method
| Data b pp | This paper method (8 neighborhood) psnr | [8] method psnr | IWDW2007 (5,3) wavelet method psnr | GL | GR |
| 0.01 | 67.1192({-4 5}，18) | 64.5 | / | 0 | 0 |
| 0.02 | 63.7495({-4 3}，20) | 61.0 | / | 0 | 0 |
| 0.03 | 61.7947({-3 3}，21) | 59.5 | / | 0 | 0 |
| 0.04 | 60.2693({-3 2}，22) | 58.0 | / | 0 | 0 |
| 0.05 | 59.1622({-3 2}，27) | 57.2 | 54.1 | 0 | 0 |
| 0.1 | 55.3834({-2 1}，35) | 54.0 | 51.2 | 0 | 0 |
| 0.2 | 50.8388({-1 0}，105) | 50.2 | 47.8 | 0 | 0 |
| 0.3 | 47.1297({-2?-1?0?1}，75) | 46.1 | 45.4 | 0 | 0 |
| 0.4 | 45.1135({-2?-?1?0?1}，300) | 44.3 | 43.4 | 0 | 0 |
| 0.5 | 43.0544({-3?-2?-1?0?1?2}，300) | 42.5 | 41.5 | 0 | 0 |
| 0.6 | 41.4742 ({-4，-3?-2?-1?0?1?2，3}，300) | 41.0 | / | 0 | 0 |
| 0.7 | 39.8815 ({-5，-4，-3?-2?-1?0?1?2，3，4}，550) | 39.0 | / | 0 | 0 |
In the table 1,67.1192 (4 5}, 18) represent with the inventive method at fluctuating threshold value T
FBe 18, the PSNR=67.1192 when optimum embedding sequence t is t={-4 5}, the rest may be inferred for other.The inventive method does not have to use the CrossSet and the DotSet of [8], and is to use common serial scan.This method is not used ordering yet, and is to use threshold value T, and method is simpler.Therefore this method is fully different with [8], but the result is better than [8], especially when little embedded quantity.From the experimental result of table 1 as can be seen, the inventive method is when embedding capacity during at 0.01bpp-0.7bpp, and PSNR all is higher than other two kinds of methods, and the embedding capacity is littler, and advantage of the present invention is bigger.
Method with document [1]-[7] in the inventive method, the background technology embeds data to Lena.bmp 512*512 image, experimental result is represented with Fig. 2, as can be seen from Figure 2, under identical embedding capacity, the embedding result of the inventive method is all better than document [1]-[7] method.
Among Fig. 3 (a) and (b), (c), (d) be respectively 512 * 512 Lena test with figure and grey level histogram, 512 * 512 Barbara test with figure and grey level histogram, 512 * 512 Baboon test with figure and grey level histogram, 960 * 768Woman (JPEG2000) test with figure and grey level histogram.Respectively four kinds of universal tests of Fig. 2 are embedded data with figure with the inventive method and document [2] [6] [8] method, result is as shown in table 2.
Table 2
As can be seen from Table 2, to different images, under difference embedding capacity, the inventive method all is better than other method.
As seen compared with prior art, the inventive method is under identical embedding capacity, and PSNR can reach higher level, and solves harmless recovery problem easily.
Claims (10)
1. the image lossless data embedding grammar of a dual threshold is characterized in that, this method is a feature with the gray scale predicated error, in conjunction with the embedding threshold value T and the fluctuating threshold value T that set
F, adopt the right principle of histogram to embed data, this method may further comprise the steps:
1) sets embedding threshold value T and fluctuating threshold value T according to the embedding capacity
F, will embed threshold value T as embedding starting point, variable P=T is set;
2) sequential scanning is by the embedding image from left to right, from top to bottom with image window, and the gray scale predicated error y of center pixel and neighborhood fluctuating value F in each scanning position computed image window are if neighborhood fluctuating value F is less than or equal to fluctuating threshold value T
F, then relatively gray scale predicated error y and P, the structure histogram is right, embeds data, determines the gray-scale value of center pixel in the image window according to embedding gray scale predicated error y ' after the data, and image window moves to next scanning position then; Otherwise image window directly moves to next scanning position;
3) judge to treat whether embedding finishes the embedding data,, P at this moment as embedding halt S, is finished if embedding is intact; Otherwise, by positive and negative alternately and the principle that progressively reduces of absolute value reset P, go to step 2) continue to embed and treat the embedding data.
2. the image lossless data embedding grammar of dual threshold as claimed in claim 1, it is characterized in that: described gray scale predicated error y is the actual grey value of center pixel in the image window and predicts the poor of gray-scale value, wherein, described prediction gray-scale value is rounded downwards after weighted mean by the gray-scale value of each neighborhood territory pixel of described center pixel and obtains.
3. the image lossless data embedding grammar of dual threshold as claimed in claim 2 is characterized in that: described neighborhood fluctuating value F is on average obtained with the gray value differences square rear weight of its each neighborhood territory pixel respectively by the prediction gray-scale value of described center pixel.
4. the image lossless data embedding grammar of dual threshold as claimed in claim 1, it is characterized in that: in step 2) in adopt one by one method to calculate gray scale predicated error y, promptly after the center pixel of this scanning position embeds data, embedding the gray-scale value after the data, as the neighborhood territory pixel value of the center pixel of next scanning position.
5. the image lossless data embedding grammar of dual threshold as claimed in claim 1 is characterized in that: described fluctuating threshold value T
FWith the positive correlation of the capacity of embedding, embed the Y-PSNR and the fluctuating threshold value T of back image
FNegative correlation.
6. the image lossless data embedding grammar of dual threshold as claimed in claim 1, it is characterized in that: described image window is 3 * 3 image windows, adopts 8 neighborhoods to calculate gray scale predicated error y and neighborhood fluctuating value F.
7. the image lossless data embedding grammar of dual threshold as claimed in claim 1 is characterized in that: step 2) in, adopt displacement method structure histogram right.
8. the image lossless data embedding grammar of dual threshold as claimed in claim 1, it is characterized in that: step 2) described in embed data method be: when P 〉=0, the gray scale predicated error y ' that embeds after the data equals gray scale predicated error y and the current embedding data sum for the treatment of; When P＜0, embed gray scale predicated error y ' after the data and equal gray scale predicated error y and currently treat the poor of embedding data; The described current embedding data for the treatment of are a bit.
9. the image lossless data embedding grammar of dual threshold as claimed in claim 2, it is characterized in that: in step 2) in, determine that according to the gray scale predicated error y ' after the embedding data method of the gray-scale value of center pixel in the image window is: the gray-scale value that embeds rear center's pixel equals described prediction gray-scale value
With the gray scale predicated error y ' sum after the embedding data.
10. the image lossless data embedding grammar of dual threshold as claimed in claim 1 is characterized in that: comprise that also wherein, the histogram adjustment information is with the described embedding data embedded images for the treatment of accordingly to the step of the histogram adjustment of gray scale predicated error.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010102004282A CN102262773A (en) | 2010-05-29 | 2010-05-29 | Dual-threshold image lossless data embedding method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010102004282A CN102262773A (en) | 2010-05-29 | 2010-05-29 | Dual-threshold image lossless data embedding method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102262773A true CN102262773A (en) | 2011-11-30 |
Family
ID=45009389
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010102004282A Pending CN102262773A (en) | 2010-05-29 | 2010-05-29 | Dual-threshold image lossless data embedding method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102262773A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107358147A (en) * | 2017-05-22 | 2017-11-17 | 天津科技大学 | Face recognition features' extraction algorithm based on local circulation graph structure |
| CN107689050A (en) * | 2017-08-15 | 2018-02-13 | 武汉科技大学 | A kind of depth image top sampling method based on Color Image Edge guiding |
| CN112333348A (en) * | 2020-10-26 | 2021-02-05 | 吉林大学 | Reversible data hiding method and system based on prediction error |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1379324A (en) * | 2002-05-09 | 2002-11-13 | 宣国荣 | Digital watermark method based on integer wavelet without damage to image |
| CN101246588A (en) * | 2008-03-20 | 2008-08-20 | 复旦大学 | Self-adapting watermarking algorithm of colorful image hypercomplex number spacing |
| US20080199093A1 (en) * | 2007-02-19 | 2008-08-21 | New Jersey Institute Of Technology | Appratus and method for reversible data hiding for jpeg images |
| CN101430786A (en) * | 2008-11-13 | 2009-05-13 | 哈尔滨工程大学 | Vector map lossless data hiding method based on vision perception characteristic |
| US20100098287A1 (en) * | 2008-10-17 | 2010-04-22 | Guorong Xuan | Reversible data hiding |
-
2010
- 2010-05-29 CN CN2010102004282A patent/CN102262773A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1379324A (en) * | 2002-05-09 | 2002-11-13 | 宣国荣 | Digital watermark method based on integer wavelet without damage to image |
| US20080199093A1 (en) * | 2007-02-19 | 2008-08-21 | New Jersey Institute Of Technology | Appratus and method for reversible data hiding for jpeg images |
| CN101246588A (en) * | 2008-03-20 | 2008-08-20 | 复旦大学 | Self-adapting watermarking algorithm of colorful image hypercomplex number spacing |
| US20100098287A1 (en) * | 2008-10-17 | 2010-04-22 | Guorong Xuan | Reversible data hiding |
| CN101430786A (en) * | 2008-11-13 | 2009-05-13 | 哈尔滨工程大学 | Vector map lossless data hiding method based on vision perception characteristic |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107358147A (en) * | 2017-05-22 | 2017-11-17 | 天津科技大学 | Face recognition features' extraction algorithm based on local circulation graph structure |
| CN107689050A (en) * | 2017-08-15 | 2018-02-13 | 武汉科技大学 | A kind of depth image top sampling method based on Color Image Edge guiding |
| CN107689050B (en) * | 2017-08-15 | 2020-11-17 | 武汉科技大学 | Depth image up-sampling method based on color image edge guide |
| CN112333348A (en) * | 2020-10-26 | 2021-02-05 | 吉林大学 | Reversible data hiding method and system based on prediction error |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8175324B2 (en) | Reversible data hiding | |
| Feng et al. | Reversible watermarking: Current status and key issues. | |
| CN108280797B (en) | Image digital watermarking algorithm system based on texture complexity and JND model | |
| CN101572819B (en) | Reversible image watermark method based on quantized DCT coefficient zero values index | |
| CN102801977B (en) | Method for embedding robust digital watermark in H.264 based on video complexity analysis | |
| CN103606136B (en) | Based on the video super resolution of key frame and non local constraint | |
| CN110191343B (en) | Adaptive video watermark embedding and extracting method based on variance analysis | |
| CN102324037A (en) | Shot boundary detection method based on support vector machine and genetic algorithm | |
| CN102262773A (en) | Dual-threshold image lossless data embedding method | |
| CN102223561B (en) | Blind watermark embedding and extracting method of stereoscopic video image | |
| Sheng-Li et al. | A digital watermarking algorithm based on region of interest for 3D image | |
| CN113099067A (en) | Reversible information hiding method and system based on pixel value sequencing prediction and diamond prediction | |
| CN116452401A (en) | Reversible robust watermark embedding and extraction model construction method for resisting image attack | |
| Xuan et al. | Double-threshold reversible data hiding | |
| CN101833745A (en) | Method for detecting embedding and extracting of multiple binary embedded watermarks of digital image | |
| EP1665123B1 (en) | Methods and apparatus for reversible data hiding through histogram modification | |
| Weng et al. | Invariability of Mean Value Based Reversible Watermarking. | |
| CN102760280B (en) | High-capacity reversible watermark embedding and extracting method as well as implement system thereof | |
| Lin et al. | A reversible data hiding scheme for block truncation compressions based on histogram modification | |
| Naheed et al. | Lossless data hiding using optimized interpolation error expansion | |
| CN109829846B (en) | Digital image blind watermarking method based on two-dimensional discrete cosine transform | |
| CN103544717A (en) | Two-phase three-dimensional image compression encoding method based on SIFT feature | |
| Yang | Robust Watermarking Scheme Based on Radius Weight Mean and Feature‐Embedding Technique | |
| Aekawit et al. | Expand variance mean sorting for reversible watermarking | |
| CN102314668A (en) | Difference-expansion digital-watermark-embedding improvement method for enhancing quality of watermark-embedded image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20111130 |