CN100481867C - A digital watermark embedding and extraction algorithm for space domain and small wave domain double sub-set - Google Patents

Abstract

The related digital watermarking embedded and extract algorithm based on space and wavelet field comprises: for embedding, using Walsh code to amplify frequency the original watermarking information into two groups of orthogonal information sequence, one is the space-field information sequence group with i orthogonal information sequences, another is the wavelet-field information sequence with j orthogonal information sequences; then, embedding first in space field then in wavelet field; for extracting, first in wavelet field then in space field, combining two groups of the two sequences by equal-gain technology to recover original information. This invention improves system robustness and comprehensive anti-interference performance.

Description

The digital watermarking embedding and the extraction algorithm of a kind of spatial domain and wavelet field two-set diversity

Technical field

The present invention relates to the digital watermarking embedding and the extraction algorithm of a kind of spatial domain and wavelet field two-set diversity, combine the diversity receiving technology of communication system simultaneously, is a kind of multimedia data protection method, belongs to the multimedia digital signal process field.

Technical background

In recent years, along with popularizing of computer and Intemet technology, and the development of multi-medium data storage and transmission technology, the transmission of digital product and sharedly become very easy, also make the bootlegging of digital product become very easy, so brought a series of problems about dispute over copyright.In order to protect copyright owner's interests, press for the intellectual property that finds a kind of effective method to protect digital product.Digital watermarking is exactly being applicable to digital picture of producing under such background, audio frequency, and digital products such as video carry out copyright protection, the technology of issue tracking and authentication.Digital watermarking mainly contains two big performance requirements, and one is invisibility, i.e. watermark is embedded into the decline that can not cause the perceived quality of original works in the copyright; One is robustness, and promptly watermarked copyright still can detect the existence of watermark or correctly recover the ability of watermark information after the attack through conventional signal processing and malice.It is the focus and emphasis of research both at home and abroad at present that the digital watermarking that is used for carrying out the multimedia data contents copyright protection requires watermark that the robustness of height, the robustness that improves watermark are arranged.

Attack to watermark mainly contains two class processing methods, and a class is to make the watermark extracting failure by the watermark information of removing or destroying in the digital product, mainly contains compression, filtering, modes such as making an uproar; Another kind of processing is by the shearing to digital product, convergent-divergent, and processing such as rotation can't realize when making watermark extracting synchronously, thereby make the watermark extracting failure.For present watermarking algorithm, transform-domain algorithm has compression, filtering, and the strong robustness that attacked by noise etc. are handled, and relatively poor to the geometric transformation attack robust; The robust watermarking at the geometric transformation attack of spatial domain design has robustness to geometric transformation, but lacks the superiority of transform-domain algorithm.The deficiency that existing digital watermarking embedding and extraction algorithm exist is to be difficult to realize simultaneously two classes being attacked the water mark method that all has better robustness.

Summary of the invention

The objective of the invention is to propose a kind ofly can resist geometric transformation and attack digital watermarking embedding and the extraction algorithm that can resist conventional attack again, it has higher robustness, can well resist this two class and attack, with the copyright of protection digital product.

For realizing such purpose, the present invention is performed such: generate two groups of orthogonal information sequences after watermark information is used the Walsh sign indicating number spread spectrum of quadrature, it is the spatial information (si) sequence set, comprise i quadrature information sequence and wavelet domain information sequence set, comprise j quadrature information sequence (i wherein, j is the integer more than or equal to 1), the elder generation spatial domain embeds wavelet field again and embeds, during watermark extracting, earlier extract the wavelet domain information sequence set from wavelet field, extract the spatial information (si) sequence set again from the spatial domain, use these the two groups of orthogonal information sequences of diversity receiving technology equal gain combining in the communication system again, thereby recover watermark information.Method of the present invention mainly comprises the embedding of watermark and extracts the two large divisions that the concrete operations step is as follows:

First: the embedding of watermark

(1) generation of watermark sequence: generate the watermark information of two-value random sequence P as embedded images, Walsh sign indicating number with quadrature carries out spread spectrum to this watermark sequence P, generate two groups of orthogonal information sequence set, wherein the spatial information (si) sequence set comprises i orthogonal information sequence, and the wavelet domain information sequence set comprises j orthogonal information sequence.The information sequence number of each information sequence group is adjustable, and according to the difference that the robustness of resisting various attack requires, every group spreading factor is also adjustable.

(2) spatial domain embeds: according to the mankind's visual characteristic, find the not too responsive zone of human eye perceives in the image, select the image texture close quarters among the present invention, the spatial information (si) sequence set is directly joined in the pixel value of image T, generate image T ₁The embedding rule is: I '=I (1+ α _id _i), wherein, I ' is the pixel value after watermarked, I is the pixel value of original image, d _iBe the information sequence that embeds, α _iBe intensity factor, the embedment strength of control watermark.

(3) wavelet field embeds: with the above-mentioned watermarked image T in the spatial domain ₁Carry out k level wavelet transformation, obtain T _1DWT, select LH at different levels, HL, the HH wavelet sub-band embeds the wavelet domain information sequence set that has generated, obtains

The embedding rule is: $W_j^{\′}=W_j(1+{\mathrm{\α}}_j{d}_j),$ Wherein,

Be the wavelet coefficient after watermarked, W _jBe not watermarked wavelet coefficient, d _jBe the information sequence that embeds, α _jBe intensity factor, the embedment strength of control watermark.Respectively at LH at different levels, HL, the HH wavelet sub-band embeds, and has both guaranteed the robustness of watermark to various conventional attacks, has also guaranteed the statistical independence that each orthogonal sequence embeds.

(4) contain the generation of watermarking images: to above-mentioned

Carry out inverse wavelet transform, then can obtain simultaneously at spatial domain and the watermarked image T ' of wavelet field.

Second portion: the extraction of watermark

(1) wavelet field is extracted: to what receive, may through attack handle contain watermarking images T ^*Carry out k level wavelet transformation, obtain T ^* _DWT, respectively from the LH at different levels of correspondence, HL, the HH wavelet sub-band extracts the wavelet domain information sequence set, obtains T ^* _1DWT, carry out inverse wavelet transform again, obtain image T ^* ₁Extract rule for embedding the inverse transformation of rule: $d_j=\frac{(\frac{W_j^{\′}}{W_j}-1)}{{\mathrm{\α}}_j}.$

(2) extract in the spatial domain: for the above-mentioned image T that extracts watermark in wavelet field ^* ₁, identical rule finds the respective texture compact district of image when embedding according to the spatial domain, utilizes the inverse transformation of embedding formula from its pixel value $d_i=\frac{(\frac{I\′}{I}-1)}{{\mathrm{\α}}_i}$ Extract the spatial information (si) sequence set.

(3) merge processing: have irrelevance between each information sequence of above-mentioned wavelet domain information sequence set that extracts and spatial information (si) sequence set from image, utilize the equal gain combining technology that these multiple signals are merged, the merging formula is $d=\underset{m=1}{\overset{i+j}{\mathrm{\Σ}}}{\mathrm{Gd}}_m,$ Wherein G is the equal gain combining coefficient, and d is the output information sequence after merging, d _mBe respectively each information sequence that extracts, 1≤m≤i+j, thus recover former watermark information.

Adopt the digital watermarking embedding of spatial domain of the present invention and wavelet field to compare with existing water mark method with extraction algorithm, its difference is:

(1) is two groups of orthogonal information sequences with a watermark information sequence spread spectrum among the present invention, and is embedded into the spatial domain and the wavelet field of original image respectively.Only the watermark of carrying out at transform domain embeds and extracts, the geometric transformation of carrying out in spatial domain (is sheared, convergent-divergent, rotations etc.) robustness is very poor, even when accurately knowing the geometric transformation type of digital picture experience, the spectral coefficient that the image of returning by how much inverse transformations passes through wavelet transformation more also can great changes have taken place, and the watermark that then extracts just has very mistake even extracts less than watermark at all, thus among the present invention the while in the spatial domain also embed watermark information.Consider directly on image pixel embed watermark information to the influence of image invisibility, the number that can regulate the spatial information (si) sequence with and spreading factor.Different qualities when embedding detection for each subband of wavelet field, the DWT low frequency is big to the invisibility influence, a little less than the attack opposing that changes a lot of pixel values, robustness to high-pass filtering class image processing is also very poor, DWT high frequency invisibility is good, capacity that can be watermarked is big, but low-pass filtering class image processing there is not robustness again, so consider image is carried out k level wavelet transformation, be chosen in wavelet sub-band LH, HL, embed watermark information among the HH then both can well be realized invisibility, had guaranteed that again watermark is to various processed conventionally robustnesss, statistical independence when the not plyability of these three frequency ranges can also guarantee the extraction of quadrature information sequence simultaneously is to realize the improvement of Diversity Reception to system robustness.

(2) the present invention utilizes the diversity receiving technology of communication system, statistical independence by each road watermark information sequence of extracting from the not overlapping subband of spatial domain and wavelet field respectively, merge each information sequence with the equal gain combining technology, completely orthogonal frequency expansion sequence does not exist multiple access to disturb when guaranteeing to merge, thereby can better recover watermark information.When image is subjected to various attack, because the information sequence of spatial domain and wavelet field can provide the information of original watermark, use diversity receiving technology, can improve the average signal-to-noise ratio of this watermaking system model to a great extent, thereby improve the robustness of watermark.

In sum, digital watermarking embedding of the present invention and its advantage of extraction algorithm are: the two-set diversity that utilizes spatial domain and wavelet field, by utilization quadrature Walsh sign indicating number spread spectrum, while embeds the diversity receiving technology of information sequence and communication system in spatial domain and wavelet field, make watermark not only be subjected to conventional attack, as compression, filtering, making an uproar etc., the robustness of doing well, be subjected to the geometric transformation attack, as shearing convergent-divergent, rotation etc., good robustness is also arranged, improved the robustness of watermaking system to a great extent.

Description of drawings

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

Fig. 2 is a watermark extracting flow process of the present invention.

The embedded location schematic diagram of wavelet field when Fig. 3 is the secondary wavelet transformation.

Fig. 4 is original image (Lena 512 * 512).

Embodiment

Below in conjunction with accompanying drawing embodiments of the present invention are described in detail.

Fig. 1 is the embedding flow chart of watermark, and Fig. 2 is the extraction flow chart of watermark, and the inventive method comprises following concrete steps according to Fig. 1 Fig. 2 flow process:

1. the generation of quadrature information sequence: water intaking official seal breath length is 128bit when specifically implementing.According to Fig. 1, at first generate two-value random sequence P, P={p _i| p _i=± 1,1≤i≤128}, promptly watermark information is 128bit, again P is carried out quadrature Walsh sign indicating number spread processing.For k level wavelet transformation, the frequency coefficient of each k level subband is $w_{\mathrm{ij}},(1\≤i\≤\frac{M}{2^k},1\≤j\≤\frac{N}{2^k}),$ Promptly be at the contained frequency coefficient of each k level subband

Individual, wherein M * N is the size of image, when specifically implementing, the present invention selects the secondary wavelet transformation for use, the Lena image of original images by using 512 * 512, as shown in Figure 4, i.e. M=N=512 is so the contained frequency coefficient of each secondary subband is 128 * 128, embed at the secondary subband during concrete enforcement, promptly embedded location is LH ₂, HL ₂, HH ₂So the spreading factor that gets the Walsh sign indicating number of wavelet domain information sequence set is 2 ⁷, obtain the sequence d of quadrature _j, 1≤j≤3, the bit of each information sequence is 128 * 2 ⁷, assurance is identical with the number of the frequency coefficient of secondary wavelet sub-band.Simultaneously P is carried out quadrature Walsh sign indicating number spread spectrum, generate the spatial information (si) sequence set of quadrature, by noted earlier, the size of spreading factor and i value can require and select for use the gamma characteristic of image and texture features to adjust according to the robustness of system, can be identical with the spreading factor of wavelet domain information sequence also can be different.In the specific implementation, adopt identical spreading factor 2 ⁷, generate spatial information (si) sequence set d _i,,, get i=1 for well guaranteeing invisibility according to the texture features of Lena image.

2. according to flow chart shown in Figure 1, at first embed information sequence group d in the spatial domain _i, in the texture compact district of image T, find and the identical pixel of the contained information bit number of spatial information (si) sequence set, d _iDirectly join in these pixel values, generate image T ₁The embedding rule is: I '=I (1+ α _id _i), wherein, I ' is the pixel value after watermarked, I is the pixel value of original image, α _iBe intensity factor, the embedment strength of controlled air space watermark, α _iSpan be 0.1～0.9, select α according to the requirement compromise of the requirement of invisibility and robustness _iSize.Then with T ₁Carry out the secondary wavelet transformation, select wavelet sub-band LH ₂, HL ₂, HH ₂, the corresponding d of embedding is distinguished as shown in Figure 3 in its position _j(1≤j≤3), the embedding rule is $W_j^{\′}=W_j(1+{\mathrm{\α}}_j{d}_j),$ Wherein

Be the wavelet coefficient behind the embedding information sequence, W _jBe former wavelet coefficient, α _jBe intensity factor, control the embedment strength of each wavelet sub-band watermark, j=1,2,3, represent LH respectively ₂, HL ₂, HH ₂Embedding character, consider the requirement of invisibility and robustness simultaneously, regulate corresponding α according to the character of each subband _jSize.

3. according to flow chart shown in Figure 2, to the watermarked image T that may be subjected to geometric attack or conventional attack ^*Carry out watermark extracting.Because pixel value is bigger to the influence of wavelet coefficient, so carry out the watermark extracting of wavelet field earlier.With image T ^*All carry out the secondary wavelet transformation with original image T, select wavelet sub-band LH ₂, HL ₂, HH ₂, utilize the inversionization that embeds rule, $d_j^{\′}=\frac{(\frac{W_j^{\′}}{W_j}-1)}{{\mathrm{\α}}_j},$ Extract the wavelet domain information sequence set $d_j^{\′}(1\≤j\≤3),$ Carry out inverse wavelet transform again, obtain image T ^* ₁Seek the principle that the original image watermark embeds pixel when embedding, find image T according to the spatial domain ^* ₁Corresponding pixel points, according to the pixel value of original image, utilize the inverse transformation of embedding formula, $d_i^{\′}=\frac{(\frac{I^{\′}}{I}-1)}{{\mathrm{\α}}_i},$ Extract the spatial information (si) sequence set

4. according to flow process shown in Figure 2, to the information sequence group that has extracted

With $d_j^{\′}(1\≤j\≤3),$ Utilize the diversity receiving technology in the communication system, equal gain combining, the merging formula is: $d=\underset{m=1}{\overset{i+j}{\mathrm{\Σ}}}{\mathrm{Gd}}_m,$ Wherein G is the equal gain combining coefficient, and d is the output information sequence after merging, d _mBe respectively each information sequence that extracts, 1≤m≤i+j, thus recover former watermark information P.Principle according to equal gain combining: for the separate received signal in N road, corresponding channel parameter estimation is all done on each road then all N path channels parameters and N road received signal are sent into multipath receiver decoding, because the present invention utilizes the spatial information (si) sequence set with the wavelet domain information sequence set thereby different attacks to be estimated to recover watermark information, so in the invention process, select for use equal gain combining right With $d_j^{\′}(1\≤j\≤3)$ Merge processing, obtain watermark information.

5. the flow chart shown in Figure 2 according to Fig. 1 embeds and extracts watermark, owing to utilized the character of spatial domain watermark and frequency domain water mark among the present invention simultaneously, and utilized the equal gain combining technology that watermark is subjected to the extraction influence that various attack brings simultaneously and carried out parameter Estimation, make watermarking algorithm of the present invention attack, as shearing rotation in opposing geometric transformation, convergent-divergent etc., and conventional attack, as compression, filtering, and attacked by noise etc., all have stronger robustness.

Claims (5)

1. the digital watermarking embedding and the extracting method of spatial domain and wavelet field two-set diversity is characterized in that: comprise following concrete steps:

(1) watermark information spread processing: watermark information generates two groups of orthogonal information sequences with quadrature Walsh sign indicating number spread spectrum: the spatial information (si) sequence set, comprise i quadrature information sequence, with the wavelet domain information sequence set, comprise j quadrature information sequence, i wherein, j is the integer more than or equal to 1, according to the different robustness requirements of resist geometric attacks and conventional attack, can adaptive adjustment every group spreading factor;

(2) the spatial domain watermark embeds: utilize the gray feature and the textural characteristics in gray level image spatial domain, find the texture compact district of image, with the property taken advantage of rule above-mentioned spatial information (si) sequence set is embedded wherein;

(3) the wavelet field watermark embeds: the image of spatial domain embedding carries out k level wavelet transformation, and wherein k is the integer more than or equal to 1, selects LH at different levels, HL, and the HH wavelet sub-band embeds above-mentioned wavelet domain information sequence set with the property taken advantage of rule;

(4) containing watermarking images generates: the above-mentioned image of finishing the wavelet field embedding is carried out inverse wavelet transform, generate the image of moisture official seal breath;

(5) watermark extracting: during watermark extracting, earlier image is carried out k level wavelet transformation, corresponding wavelet sub-band when embedding utilizes the inverse transformation of embedding formula to extract the wavelet domain information sequence set respectively, behind the inverse wavelet transform again the respective texture compact district in the spatial domain extract the spatial information (si) sequence set;

(6) merge: the spatial information (si) sequence set that extracts and wavelet domain information sequence set as reception information, are utilized the two-set diversity of spatial domain and wavelet field, merge this two groups of irrelevant information sequences with the equal gain combining technology, its equal gain combining formula is: $d=\underset{m=1}{\overset{i+j}{\mathrm{\Σ}}}{\mathrm{Gd}}_m,$ Wherein G is for waiting gain weight coefficient, and d is the output information sequence after merging, d _mBe respectively each information sequence that extracts, 1≤m≤i+j, thus recover former watermark information.

2. the digital watermarking embedding and the extracting method of a kind of spatial domain according to claim 1 and wavelet field two-set diversity, it is characterized in that: the watermark information sequence generates two groups of orthogonal information sequence set through quadrature Walsh sign indicating number spread spectrum, i.e. spatial information (si) sequence set and wavelet domain information sequence set.

3. the digital watermarking embedding and the extracting method of a kind of spatial domain according to claim 1 and wavelet field two-set diversity, it is characterized in that: according to the different requirements of resist geometric attacks and conventional attack, can adaptive adjustment spatial information (si) sequence set and the spreading factor of wavelet domain information sequence set.

4. the digital watermarking embedding and the extracting method of a kind of spatial domain according to claim 1 and wavelet field two-set diversity is characterized in that: during embedding, embed in the spatial domain earlier, carry out wavelet field then and embed, during extraction, extract in wavelet field earlier, extract in the spatial domain.

5. the digital watermarking embedding and the extracting method of a kind of spatial domain according to claim 1 and wavelet field two-set diversity is characterized in that: spatial information (si) sequence that extracts and wavelet domain information sequence are carried out the equal gain combining processing, recover former watermark information.

Images