CN1514409A

CN1514409A - Wavelet Domain Digital Watermarking Method Based on Image Target Area

Info

Publication number: CN1514409A
Application number: CNA031344372A
Authority: CN
Inventors: 郭宝龙; 郭磊
Original assignee: Xidian University
Current assignee: Xidian University
Priority date: 2003-07-28
Filing date: 2003-07-28
Publication date: 2004-07-21
Anticipated expiration: 2023-07-28
Also published as: CN1256705C

Abstract

The invention relates to a wavelet domain digital watermarking method based on image target area. This method uses image understanding technology to embed the watermark into the most visually important part of the image, and realizes the blind detection of the watermark through the interrelationship between the watermark and the features of the target area of the image. The key technology is to decompose the original image by wavelet to determine the embedding domain of the watermark. When embedding the watermark, perform BCH error correction coding on the watermark, select high-frequency coefficients in the embedding domain to quantize it and embed the watermark, and perform wavelet inverse transform on the wavelet coefficients embedded in the watermark to obtain the image embedded with the watermark; when extracting the watermark, use BCH The code is decoded to obtain the extracted watermark, and the extracted watermark and the original watermark are correlated according to the normalized cross-correlation function to obtain the judgment result of the watermark. It has the advantages of good blind detection performance of watermark, high subjective image quality, and strong ability to resist clipping attacks, etc., and can be used for multimedia information security protection in the network.

Description

Wavelet field digital watermark method based on image target area

Technical field:

The present invention relates to technical field of image processing, a kind of specifically wavelet field digital watermark method based on image target area can be used for the multi-media information security protection in the network.

Background technology:

Along with network technology and rapid development of multimedia,, but also provide convenience simultaneously for the digital publishing rights abuse for the transmission and the utilization of numerical information brought very big facility.At present arbitrarily copy on the Internet behavior of various images constituted serious infringement to works human rights reproduction right, cause a big obstacle of the information industry development.Someone's method of having proposed a kind of digital watermarking addresses this problem for this reason.So-called digital watermark technology is meant and embeds and be present in invisible identification code or mark in the data.Existing water mark method roughly is divided into two classes: a class is the spatial domain method, as Raymond B.Wolfgang, Christine I.Podilchukand Edward J.Delp.Perceptual watermarks for digital images and video.Proceedings of theIEEE, Special issue on Identification and Protection of MultimediaInformation, 87 (7): 1108-1126, July, 1999. documents are described; Another kind of is the transform domain method, as Liehua Xie, Gonzalo R.Arce.A class of authentication digitalwatermarks for secure multimedia communication[J] .IEEE Trans onImage Processing, 2001,10 (11): 1754-1764.) document is described.The spatial domain method lacks robustness, and the transform domain method is then utilized the visually-perceptible model easily.Spread spectrum watermark method (the I.J.Cox that people such as the graduate I.J.Cox of NEC propose, J.Kilian, T.Leighton, andT.Shamoon.Secure spread spectrum watermarking for multimedia[J] .IEEE Trans on Image Processing, 1997,6 (12): 1673-1687.), be digital watermarking field classic methods the most, comprising watermark embedding and two kinds of processes of watermark detection, as Fig. 6, shown in Figure 7.

Watermark embed process shown in Figure 6 is:

(1) be that seed produces pseudo-random sequence with the key, this sequence has Gauss N (0,1) and distributes;

(2) image is the discrete surplus conversion DCT that revolves;

(3) with the discrete surplus conversion DCT coefficient that revolves of 1000 maximums of pseudorandom gaussian sequence modulation (stack) this image except that direct current (DC) component, embedding formula is:

y _f＝x _i(1+α _lw _l)

Y wherein _lDCT coefficient after the representative modulation, x _lRepresent original DCT coefficient, w _lRepresent the watermark deal, α _lBe modulation factor;

(4) the discrete surplus conversion DCT coefficient that revolves to the image after the modulation carries out inverse discrete cosine transform IDCT, obtains containing watermarking images.

Watermark detection process shown in Figure 7 is:

(1) watermarking images that contains of original image and possibility distortion is done the discrete surplus conversion dct transform that revolves respectively, obtain the discrete surplus conversion DCT coefficient x that revolves of image respectively _iAnd the discrete surplus conversion DCT coefficient that revolves of the watermarking images of possibility distortion

(2) the anti-process by embedding formula can get:

{\tilde{w}}_{i} = \frac{{\tilde{y}}_{i} - x_{i}}{α_{i} . x_{i}}

(3) when the detection of watermark, ask original watermark

W = (w_{1}, w_{k}, . . ., w_{N_{w}})

With the watermark of extracting

\tilde{W} = ({\tilde{w}}_{1}, {\tilde{w}}_{k}, . . ., {\tilde{w}}_{N_{w}})

Between similarity, with following similarity criterion:

sim (W, \tilde{W}) = \frac{W . \tilde{W}}{\sqrt{\tilde{W} . \tilde{W}}}

If the response of detecting device greater than some threshold values, is taken as 6 by calculating in the Cox algorithm, then think to detect watermark.

Though this spread spectrum water mark method has proposed to strengthen the cardinal principle of watermark robustness and anti-attack method, promptly watermark signal should embed human eye vision part and parcel in the raw data, and has certain security, but still has the following disadvantages:

(1) this method is regarded image as whole scene, does not take into full account most important sensation target zone in the image, has significant limitation;

(2) employing of this method is on the low frequency component with the important component visual of watermark embedded images, not quite identical with people's visual effect, most important parts should be the outstanding target area of image on the image vision, and this target area can not fully be reflected the conversion DCT low frequency deal from discrete surplus the revolving of image;

(3) this method requires accurate localization is carried out in the position of watermark when watermark extracting, be the synchronous of watermark information, need utilize original image, if original image can not obtain, then the view data in the watermark detection process presents noise signal with respect to watermark information, be that view data itself can produce very big interference to watermark detection, cause the extraction difficulty of watermark;

(4) watermark that embeds of this method is distributed in the whole spatial domain of image, make to the robustness of shearing attack very a little less than, as shown in Figure 8, the left side figure among Fig. 8 be an original image, right side figure is a watermarking images.

The invention technology contents:

The objective of the invention is to overcome the deficiency of above-mentioned prior art, propose a kind of wavelet field digital watermark method, to realize copyright protection the Internet images based on image target area.

The present invention is when design technology project, take into full account the time to the image embed watermark on the Internet, the target area vision is very important, and the background area smoothly is difficult for embed watermark, and the entire image embed watermark easily causes all multifactor of shearing background area havoc watermark.Its technical thought is to adopt the image understanding technology, the most important part of vision that watermark is embedded into image reality, to improve the effect that watermark can be resisted common attack, adopting target area feature with watermark and image to be closely connected simultaneously is in the same place, and in the process of embed watermark, make the interior watermark deal that only embeds seldom of subrange of image, with the blind Detecting of realization watermark and the subjective quality of assurance image.Realize that the object of the invention key problem in technology is that original image is carried out wavelet decomposition; Determine the embedding territory of watermark; When watermark embeds, earlier the BCH Error Correction of Coding is carried out in watermark, in embedding the territory, select high frequency coefficient that it is quantized the back embed watermark.The concrete steps that its watermark embeds, detects are as follows:

1. watermark embed step:

(1) to image block, and the embedding territory of definite watermark, be about to original image and carry out wavelet decomposition earlier, make an image in the spatial domain corresponding with three blocks of images in the high-frequency sub-band of image wavelet territory, characteristic according to the image high frequency coefficient, again the variance of the wavelet coefficient of each image block is asked average, in the wavelet field of entire image, this average is carried out cluster, and obtaining sensation target zone and background through last handling process, the set of the high frequency wavelet coefficient that this sensation target zone is corresponding is the embedding territory of watermark;

(2) produce the binary watermarking random series with secret key, carry out Error Correction of Coding with BCH code;

(3) in embedding territory Ω, adopt the quantification modulation to embed a watermark corresponding to each spatial domain image block;

(4) in each image block that embeds territory Ω correspondence, embed all watermarks, the image wavelet coefficient after obtaining embedding successively;

(5) wavelet coefficient to embed watermark carries out inverse wavelet transform, obtains the image of embed watermark;

2. the watermark extracting step is:

(1) carries out piecemeal to containing watermarking images, determine the embedding territory Ω of watermark, promptly will contain watermarking images earlier and carry out wavelet decomposition, characteristic according to the image high frequency coefficient, again it is carried out cluster, and obtaining sensation target zone and background through last handling process, the set of the high frequency wavelet coefficient that this sensation target zone is corresponding is the embedding territory of watermark;

(2) use the secret key identical to produce watermark random sequence with the watermark embedding;

(3) embedding corresponding to containing watermark of each piece extraction in the watermarking images spatial domain;

(4) in each image block that embeds territory Ω correspondence, extract all watermarks successively, and decode the watermark that obtains extracting with BCH code;

(5) watermark and the original watermark that extracts carried out coherent detection by Normalized Cross Correlation Function NC, obtain the court verdict of watermark, promptly

NC = \frac{Σ_{i = 1}^{N_{w}} w (i) \hat{w} (i)}{\sqrt{Σ_{i = 1}^{N_{w}} w^{2} (i)} \sqrt{Σ_{i = 1}^{N_{w}} {\hat{w}}^{2} (i)}}

In the formula: w (i) represents original watermark

The watermark that expression is extracted.

The present invention really is embedded into the most important target area of image vision to watermark, thereby can resists shearing attack effectively owing to adopt the image understanding technology; Owing to adopting the embedding of watermark and the target area feature of image are combined closely simultaneously, so can realize the blind Detecting (promptly not needing original image) of watermark; In addition owing to adopted wavelet transform DWT, therefore can with Joint Photographic Experts Group JPEG2000 compatibility.

Diagram shows that watermark is embedded in the important goal zone of image among the present invention, and watermark energy is very little in subrange, and the vision distortion that causes is also little.

Description of drawings:

Fig. 1 is telescopiny figure of the present invention

Fig. 2 is testing process figure of the present invention

Fig. 3 is the watermark embedding territory figure that obtains of the present invention.1 is in the original image spatial domain one among the figure; 2,3,4 be respectively original image, in the 1 pairing image wavelet territory high-frequency sub-band of original image spatial domain three through after the one-level wavelet decomposition; Variance to image wavelet high frequency coefficient piece is carried out cluster, obtains 5,6,7, and they comprise a plurality of small echo high frequency coefficient pieces, constitutes the embedding territory of watermark.

Fig. 4 is the telescopiny figure of a watermark of the present invention

Fig. 5 is the embedding instance graph of watermark of the present invention

Fig. 6 is prior art watermark embed process figure

Fig. 7 is the prior art watermark detection process

Fig. 8 is the prior art simulation result

Fig. 9 is the present invention and prior art effect comparison diagram

Embodiment:

With reference to Fig. 1, Fig. 3, Fig. 4, watermark embed step of the present invention is:

The first step is obtained the embedding territory of watermark

Original image by piecemeal shown in Figure 3, through the one-level wavelet transformation, is made in the spatial domain one 1 in corresponding wavelet field medium-high frequency subband corresponding three 2,3,4.The wavelet coefficient variance of the respective image piece among subband HL, LH, the

HH

2,3,4 is asked average, in the wavelet field of entire image, this average is carried out cluster, the pairing small echo high frequency coefficient of the set that the cluster centre that obtains is bigger piece is the embedding territory of watermark

If it is Ω that watermark embeds domain representation, total num piece in the spatial domain of Ω correspondence, wherein one is expressed as U _{K, block}, U _{K, block}Corresponding three 2,3,4 in Ω among subband HL, LH, the HH are respectively V _{K, block_HL}, V _{K, block_LH}, V _{K, block_HH}

Second step, use secret key K to produce watermark frequency expansion sequence W, and carry out Bose-Chaudhuri-Hocquenghem Code,

Watermark length is N _w=num, value is-1 and+1, average is 0.Be W=(w ₁, w _k..., w _Num), w _k∈ 1 ,+1};

The 3rd step embedded among the Ω of territory in watermark, corresponding to watermark of each piece embedding of target area in the image;

Consider image block U _{K, block}, coefficient block corresponding in Ω is respectively V _{K, block_HL}, V _{K, block_LH}, V _{K, block_HH}, then the watermark bit that will embed is w _kConcrete grammar is as follows:

1) at V _{K, block_HL}, V _{K, block_LH}And V _{K, block_HH}In find the wavelet coefficient of numerical value maximum, be designated as V _{K, block_HL, x}, and the coefficient that is located at relevant position in LH and the HH subband is designated as V respectively _{K, block_LH, x}, V _{K, block_HH, x}

2) to 1) in three coefficients arrange from small to large;

V _{k，block_HH，x}≤V _{k，block_LH，x}≤V _{k，block_HL，x}

3) to middle coefficient V _{K, block_LH, x}Quantize with embed watermark according to Fig. 4.Earlier V _{K, block_HH, x}And V _{K, block_HL, x}Between distance by the step delta five equilibrium:

Δ=(V _{K, block_HL, x}-V _{K, block_HH, x})/(2Q-1) wherein, Q is a quantization parameter, the size given by the user,

If w _k=1, V _{K, block_LH, x}Be quantified as from the represented numerical value of the nearest solid line of itself; If w _k=-1, then V _{K, block_LH, x}Be quantified as from the represented numerical value of the nearest dotted line of itself.

The 4th step embedded all watermarks successively in all images piece of watermark embedding territory Ω correspondence,

Image wavelet coefficient after obtaining embedding;

The 5th step, the wavelet coefficient behind the embed watermark is carried out secondary discrete wavelet inverse transformation, obtain the image of embed watermark.

With reference to Fig. 2, watermark detection of the present invention does not need original image, but need use secret key K and the given quantization parameter Q of user, and concrete steps are as follows:

The first step obtains Ω (because this important images feature of target area has unchangeability before and after watermarking images is attacked, therefore can with identical method) to containing watermarking images identical method when embedding;

In second step, use secret key K to produce original watermark sequence W;

In the 3rd step, in embedding territory Ω, each

piece

2,3,4 extracts a watermark successively in the watermarking images spatial domain corresponding to containing, and considers that wherein one is expressed as

Then in Ω, correspondingly get 2,3,4 three coefficient block, be expressed as respectively Watermark bit w then _kLeaching process as follows:

1) exists With

In find the wavelet coefficient of numerical value maximum, be designated as Corresponding coefficient is designated as respectively in LH and HH subband

2) to 1) in three coefficients arrange from small to large;

3) to middle coefficient Value judge, earlier right

Between distance quantize with Q, obtain Δ, then

If the r value is odd number, then w _k=1, the r value is even number, then w _k=-1;

In the 4th step, in embedding territory Ω, extract watermark in all images piece of correspondence, and carry out BCH decoding, the watermark that obtains extracting

In the 5th step, the watermark of original watermark W and extraction is adopted in the watermark judgement

Between Normalized Cross Correlation Function NC carry out coherent detection, obtain the court verdict of watermark, promptly

NC = \frac{Σ_{i = 1}^{N_{w}} w (i) \hat{w} (i)}{\sqrt{Σ_{i = 1}^{N_{w}} w^{2} (i)} \sqrt{Σ_{i = 1}^{N_{w}} {\hat{w}}^{2} (i)}}

With reference to Fig. 5, a is the embedding original image of watermark, and b is the embedding territory of acquisition watermark, and is partitioned into its pairing image space zone, and c is the image behind the embed watermark, and d is the error image that contains watermarking images and original image.

With reference to Fig. 9, (a) be the image of embed watermark of the present invention and watermark distribution at image space; Figure (b) be the image of spread-spectrum watermark algorithm embed watermark of classics and the watermark distribution at image space.Can see that watermark is embedded in the important goal zone of image among the present invention, and within the scope of part, watermark energy is very little, thereby the vision distortion that watermark causes is also less.

Claims

1. A wavelet domain digital watermarking method based on an image target area, including two processes of watermark embedding and watermark detection, wherein:

(1) The watermark embedding steps are:

1) Divide the original image into blocks, and determine the embedding domain (Ω) of the watermark, that is, the original image is decomposed by wavelet first, so that a piece of image (1) in the air domain and the high frequency subband (HL) and (LH) of the image wavelet domain , (HH) corresponding to the three images (2), (3), (4), according to the characteristics of the high-frequency coefficients of the image, the wavelet coefficients of (2), (3), (4) image blocks The variance is averaged, and the average number is clustered in the wavelet domain of the entire image, and the visual target area and background are obtained through the post-processing process. The set of high-frequency wavelet coefficients corresponding to the visual target area is the embedded domain of the watermark;

2) Use the secret key to generate a binary watermark random sequence, and use the BCH code for error correction coding;

3) In the embedding domain (Ω), use quantized modulation to embed a one-bit watermark for each spatial domain image block (2), (3), (4);

4) sequentially embed all watermarks in each image block (2), (3), (4) corresponding to the embedding domain (Ω), and obtain the embedded image wavelet coefficients;

5) Perform wavelet inverse transform on the wavelet coefficients embedded in the watermark to obtain the image embedded in the watermark;

(2) The watermark extraction steps are:

1) Divide the watermarked image into blocks and determine the embedding domain (Ω) of the watermark, that is, decompose the watermarked image by wavelet first, then cluster it according to the characteristics of the high-frequency coefficients of the image, and obtain Visual target area and background, the set of high-frequency wavelet coefficients corresponding to the visual target area is the embedding domain of the watermark;

2) Use the same key as the watermark embedding to generate a watermark random sequence;

3) In the embedding domain (Ω), extract one-bit watermark corresponding to each block (2), (3), (4) in the spatial domain of the watermarked image;

4) Extract all watermarks in the image blocks (2), (3) and (4) corresponding to the embedded domain in turn, and decode them with BCH codes to obtain the extracted watermarks;

5) Correlation detection is performed on the extracted watermark and the original watermark according to the normalized cross-correlation function NC, and the judgment result of the watermark is obtained, namely

NC NC = = \frac{{Σ Σ}_{i i = = 11}^{{N N}_{w w}} w w ((i i)) \overset{^^}{w w} ((i i))}{\sqrt{{Σ Σ}_{i i = = 11}^{{N N}_{w w}} {w w}^{22} ((i i))} \sqrt{{Σ Σ}_{i i = = 11}^{{N N}_{w w}} {\overset{^^}{w w}}^{22} ((i i))}}

In the formula: w(i) represents the original watermark

Represents the extracted watermark.

2. the wavelet domain digital watermarking method of image target area according to claim 1 is characterized in that in embedding domain (Ω), adopts quantized modulation to embed one-bit watermark corresponding to each air domain image block, is to embedding domain ( Ω) corresponding to a piece of image in the spatial domain is expressed as U _{k, block} , U _k, the image blocks corresponding to the subbands (HL), (LH), (HH) of the block in the embedding domain (Ω) (2), The coefficients of (3) and (4) are respectively expressed as V _{k, block_HL} , V _{k, block_LH} , V _{k, block_HH} , and the watermark bit to be embedded is w _k , the specific method is as follows:

1) Find the wavelet coefficient with the largest value in V _{k, block_HL} , V _{k, block_LH} and V _{k, block_HH} , record it as V _{k, block_HL, x} , and set it at the corresponding position in the subband (LH) and (HH) The coefficients are recorded as V _{k, block_LH, x} , V _{k, block_HH, x} respectively;

2) Arrange the three coefficients in 1) from small to large, namely

V _{k, block_HH, x} ≤ V _{k, block_LH, x} ≤ V _{k, block_HL, x}

3) Quantize the intermediate coefficients V _{k, block_LH, x} to embed the watermark, that is, first put V _{k, block_HH, x}

and the distance between Vk _{, block_HL, x} is equally divided by the step size Δ:

Δ=(V _{k, block_HL, x} −V _{k, block_HH, x} )/(2Q-1)

Q: It is a quantization parameter, the size is given by the user

3. the wavelet domain digital watermarking method of the image target area according to claim 1 is characterized in that in the embedding domain (Ω), corresponding to the method of extracting one-bit watermark in the watermarked image space domain, set ( Ω) corresponding to an image in the spatial domain is expressed as

The three coefficient blocks (2), (3), and (4) corresponding to Ω are expressed as

Then the extraction process of the watermark bit w _k is as follows:

1) at

and Find the wavelet coefficient with the largest value in , denoted as The corresponding coefficients in the LH and HH subbands are denoted as

2 pairs

and The three coefficients are arranged from small to large, namely

4) For intermediate coefficients value to be judged, that is, first to

The distance between is quantified by Q to get Δ, then If the value of r is an odd number, then w _k =1, and if the value of r is an even number, then w _k =-1.