CN103617591A

CN103617591A - Dual-color image blind watermarking method based on QR decomposition and compensation

Info

Publication number: CN103617591A
Application number: CN201310682411.9A
Authority: CN
Inventors: 苏庆堂; 赵云峰; 贾韶丽
Original assignee: Ludong University
Current assignee: Ludong University
Priority date: 2013-12-16
Filing date: 2013-12-16
Publication date: 2014-03-05
Anticipated expiration: 2033-12-16
Also published as: CN103617591B

Abstract

The invention discloses a dual-color image blind watermarking method based on QR decomposition and compensation, which aims to change the current situation that the current digital watermarking is mostly a binary or grayscale image and meet the requirement that a color digital image serves as digital watermarking. According to the technical key point, color image watermarking is embedded into an orthogonal matrix Q obtained by QR decomposition; visual distortion caused by watermarking embedding is compensated by utilizing a triangular matrix R; and assistance of the original watermarking or the original host image is not needed during watermarking extraction, so that blind detection is realized. The method is simple and quick, has the advantages of high watermarking invisibility, strong robustness and high capacity, and is applicable to copyright protection of the color digital image serving as the digital watermarking.

Description

A kind of two coloured image blind watermark methods that decompose and compensate based on QR

Technical field

The invention belongs to field of information security technology, relate to large capacity color digital image as the copyright protection of digital watermarking.

Background technology

Most image digital watermark method of today is using two-value or gray level image as digital watermarking, and fewer using color digital image as digital watermarking, its main cause is during using color digital image as watermark, its quantity of information containing is 3 times of same size gray level image, be 24 times of bianry image, so increased the difficulty that watermark embeds; Meanwhile, the existing digital watermark method that bianry image is embedded to gray level image can not meet that to take color digital image be main copyright protection well, therefore how color digital image is become to one of problem demanding prompt solution as digital watermarking.

At present, what in digital watermarking field, most of researcher paid close attention to is blind watermark method rather than non-blind watermark method, this is because non-blind watermark need to just can be carried out detection or the extraction of watermark by means of the help of original host image or original watermark image, and the method in routine testing, some is loaded down with trivial details, therefore, in digital watermark technology research in recent years, blind Detecting digital watermark method becomes the main flow of digital watermark technology development gradually.Regrettably; the copyright protection information comprising due to color digital image is very large; with non-blind watermark method, can embed more easily or extract watermark; the blind watermark of right and wrong of existing a lot of Watermarking for Color Images method research; therefore, the blind extraction that realizes color watermark image that how to change the status quo is one of difficult point of present image digital watermark technology.

Summary of the invention

The object of this invention is to provide a kind of two coloured image blind watermark methods that decompose and compensate based on QR, comprise concrete watermark embed process and leaching process, the detailed process that its watermark embeds is described below:

The first step: the pre-service of color watermark image: by dimension-reduction treatment, one width is big or small for 24 original color watermarking images W of N * N are divided into R, G, tri-watermark components of B, and each watermark component is carried out to the Arnold scramble conversion based on key K a; Then, each pixel transitions is become to two value informations of 8 bits, being finally combined to form length is the watermark sequence W of LW;

Second step: the piece of colored host image is processed: at rgb color space, host image H is divided into R, G, B tri-tomographic images, and each tomographic image is divided into non-overlapped 4 * 4 block of pixels;

The 3rd step: utilize the MD5 function pseudorandom based on key K h to select LW block of pixels as the embedded location of watermark, LW represents the total length of watermark herein;

The 4th step: choose a block of pixels

according to formula (1), carry out QR and decompose acquisition orthogonal matrix

and upper triangular matrix

, i herein, j represents respectively line number and the row number at this block of pixels place;

(1)

The 5th step: according to formula (2), (3), by matrix

second element of first row

with the 3rd of first row

element is revised as respectively

,

carry out embed watermark

, obtain the orthogonal matrix after embed watermark ;

(2)

(3)

Herein,

;

The 6th step: utilize formula (4) to obtain the matrix after embed watermark

;

(4)

The 7th step: utilize rmatrix compensate because of qthe visual distortion that causes of modification;

Order

,

, according to formula (5), (6), (7), (8), obtain respectively four elements of the first row in matrix R

,

,

,

index word be respectively

;

(5)

(6)

(7)

(8)

Utilize formula (9), (10), (11), (12) to obtain the rear matrix of compensation four elements of the first row

;

(9)

(10)

(11)

(12)

The 8th step: utilize formula (13) to carry out contrary QR conversion, the image block containing watermark being compensated

;

(13)

The 9th step: choose the final image block containing watermark according to formula (14) ;

(14)

The tenth step: repeat the 4th step to the nine steps, until all watermark informations are embedded in host image;

The 11 step: reconfigure R, G, B tri-tomographic images containing watermark, obtain the final image H ' containing watermark.

The detailed process of watermark extracting of the present invention is as follows:

The first step: will be divided into R, G, tri-layered images of B containing the image H ' of watermark, and each tomographic image is divided into non-overlapped 4 * 4 block of pixels;

Second step: utilize the MD5 function pseudorandom based on key K h to select LW block of pixels that contains watermark;

The 3rd step: will contain the block of pixels of watermark

carry out QR decomposition, obtain its orthogonal matrix, i herein, j represents respectively line number and the row number at this block of pixels place;

The 4th step: according to formula (15), utilize

second element of middle first row

with the 3rd element

extract binary watermarking information

;

(15)

The 5th step: repeat the 3rd step, the 4th step, until extract all watermark informations, and convert every eight one group of extracted binary watermarking information to metric pixel value;

The 6th step: the contrary Arnold carrying out based on key K a converts and obtain layering watermark;

The 7th step: combination layering watermark forms final extraction watermark W '.

The method simple and fast; there is higher watermark invisibility, strong robustness and large capacity; the present invention reaches following effect: using color digital image as digital watermarking, also " have no to discover " and be embedded into host image; while extracting watermark, do not need the help of original host image or original watermark image; can the embedded watermark of rapid extraction from various under fire images, this invention is applicable to color digital image as the copyright protection of digital watermarking.

Accompanying drawing explanation

Fig. 1 (a), Fig. 1 (b), Fig. 1 (c), Fig. 1 (d) are four width original color host images.

Fig. 2 (a), Fig. 2 (b) are two width color watermark images.

Fig. 3 (a), Fig. 3 (b) are embedded into the rear resulting watermarking images that contains of host image Fig. 1 (a), Fig. 1 (b) successively by the watermark shown in Fig. 2 (a), and its PSNR value is 37.8827dB, 35.0463dB successively; Fig. 3 (c), Fig. 3 (d) are embedded into the rear resulting watermarking images that contains of host image Fig. 1 (c), Fig. 1 (d) successively by the watermark shown in Fig. 2 (b), and its PSNR value is 38.4577dB, 39.1066dB successively.

Fig. 4 (a), Fig. 4 (b) are the watermarks of extracting from Fig. 3 (a), Fig. 3 (b) successively, its NC value is respectively 1.0000,1.0000, Fig. 4 (c), Fig. 4 (d) are the watermarks of extracting from Fig. 3 (c), Fig. 3 (d) successively, and its NC value is respectively 0.9979,1.0000.

Fig. 5 (a), Fig. 5 (b), Fig. 5 (c), Fig. 5 (d), Fig. 5 (e), Fig. 5 (f), Fig. 5 (g) be by Fig. 3 (a) carry out successively JPEG, add make an uproar, watermarking images that filtering, sharpening, fuzzy, shearing, convergent-divergent extract after attacking, its NC value is respectively 0.9406,0.9689,0.9517,0.9499,0.9966,0.9433,0.9766.

Fig. 6 (a), Fig. 6 (b), Fig. 6 (c), Fig. 6 (d), Fig. 6 (e), Fig. 6 (f), Fig. 6 (g) be by Fig. 3 (c) carry out successively JPEG, add make an uproar, watermarking images that filtering, sharpening, fuzzy, shearing, convergent-divergent extract after attacking, its NC value is respectively 0.9916,0.9078,0.9608,0.9980,0.8994,0.8752,0.9389.

embodiment

The first step: the pre-service of color watermark image: 24 original color watermarking images W that are 32 * 32 by a width size by dimension-reduction treatment are divided into R, G, tri-watermark components of B, and each watermark component is carried out to the Arnold scramble conversion based on key K a; Then, each pixel transitions is become to two value informations of 8 bits, be finally combined to form length and be 24576 watermark sequence w;

Second step: the piece of colored host image is processed: at rgb color space, 512 * 512 host image H are divided into R, G, B tri-tomographic images, and each tomographic image is divided into non-overlapped 4 * 4 block of pixels;

The 3rd step: utilize the MD5 function pseudorandom based on key K h to select 24576 block of pixels as the embedded location of watermark;

The 4th step: the block of pixels of choosing is carried out to QR and decompose acquisition its orthogonal matrix Q and upper triangular matrix R;

If an original pixels piece is, QR after decomposing its orthogonal matrix Q and upper triangular matrix R are respectively

,

;

The 5th step: utilize threshold value T=0.04 to revise second element of Q matrix first row

with the 3rd

unit is embed watermark usually (establishing this watermark information is herein 0), the orthogonal matrix of above-mentioned the 4th step gained is revised as:

;

The 6th step: carry out contrary QR conversion, obtain containing the image block of watermark be:

；

The 7th step: to R matrix compensation because of qthe visual distortion that causes of modification, the upper triangular matrix R compensation of above-mentioned the 4th step gained is revised as:

;

The 8th step: carry out contrary QR conversion, the image block containing watermark being compensated:

；

The 9th step: choose the final image block containing watermark :

；

The concrete steps of watermark extracting of the present invention are described below:

Second step: utilize the MD5 function pseudorandom based on key K h to select 24576 block of pixels that contain watermark;

The 3rd step: by a block of pixels that contains watermark

carry out QR decomposition, obtain its orthogonal matrix

;

The 4th step: utilize

second element of middle first row with the 3rd element

₁extract contained binary watermarking information

=0;

The method simple and fast, has higher watermark invisibility, strong robustness and high power capacity, is highly suitable for coloured image as the copyright protection of digital watermarking.

validation verification of the present invention

In order to prove validity of the present invention, 24 standard pictures that the four width sizes of selection as shown in Fig. 1 (a), Fig. 1 (b), Fig. 1 (c), Fig. 1 (d) are 512 * 512 are as host image, and 24 coloured images that are 32 * 32 by two width sizes as shown in Fig. 2 (a), Fig. 2 (b) are as digital watermarking, setting threshold T=0.04 verifies simultaneously.Shown in Fig. 3 (a), Fig. 3 (b) is watermark Fig. 2 (a) to be embedded into the watermarking images that contains of Fig. 1 (a), Fig. 1 (b) gained, its PSNR is respectively 37.8827dB, 35.0463dB, shown in Fig. 3 (c), Fig. 3 (d) is watermark Fig. 2 (b) to be embedded into the watermarking images that contains of Fig. 1 (c), Fig. 1 (d) gained, its PSNR is respectively 38.4577dB, 39.1066dB, can find out that embedded watermark has good invisibility;

Meanwhile, Fig. 4 (a), Fig. 4 (b) is from Fig. 3 (a), the watermark of extracting in Fig. 3 (b), its NC value is all 1.0000, Fig. 4 (c), Fig. 4 (d) is from Fig. 3 (c), the watermark of extracting in Fig. 3 (d), its NC value is respectively 0.9979,1.0000, Fig. 5 (a), Fig. 5 (b), Fig. 5 (c), Fig. 5 (d), Fig. 5 (e), Fig. 5 (f), Fig. 5 (g) carries out JPEG successively by Fig. 3 (a), add and make an uproar, filtering, sharpening, fuzzy, shear, the watermarking images that convergent-divergent extracts after attacking, its NC value is respectively 0.9406, 0.9689, 0.9517, 0.9499, 0.9966, 0.9433, 0.9766, Fig. 6 (a), Fig. 6 (b), Fig. 6 (c), Fig. 6 (d), Fig. 6 (e), Fig. 6 (f), Fig. 6 (g) carries out JPEG successively by Fig. 3 (c), add and make an uproar, filtering, sharpening, fuzzy, shear, the watermarking images that convergent-divergent extracts after attacking, its NC value is respectively 0.9916, 0.9078, 0.9608, 0.9980, 0.8994, 0.8752, 0.9389, as can be seen here, the digital watermarking image extracting has good identifiability, illustrates that the method has stronger robustness, can extract well embedded color watermark and carry out copyright protection.

Claims

1. based on QR, decompose and two coloured image blind watermark methods that compensate, it is characterized in that: by specific watermark embed process and leaching process, realize, the detailed process that its watermark embeds is described below:

The 4th step: choose a block of pixels