CN102314669A - DCT (discrete cosine transform)-based anti-geometric-attack zero-digital-watermarking method for medical image - Google Patents

Abstract

The invention relates to a DCT (discrete cosine transform)-based anti-geometric-attack zero-digital-watermarking method for a medical image, wherein the embedding of watermarks is firstly carried out, and the extracting of the watermarks is then carried out. The method comprises the following steps: (1) carrying out full-image DCT on an original medical image, and extracting a vector which can represent the important visual features of the original image from transformed coefficients; (2) obtaining a binary logic sequence through the Hash function by using the eigenvector and the watermarks to be embedded; (3) carrying out the full-image DCT on a medical image to be tested, and finding out a visual eigenvector of the image to be tested; and (4) extracting the watermarks by using the characteristics of the Hash function and the binary logic sequence which is obtained when the watermarks are embedded. Based on the DCT digital-watermarking technology, the method provided by the invention is fast for the embedding of the watermarks under the condition of not influencing on the quality of original medical images, has the characteristics of rapid calculating speed, high accuracy, good compatibility, strong attack resistance and the like, and has a high practical value in aspects, such as protection of patient privacy and the like.

Description

A kind of medical image remainder word water mark method based on the DCT resist geometric attacks

Technical field

The invention belongs to field of multimedia signal processing, relate to a kind of medical image digital watermark technology based on dct transform and Image Visual Feature, specifically is a kind of medical image remainder word water mark method based on the DCT resist geometric attacks.

Background technology

At present, medical image accounts for 70%～80% of whole hospital medical information, and the digital content management system has brought into play more and more important effect in the medical system in modern times, but along with the applying of network, its information security issue comes out gradually.

When medical image carries out remote transmission on network, be recorded in the patient's on the medical picture personal information, revealed easily.If be embedded in personal information in the medical picture as digital watermarking, just can solve this difficult problem preferably, this watermark is called the medical image digital watermarking.

The research to medical image digital watermarking field at present mainly concentrates on spatial domain and two aspects of transform domain (DCT, DFT and DWT), they respectively the value of some coefficients of gray scale or the transform domain of some pixel through the change spatial domain come embed watermark.Wherein cosine transform (Discrete Cosine Transform, DCT) territory water mark method is because its calculated amount is less; And with ID compression standard (JPEG; MPEG) compatibility, that studies at present is many, is the focus of existing most frequency field Study of Watermarking.

In view of singularity requirement to medical image focal zone protection, in the general document normal select with watermark information be embedded into image non-area-of-interest (Region of Non-Interest, RONI).(Region of Interest ROI) refers to the focal zone that those comprise important pathological characters or diagnosis and treatment information to area-of-interest in the medical image, if in this zone embed watermark, the diagnosis that then might make the mistake.But often people will spend long time and energy when seeking ROI, and in case select wrongly, then might disturb doctor's diagnosis.

In medical image digital watermarking research field; Up to now geometric attack is still a more insoluble problem, as for being highly resistant to conventional attack and geometric attack simultaneously, the water mark method research of these two kinds of attack types; Do not appear in the newspapers as yet at present, still belong to blank.And in the practical application, the medical digital watermarking images usually receives this two kinds of attacks simultaneously.

Summary of the invention

The purpose of this invention is to provide a kind of medical image remainder word water mark method based on the DCT resist geometric attacks; Visual feature vector, encryption technology and third-party notion through with medical image combine; Need not carry out choosing of area-of-interest; Thereby solved the agility problem of watermark embedding, extraction, had very desirable robustness and invisibility, with the copyright of protection medical image and the crypticity of sufferer information; Solve the hiding property of patient information and the sensitive question of medical image effectively, solve the resistance geometric attack and resistance conventional attack problem that occur in the medical image applications simultaneously.

To achieve these goals; The present invention is performed such: based on the full figure dct transform, in dct transform coefficient, extract the medical image visual feature vector of a resist geometric attacks; And digital watermark and cryptography combined, realized the anti-geometry and the conventional attack of digital watermarking.The method that the present invention adopted comprises watermark embedding and watermark extracting two large divisions; First is a watermarking algorithm; Comprise: (1) is through carrying out the full figure dct transform; Obtain a visual feature vector V (j) of medical image, (2) generate a two-valued function sequence key (j) according to watermark W (j) and visual feature of image vector V (j).Second portion is the watermark extracting algorithm, comprising: the visual feature vector V ' that medical image to be measured is obtained in (3) (j), (4) utilize two-valued function sequence key (j) and testing image visual feature vector V ' (j), extract watermark W ' (j).

Method of the present invention is elaborated as follows at present:

At first with one group of binary pseudo-random W that can represent sufferer information, W={w (j) | w (j)=0,1; 1≤j≤L} is as digital watermarking, original image be designated as F={f (i, j) | f (i, j) ∈ R; 1≤i≤N1,1≤j≤N2) }, wherein, (i j) representes the grey scale pixel value of watermark sequence and primitive medicine image respectively, establishes N1=N2=N for w (j) and f.

First: watermarking algorithm

1), obtains the visual feature vector V (j) of medical image through carrying out the full figure dct transform.

(i j) carries out the full figure dct transform, obtains DCT matrix of coefficients FD (i to former figure F earlier; J), again to DCT matrix of coefficients FD (i, j); In the Low Medium Frequency coefficient, get preceding L coefficient, and obtain this visual feature of image vector V (j) through the computing of DCT coefficient symbols; Specific practice be when the DCT coefficient for we are with " 1 " expression just or zero the time, with " 0 " expression, program description is following when negative for coefficient:

FD(i，j)＝DCT2(F(i，j))

V(j)＝-Sign(FD(i，j))

2) generate a two-valued function sequence key (j) according to watermark W (j) and visual feature of image vector V (j).

$\mathrm{key}\left(j\right)=V\left(j\right)\⊕W\left(j\right)$

Key (j) is by visual feature of image vector V (j) and watermark W (j), and the Hash function commonly used through cryptography generates.Preserve key (i), need use when extracting watermark afterwards.Through key (j) is applied for to the third party as key,, thereby reach the purpose of protecting medical image with the entitlement and the right to use of acquisition medical image.

Second portion: watermark extracting algorithm

3) the visual feature vector V ' that obtains medical image to be measured (j).

If medical image to be measured be F ' (i, j), through obtain behind the full figure dct transform DCT matrix of coefficients be FD ' (i, j), by above-mentioned Step1 method, the visual feature vector V ' that tries to achieve testing image (j);

FD’(i，j)＝DCT2(F’(i，j))

V’(j)＝-Sign(FD’(i，j))

4) in testing image, extract watermark W ' (j).

According to the visual feature vector V ' of key (j) that generates when the embed watermark and testing image (j), utilize watermark W ' that Hash character can extract testing image (j)

$W^{,}\left(j\right)=\mathrm{key}\left(j\right)\⊕V^{,}\left(j\right)$

Differentiate the entitlement of testing image and the safety issue of hidden danger information according to W (j) and W ' degree of correlation (j) again.

The present invention and existing medical science digital watermark relatively have following advantage:

Because the present invention is based on the digital watermark technology of dct transform, it is fast to have computing velocity, and precision is high, and compatibility is preferably arranged, and stronger resist geometric attacks ability and anti-conventional attack ability are arranged; Do not need artificial the choosing of area-of-interest of carrying out, thereby solved the agility problem that watermark embeds; The watermark that embeds is a kind of zero watermark, does not influence primitive medicine picture quality, aspect medical, have very high practical value, and this algorithm is applicable to other field; Utilize third-party notion, adapted to the practicability and the standardization of the network promotion now; Below from the explanation of theoretical foundation and test figure:

1) discrete cosine transform

DCT is used for the standard that picture coding is present widely used JPEG compression and MPEG-1/2.DCT is the suboptimum orthogonal transformation that is only second to Karhunen-Loeve transformation that draws for a short time in the Minimum Mean Square Error condition, is a kind of harmless chief of a tribe's conversion.Its fast operation, precision is high, is celebrated with the ability of extracting characteristic component and the optimum balance between the arithmetic speed.

2-D discrete cosine direct transform (DCT) formula is following:

$F(u,v)=c\left(u\right)c\left(v\right)\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}f(x,y)\cos \frac{\mathrm{\π}(2x+1)u}{2M}\cos \frac{\mathrm{\π}(2y+1)v}{2N}$

u＝0，1，Λ，M-1；v＝0，1，Λ，N-1；

In the formula

$c\left(u\right)=\left\{\begin{array}{cc}\sqrt{1/M}& u=0\\ \sqrt{2/M}& u=\mathrm{1,2},\mathrm{\Λ},M-1\end{array}\right.$ $c\left(v\right)=\left\{\begin{array}{cc}\sqrt{1/N}& v=0\\ \sqrt{2/N}& v=\mathrm{1,2},\mathrm{\Λ},N-1\end{array}\right.$

2-D discrete cosine inverse transformation (IDCT) formula is following:

$f(x,y)=\underset{u=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{v=0}{\overset{N-1}{\mathrm{\Σ}}}c\left(u\right)c\left(v\right)F(u,v)\cos \frac{\mathrm{\π}(2x+1)u}{2M}\cos \frac{\mathrm{\π}(2y+1)v}{2N}$

x＝0，1，Λ，M-1；y＝0，1，Λ，N-1

X wherein, y is the spatial domain sampled value; U, v are the frequency field sampled value, and digital picture is represented with the pixel square formation usually, i.e. M=N

Can know that from top formula the coefficient symbols of DCT is relevant with the phase place of component.

2) choosing method of medical image vision principal character vector

The main cause of present most of medical image watermarking algorithm resist geometric attacks ability is: people are embedded in digital watermarking in pixel or the conversion coefficient, and the slight geometric transformation of medical image usually causes the bigger variation of having of pixel value or transform coefficient values.So just, can make the watermark of embedding just under attack very easily.If can find the visual feature vector of reflection image geometry characteristics, when little geometric transformation took place image, tangible sudden change can not take place in this visual feature of image value so.Hayes research shows that as far as characteristics of image, phase place is more important than amplitude.We find through observing a large amount of full figure DCT data (Low Medium Frequency); When a medical image is carried out common geometric transformation; Some variations possibly take place in the Low Medium Frequency coefficient magnitude, but its coefficient symbols remains unchanged basically, and we choose some experimental datas and see shown in the table 1.The primitive medicine image that is used as test in the table 1 is Fig. 1 (a), is piece image medical image (128x128).What the 1st row showed in the table is medical image type under attack, and the medical image that receives behind the conventional attack is seen Fig. 1 (b)-(d), and the medical image that receives behind the geometric attack is seen Fig. 2 (a)-(d).The 3rd is listed as the 11st row, and this is FD (1,1)-nine Low Medium Frequency coefficients of FD (3,3) of in the DCT matrix of coefficients, getting, wherein the DC component value of coefficient F (1,1) expression medical image.For conventional attack, these Low Medium Frequency coefficient values FD (1,1)-FD (3,3) remains unchanged and primitive medicine image value approximately equal basically; For geometric attack, the part coefficient has bigger variation, but we can find that medical image is when receiving geometric attack, and the size of part DCT Low Medium Frequency coefficient has taken place to change but its symbol does not change basically.With " 1 " expression (containing value is zero coefficient), negative coefficient is with " 0 " expression, so for the primitive medicine image with positive DCT coefficient for we; FD in the DCT matrix of coefficients (1,1)-FD (3,3) coefficient; Corresponding coefficient symbols sequence is: " 1,100 01001 "; See the 12nd row of table 1, we observe these row and can find, no matter conventional attack still is this symbol sebolic addressing of geometric attack keeps similar with the primitive medicine image energy; All very greatly 1.0 (see table 1 the 13rd row), (having got 9 DCT coefficient symbols here for the purpose of convenient) with primitive medicine image normalization related coefficient.

Table 1 image full figure dct transform Low Medium Frequency part coefficient and receive different the attack after changing value

^*The 1.0e+002 of dct transform coefficient unit

The related coefficient of the different medical image proper vectors of table 2 (vector length 32bit)

	Pa	Pb	Pc	Pd	Pe	Pf	Pg	Ph
									Pa	1.00	0.34	0.00	0.31	-0.17	-0.24	0.18	0.38
Pb	0.34	1.00	0.32	0.01	-0.04	0.30	-0.25	0.32
									Pc	0.00	0.32	1.00	-0.19	0.19	0.25	0.31	0.00
Pd	0.31	0.01	-019	1.00	-0.01	-0.05	0.00	0.31
									Pe	-017	-0.04	0.19	-0.01	1.00	-0.09	0.01	0.06
Pf	-0.24	0.30	0.25	-0.05	-0.09	1.00	-0.18	-0.13
									Pg	0.18	-0.25	0.31	0.00	0.01	-0.18	1.00	-0.06
Ph	0.38	0.32	0.00	0.31	0.06	-0.13	-0.06	1.00

Prove that for further the dct transform coefficient symbol sebolic addressing of full figure is a vision key character that belongs to this figure, we different test patterns (seeing Fig. 3 (a)-(g)), carry out the full figure dct transform again according to the method described above; Obtain corresponding DCT coefficient; F (1,1)-F (4,8); And obtain the related coefficient with the symbol sebolic addressing of former figure, result of calculation is as shown in table 2.

Can find out that from table 2 between the different medical images, it is bigger that symbol sebolic addressing differs, the degree of correlation is less, less than 0.5.

This explains that more the symbol sebolic addressing of DCT coefficient can reflect the main visual signature of this medical image.After watermarking images received conventional attack and geometric attack to a certain degree, this vector was constant basically, and this also meets the DCT ability that " very strong extraction characteristics of image arranged ".According to human vision property (HVS), the Low Medium Frequency signal is bigger to people's visual impact, is representing the principal character of medical image.Therefore the visual feature vector of our selected medical image is the symbol of Low Medium Frequency coefficient; It is relevant with the robustness of the quantity of information of the size of the primitive medicine image that carries out the full figure dct transform and disposable embedding and requirement that the number of Low Medium Frequency coefficient is selected; The L value is more little; The quantity of information of disposable embedding is few more, but robustness is high more.In the test of back, the length that we choose L is 32.

In sum, we are through the analysis to full figure DCT coefficient, and the symbol sebolic addressing of Low Medium Frequency coefficient capable of using obtains a kind of method that obtains the medical image visual feature vector.

Description of drawings

Fig. 1 (a) is the primitive medicine image.

Fig. 1 (b) is the image that disturbs through Gauss.

Fig. 1 (c) is the image of attacking through JPEG.

Fig. 1 (d) is the image through medium filtering.

Fig. 2 (a) is the image through rotational transform.

Fig. 2 (b) is the image through convergent-divergent 2.0.

Fig. 2 (c) is the image through convergent-divergent 0.5.

Fig. 2 (d) is the image through vertical moving.

Fig. 3 (a) is standardized test chart MRI_1.

Fig. 3 (b) is standardized test chart MRI_2.

Fig. 3 (c) is standardized test chart MRI_3.

Fig. 3 (d) is standardized test chart Engine.

Fig. 3 (e) is standardized test chart Head.

Fig. 3 (f) is standardized test chart Teddy bear.

Fig. 3 (g) is standardized test chart Mri_1back1.

Fig. 3 (h) is standardized test chart Mri_1back2.

The watermarking images of Fig. 4 (a) when not disturbing.

The watermark detection of Fig. 4 (b) when not disturbing.

Watermarking images when Fig. 5 (a) has Gauss to disturb (Gauss's interference strength is 3%).

Watermark detection when Fig. 5 (b) has Gauss to disturb.

Watermarking images (compression quality is 4%) after Fig. 6 (a) JPEG compression.

Watermark detection after Fig. 6 (b) JPEG compression.

Watermarking images behind Fig. 7 (a) medium filtering (through 10 filtering of [3,3]).

Watermark detection behind Fig. 7 (b) medium filtering.

Watermarking images behind Fig. 8 (a) rotation 20 degree.

Watermark detection behind Fig. 8 (b) rotation 20 degree.

Fig. 9 (a) zoom factor is 4.0 watermarking images.

Fig. 9 (b) zoom factor is 4.0 image watermark detection.

Figure 10 (a) zoom factor is 0.5 watermarking images.

Figure 10 (b) zoom factor is 0.5 image watermark detection.

Image after Figure 11 (a) vertical moving 10%.

Watermark detection after Figure 11 (b) vertical moving 10%.

Figure 12 (a) shears 20% watermarking images.

Figure 12 (b) shears 20% image watermark detection.

Embodiment

Below in conjunction with accompanying drawing the present invention is described further and uses 1000 groups of independently binary pseudo-random (value is+1 or 0); Every group of sequence length is 32bit; In these 1000 groups of data, appoint and extract one group (selecting the 500th group here), as the watermark sequence that embeds.Used medical image is seen Fig. 4 (a) in the experiment, is the sectioning image (128x128) of a width of cloth brain.If former figure be expressed as F (i, j), 1≤i≤128,1≤j≤128 wherein; Corresponding full figure DCT matrix of coefficients be FD (i, j), getting its Low Medium Frequency coefficient is Y (j); 1≤j≤L, the DC component of first value Y (1) representative image, from low to high frequency order is arranged then.Consider the capacity of robustness and disposable embed watermark, we select 4x8=32 coefficient of medium and low frequency to do proper vector, i.e. L=32; The DCT matrix of coefficients of choosing be FD (i, j), 1≤i≤4,1≤j≤8.Through watermarking algorithm detect W ' (j) after, judged whether the watermark embedding through calculating W (j) with W ' normalized correlation coefficient NC (Normalized Cross Correlation) (j) again.

Fig. 4 (a) is the watermarking images that does not add when disturbing;

Fig. 4 (b) does not add when disturbing, and the output of watermark detector can be seen NC=1.0, obviously detects the existence of watermark.

Below we judge the anti-conventional attack ability and the resist geometric attacks ability robustness of this digital watermark method through concrete test.

Test the ability of the anti-conventional attack of this watermarking algorithm earlier.

(1) adds Gaussian noise

Use imnoise () function in watermarking images, to add gaussian noise.

Fig. 5 (a) is for the watermarking images when Gaussian noise intensity is 3%, and is visually very fuzzy;

The output of Fig. 5 (b) watermark detector can clearly detect the existence of watermark, NC=0.87.

Table 3 is the anti-Gauss of watermark detection data when disturbing.Can see from experimental data, when Gaussian noise intensity when being 25%, watermarking images PSNR reduces to 0.11dB, at this moment detects watermark, related coefficient NC=0.64 still can detect the existence of watermark. this explanation adopts this invention that good anti-Gaussian noise ability is arranged.

The anti-Gaussian noise interfering data of table 3 watermark

Noise intensity (%)	1	3	5	10	15	20	25
								PSNR(dB)	12.40	7.90	5.89	3.25	1.69	0.72	0.11
NC	0.94	0.87	0.82	0.75	0.69	0.67	0.64

Adopt image compression quality percentage watermarking images to be carried out the JPEG compression as parameter;

Fig. 6 (a) is that compression quality is 4% image, and blocking artifact has appearred in this figure;

Fig. 6 (b) is the response of watermark detector, NC=0.81, and it is obvious to detect effect.

Table 4 is the test figure of the anti-JPEG of watermarking images.When compression quality is very poor, compression quality is 4% o'clock, still can record the existence of watermark, NC=0.81.

The experimental data of the anti-JPEG compression of table 4 watermark

Compression quality (%)	4	8	10	20	40	60	80
								PSNR(dB)	17.61	19.99	20.98	23.04	25.06	26.52	29.27
NC	0.81	0.94	0.71	0.75	1.0	1.0	1.0

Fig. 7 (a) is that the medium filtering parameter is [3x3], and the filtering multiplicity is 10 medical image, and bluring has appearred in image;

Fig. 7 (b) is the response of watermark detector, NC=0.94, and it is obvious to detect effect.

Table 5 is the anti-medium filtering ability of watermarking images, and it can be seen from the table, when the medium filtering parameter is [7x7], the filtering multiplicity is 20 o'clock, still can record the existence of watermark, NC=0.69.

The anti-medium filtering experimental data of table 5 watermark

Watermark resist geometric attacks ability:

(1) rotational transform

Fig. 8 (a) is 20 ° of watermarking images rotations, the PSNR=12.38dB of watermarking images at this moment, and signal to noise ratio (S/N ratio) is very low;

Fig. 8 (b) is the watermarking images of detection, can obviously detect the NC=0.83 that exists of watermark.

Table 6 is the anti-rotation of watermark challenge trial data.Can see in the table that when watermarking images rotates 35 ° NC=0.79 still can detect watermark and exist; The resist geometric attacks algorithm that people such as Pitas propose embeds watermark in the garden ring of DFT amplitude spectrum, can only resist the rotation that is not more than 3 degree.

Experimental data is attacked in the anti-rotation of table 6 watermark

(2) scale transformation

Fig. 9 (a) is the watermarking images when zoom factor 4.0, at this moment center image big than former figure;

Fig. 9 (b) is a watermarking detecting results, can detect the existence of watermark, NC=1.0.

Figure 10 (a) is 0.5 watermarking images for zoom factor, at this moment center image little a lot of than former figure;

Figure 10 (b) is a watermarking detecting results, can obviously detect the NC=1.0 that exists of watermark.

Table 7 is watermark convergent-divergent challenge trial data, from table 7 can see when the watermarking images zoom factor little to 0.2 the time, related coefficient NC=0.87 still can record watermark.The method of in DFT, inserting template of employings such as Pereira can only be resisted zoom factor and be not less than 0.65 convergent-divergent, explains that this invention has stronger nonshrink exoergic power.

Table 7 watermark convergent-divergent is attacked experimental data

Zoom factor	0.2	0.5	0.8	1.0	1.2	2.0	4.0
								NC	0.87	1.0	1.0	1.0	1.0	1.0	1.0

(3) translation transformation

Figure 11 (a) moves down 10% situation for image level, PSNR=11.69dB at this moment, and signal to noise ratio (S/N ratio) is very low;

Figure 11 (b) is watermark detector output, can obviously detect the NC=0.81 that exists of watermark.

Table 8 is the anti-translation challenge trial of watermark data.From table, learn, still can detect the existence of watermark, so this digital watermarking has stronger anti-translation capability when level or vertical moving 10%.

Experimental data is attacked in the anti-translation of table 8 watermark

(4) shear test

Figure 12 (a) is for to shear 20% situation to watermarking images, and at this moment image has been sheared 1/5th;

Figure 12 (b) is its watermark detection situation, can obviously detect the existence of watermark, NC=0.87.

Table 9 is watermark cut-through resistance test data, and test figure can learn that this algorithm has certain anti-shear ability from table.

The anti-shearing attack experimental data of table 9 watermark

Through above description of test, the embedding grammar of this watermark has stronger anti-conventional attack ability and geometric attack ability, and the embedding of watermark do not influence the value of medical image, is a kind of zero watermark.

Claims (1)

1. medical image remainder word water mark method based on the DCT resist geometric attacks; It is characterized in that: based on the full figure dct transform; Obtain visual feature vector, and encryption technology and digital watermark are combined, realized the anti-geometry and the conventional attack of medical image digital watermarking; This digital watermark method is divided into two parts, amounts to four steps:

First is a watermark embedding method: through the embedding operation to watermark, obtain corresponding two-valued function sequence key (j);

1) the primitive medicine image is carried out the full figure dct transform, from the DCT coefficient, obtain the visual feature vector V (j) of this figure according to the symbol sebolic addressing of Low Medium Frequency coefficient;

2) utilize Hash function and the watermark W (j) that will embed, obtain a two-valued function sequence key (j), $\mathrm{Key}\left(j\right)=V\left(j\right)\⊕W\left(j\right);$

Preserve key (j), will use when extracting watermark below, through applying for as key key (j) to the third party, to obtain entitlement the primitive medicine image;

Second portion is a watermark extracting method: the visual feature vector V ' through two-valued function sequence key (j) and testing image (j) extracts watermark W ' (j);

3) medical image to be measured is carried out the full figure dct transform, in the DCT coefficient, the visual feature vector V ' that goes out testing image according to the symbol extraction of Low Medium Frequency coefficient (j);

4) utilize Hash function character to extract watermark; (j) carries out degree of correlation test with W (j) and W ', confirms the entitlement of medical image.

Images