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 Zero Digital Watermarking Method for Medical Images Based on DCT Against Geometric Attacks

technical field

The invention belongs to the field of multimedia signal processing, and relates to a medical image digital watermarking technology based on DCT transformation and image visual features, in particular to a medical image zero digital watermarking method based on DCT against geometric attack.

Background technique

At present, medical images account for 70% to 80% of the medical information in the whole hospital. The digital information management system plays an increasingly important role in the modern medical system. However, with the popularization and application of the network, its information security problems are gradually exposed.

When medical images are transmitted remotely on the network, the personal information of patients recorded on the medical images is easily leaked. If the personal information is embedded in the medical image as a digital watermark, this problem can be better solved. This kind of watermark is called medical image digital watermark.

At present, the research in the field of digital watermarking of medical images mainly focuses on the two aspects of space domain and transform domain (DCT, DFT and DWT). to embed the watermark. Among them, the cosine transform (Discrete Cosine Transform, DCT) domain watermarking method, due to its small amount of calculation and compatibility with international data compression standards (JPEG, MPEG), is currently being studied more, and it is most of the existing frequency domain digital watermarking methods. Algorithm research hotspot.

In view of the special requirements for the protection of medical image lesion areas, the general literature often chooses to embed watermark information into the non-interest region of the image (Region of Non-Interest, RONI). The Region of Interest (ROI) in medical images refers to those lesion areas that contain important pathological features or diagnosis and treatment information. If a watermark is embedded in this area, it may cause wrong diagnosis. But people often spend a lot of time and energy when looking for ROI, and once the wrong choice is made, it may interfere with the doctor's diagnosis.

In the field of medical image digital watermarking research, geometric attack is still a relatively difficult topic so far. As for the watermarking method that can effectively resist conventional attack and geometric attack at the same time, the research on these two attack types of watermarking methods has not been reported yet, and it is still blank. . However, in practical applications, medical digital watermarked images are often subject to both attacks at the same time.

Contents of the invention

The purpose of the present invention is to provide a zero-digital watermarking method for medical images based on DCT anti-geometric attack. By organically combining the visual feature vector of medical images, encryption technology and the concept of a third party, it is not necessary to select the region of interest. This solves the problem of fast watermark embedding and extraction, and has ideal robustness and invisibility to protect the copyright of medical images and the confidentiality of patient information, effectively solving the concealment of patient information and the privacy of medical images. Sensitivity issues, while solving the problem of resisting geometric attacks and resisting conventional attacks in medical image applications.

In order to achieve the above object, the present invention is carried out as follows: based on the full-image DCT transformation, a medical image visual feature vector that is resistant to geometric attacks is extracted from the DCT transformation coefficients, and the watermarking technology is organically combined with cryptography to realize digital Resistance of watermarks to geometric and conventional attacks. The method adopted in the present invention includes two parts: watermark embedding and watermark extraction. The first part is the watermark embedding algorithm, including: (1) obtaining a visual feature vector V(j) of the medical image by performing full-image DCT transformation, (2 ) Generate a binary logic sequence key(j) according to the watermark W(j) and the visual feature vector V(j) of the image. The second part is the watermark extraction algorithm, including: (3) finding the visual feature vector V'(j) of the medical image to be tested, (4) using the binary logic sequence key(j) and the visual feature vector V' of the image to be tested (j), extract the watermark W'(j).

Now the method of the present invention is described in detail as follows:

First, a group of binary pseudo-random sequences W that can represent patient information, W={w(j)|w(j)=0, 1; 1≤j≤L} are used as digital watermarks, and the original image is recorded as F= {f(i, j)|f(i, j)∈R; 1≤i≤N1, 1≤j≤N2)}, where w(j) and f(i, j) denote the watermark sequence and the original For the pixel gray value of the medical image, set N1=N2=N.

Part 1: Watermark Embedding Algorithm

1) Obtain the visual feature vector V(j) of the medical image by performing full-image DCT transformation.

First perform full-image DCT transformation on the original image F(i, j) to obtain the DCT coefficient matrix FD(i, j), and then take the first L coefficients among the low-intermediate frequency coefficients for the DCT coefficient matrix FD(i, j) , and obtain the visual feature vector V(j) of the image through the DCT coefficient symbol operation. The specific method is that when the DCT coefficient is positive or zero, we use "1" to represent it, and when the coefficient is negative, we use "0" to represent it. The procedure is described as follows :

FD(i,j)=DCT2(F(i,j))

V(j)=-Sign(FD(i,j))

2) Generate a binary logic sequence key(j) according to the watermark W(j) and the visual feature vector V(j) of the image.

$\mathrm{<span\; class="notranslate">\; key</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CirclePlus;</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span>$

Key(j) is generated by the visual feature vector V(j) and watermark W(j) of the image through the Hash function commonly used in cryptography. Save key(i), it will be used later when extracting the watermark. By using key(j) as a key to apply to a third party to obtain the ownership and use rights of medical images, so as to achieve the purpose of protecting medical images.

Part II: Watermark Extraction Algorithm

3) Obtain the visual feature vector V'(j) of the medical image to be tested.

Suppose the medical image to be tested is F'(i, j), the DCT coefficient matrix obtained after DCT transformation of the whole image is FD'(i, j), and the visual feature vector V'( j);

FD'(i,j)=DCT2(F'(i,j))

V'(j)=-Sign(FD'(i,j))

4) Extract the watermark W'(j) from the image to be tested.

According to the key(j) generated when embedding the watermark and the visual feature vector V'(j) of the image to be tested, the watermark W'(j) of the image to be tested can be extracted by using the Hash property

${\mathrm{<span\; class="notranslate">\; W</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; key</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CirclePlus;</span>{\mathrm{<span\; class="notranslate">\; V</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span>$

Then according to the degree of correlation between W(j) and W'(j), the ownership of the image to be tested and the security of hidden information are judged.

Compared with the existing medical watermarking technology, the present invention has the following advantages:

Since the present invention is a digital watermarking technology based on DCT transformation, it has fast calculation speed, high precision, good compatibility, and strong anti-geometric attack ability and anti-conventional attack ability; selected, thus solving the problem of the quickness of watermark embedding; the embedded watermark is a kind of zero watermark, which does not affect the quality of the original medical image, and has high practical value in medical treatment, and the algorithm can be applied to other fields; using third-party The concept of is adapted to the practicality and standardization of today's network promotion; the following is an explanation from the theoretical basis and experimental data:

1) Discrete cosine transform

The use of DCT in image coding is currently widely used in JPEG compression and MPEG-1/2 standards. DCT is the second-best orthogonal transform next to K-L transform obtained under the minimum mean square error condition, and is a lossless chieftain transform. It has fast operation speed and high precision, and is famous for the best balance between the ability to extract feature components and operation speed.

The two-dimensional discrete cosine transform (DCT) formula is as follows:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span>\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; cos</span>}\frac{\mathrm{<span\; class="notranslate">\; \&pi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; u</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; m</span>}}\mathrm{<span\; class="notranslate">\; cos</span>}\frac{\mathrm{<span\; class="notranslate">\; \&pi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; v</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}$

u=0,1,Λ,M-1; v=0,1,Λ,N-1;

In the formula

$\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\sqrt{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; m</span>}}& \mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}\\ \sqrt{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; m</span>}}& \mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right.$ $\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\sqrt{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; N</span>}}& \mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}\\ \sqrt{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; N</span>}}& \mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}\end{array}\right.$

The two-dimensional inverse discrete cosine transform (IDCT) formula is as follows:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; u</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; v</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; cos</span>}\frac{\mathrm{<span\; class="notranslate">\; \&pi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; u</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; m</span>}}\mathrm{<span\; class="notranslate">\; cos</span>}\frac{\mathrm{<span\; class="notranslate">\; \&pi;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; v</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; N</span>}}$

x=0,1,Λ,M-1; y=0,1,Λ,N-1

Among them, x, y are sampling values in the space domain; u, v are sampling values in the frequency domain, and usually digital images are represented by a square matrix of pixels, that is, M=N

It can be seen from the above formula that the coefficient sign of DCT is related to the phase of the component.

2) Selection method of main feature vectors of medical image vision

The main reason why most medical image watermarking algorithms have poor resistance to geometric attacks is that people embed digital watermarks in pixels or transform coefficients, and slight geometric transformations of medical images often lead to large changes in pixel values or transform coefficient values. This makes the embedded watermark vulnerable to attack. If the visual feature vector reflecting the geometric characteristics of the image can be found, then when the image undergoes a small geometric transformation, the visual feature value of the image will not change significantly. Hayes research shows that phase is more important than magnitude for image features. After observing a large number of full-image DCT data (low intermediate frequency), we found that when a common geometric transformation is performed on a medical image, the size of the low intermediate frequency coefficient may change, but the sign of the coefficient remains basically unchanged. We select some experiments The data are shown in Table 1. The original medical image used for testing in Table 1 is Figure 1(a), which is an image medical image (128x128). The first column in the table shows the types of attacks on medical images. See Figure 1(b)-(d) for medical images subjected to conventional attacks, and Figure 2(a)-(d) for medical images subjected to geometric attacks. . Columns 3 to 11, this is the nine low-frequency coefficients of FD(1,1)-FD(3,3) taken in the DCT coefficient matrix, where the coefficient F(1,1) represents the DC component of the medical image value. For conventional attacks, these low intermediate frequency coefficient values FD(1,1)-FD(3,3) remain basically unchanged, and are approximately equal to the original medical image values; for geometric attacks, some coefficients have a large change, but we can find , when the medical image is attacked geometrically, the size of some DCT low-intermediate frequency coefficients changes but its sign basically does not change. We represent positive DCT coefficients with "1" (including coefficients with a value of zero), and negative coefficients with "0", then for the original medical image, FD(1,1)-FD in the DCT coefficient matrix (3, 3) coefficient, the corresponding coefficient symbol sequence is: "1100 01001", see the 12th column of Table 1, we observe this column and find that the symbol sequence and the original medical image can remain similar regardless of the conventional attack or the geometric attack , the normalized correlation coefficient with the original medical image is 1.0 (see column 13 of Table 1), (9 DCT coefficient symbols are taken here for convenience).

Table 1. Coefficients of low and intermediate frequency parts of the DCT transformation of the whole image and the change values after different attacks

^* DCT transformation coefficient unit 1.0e+002

Table 2 Correlation coefficients of different medical image feature vectors (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

In order to further prove that the DCT transformation coefficient symbol sequence of the whole image is a visually important feature of the image, we take different test images (see Figure 3(a)-(g)) and perform the DCT transformation of the whole image according to the above method, The corresponding DCT coefficients, F(1,1)-F(4,8), are obtained, and the correlation coefficient with the symbol sequence of the original image is obtained, and the calculation results are shown in Table 2.

It can be seen from Table 2 that, among different medical images, the sign sequence has a large difference, and the correlation is small, less than 0.5.

This further shows that the sign sequence of DCT coefficients can reflect the main visual features of the medical image. When the watermark image is subjected to a certain degree of conventional attack and geometric attack, the vector is basically unchanged, which also conforms to DCT's ability to extract image features very strongly. According to the characteristics of human vision (HVS), low-intermediate frequency signals have a greater impact on human vision and represent the main characteristics of medical images. Therefore, the visual feature vector of the medical image we choose is the symbol of the low intermediate frequency coefficient, the number of low intermediate frequency coefficients is selected and the size of the original medical image for full-image DCT transformation, as well as the amount of information embedded at one time and the required robustness The smaller the L value is, the less information is embedded at one time, but the higher the robustness. In the following experiments, we choose the length of L to be 32.

To sum up, through the analysis of the DCT coefficients of the whole image, we can use the symbol sequence of the low-intermediate frequency coefficients to obtain a method to obtain the visual feature vector of the medical image.

Description of drawings

Figure 1(a) is the original medical image.

Figure 1(b) is the image after Gaussian interference.

Figure 1(c) is the image after JPEG attack.

Figure 1(d) is the median filtered image.

Figure 2(a) is the image after rotation transformation.

Figure 2(b) is the image scaled by 2.0.

Figure 2(c) is the image scaled by 0.5.

Figure 2(d) is a vertically shifted image.

Figure 3(a) is the standard test chart MRI_1.

Figure 3(b) is the standard test chart MRI_2.

Figure 3(c) is the standard test chart MRI_3.

Figure 3(d) is the standard test diagram Engine.

Figure 3(e) is the standard test chart Head.

Figure 3(f) is the standard test pattern Teddy bear.

Figure 3(g) is the standard test graph Mri_1back1.

Figure 3(h) is the standard test graph Mri_1back2.

Figure 4(a) Watermarked image without disturbance.

Figure 4(b) Watermark detection without interference.

Figure 5(a) Watermark image with Gaussian interference (Gaussian interference strength is 3%).

Figure 5(b) Watermark detection with Gaussian interference.

Figure 6(a) Watermarked image after JPEG compression (compression quality is 4%).

Figure 6(b) Watermark detection after JPEG compression.

Figure 7(a) Watermarked image after median filtering (filtered 10 times by [3, 3]).

Figure 7(b) Watermark detection after median filtering.

Figure 8(a) The watermarked image rotated by 20 degrees.

Figure 8(b) Watermark detection after rotating 20 degrees.

Figure 9(a) Watermarked image with scaling factor 4.0.

Figure 9(b) Image watermark detection with scaling factor 4.0.

Figure 10(a) Watermarked image with a scaling factor of 0.5.

Figure 10(b) Image watermark detection with a scaling factor of 0.5.

Figure 11(a) Image after vertical shift of 10%.

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

Figure 12(a) crops 20% of the watermarked image.

Figure 12(b) Image watermark detection with 20% cut.

Detailed ways

Below in conjunction with accompanying drawing, the present invention is further described and uses 1000 groups of independent binary pseudo-random sequences (values are +1 or 0), and the length of each group of sequences is 32bit. In these 1000 groups of data, a group (selected here) Group 500), as an embedded watermark sequence. The medical image used in the experiment is shown in Figure 4(a), which is a sliced image of the brain (128x128). Let the original image be expressed as F(i, j), where 1≤i≤128, 1≤j≤128, the corresponding full-image DCT coefficient matrix is FD(i,j), and its low-intermediate frequency coefficient is Y(j) , 1≤j≤L, the first value Y(1) represents the DC component of the image, and then arranged in order from low to high frequency. Considering the robustness and the capacity of embedding the watermark at one time, we choose 4x8=32 coefficients of medium and low frequencies as the feature vector, that is, L=32; the selected DCT coefficient matrix is FD(i, j), 1≤i≤4 , 1≤j≤8. After W'(j) is detected by the watermark algorithm, the normalized correlation coefficient NC (Normalized Cross Correlation) between W(j) and W'(j) is calculated to determine whether there is a watermark embedded.

Figure 4(a) is the watermarked image without interference;

Figure 4(b) When no interference is added, the output of the watermark detector can be seen that NC = 1.0, and the existence of the watermark is clearly detected.

Next, we judge the robustness of the digital watermarking method against conventional attacks and against geometric attacks through specific experiments.

First test the ability of the watermarking algorithm to resist conventional attacks.

(1) Add Gaussian noise

Use the imnoise() function to add Gaussian noise to the watermarked image.

Figure 5(a) is the watermark image when the intensity of Gaussian noise is 3%, which is visually blurred;

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

Table 3 is the detection data when the watermark resists Gaussian interference. As can be seen from the experimental data, when the Gaussian noise intensity is as high as 25%, the PSNR of the watermark image drops to 0.11dB. At this time, the watermark is detected, and the correlation coefficient NC=0.64, and the existence of the watermark can still be detected. This shows that the invention is effective Good ability to resist Gaussian noise.

Table 3 Watermark anti-Gaussian noise interference data

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 (2) JPEG compression processing

Using the image compression quality percentage as a parameter to perform JPEG compression on the watermarked image;

Figure 6(a) is an image with a compression quality of 4%, and the block effect has already appeared in this figure;

Figure 6(b) is the response of the watermark detector, NC=0.81, the detection effect is obvious.

Table 4 is the test data of anti-JPEG watermark image. When the compression quality is very poor and the compression quality is 4%, the existence of the watermark can still be detected, NC=0.81.

Table 4 Experimental data of watermark anti-JPEG compression

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 (3) Median filter processing

Figure 7(a) is a medical image with a median filter parameter of [3x3] and a filter repetition number of 10, and the image has been blurred;

Figure 7(b) is the response of the watermark detector, NC=0.94, the detection effect is obvious.

Table 5 shows the anti-median filter capability of the watermark image. It can be seen from the table that when the median filter parameter is [7x7] and the number of filter repetitions is 20, the existence of the watermark can still be detected, NC=0.69.

Table 5 Watermark anti-median filtering experimental data

Watermark anti-geometric attack ability:

(1) Rotation transformation

Fig. 8(a) is that the watermark image is rotated 20°, at this time, the PSNR of the watermark image is 12.38dB, and the signal-to-noise ratio is very low;

Figure 8(b) is the detected watermark image, and the existence of the watermark can be clearly detected with NC=0.83.

Table 6 is the watermark anti-rotation attack test data. It can be seen from the table that when the watermark image is rotated by 35°, NC=0.79, the existence of the watermark can still be detected; the anti-geometric attack algorithm proposed by Pitas et al., which embeds the watermark in the ring of the DFT amplitude spectrum, can only resist no more than 3 degrees of rotation.

Table 6 Watermark anti-rotation attack experimental data

(2) Zoom transformation

Figure 9(a) is the watermark image when the zoom factor is 4.0, and the central image is larger than the original image at this time;

Figure 9(b) is the result of watermark detection, the existence of watermark can be detected, NC=1.0.

Figure 10(a) is a watermark image with a scaling factor of 0.5, at this time the center image is much smaller than the original image;

Fig. 10(b) is the watermark detection result, it can be clearly detected that the existence of the watermark NC=1.0.

Table 7 shows the test data of watermark scaling attack. It can be seen from Table 7 that when the watermark image scaling factor is as small as 0.2, the correlation coefficient NC=0.87, and the watermark can still be measured. The method of embedding templates in DFT adopted by Pereira et al. can only resist scaling when the scaling factor is not less than 0.65, which shows that the invention has strong anti-scaling ability.

Table 7 Watermark Scaling Attack Experimental Data

scaling 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

Fig. 11 (a) is the situation that image level moves down 10%, at this moment PSNR=11.69dB, signal-to-noise ratio is very low;

Figure 11(b) is the output of the watermark detector, and the existence of the watermark can be clearly detected with NC=0.81.

Table 8 is the watermark anti-translation attack test data. It is known from the table that when the horizontal or vertical movement is 10%, the existence of the watermark can still be detected, so the digital watermark has strong anti-translation ability.

Table 8 Watermark anti-translation attack experimental data

(4) Shear test

Fig. 12(a) is the situation of cutting 20% of the watermark image, at this moment the image has been cut off by one-fifth;

Figure 12(b) shows the watermark detection situation, the existence of the watermark can be clearly detected, NC=0.87.

Table 9 shows the watermark anti-cutting test data. From the test data in the table, it can be known that the algorithm has a certain anti-shearing ability.

Table 9 Watermark anti-shearing attack experimental data

The above experiments show that the watermark embedding method has a strong ability to resist conventional attacks and geometric attacks, and the embedding of the watermark does not affect the value of the medical image, which is a zero watermark.

Claims (1)

1. A medical image zero digital watermarking method based on DCT anti-geometric attack, characterized in that: based on the full-image DCT transformation, the visual feature vector is obtained, and the encryption technology and watermarking technology are organically combined to realize the digital watermarking of medical images Anti-geometric and conventional attacks, the digital watermarking method is divided into two parts, a total of four steps: The first part is the watermark embedding method: through the embedding operation of the watermark, the corresponding binary logic sequence key(j) is obtained; 1) Perform full-image DCT transformation on the original medical image, and obtain the visual feature vector V(j) of the image from the DCT coefficients according to the symbol sequence of the low-intermediate frequency coefficients; 2) Use the Hash function and the watermark W(j) to be embedded to obtain a binary logic sequence key(j), $\mathrm{key}\left(j\right)=V\left(j\right)\&CirclePlus;W\left(j\right);$ Save the key(j), which will be used when extracting the watermark below, and apply to the third party by using key(j) as the key to obtain the ownership of the original medical image; The second part is the watermark extraction method: extract the watermark W'(j) through the binary logic sequence key(j) and the visual feature vector V'(j) of the image to be tested; 3) Carry out full-image DCT transformation on the medical image to be tested, and extract the visual feature vector V'(j) of the image to be tested according to the sign of the low intermediate frequency coefficient among the DCT coefficients; 4) Use the Hash function property to extract the watermark, Correlation test is performed on W(j) and W'(j) to determine the ownership of the medical image.

Images