CN108876696A - A kind of medical image robust watermarking method based on SIFT-DCT - Google Patents

Abstract

The invention discloses a medical image robust watermarking method based on SIFT-DCT, which belongs to the field of multimedia signal processing. The steps of the present invention are as follows: first, the watermark is symbolically encrypted in the frequency domain by utilizing the properties of the Logistic Map; then a feature vector is extracted by performing SIFT-DCT transformation on the medical image to embed the watermark, and the feature vector is associated with the binary watermark Obtain a binary logic sequence, and store the binary sequence in a third party; then extract its feature vector by performing SIFT-DCT transformation on the medical image to be tested, and associate it with the binary sequence stored in the third party for watermarking extract. The present invention is a medical image digital watermarking technology based on SIFT-DCT, which has better robustness, and is particularly prominent against geometric attacks such as rotation, translation, and shearing. The embedding of the watermark does not change the content of the original encrypted data, and is a kind of Zero watermark technology.

Description

A robust watermarking method for medical images based on SIFT-DCT

technical field

The present invention relates to a medical image robust digital watermarking technology based on SIFT-DCT transformation, chaotic mapping and image feature vectors, specifically a SIFT-DCT-based medical image robust watermarking method, which is a multimedia data protection method, It belongs to the field of multimedia signal processing.

technical background

The development of medicine is gradually changing from traditional medicine to telemedicine, which makes a large number of medical images transmitted and shared in the network; in order to solve the problems that medical images may be tampered with and stolen during the transmission and sharing process, it is necessary to process the original medical images; The combination of zero watermark technology and perceptual hash technology, as a security technology for information security, can not only ensure safe transmission, but also realize information authentication, which is very important in practical applications.

Digital watermarking technology was originally used for copyright protection of digital media. Now, using the invisibility and robustness of digital watermarking, it can hide the patient's personal information in its medical images to ensure its safe transmission on the Internet. . Therefore, when digital images are widely used in network transmission, research on digital watermarking algorithms for medical images has become extremely important; through unique invisibility, robustness and other characteristics, the privacy of patients is protected, and zero Watermarking can avoid tampered medical data, thus enabling relevant patient information required for remote medical diagnosis.

At present, there are few researches on digital watermarking algorithms for medical images, and even less research results on zero-watermarking algorithms for medical data resistant to geometric attacks. In the future, there will be a large amount of medical data transmission problems, so it is of great significance to study how to embed digital robust watermarks in medical data, and for medical data, it is generally not allowed to modify its content. This makes it difficult to embed watermarks in medical data.

In short, the method of embedding a digital watermark that can resist geometric attacks such as rotation, scaling, translation, shearing, and distortion in SIFT-DCT-based medical images is still blank, and there is no public report.

Contents of the invention

The present invention is a SIFT-DCT-based robust watermarking method for medical images, which makes up for the traditional digital watermarking method that cannot be used for medical images by combining the feature vectors, cryptography, hash functions and zero watermarking technology of medical images. The shortcomings of the protection, with strong robustness and invisibility, can protect the patient's private information and the data security of medical images at the same time.

In order to achieve the above object, the present invention is carried out as follows: based on the full-image SIFT transformation, DCT transformation is performed on the normalized feature descriptor matrix descrips in the SIFT transformation, and an anti-geometric attack is extracted from the coefficient matrix after the DCT transformation. Medical image visual feature vector, and organically combine watermarking technology with chaotic encryption, Hash function and "third-party concept", and realize the anti-geometric attack and conventional attack of digital watermarking. The method adopted in the present invention includes five parts: feature vector extraction based on SIFT-DCT, watermark encryption, watermark embedding, watermark extraction and watermark decryption.

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

Select a meaningful binary text image as the watermark embedded in the medical image, recorded as W={w(i,j)|w(i,j)=0,1; 1≤i≤M1,1≤j≤M2 }. At the same time, we select a 512*512 medical image as the original medical image, denoted as I(i,j), W(i,j) and I(i,j) respectively represent the pixel gray value of the watermark and the original medical image .

The first part: Under the SIFT-DCT transformation, extract the feature vector of the medical image

1) Perform SIFT transformation on the original medical image I(i,j) to obtain the feature descriptor matrix descriptions;

2) Perform DCT transformation on the descriptions matrix to obtain the coefficient matrix F(i,j);

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

3) Select 4*8 modules in F(i,j) to form a new matrix A(i,j), and calculate the mean value m of A matrix;

4) Use the hash function to generate the characteristic binary sequence V(i,j) of the 32-bit medical image Part II: Encryption of the watermark

5) Obtain binary chaotic sequence

First generate a chaotic sequence X(i,j) according to the initial value x ₀ , then assign the elements greater than or equal to 0.5 to "1", and assign the rest to "0" to obtain a binary chaotic sequence k(n);

6) Get the chaotic encrypted watermark

The encrypted watermark EW(i,j) is obtained by bit-by-bit XOR operation of binary watermark W(i,j) and binary chaotic sequence k(n);

Part III: Watermark Embedding

7) Execute the bit-by-bit XOR operation of the feature vector V(i,j) and the encrypted watermark EW(i,j) to embed the watermark into the medical image and obtain the logical key Key(i,j) at the same time ;

Save Key(i,j), which will be used later when extracting the watermark. By applying Key(i,j) as a key to a third party, the ownership and use right of the original medical image can be obtained, so as to achieve the purpose of protecting the medical image;

Part IV: Watermark Extraction

8) The feature vector of the medical image I'(i,j) to be tested

Perform SIFT processing on the medical image to be tested, obtain the descrips matrix, and then perform DCT transformation to obtain the coefficient matrix F'(i,j), select the 4*8 modules in the coefficient, and obtain the visual feature sequence of the medical image to be tested through the hash function V'(i,j);

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

9) Extract watermark EW'(i,j)

Perform XOR operation on the feature vector V'(i,j) of the encrypted image to be tested and the logical key Key(i,j) to extract the encrypted watermark EW'(i,j);

This algorithm only needs the key Key(i,j) when extracting the watermark, and does not require the participation of the original image. It is a zero-watermark extraction algorithm;

Part V: Decryption of Watermark

10) Obtain binary chaotic encrypted sequence k(n)

Using the same method as watermark encryption, the same binary chaos matrix k(n) is obtained;

11) Restore the extracted encrypted watermark

The restored watermark W'(i,j) is obtained by XORing the binary chaotic sequence k(n) and the extracted encrypted watermark EW'(i,j);

Determine the ownership of the medical image and the embedded watermark information by calculating the correlation coefficient NC of W(i,j) and W'(i,j);

Innovation point of the present invention:

This algorithm is based on SIFT and DCT, taking into account the advantages of SIFT's fast calculation speed and strong ability to resist geometric attacks, and the characteristics of DCT's ergodicity and robustness, and extracts features from medical images. As a special class of images, medical images require the integrity of the original data. Due to the adoption of zero-watermark embedding technology, this algorithm solves the defects caused by traditional watermark embedding technology to modify the original image data, and ensures the quality of medical images. Using the concept of a third party, it adapts to the practicality and standardization of today's network technology.

The following is explained from the theoretical basis and experimental data:

1) SIFT algorithm

SIFT (Scale Invariant Feature Transform) is a local feature extraction and matching method that researchers are more enthusiastic about in recent years. Its function is to find some "key points" similar to the fingerprint of the image from the image matrix. This algorithm is a local feature algorithm, which searches for extreme points in the scale space to extract position, scale, and rotation invariants. The SIFT algorithm extracts features from images generally includes the following steps:

①Extreme value detection in scale space: search for image positions on all scales, and use local extremum in DOG (Difference-of-Gaussian) scale space as candidate points to determine the location and scale. The scale space of a two-dimensional image is defined as:

L(x,y,σ)=G(x,y,σ)*I(x,y)

Among them, G(x, y, σ) is a scale-variable Gaussian function, and (x, y) is a spatial scale coordinate. In order to efficiently detect stable keypoints in the rescale space, the Difference of Gaussian space (DOG scale-space) is proposed. Generated using Gaussian difference kernels of different scales and image convolution.

D(x,y,σ)=(G(x,y,kσ)-G(x,y,σ))*I(x,y)

=L(x,y,kσ)-L(x,y,σ)

In order to find the extremum points, each sample point is compared with all its neighbors. The middle detection point is compared with its 8 adjacent points of the same scale and 9*2 points corresponding to the upper and lower adjacent scales, a total of 26 points are compared to ensure that extreme points are detected in both scale space and two-dimensional space.

② Key point positioning: at each candidate position, the position and scale are determined through a fine-fitting model. At the same time, edge effects are removed to enhance matching stability and improve noise immunity.

③Direction determination: Sampling in the neighborhood centered on the key point, and use the histogram to count the gradient direction of the neighborhood pixels.

④ Key point descriptor: In the neighborhood around each key point, a local coordinate system is established with the main direction of the key point as 0 degrees to ensure rotation invariance.

2) discrete cosine transform

in, f(x, y) is the pixel value at point (x, y), F(u, v) is the 2D-DCT transformation coefficient of f(x, y); x, y is the sampling value in the space domain; u, v Sample values for the frequency domain.

2) Logistic Map

Logistic Map is one of the most famous chaotic maps. It is a simple dynamic nonlinear regression with chaotic behavior. Its mathematical definition can be expressed as follows:

x _k+1 ＝μ x _k (1-x _k )

Among them, x(k) belongs to (0, 1), 0<u<=4; experiments show that when 3.5699456<u<=4, the logistic map enters the chaotic state, and the Logistic chaotic sequence can be used as an ideal key sequence.

3) Selection method of medical image feature vector

The main reason why most medical image watermarking algorithms have poor anti-geometric attack ability is that people embed digital watermarks in pixels or transformation coefficients, and slight geometric transformations of medical images often lead to large changes in pixel values or transformation coefficient values. It will make the embedded watermark very easy to be attacked. If the eigenvectors that reflect the geometric characteristics of medical images can be found, then when the image undergoes a small geometric transformation, the eigenvalues of the image will basically not change significantly. Through observing to the SIFT data (descrips matrix) of a large amount of medical images, we find that when carrying out medical image processing by the method for the present invention, when carrying out common geometric transformation to a medical image, some changes may take place in the numerical size of the descriptions matrix , but the degree of curvature of all data curves remains basically the same. Combined with the characteristics of DCT as long as the energy is concentrated in low frequency, according to the human visual characteristics (HVS), the low-intermediate frequency signal has a greater impact on human vision, representing the main features of the image, so the SIFT-DCT transformation of the medical image we selected The final data is used as a visual feature vector.

Description of drawings

Figure 1 is the original medical image.

Figure 2 is the original watermarked image.

Figure 3 is the encrypted watermark image.

Figure 4 is the extracted watermark without interference.

Fig. 5 is a medical image with a Gaussian noise interference intensity of 10%.

Figure 6 is the watermark extracted when the interference strength of Gaussian noise is 10%.

Figure 7 is a JPEG compressed medical image (20% compression quality).

Figure 8 is the extracted watermark when the compression quality is 20% of the JPEG compression.

Fig. 9 is a medical image after median filtering (the window size is [3x3], and the number of filtering is 10 times).

Figure 10 is [3x3], the watermark extracted after median filtering 10 times.

Figure 11 is the medical image after median filtering (the window size is [5x5], and the number of filtering is 10 times).

Figure 12 is [5x5], the watermark extracted after median filtering 10 times.

Figure 13 is a medical image rotated 30° clockwise.

Figure 14 is the extracted watermark when rotated 30° clockwise.

Figure 15 is a medical image rotated 40° counterclockwise.

Figure 16 is the extracted watermark when it is rotated counterclockwise by 40°.

Fig. 17 is a medical image scaled 1.2 times.

Figure 18 is the extracted watermark when zoomed in by 1.2 times.

Fig. 19 is a medical image horizontally shifted to the left by 30%.

Figure 20 is the watermark extracted when horizontally shifted to the left by 30%.

Fig. 21 is a medical image vertically shifted down by 30%.

Figure 22 is the extracted watermark when moving down 30% vertically.

Fig. 23 is a medical image cropped 30% along the X axis.

Figure 24 is the watermark extracted when cutting 30% along the X axis.

Figure 25 is a medical image cropped 15% along the Y axis.

Fig. 26 is the watermark extracted when cutting 15% along the Y axis.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings. The object of the experimental test is a 512×512 abdominal medical image, as shown in FIG. 1 , represented by I(i,j), where 1≤i, j≤512. Select a meaningful binary image as the original watermark, recorded as: W={w(i,j)|w(i,j)=0,1; 1≤i≤M1,1≤j≤M2}, see Figure 2, where the size of the watermark is 32×32. First, the SIFT algorithm is used to process the original image. Among the three generated matrices, the descrips feature descriptor matrix is selected for DCT transformation. Considering the robustness and the capacity of one-time embedding watermark, we take 32 coefficients, that is, a 4 *8 modules. Set the initial value of the chaos coefficient to 0.2, the growth parameter to 4, and the number of iterations to 32. Then perform chaotic encryption on the original watermark, and the encrypted watermark is shown in Figure 3. After detecting W'(i,j) through the watermark algorithm, we judge whether there is a watermark embedded by calculating the normalized correlation coefficient NC (Normalized Cross Correlation). When the value is closer to 1, the similarity is higher, so Judging the robustness of the algorithm. The distortion degree of the picture represented by PSNR, when the PSNR value is larger, the distortion degree of the picture is smaller.

Figure 4 is the watermark extracted without interference, it can be seen that NC=1.00, the watermark can be extracted accurately.

Next, we judge the anti-conventional attack ability and anti-geometric attack ability of the digital watermarking method 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 watermark.

Table 1 is the experimental data of watermark anti-Gaussian noise interference. It can be seen from the table that when the Gaussian noise intensity is as high as 15%, the PSNR of the image after the attack drops to 10.98dB. At this time, the extracted watermark, the correlation coefficient NC=0.62, can still accurately extract the watermark, and the overall data is uniform Around 0.6. This shows that adopting this invention can resist Gaussian noise.

Fig. 5 is a medical image when the Gaussian noise intensity is 10%, which is visually different from the original abdominal medical image;

Figure 6 is the watermark extracted when the intensity of Gaussian noise is 10%, NC=0.64.

Table 1 Watermark anti-Gaussian noise interference data

Noise intensity (%) 3 5 8 10 12 15 18 PSNR(dB) 17.30 15.20 13.35 12.50 11.81 10.98 10.33 NC 0.56 0.54 0.59 0.64 0.59 0.62 0.59

(2) JPEG compression processing

The percentage of image compression quality is used as a parameter to perform JPEG compression on abdominal medical images; Table 2 shows the experimental data of watermark resistance to JPEG compression. When the compression quality is only 5%, the image quality is low, but the watermark can still be extracted, NC=0.62.

Figure 7 is a medical image with a compression quality of 20%;

Figure 8 is the extracted watermark with 20% compression quality, NC=0.80, the watermark can be extracted accurately.

Table 2 Watermark anti-JPEG compression experiment data

Compression quality (%) 5 10 15 20 25 30 40 50 PSNR(dB) 27.00 29.41 32.91 35.35 36.67 40.12 30.82 36.67 NC 0.62 0.48 0.51 0.80 0.89 0.79 0.73 0.89

(3) Median filter processing

Table 3 shows the watermark anti-median filtering ability of medical images. It can be seen from the table that when the median filtering parameter is [3x3] and the number of filtering repetitions is 15, the presence of watermark can still be detected, NC=0.62.

Figure 9 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 10 is the watermark extracted when the median filtering parameter is [3x3] and the number of filtering repetitions is 10.

NC=0.62, the watermark can be extracted.

Figure 11 is a medical image with a median filter parameter of [5x5] and a filter repetition number of 10;

Figure 12 is the watermark extracted when the median filter parameter is [5x5] and the number of filter repetitions is 10.

NC=0.62, the watermark can be extracted.

Table 3 Watermark anti-median filtering experimental data

Watermark anti-geometric attack capability

(1) Rotation transformation

Table 4 is the watermark anti-rotation attack experimental data. It can be seen from the table that when the image is rotated clockwise by 45°, NC=0.81, the watermark can still be extracted.

Figure 13 is a medical image rotated 30° clockwise;

Figure 14 is the extracted watermark clockwise rotated 30°, NC=0.81, the watermark can be extracted accurately.

Figure 15 is a medical image rotated 40° counterclockwise;

Figure 16 is the extracted watermark rotated 40° counterclockwise, NC=0.79, the watermark can be extracted accurately.

Table 4 Watermark anti-rotation attack experimental data

Degrees of rotation° -40 -20 -10 10 30 45 PSNR(dB) 13.72 14.71 24.51 16.23 14.05 13.7 NC 0.79 0.58 0.79 0.80 0.81 0.81

Note: Negative is counterclockwise, positive is clockwise (2) scaling transformation

Table 5 shows the watermark anti-scaling attack experimental data of medical images. It can be seen from Table 5 that when the scaling factor is as small as 0.5, the correlation coefficient NC=0.63, and the watermark can be extracted.

Fig. 17 is a zoomed medical image (the scaling factor is 1.2);

Figure 18 is the watermark extracted after scaling attack, NC=0.73, the watermark can be extracted accurately.

Table 5 Watermark anti-scaling attack experimental data

scaling factor 0.5 0.8 1.2 1.5 2.0 2.5 NC 0.63 0.62 0.73 0.61 0.56 0.62

(3) Translation transformation

Table 6 is the experimental data of watermark anti-translation transformation. It is known from the table that when the image data moves 35% horizontally or vertically, the NC value is higher than 0.5, and the watermark can be extracted accurately, so the watermark method has a strong ability to resist translation transformation.

Fig. 19 is the image after the medical image is shifted vertically to the left by 30%;

Figure 20 is the watermark extracted after shifting vertically to the left by 30%, the watermark can be extracted accurately, NC=0.90.

Fig. 21 is the image after the medical image is vertically shifted down by 30%;

Figure 22 is the watermark extracted after moving down 30% vertically, the watermark can be extracted accurately, NC=0.81.

Table 6 Experimental data of watermark anti-translation transformation

(4) Cut attack

Table 7 shows the watermark anti-shearing attack experimental data. It can be seen from the table that when the medical image is cut along the coordinate axis X or Y, and the cutting amount is 30%, the NC value is greater than 0.5, and the watermark can still be extracted, indicating that the The watermarking algorithm has a strong ability to resist shearing attacks.

Fig. 23 is a medical image cut by 30% along the X axis;

Figure 24 is the watermark extracted after cutting 30% along the X axis, the watermark can be extracted accurately, NC=0.81.

Fig. 25 is a medical image cut by 15% along the Y axis;

Fig. 26 is the watermark extracted after cutting 15% along the Y axis, the watermark can be extracted accurately, NC=0.81.

Table 7 Watermark anti-shearing attack experimental data

Claims (1)

1. A medical image robust watermarking method based on SIFT-DCT, which is characterized in that: based on SIFT-DCT transformation, the feature vector of the anti-geometric attack of the medical image is obtained, and combined with the watermarking technology, the medical image zero The anti-geometric attack and conventional attack of watermarking, the realization method of medical image digital watermarking is divided into three parts, a total of nine steps: The first part is the feature extraction of medical images: 1) Use SIFT to process medical images to obtain three feature matrices: image (image matrix), descriptions (normalized feature descriptor), locs (key point information); 2) Perform DCT transformation on the describes matrix to obtain F(i,j); 3) Obtain the feature sequence V(i,j) by using the Hash function operation on F(i,j); The second part is the encryption and embedding of the watermark: 4) Generate a chaotic sequence through the Logistic Map, and obtain k(n) after binarization; 5) XOR the chaotic sequence k(n) of the binary watermark W(i,j) to realize the encryption of the watermark, and obtain the encrypted watermark sequence EW(i,j), EW(i,j)=W(i,j) j)⊕k(n); 6) Generate a binary logical key sequence Key(i,j) according to the encrypted watermark sequence EW(i,j) and the extracted medical image feature sequence V(i,j), and then convert the binary logical sequence Key( i,j) There is a third party, Key(i,j)=V(i,j)⊕EW(i,j); The third part is the extraction of the watermark: 7) Find the feature sequence V'(i, j) of the medical image to be tested; 8) Use the binary logic key sequence Key(i,j) existing in the third party and the feature vector V'(i,j) of the medical image to be tested to extract the encrypted watermark EW'(i,j), EW' (i,j)=Key(i,j)⊕V'(i,j); 9) Decrypt the extracted watermark to get W'(i,j), W'(i,j)=k(n)⊕EW'(i,j); 10) Calculate the normalized correlation coefficient of W(i,j) and W'(i,j) to find the NC value and measure the robustness of the algorithm.