A medical image zero watermarking method based on multi-feature points and DNA encryption

Abstract

The application is based on a medical image zero watermarking method based on multi-feature points and DNA encryption, first, according to the characteristics of the medical image, the ROI and RONI of the medical image are accurately separated; then the four-dimensional features of CNN, variance delta, LBP and Gabor in the ROI region are extracted by using the newly proposed high-discrimination feature extraction method, and then the zero watermarking image is constructed, which can fully utilize the characteristics of zero watermarking not truly embedding any information, and ensure that the important area of the ROI of the medical image is not modified, which has high invisibility and can resist conventional attacks, and has good robustness; at the same time, the medical image is Zig-Zag block scrambling and DNA encryption, to prevent patient privacy leakage. It can meet the requirements of confidentiality, integrity and authentication of medical images, and solve the problems of privacy leakage, illegal copying and other problems existing in the transmission of medical image data in remote medical treatment.

Description

Technical Field

[0001] This invention relates to multi-feature extraction, zero-watermark registration and verification, and DNA encryption and decryption of carrier images, specifically to a zero-watermark method for medical images based on multi-feature points and DNA encryption. Background Technology

[0002] While 5G networks offer enhanced capabilities for telemedicine, medical image data also faces significant threats to secure transmission, such as malicious tampering, illegal copying, and privacy breaches. Protecting patient privacy has become extremely important in the medical field, and embedding watermarks in medical images is one effective method for protecting patient privacy. Previously, most watermarks were applied in the plaintext domain, posing a potential risk of exposing medical image information. Furthermore, watermarking methods that modify pixel values ​​are unsuitable for medical images because they alter image integrity. Subsequently, many methods for encrypting medical images have been developed, but not all encryption methods meet robustness requirements.

[0003] Dong et al. proposed a robust zero-watermarking scheme in the DWT-DFT encryption domain, which constructs and extracts watermarks without modifying pixel values. First, the original medical image and the watermark image are encrypted; then, low-frequency coefficients of the DWT-DFT are extracted as feature vectors of the encrypted medical image; finally, zero-watermarking technology is used in the watermark construction and extraction stages to ensure the integrity of the medical image. Using the "db2" wavelet transform as an example, experiments were conducted on the visual quality and robustness of this watermarking scheme. The results show that the algorithm not only has good watermark robustness but also ideal computational speed in the homomorphic encryption domain. Xiao et al. proposed a robust algorithm based on encrypted medical images. First, the medical image is encrypted using a combination of DWT-DCT and Tent mapping; then, features of the encrypted medical image are extracted using DWT-Gabor transform; finally, the watermark is embedded into the encrypted medical image using zero-watermarking technology. Experiments show that the algorithm has good robustness and invisibility, resisting a certain degree of conventional and geometric attacks, especially translation and cropping attacks. Qin et al. proposed a robust zero-watermarking algorithm based on Curvelet-DCT and RSA pseudo-random sequences. First, a DCT transform is performed on the Coarse layer of the Curvelet transform to extract the most energy-concentrated visual feature vectors from medical images. Then, a pseudo-random sequence generated by the RSA algorithm is used to encrypt the watermark, improving patient privacy security. Experimental results show that the algorithm has good robustness against both conventional and geometric attacks.

[0004] However, the above algorithm still has the following problems:

[0005] (1) Most of them are based on various transformations to extract low-frequency features, such as DWT, DCT, DFT and Curvelet. The extracted features have a certain stability, but in special cases, such as when the number of protected images is large or the similarity of the images is high, false alarms may occur.

[0006] (2) Combining only some simple scrambling encryption techniques, such as RSA and Tent, does not provide high security;

[0007] (3) Only the robustness of the watermark is considered, without paying attention to the security of the medical image itself and the special characteristics of the medical image. Summary of the Invention

[0008] The present invention aims to solve the aforementioned technical problems existing in the prior art by providing a zero-watermarking method for medical images with multi-feature point and DNA encryption.

[0009] The technical solution of this invention is: a zero-watermarking method for medical images based on multiple feature points and DNA encryption, characterized by proceeding through the following steps:

[0010] Step 1. Carrier medical image region segmentation

[0011] For an M×N medical image C, global thresholding is used for edge detection. The image is segmented along its internal edge information. The positions of the maximum and minimum values ​​of the horizontal and vertical coordinates are recorded and used as the four vertices of a rectangle. The inside of the rectangle is the ROI, and the outside is the RONI. The main information of the medical image is located in the ROI.

[0012] Step 2. Multi-feature extraction and encryption

[0013] Step 2.1 Extract ROI features based on CNN to form a binary sequence F1 of feature vectors;

[0014] The horizontal, vertical and diagonal features of the ROI are obtained using CNN, and a max pooling layer of size a×a is obtained. The mean of the max pooling layer, mean(max_pool), is calculated according to formula (1). The values ​​in the max pooling layer are compared with the mean. According to formula (2), if it is greater than or equal to the mean, it is 1; otherwise, it is 0. The CNN binary feature sequence F1 is obtained.

[0015]

[0016]

[0017] Step 2.2 Extract ROI features based on variance δ, LBP, and Gabor to form binary sequences F2, F3, and F4 of each feature vector;

[0018] Step 2.2.1 Divide the ROI into two equal layers;

[0019] Step 2.2.2 Calculate the variance-based binary feature sequence F2

[0020] The variance of each layer is calculated according to formula (3,4), where μ k m is the average value of pixels in each layer. ij N represents the pixels of each layer of the image. k For P k Total number of pixels in P k (i) is P k The i-th pixel in the middle, k = 1, 2; then according to formula (5), the variance of the first layer is subtracted from the variance of the second layer to obtain the variance residual; calculate the mean of the residual, compare each residual with the mean of the residual, according to formula (6), if it is greater than or equal to the mean, it is 1, otherwise it is 0, to obtain the binary feature sequence F2 of the variance;

[0021]

[0022]

[0023] Res2 k =δ 1k -δ 2k (5)

[0024]

[0025] Step 2.2.3 Calculate the LBP-based binary feature sequence F3

[0026] Calculate the LBP feature operator according to formula (7,8); then subtract the LBP feature operator of the first layer from the LBP feature operator of the second layer according to formula (9) to obtain the LBP residual; calculate the mean of the LBP residual, compare each residual with the mean of the residual, and according to formula (10), if it is greater than or equal to the mean, it is 1, otherwise it is 0, to obtain the binary feature sequence F3 of LBP;

[0027]

[0028]

[0029] In the formula, X k (i,j) represents m in each layer. ij LBP feature operator, pixel o k and o y These represent the gray values ​​of the nearby pixels and the center pixel, respectively, and the value of the function l(x) is 1 or 0;

[0030] Res3k =X 1k -X 2k (9)

[0031]

[0032] Step 2.2.4 Calculate the Gabor-based binary feature sequence F4

[0033] According to formula (11), the Gabor features on each layer are calculated, where Q k (i,j,θ,w,σ) represents the middle pixel m of each layer. ij The Gabor filter function, i m The imaginary part value is The term w in the formula represents the frequency of the sine wave used in the Gaussian filter; according to formula (12), the Gabor feature of the first layer is subtracted from the Gabor feature of the second layer to obtain the Gabor residual; the mean of the residual is calculated, and each residual is compared with the mean of the residual. According to formula (13), if it is greater than or equal to the mean, it is 1, otherwise it is 0, and the binary feature sequence F4 of Gabor can be obtained.

[0034]

[0035] Res4 k =Q 1k -Q 2k (12)

[0036]

[0037] Step 2.3 Connect the binary sequences F1, F2, F3, and F4 of each feature vector according to formula (14) to form the final ROI feature vector F;

[0038] F = F1||F2||F3||F4 (14)

[0039] In the formula, || indicates that the two eigenvectors are connected end to end;

[0040] Step 2.4 Encrypt feature vector

[0041] Step 2.4.1 Generate a random sequence based on the Sine map, and record the initial value as the sequence key Key1 = (s0, α), where α represents the system control parameter, α ∈ [0, 4], s n Sine is in a chaotic state when α∈(-1,1) and when α∈(3.48,4].

[0042] s n+1 =αsin(πs) n ) / 4 (15)

[0043] Step 2.4.2 Number the random sequence and feature vector F in order, then sort the random sequence in ascending order, and scramble the sequence position of feature vector F according to the sorted random sequence number to obtain the encrypted feature vector F';

[0044] Step 3. Encrypt the watermark

[0045] Step 3.1 Generate a random sequence based on the Logistic map. The length of the sequence is the same as the number of pixels in the watermark image. Record the initial value as the sequence key Key2 = (l0, β), where β represents the system control parameter. When the parameter l n The sequence is chaotic when β∈(0,1) and β∈(3.57,4].

[0046] l n+1 =βl n (1-l n (16)

[0047] Step 3.2: Convert the one-dimensional random sequence generated by the Logistic map into a two-dimensional matrix, and number the two-dimensional matrix and the watermark image in sequence; then reorder the two-dimensional matrix in ascending order, and set the pixel positions of the random watermark image according to the sorted random sequence number.

[0048] Step 3.3 Reduce the scrambled watermark image to one dimension to obtain the encrypted watermark sequence W';

[0049] Step 4. Construct a watermark-free image

[0050] According to formula (17), the encrypted watermark sequence W' and the encrypted feature vector F' are XORed to obtain the zero watermark image ZM;

[0051]

[0052] Step 5. Encryption of DNA in Carrier Medical Images

[0053] Step 5.1 Divide the M×N medical image C into m×n non-overlapping sub-blocks, and use Zig-Zag scrambling between the blocks to obtain a grayscale image C';

[0054] Step 5.2 Represent each pixel in the grayscale image C' using 8 bits to obtain a binary image C′ of size M×N×8. b , the binary image C′ b Divided into groups of 8 bits;

[0055] Step 5.3 Generate three random sequences based on the Chen hyperchaotic sequence according to formula (18). The initial value is recorded as the sequence key Key3 = (a, b, c, x, y, z). When a = 35, b = 3, and c ∈ [20, 28.4], the system is in a chaotic state.

[0056]

[0057] Step 5.4 According to formula (19), the Chen hyperchaotic sequence... The first M×N×8 elements are converted into a binary sequence X={X j}(j=1,2...M×N×8), divide the sequence X into 8-bit groups;

[0058]

[0059] Step 5.5 Based on formula (20), the Chen hyperchaotic sequence is... The values ​​of the first M×N elements are converted to 1-8, resulting in the sequence Y = {Y} k}(k=1,2...M×N);

[0060]

[0061] Step 5.6 According to formula (21), the Chen hyperchaotic sequence is... The values ​​of the first M×N elements are converted to 1-8, resulting in the sequence Z = {Z q}(q=1,2...M×N);

[0062]

[0063] Step 5.7 Based on sequence Y, pair the binary image C′ b DNA encoding was performed, and sequence X was encoded based on sequence Z, resulting in the encoded C′. b The base AGCT and the encoded sequence X base CGAT;

[0064] Step 5.8 Encode C′ according to formula (22) b The bases AGCT and X, which encode the sequence X, are XORed according to the DNA XOR rules to obtain the XOR sequence CACA.

[0065]

[0066] Step 5.9 According to formula (23), add the elements in sequence Y and sequence Z in order and convert them to the range of 1-8 to obtain sequence A;

[0067] A = mod((Y+Z), 8) + 1 (23)

[0068] Step 5.10: Encode the XOR sequence CACA based on sequence A into DNA, convert the resulting 8-bit binary number into decimal to replace the pixels of the original medical image, and obtain the encrypted medical image C°.

[0069] The extraction of the original medical image and watermark is performed in the following steps:

[0070] Step 6: Extract the raw medical image

[0071] Step 6.1 Convert the pixels of the M×N encrypted medical image C° from decimal to binary. Each pixel is represented by 8 bits of binary code, resulting in a encrypted binary medical image. The size is M×N×8, with each group consisting of 8 bits.

[0072] Step 6.2 Based on the sequence key Key3=(a,b,c,x,y,z), generate a random sequence based on Chen map according to formula (18).

[0073] Step 6.3 According to formula (19), the Chen hyperchaotic sequence is... The first M×N×8 elements are converted into a binary sequence X, and the sequence X is divided into groups of 8 bits each.

[0074] Step 6.4 According to formula (20), the Chen hyperchaotic sequence... The first M×N elements are converted to 1-8 to obtain sequence Y;

[0075] Step 6.5 According to formula (21), the Chen hyperchaotic sequence... The first M×N elements are transformed into 1-8 to obtain sequence Z;

[0076] Step 6.6 According to formula (23), add the elements in sequence Y and sequence Z in order to convert them to 1-8, and obtain sequence A;

[0077] Step 6.7 Based on sequence A pair DNA encoding; DNA encoding of sequence X based on sequence Z, respectively obtaining the encoded sequences. The base CACA and the encoded sequence X base CGAT;

[0078] Step 6.8 will encode the data. The bases CACA and X, which encode the sequence X, are XORed according to the DNA XOR rules to obtain the XOR sequence AGCT.

[0079]

[0080] Step 6.9 Based on sequence Y, the XOR sequence AGCT is encoded into DNA. The resulting 8-bit binary is converted into decimal and used to replace the pixels of the encrypted medical image, resulting in the encrypted medical image C'.

[0081] Step 6.10 Divide the M×N medical carrier image C' into m×n non-overlapping blocks, and use Zig-Zag inverse scrambling between the blocks to obtain the original medical image C;

[0082] Step 7. Extract the watermark

[0083] Step 7.1 For the obtained M×N medical image C, perform edge detection using a global threshold, segment along the internal edge information of the medical image, and record the four vertices where the maximum and minimum values ​​of the horizontal and vertical coordinates are located, which are used as the four vertices of a rectangle. The inside of the rectangle is the ROI, and the outside is the RONI. The main information of the medical image is in the ROI.

[0084] Step 7.2 Multi-feature extraction and encryption

[0085] Following steps 2.1-2.3, the final ROI feature vector F is obtained;

[0086] Step 7.3 Encrypt feature vector

[0087] Step 7.3.1 Based on the sequence key Key1=(s0,α), generate a random sequence based on the Sine map according to formula (15);

[0088] Step 7.3.2 Number the random sequence and feature vector F in order, then sort the random sequence in ascending order, and scramble the position of the feature sequence according to the random sequence number after sorting to obtain the encrypted feature vector F';

[0089] Step 8. Verify zero watermark

[0090] According to formula (25), the zero-watermark image ZM is XORed with the encrypted feature vector F' to obtain the encrypted watermark image;

[0091]

[0092] Step 9. Decrypting the watermark image

[0093] Step 9.1 Using the sequence key Key2=(l0,β), generate a random sequence based on the Logistic map according to formula (16);

[0094] Step 9.2 Convert the random sequence into a two-dimensional matrix, number the two-dimensional matrix and the watermark image W' in sequence; then reorder the two-dimensional matrix in ascending order, and scramble the ciphertext watermark W' according to the numbering of the sorted matrix to obtain the watermark W.

[0095] This invention first precisely separates the Region of Interest (ROI) and Ronier Area (RONI) of medical images based on their characteristics. Then, using a proposed high-discrimination feature extraction method, it extracts four-dimensional features (CNN, variance δ, LBP, and Gabor) from the ROI region to construct a zero-watermark image. This fully leverages the characteristic of zero-watermarking—that it doesn't actually embed any information—ensuring that important regions of the medical image's ROI remain unmodified. This provides high invisibility and resistance to conventional attacks, exhibiting good robustness. Simultaneously, the medical image undergoes Zig-Zag block scrambling and DNA encryption to prevent patient privacy leaks. This method meets the requirements for confidentiality, integrity, and authentication of medical images, solving problems such as privacy leaks and unauthorized copying in remote medical image data transmission. Attached Figure Description

[0096] Figure 1 This is a schematic diagram illustrating the construction process of multi-feature zero watermark and DNA encryption in an embodiment of the present invention.

[0097] Figure 2 This is a schematic diagram of the decryption process of multi-feature zero watermark and DNA encryption in an embodiment of the present invention.

[0098] Figure 3 This is a carrier image according to an embodiment of the present invention.

[0099] Figure 4 This is a grayscale brain medical image global threshold detection image according to an embodiment of the present invention.

[0100] Figure 5 This is the ROI after grayscale brain medical image segmentation in an embodiment of the present invention.

[0101] Figure 6 This is a schematic diagram of grayscale brain medical image layering according to an embodiment of the present invention.

[0102] Figure 7 This is a grayscale brain medical encrypted image according to an embodiment of the present invention.

[0103] Figure 8 This is a grayscale brain medical decryption image according to an embodiment of the present invention.

[0104] Figure 9 This is a watermark image according to an embodiment of the present invention.

[0105] Figure 10 This is an encrypted watermark image according to an embodiment of the present invention.

[0106] Figure 11 This is a watermark decryption image according to an embodiment of the present invention.

[0107] Figure 12 This is a schematic diagram of Zig-Zag encryption according to an embodiment of the present invention.

[0108] Figure 13 This is a schematic diagram of DNA encryption according to an embodiment of the present invention.

[0109] Figure 14 This is a schematic diagram of DNA decryption according to an embodiment of the present invention. Specific implementation methods

[0110] This invention provides a zero-watermarking method for medical images based on multiple feature points and DNA encryption. Figure 1 As shown, proceed with the following steps in sequence:

[0111] Step 1. Carrier medical image region segmentation

[0112] For example Figure 3 The M×N (256×256) grayscale medical image C shown is used for edge detection using a global threshold (set to 80), as follows: Figure 4 As shown, the medical image is segmented along its internal edge information. The ginput function is used to determine the two diagonal points of the pathological region in the medical image. The locations of the maximum and minimum values ​​of the horizontal and vertical coordinates are recorded and used as the four vertices of a rectangle. The interior of the rectangle is the Region of Interest (ROI), and the exterior is the Region of Interest (RONI). The main information of the medical image is located in the ROI, as shown below. Figure 5 As shown;

[0113] Step 2. Multi-feature extraction and encryption

[0114] Step 2.1 Calculate the F1 score of the CNN-based binary feature sequence.

[0115] The horizontal, vertical and diagonal features of the ROI are obtained using CNN, and a max pooling layer of size a×a is obtained. The mean of the max pooling layer, mean(max_pool), is calculated according to formula (1). The values ​​in the max pooling layer are compared with the mean. According to formula (2), if it is greater than or equal to the mean, it is 1; otherwise, it is 0. The CNN binary feature sequence F1 is obtained.

[0116]

[0117]

[0118] Step 2.2 Extract ROI features based on variance δ, LBP, and Gabor to form binary sequences F2, F3, and F4 of each feature vector;

[0119] Step 2.2.1 Divide the ROI into two equal layers;

[0120] Step 2.2.2 Calculate the variance-based binary feature sequence F2

[0121] The variance of each layer is calculated according to formula (3,4), where μ km is the average value of pixels in each layer. ij N represents the pixels of each layer of the image. k For P k Total number of pixels in P k (i) is P k The i-th pixel in the middle, k = 1, 2; then according to formula (5), the variance of the first layer is subtracted from the variance of the second layer to obtain the variance residual; calculate the mean of the residual, compare each residual with the mean of the residual, according to formula (6), if it is greater than or equal to the mean, it is 1, otherwise it is 0, to obtain the binary feature sequence F2 of the variance;

[0122]

[0123]

[0124] Res2 k =δ 1k -δ 2k (5)

[0125]

[0126] Step 2.2.3 Calculate the LBP-based binary feature sequence F3

[0127] Calculate the LBP feature operator according to formula (7,8); then subtract the LBP feature operator of the first layer from the LBP feature operator of the second layer according to formula (9) to obtain the LBP residual; calculate the mean of the LBP residual, compare each residual with the mean of the residual, and according to formula (10), if it is greater than or equal to the mean, it is 1, otherwise it is 0, to obtain the binary feature sequence F3 of LBP;

[0128]

[0129]

[0130] In the formula, X k (i,j) represents m in each layer. ij LBP feature operator, pixel o k and o y These represent the gray values ​​of the nearby pixels and the center pixel, respectively, and the value of the function l(x) is 1 or 0;

[0131] Res3 k =X 1k -X 2k (9)

[0132]

[0133] Step 2.2.4 Calculate the Gabor-based binary feature sequence F4

[0134] According to formula (11), the Gabor features on each layer are calculated, where Q k (i,j,θ,w,σ) represents the middle pixel m of each layer. ij The Gabor filter function, i m The imaginary part value is The term w in the formula represents the frequency of the sine wave used in the Gaussian filter; according to formula (12), the Gabor feature of the first layer is subtracted from the Gabor feature of the second layer to obtain the Gabor residual; the mean of the residual is calculated, and each residual is compared with the mean of the residual. According to formula (13), if it is greater than or equal to the mean, it is 1, otherwise it is 0, and the binary feature sequence F4 of Gabor can be obtained.

[0135]

[0136] Res4 k =Q 1k -Q 2k (12)

[0137]

[0138] Step 2.3 Connect the binary sequences F1, F2, F3, and F4 of each feature vector according to formula (14) to form the final ROI feature vector F;

[0139] F = F1||F2||F3||F4 (14)

[0140] In the formula, || indicates that the two eigenvectors are connected end to end;

[0141] Step 2.4 Encrypt feature vector

[0142] Step 2.4.1 Generate a random sequence based on the Sine map, and record the initial value as the sequence key Key1 = (s0, α), where α represents the system control parameter, α ∈ [0, 4], s n Sine is in a chaotic state when α∈(-1,1) and when α∈(3.48,4].

[0143] s n+1 =αsin(πs) n ) / 4 (15)

[0144] For example, when Key1 = (s0, α) = (0.99, 4), a random sequence of length 8 is generated:

[0145] -0.14 0.45 -0.36 -0.71 0.36 0.68 0.12 -0.58

[0146] Step 2.4.2 Number the random sequence and feature vector F in order, then sort the random sequence in ascending order, and scramble the sequence position of feature vector F according to the sorted random sequence number to obtain the encrypted feature vector F';

[0147] Table 1 and Table 2 show the original and scrambled sequences, respectively.

[0148] Table 1 Examples of Original Sine Random Sequences and Eigenvectors

[0149] serial number 1 2 3 4 5 6 7 8 Original random sequence -0.14 0.45 -0.36 -0.71 0.36 0.68 0.12 -0.58 Feature sequence F 1 1 0 0 1 0 0 1

[0150] Table 2 Random Sequences and Eigenvectors After Scrambling

[0151] Sorted random sequence -0.71 -0.58 -0.36 -0.14 0.12 0.36 0.45 0.68 serial number 4 8 3 1 7 5 2 6 Scrambled feature sequence F' 0 1 0 1 0 1 1 0

[0152] Step 3. Encrypt the watermark

[0153] Step 3.1 Generate a random sequence based on the Logistic map. The length of the sequence is the same as the number of pixels in the watermark image. Record the initial value as the sequence key Key2 = (l0, β), where β represents the system control parameter. When the parameter l n The sequence is chaotic when β∈(0,1) and β∈(3.57,4].

[0154] l n+1 =βl n (1-l n (16)

[0155] For example, when Key2 = (l0, β) = (0.15, 3.7), a random sequence of length 9 is generated: 0.56 0.42 0.31 0.06 0.79 0.29 0.88 0.25 0.63

[0157] Step 3.2: Convert the one-dimensional random sequence generated by the Logistic map into a two-dimensional matrix, and number the two-dimensional matrix and the watermark image in sequence; then reorder the two-dimensional matrix in ascending order, and set the pixel positions of the random watermark image according to the sorted random sequence number.

[0158] For example, the above one-dimensional random sequence can be transformed into a two-dimensional matrix as follows: Assuming the watermark image is Its sequential number is Reorder the two-dimensional matrix in ascending order. Update the index position according to the sorted matrix. The watermark image is scrambled based on the new number position.

[0159] Step 3.3 Reduce the scrambled watermark image to one dimension to obtain the encrypted watermark sequence W'; Figure 9 Figure 10 These are the watermark image and the scrambled watermark image of the present invention, respectively.

[0160] Step 4. Construct a watermark-free image

[0161] According to formula (17), the encrypted watermark sequence W' and the encrypted feature vector F' are XORed to obtain the zero watermark image ZM;

[0162]

[0163] Step 5. Encryption of DNA in Carrier Medical Images

[0164] Step 5.1 Divide the M×N medical image C into m×n non-overlapping sub-blocks, and scramble the blocks using Zig-Zag (e.g., ...). Figure 12 As shown, the grayscale image C' is obtained;

[0165] Step 5.2 Represent each pixel in the grayscale image C' using 8 bits to obtain a binary image C of size M×N×8 (256×256×8). b ', will the binary image C b Divide into groups of 8 bits, and assume that one of the 8 bits in a group is 1000 11 01;

[0166] Step 5.3 Generate three random sequences based on the Chen hyperchaotic sequence according to formula (18). The initial value is recorded as the sequence key Key3 = (a, b, c, x, y, z). When a = 35, b = 3, and c ∈ [20, 28.4], the system is in a chaotic state.

[0167]

[0168] Step 5.4 According to formula (19), the Chen hyperchaotic sequence... The first M×N×8 elements are converted into a binary sequence X={X j}(j=1,2...M×N×8), divide the sequence X into 8-bit groups;

[0169]

[0170] Step 5.5 Based on formula (20), the Chen hyperchaotic sequence is... The values ​​of the first M×N elements are converted to 1-8, resulting in the sequence Y = {Y} k}(k=1,2...M×N);

[0171]

[0172] Step 5.6 According to formula (21), the Chen hyperchaotic sequence is... The values ​​of the first M×N elements are converted to 1-8, resulting in the sequence Z = {Z q}(q=1,2...M×N);

[0173]

[0174] Step 5.7 Based on sequence Y, pair the binary image C′ b DNA encoding was performed, and sequence X was encoded based on sequence Z, resulting in the encoded C′. b The base AGCT and the encoded sequence X base CGAT;

[0175] like Figure 13 As shown: For example, based on the k-th element in sequence Y, 6 pairs of binary images C′ b A set of 10 00 1101 is encoded using DNA. Each two bits in 10 00 11 01 can be encoded using DNA encoding rules. 10, using the DNA encoding rules shown in Table 3, yields base A, and so on. The remaining 6 bits yield base GCT, which is the binary medical image C′. b The bases of the corresponding group. Similarly, using element 2 of sequence Z to encode a group of 01 10 00 11 for sequence X, we can obtain CGAT, which are the bases of the corresponding group for sequence X;

[0176] Table 3 DNA Encoding Rules

[0177]

[0178] Step 5.8 Encode C′ according to formula (22) b The bases AGCT and X-base CGAT of the encoded sequence are XORed according to the DNA XOR rules shown in Table 4 to obtain the XOR sequence CACA.

[0179]

[0180] Table 4 DNA XOR Rules

[0181]

[0182] Step 5.9 According to formula (23), add the elements in sequence Y and sequence Z in order and convert them to the range of 1-8 to obtain sequence A;

[0183] A = mod((Y+Z), 8) + 1 (23)

[0184] Step 5.10: Encode the XOR sequence CACA based on sequence A into DNA, convert the resulting 8-bit binary number into decimal to replace the pixels of the original medical image, and obtain the encrypted medical image C°; transmit the encrypted medical image over the network and send it to the recipient.

[0185] For example, using element 1 of sequence A as the XOR sequence CACA for DNA encoding, one base C, according to encoding rule 1, yields 2 bits of binary 10. This process continues until the remaining 6 bits are 00 10 00. Converting each 8-bit binary number to decimal, 10 00 10 00 is converted to 136. The encoded value is then replaced with the original medical image pixels; this process is repeated until all pixels are processed, resulting in the encrypted medical image C°, as shown below. Figure 7 As shown; using a carrier image ( Figure 3 For example, Table 5 shows the diagonal correlation coefficient and information entropy before and after encryption. As can be seen from Table 5, the diagonal correlation coefficient of the present invention before encryption is slightly higher than that in reference [3], but the correlation coefficient after encryption is slightly lower than that in reference [3]. This indicates that the DNA encryption algorithm of the present invention has effectively destroyed the correlation of the original pixels and improved security. In addition, the information entropy before and after encryption in reference [3] is 7.414 and 7.456, respectively, which is basically unchanged. However, the information entropy before encryption in the present invention is only 5.968, while the information entropy after encryption is as high as 7.521. This again shows that the DNA encryption algorithm of the present invention has high security.

[0186] Table 5

[0187]

[0188] This invention extracts original medical images and watermarks, such as... Figure 2 The steps shown are as follows:

[0189] Step 6: Extract the raw medical image

[0190] Step 6.1 Convert the pixels of the M×N encrypted medical image C° from decimal to binary. Each pixel is represented by 8 bits of binary code, resulting in a encrypted binary medical image. The size is M×N×8, with each group consisting of 8 bits; for example Figure 14 As shown, assume that one of the 8-bit binary values ​​is 10 00 10 00;

[0191] Step 6.2 Based on the sequence key Key3=(a,b,c,x,y,z), generate a random sequence based on Chen map according to formula (18).

[0192] Step 6.3 According to formula (19), the Chen hyperchaotic sequence is... The first M×N×8 elements are converted into a binary sequence X, and the sequence X is divided into groups of 8 bits each.

[0193] Step 6.4 According to formula (20), the Chen hyperchaotic sequence... The first M×N elements are converted to 1-8 to obtain sequence Y;

[0194] Step 6.5 According to formula (21), the Chen hyperchaotic sequence... The first M×N elements are transformed into 1-8 to obtain sequence Z;

[0195] Step 6.6 According to formula (23), add the elements in sequence Y and sequence Z in order to convert them to 1-8, and obtain sequence A;

[0196] Step 6.7 Based on sequence A pair DNA encoding; DNA encoding of sequence X based on sequence Z, respectively obtaining the encoded sequences. The base CACA and the encoded sequence X base CGAT;

[0197] Suppose that each 2 bits in 10 00 10 00 is encoded using encoding rule 1. 10 is encoded using encoding rule 1 to obtain the base C, and so on, the remaining 6 bits are encoded to obtain the base ACA, for a total of four bases. Similarly, an element 2 in sequence Z can be encoded into a group of 01 10 00 11 in sequence X to obtain the four bases CGAT.

[0198] Step 6.8 will encode the data. The bases CACA and X, which encode the sequence X, are XORed according to the DNA XOR rules to obtain the XOR sequence AGCT.

[0199]

[0200] Step 6.9 Based on sequence Y, the XOR sequence AGCT is encoded into DNA. The resulting 8-bit binary is converted into decimal and used to replace the pixels of the encrypted medical image, resulting in the encrypted medical image C'.

[0201] For example, in sequence Y, element 6 is XORed with the bases AGCT. One base generates a 2-bit binary number, and four bases generate an 8-bit binary number. Using DNA encoding rules, 6 pairs of AGCT are encoded as 10 00 11 01. Each 8-bit binary number is converted to a decimal number; 10 00 11 01 is 141 in decimal. The encoded value is then replaced with the original medical image pixels, and this process is repeated until all pixels are processed, resulting in image C'.

[0202] Step 6.10 divides the M×N medical carrier image C' into m×n non-overlapping blocks, and performs Zig-Zag inverse scrambling on the blocks to obtain the original medical image C, as shown below. Figure 8 As shown;

[0203] Step 7. Extract the watermark

[0204] Step 7.1 As in step 1, for the obtained M×N medical image C, use global thresholding to perform edge detection, segment along the internal edge information of the medical image, and record the four vertices where the maximum and minimum values ​​of the horizontal and vertical coordinates are located, as the four vertices of a rectangle. The inside of the rectangle is the ROI, and the outside is the RONI. The main information of the medical image is in the ROI.

[0205] Step 7.2 Multi-feature extraction and encryption (zero-watermark receiver)

[0206] Following steps 2.1-2.3, the final ROI feature vector F is obtained;

[0207] Step 7.3 Encrypt feature vector

[0208] Step 7.3.1 Based on the sequence key Key1=(s0,α), generate a random sequence based on the Sine map according to formula (15);

[0209] For example, when Key1 = (s0, α) = (0.99, 4), a random sequence of length 8 is generated:

[0210] -0.14 0.45 -0.36 -0.71 0.36 0.68 0.12 -0.58

[0211] Step 7.3.2 Number the random sequence and feature vector F in order, then sort the random sequence in ascending order, and scramble the feature sequence position according to the random sequence encoding after sorting to obtain the encrypted medical feature vector F', as shown in Table 1 and Table 2.

[0212] Step 8. Verify zero watermark

[0213] According to formula (25), the zero-watermark image ZM is XORed with the encrypted feature vector F' to obtain the encrypted watermark image;

[0214]

[0215] Step 9. Decrypting the watermark image

[0216] Step 9.1 Using the sequence key Key2=(l0,β), generate a random sequence based on the Logistic map according to formula (16);

[0217] For example, when Key2 = (l0, β) = (0.15, 3.7), a random sequence of length 9 is generated: 0.56 0.12 0.31 0.06 0.79 0.42 0.88 0.25 0.63

[0219] Step 9.2 transforms the random sequence into a two-dimensional matrix, and numbers the two-dimensional matrix and the watermark image W' sequentially; then, reorders the two-dimensional matrix in ascending order, and reverses the scrambled ciphertext watermark W' based on the sorted matrix numbers to obtain the watermark W, as shown below. Figure 11 As shown.

[0220] For example, the above one-dimensional random sequence can be transformed into a two-dimensional matrix as follows: Assuming the watermark image is messed up Its sequential number is Reorder the two-dimensional matrix in ascending order. Update the index position according to the sorted matrix. The watermark image is obtained by reversing the position scrambling of the watermark image according to the new number position.

[0221] The robustness of this invention is verified through attacks with different parameters. This invention can be evaluated from the perspective of NC (Normalized Correlation).

[0222]

[0223] Where W(i,j) represents the pixel value of the original watermark image, and W'(i,j) represents the pixel value of the extracted watermark. NC mainly evaluates the similarity between the original watermark and the extracted watermark. The coefficient of NC ranges from 0 to 1. The closer the NC coefficient is to 1, the higher the accuracy of the extracted watermark and the smaller the difference between it and the original image. Figure 3 For carrier image, Figure 9 For the watermarked image, Table 6 shows the test results of the ability of conventional attacks. As can be seen from Table 6, the NC value of the present invention under the median filtering attack is 0.9874, which is higher than that of references [1,2]; under the salt and pepper noise attack, the NC value of the present invention is 0.9502, which is higher than that of reference [2] and slightly lower than that of reference [1]; under the rotation attack, the NC value of the present invention is as high as 0.9482, which is much higher than that of references [1,2].

[0224] Table 6 Comparison of NC values ​​under different attacks

[0225]

[0226]

[0227] References:

[0228] [1]XIA ZQ,WANG XY,LI XX,et.al.Efficient copyright protection for three CT images based on quaternion polar harmonic Fourier moments[J].SignalProcessing,2019,164:368-379.

[0229] [2]XIA ZQ,WANG XY,LI XX,et.al.Efficient copyright protection for three CT images based on quaternion polar harmonic Fourier moments[J].SignalProcessing,2019,164:368-379.

[0230] [3] Ren Hua, Niu Shaozhang, Ren Ruyong, Yue Zhen. Research on meaningful image encryption algorithm based on two-dimensional compressed sensing [J / OL]. Journal of Communications:

[0231] 1-12[2022-05-09].http: / / kns.cnki.net / kcms / detail / 11.2102.TN.20220429.0959.002.html.

Claims

1. A medical image zero-watermarking method based on multi-feature points and DNA encryption, characterized by The following steps are performed in sequence: Step 1. Region division of carrier medical image Edge detection is performed on the MxN medical image C using a global threshold, segmentation is performed along the internal edge information of the medical image, and the positions of the maximum and minimum values of the horizontal and vertical coordinates are recorded as the four vertices of a rectangle. The ROI is inside the rectangle, and the RONI is outside the rectangle. The main information of the medical image is located in the ROI; Step 2. Multi-feature extraction and encryption Step 2.1 Extract ROI features based on CNN to form a binary sequence F1 of feature vectors; Use CNN to obtain the horizontal, vertical and diagonal features of ROI to obtain a max_pool of a x a size. Calculate the mean of the max_pool according to formula (1). Compare the values in the max_pool with the mean. According to formula (2), if it is greater than or equal to the mean, it is 1, otherwise it is 0. Obtain the CNN binary feature sequence F1; Step 2.2 Extract ROI features based on variance δ, LBP and Gabor to form binary sequences F2, F3 and F4 of feature vectors; Step 2.2.1 Divide the ROI into 2 layers equally; Step 2.2.2 Calculate the binary feature sequence F2 based on variance The variance of each layer is calculated according to formula (3, 4), wherein μ k is the average value of each layer of pixels, m ij is the average value of each layer of pixels, m k is the total number of pixels in P k , P k is the i-th pixel in P k , k = 1, 2; the variance of the first layer is subtracted from the variance of the second layer according to formula (5), to obtain the residual of the variance; the average value of the residual is calculated, each residual is compared with the average value of the residual, and according to formula (6), if greater than or equal to the average value, 1 is obtained, otherwise 0 is obtained, to obtain the binary feature sequence F2 of the variance; Res2 k = δ 1k - δ 2k (5) Step 2.2.3 Calculate the binary feature sequence F3 based on LBP Calculate the LBP feature operator according to formulas (7, 8). Then subtract the LBP feature operator of the first layer from the LBP feature operator of the second layer to obtain the LBP residual according to formula (9). Calculate the mean of the LBP residual. Compare each residual with the mean of the residual. According to formula (10), if it is greater than or equal to the mean, it is 1, otherwise it is 0. Obtain the binary feature sequence F3 of LBP; In the formula, X k (i,j) represents m in each layer. ij LBP feature operator, pixel o k and o y These represent the gray values ​​of the nearby pixels and the center pixel, respectively, and the value of the function l(x) is 1 or 0; Res3 k = X 1k - X 2k (9) Step 2.2.4 Calculate the binary feature sequence F4 based on Gabor According to formula (11), the Gabor features on each layer are calculated, wherein Q k (i,j,θ,w,σ) represents the Gabor filter function of each sub-layer central pixel m ij m The term w in the formula represents the frequency of the sine wave used in the Gaussian filter; the Gabor residual is obtained by subtracting the Gabor feature of the second layer from the Gabor feature of the first layer according to formula (12); the mean value of the residual is calculated, each residual is compared with the mean value of the residual, and according to formula (13), if greater than or equal to the mean value, it is 1, otherwise it is 0, and the binary feature sequence F4 of the Gabor can be obtained. The term w in the formula represents the frequency of the sine wave used in the Gaussian filter; the Gabor residual is obtained by subtracting the Gabor feature of the second layer from the Gabor feature of the first layer according to formula (12); the mean value of the residual is calculated, each residual is compared with the mean value of the residual, and according to formula (13), if greater than or equal to the mean value, it is 1, otherwise it is 0, and the binary feature sequence F4 of the Gabor can be obtained.​ Res4 k = Q 1k - Q 2k (12) Step 2.3 Concatenate the binary sequences F1, F2, F3 and F4 of the feature vectors in front and back to form the final ROI feature vector F according to formula (14); F = F1||F2||F3||F4 (14) Where || represents the concatenation of the first and last feature vectors; Step 2.4 Encrypt the feature vector Step 2.4.1 generates a random sequence based on Sine map, and records the initial value as the sequence key Key1=(s0, a), where a represents a system control parameter, a∈[0, 4], s n ∈(-1, 1), and when a∈(3.48, 4], Sine is in a chaotic state; s n+1 = α sin(πs n ) / 4 (15) Step 2.4.2 Number the random sequence and the feature vector F in order, then sort the random sequence in ascending order, and rearrange the feature vector F sequence according to the sorted random sequence number to obtain the encrypted feature vector F'; Step 3. Encrypt the watermark Step 3.1 Generate a random sequence based on the Logistic map, the length of which is the same as the pixel number of the watermark image, and record the initial value as the sequence secret key Key2 = (lo, β). β represents the system control parameter. When the parameter lo n ∈(0, 1), β ∈(3.57, 4], the sequence is chaotic. l n+1 ＝βl n (1-l n ) (16) Step 3.2 Convert the one-dimensional random sequence generated by the Logistic map to a two-dimensional matrix, number the two-dimensional matrix and the watermark image in order, then reorder the two-dimensional matrix in ascending order, and rearrange the pixel position of the watermark image according to the sorted random sequence number; Step 3.3 Reduce the dimension of the rearranged watermark image to one dimension to obtain the encrypted watermark sequence W'; Step 4. Construct a zero watermark image XOR the encrypted watermark sequence W' and the encrypted feature vector F' according to formula (17) to obtain the zero watermark image ZM; Step 5. DNA encryption of carrier medical image Step 5.1 Divide the MxN medical image C into m x n non-overlapping blocks, and use Zig-Zag scrambling between the blocks to obtain a grayscale image C'; Step 5.2 Each pixel in the gray-scale image C' is represented by 8 bits to obtain a binary image C of size M x N x 8 b The binary image C is divided into 8-bit groups b ; Step 5.

3. Generating 3 random sequences based on Chen hyperchaotic sequence according to formula (18) The initial value is recorded as the sequence key Key3 = (a, b, c, x, y, z), when a = 35, b = 3, c ∈ [20, 28.4], the system is in a chaotic state; Step 5.4 Transforming the Chen hyperchaotic sequence X according to formula (19) The first MxNx8 elements are transformed into a binary sequence X = {X j (j = 1, 2... MxNx8), and the sequence X is divided into 8-bit groups; Step 5.5 Transform the Chen hyperchaotic sequence X(k) (k = 1, 2,..., M x N) according to formula (20) to obtain the sequence Y = {Y(k)} (k = 1, 2,..., M x N) The values of the first M x N elements are transformed to 1-8 to obtain the sequence Y = {Y(k)} (k = 1, 2,..., M x N) k}(k = 1, 2,..., M x N) Step 5.6 Transform the values of the first MxN elements to 1-8, obtaining the sequence Z = {Z The values of the first MxN elements are transformed to 1-8, obtaining the sequence Z = {Z q}(q = 1, 2... MxN); Step 5.7 Binary image C' based on sequence Y b DNA encoding, DNA encoding of sequence X based on sequence Z, respectively, to obtain the encoded C' b Bases AGCT and encoded sequence X bases CGAT; Step 5.8 XOR the encoded C' with the base AGCT according to formula (22) to get the encoded sequence CACA. b The bases AGCT and the encoded sequence X bases CGAT are XORed according to the DNA XOR rule to get the XOR sequence CACA. Step 5.9 Add the elements in sequence Y and sequence Z in order and convert to 1-8 according to formula (23) to obtain sequence A; A = mod((Y + Z), 8) + 1 (23) Step 5.10 DNA encode the XOR sequence CACA based on sequence A, convert the obtained 8-bit binary to decimal to replace the original medical image pixel to obtain the ciphertext medical image C°. 2.The medical image zero-watermarking method based on multi-feature points and DNA encryption of claim 1, wherein The extraction of the original medical image and the watermark is carried out in turn according to the following steps: Step 6 Extract the original medical image Step 6.1 Convert the pixels of the ciphertext medical image C° of size M x N from decimal to binary medical image C b °, then each pixel is represented by 8 bits of binary, the ciphertext binary medical image C b ° has a size of M x N x 8, with a group of 8 bits. Step 6.2 Generate a random sequence based on Chen map according to formula (18) based on the sequence key Key3 = (a, b, c, x, y, z) Step 6.3 Transform the Chen hyperchaotic sequence x(t) into a binary sequence X according to formula (19) The elements of the previous MxNx8 length are transformed into a binary sequence X, and the sequence X is divided into groups of 8 bits each. Step 6.4 Take the Chen hyperchaotic sequence according to formula (20) The former MxN length of element is converted to 1-8, and the sequence Y is obtained. Step 6.5 Transforming the Chen hyperchaotic sequence The first MxN elements are transformed to 1-8 to obtain the sequence Z; Step 6.6 Add the elements in sequence Y and sequence Z in order and convert to 1-8 according to formula (23) to obtain sequence A; Step 6.7 based on sequence A on DNA encoding; DNA encoding of sequence X based on sequence Z, respectively, to obtain the encoded sequence X base CACA and the encoded sequence X base CGAT; Step 6.8 XOR the encoded sequence X bases AGCT with the sequence X bases CACA to obtain the sequence X bases AGCT. The bases CACA and the encoded sequence X bases CGAT are XORed according to the DNA XOR rule to obtain the XOR sequence X bases AGCT. Step 6.9 DNA encode the XOR sequence AGCT based on sequence Y, convert the obtained 8-bit binary to decimal to replace the ciphertext medical image pixel to obtain the ciphertext medical image C'; Step 6.10 Divide the MxN medical carrier image C' into m x n non-overlapping blocks, and use Zig-Zag inverse permutation between blocks to obtain the original medical image C; Step 7. Extract watermark Step 7.1 For the obtained MxN medical image C, use global threshold to perform edge detection, segment along the internal edge information of the medical image, record the four vertices of the maximum value and minimum value of the horizontal and vertical coordinates as the four vertices of the rectangle, the inside of the rectangle is ROI, and the outside is RONI, and the main information of the medical image is in ROI; Step 7.2 Multi-feature extraction and encryption According to steps 2.1-2.3, the final ROI feature vector F is obtained; Step 7.3 Encrypt the feature vector Step 7.3.1 Based on sequence secret key Key1 = (s0, a), generate a random sequence based on Sine map according to formula (15); Step 7.3.2 Number the random sequence and the feature vector F in order, then sort the random sequence in ascending order, and scramble the feature sequence position according to the sorted random sequence number to obtain the encrypted feature vector F'; Step 8. Verify zero watermark According to formula (25), XOR the zero watermark image ZM and the encrypted feature vector F' to obtain the encrypted watermark image; Step 9. Watermark image decryption Step 9.1 Use sequence secret key Key2 = (l0, β), generate a random sequence based on Logistic map according to formula (16); Step 9.2 Convert the random sequence to a two-dimensional matrix, number the two-dimensional matrix and the watermark image W' in order; then reorder the two-dimensional matrix in ascending order, and inverse scramble the ciphertext watermark W' to obtain the watermark W according to the sorted matrix number.

Images