A robust watermarking method for medical images based on agast-latch

Abstract

The application relates to the technical field of image processing, and discloses a medical image robust watermarking method based on AGAST-LATCH, which comprises the following steps: AGAST feature points of a medical image are extracted by using an AGAST algorithm; the AGAST feature points are described by using a LATCH algorithm to generate a binary feature descriptor matrix; the binary feature descriptor matrix is subjected to DCT transformation, and a binary feature vector is obtained by combining a hash function; and the binary feature vector and an encrypted scrambling watermark are subjected to bit-by-bit XOR operation to embed the watermark information into the medical image. The above method can embed the watermark without modifying the medical image, avoids the digital watermark from being found and destroyed by malicious attackers, guarantees the quality and integrity of the medical image content, makes the medical image have strong robustness and invisibility, and can simultaneously protect the privacy information of patients and the data security of the medical image.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a robust watermarking method for medical images based on AGAST-LATCH. Background Technology

[0002] With the rapid development of information technology and the widespread application of the internet, traditional healthcare is gradually shifting towards telemedicine. Telemedicine allows patients to receive medical services at different times and locations, while simultaneously reducing healthcare costs and improving efficiency. The transmission and sharing of medical images has become an indispensable part of telemedicine, but it also brings a series of security issues. The sharing and transmission of medical images involves patient privacy and the security of medical data. Once leaked or tampered with, it could lead to significant privacy breaches and medical risks. For example, sensitive information such as patients' personal identification information, medical history, and disease diagnoses may be leaked, resulting in violations of patients' privacy rights; medical image data may be tampered with, leading to misdiagnosis and incorrect treatment plans.

[0003] Currently, to address the aforementioned security issues, medical images are processed, for example, using digital watermarking technology. However, research on digital watermarking algorithms for medical images is limited, particularly regarding zero-watermarking algorithms for medical data resistant to geometric attacks. Traditional digital watermarking techniques are prone to modifying the content of medical images, compromising the integrity of medical data. Therefore, developing more effective digital watermarking algorithms to ensure the quality and integrity of medical image content and improve the security of medical images is a pressing technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a robust watermarking method for medical images based on AGAST-LATCH, which can embed watermarks without modifying medical images, ensuring the quality and integrity of medical image content, and protecting patient privacy and medical image data security. The specific solution is as follows:

[0005] A robust watermarking method for medical images based on AGAST-LATCH includes:

[0006] The AGAST algorithm is used to extract AGAST feature points from medical images;

[0007] The LATCH algorithm is used to describe the AGAST feature points of the medical image, generating a binary feature descriptor matrix of the medical image;

[0008] The binary feature descriptor matrix of the medical image is subjected to DCT transformation, and the binary feature vector of the medical image is obtained by combining it with a hash function;

[0009] The binary feature vector of the medical image and the encrypted scrambled watermark are XORed bit by bit to embed the watermark information into the medical image.

[0010] Preferably, in the robust watermarking method for medical images based on AGAST-LATCH provided in the embodiments of the present invention, the AGAST algorithm is used to extract AGAST feature points from the medical image, including:

[0011] The grayscale value difference configuration space of the medical image is expanded to construct the AGAST scale space of the medical image;

[0012] Corner detection is performed using a binary tree in the AGAST scale space of the medical image to extract the corners in the medical image;

[0013] The corner points in the medical image are filtered using the non-maximum suppression method to identify the AGAST feature points of the medical image.

[0014] Preferably, in the robust watermarking method for medical images based on AGAST-LATCH provided in the embodiments of the present invention, the AGAST feature points of the medical image are described using the LATCH algorithm to generate a binary feature descriptor matrix of the medical image, including:

[0015] A detection window of a set size is selected, centered on the AGAST feature points of the medical image;

[0016] Three pixel blocks are selected in the detection window, with one pixel block designated as the main pixel block and the other two pixel blocks designated as accompanying pixel blocks.

[0017] Calculate and compare the F-norm between the main pixel block and each of the accompanying pixel blocks to obtain the value of the corresponding binary bit of the medical image;

[0018] A binary feature descriptor matrix of the medical image is generated based on the values ​​of the corresponding bits in the binary representation of the medical image.

[0019] Preferably, in the robust watermarking method for medical images based on AGAST-LATCH provided in the embodiments of the present invention, performing DCT transformation on the feature descriptor matrix of the medical image and obtaining the binary feature vector of the medical image by combining a hash function includes:

[0020] The feature descriptor matrix of the medical image is subjected to DCT transformation to obtain the coefficient matrix of the medical image;

[0021] A new matrix of the medical image is constructed by selecting modules of a set size at the low-frequency part of the coefficient matrix;

[0022] The new matrix of the medical image is symbolically transformed using a perceptual hash algorithm to obtain the binary feature vector of the medical image.

[0023] Preferably, in the robust watermarking method for medical images based on AGAST-LATCH provided in the embodiments of the present invention, before performing a bit-by-bit XOR operation on the binary feature vector of the medical image and the encrypted scrambled watermark, the method further includes:

[0024] Generate chaotic sequences;

[0025] The values ​​of the chaotic sequence are arranged in a set order, and the pixel position space of the original watermark is scrambled according to the position changes of each value before and after sorting in the chaotic sequence to obtain an encrypted scrambled watermark.

[0026] Preferably, in the above-described robust watermarking method for medical images based on AGAST-LATCH provided in this embodiment of the invention, while embedding the watermark information into the medical image, it further includes:

[0027] Generate a logical key;

[0028] Obtain the binary feature vector of the medical image to be tested;

[0029] The encrypted watermark is extracted by performing an XOR operation between the binary feature vector of the medical image to be tested and the logical key.

[0030] Preferably, in the robust watermarking method for medical images based on AGAST-LATCH provided in the embodiments of the present invention, obtaining the binary feature vector of the medical image to be tested includes:

[0031] The AGAST algorithm is used to extract AGAST feature points from the medical image under test;

[0032] The LATCH algorithm is used to describe the AGAST feature points of the medical image under test, and a binary feature descriptor matrix of the medical image under test is generated.

[0033] A DCT transformation is performed on the binary feature descriptor matrix of the medical image under test, and the binary feature vector of the medical image under test is obtained by combining it with a hash function.

[0034] Preferably, in the robust watermarking method for medical images based on AGAST-LATCH provided in the embodiments of the present invention, the AGAST algorithm is used to extract AGAST feature points of the medical image to be tested, including:

[0035] The grayscale value difference configuration space of the medical image under test is expanded to construct the AGAST scale space of the medical image under test;

[0036] Corner detection is performed using a binary tree in the AGAST scale space of the medical image under test to extract the corners in the medical image under test.

[0037] The corner points in the medical image under test are screened using the non-maximum suppression method to identify the AGAST feature points of the medical image under test.

[0038] Preferably, in the robust watermarking method for medical images based on AGAST-LATCH provided in the embodiments of the present invention, the LATCH algorithm is used to describe the AGAST feature points of the medical image to be tested, generating a binary feature descriptor matrix of the medical image to be tested, including:

[0039] The detection window of the set size is selected with the AGAST feature points of the medical image to be tested as the center;

[0040] Three pixel blocks are selected in the detection window, and one pixel block is designated as the main pixel block, while the other two pixel blocks are designated as the accompanying pixel blocks.

[0041] Calculate and compare the F-norm between the main pixel block and each of the accompanying pixel blocks to obtain the value of the corresponding binary bit of the medical image under test;

[0042] Based on the values ​​of the corresponding bits in the binary representation of the medical image under test, a binary feature descriptor matrix of the medical image under test is generated.

[0043] Preferably, in the above-described robust watermarking method for medical images based on AGAST-LATCH provided in the embodiments of the present invention, after extracting the encrypted watermark, the method further includes:

[0044] The pixel position space of the encrypted watermark is restored according to the position changes of each value before and after sorting in the chaotic sequence to obtain the restored watermark.

[0045] As can be seen from the above technical solution, the robust watermarking method for medical images based on AGAST-LATCH provided by the present invention includes: extracting AGAST feature points of the medical image using the AGAST algorithm; describing the AGAST feature points of the medical image using the LATCH algorithm to generate a binary feature descriptor matrix of the medical image; performing DCT transformation on the binary feature descriptor matrix of the medical image and obtaining the binary feature vector of the medical image by combining a hash function; and performing a bitwise XOR operation between the binary feature vector of the medical image and the encrypted scrambled watermark to embed the watermark information into the medical image.

[0046] The robust watermarking method for medical images based on AGAST-LATCH provided by this invention combines the advantages of AGAST and LATCH algorithms, such as fast computation speed and strong resistance to geometric attacks, with the ergodicity and robustness of DCT transform. It extracts and describes features from medical images, and then uses an encrypted scrambled watermark to perform calculations with the feature vectors of the medical image. This ensures the security and uncrackability of the digital watermark, embedding the watermark without modifying the medical image, preventing malicious attackers from discovering and destroying it. This effectively solves the defects caused by the modification of original medical image data in traditional digital watermarking embedding techniques, ensuring the quality of the medical image, giving it strong robustness and invisibility, and simultaneously protecting patient privacy and medical image data security. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0048] Figure 1 A flowchart of a robust watermarking method for medical images based on AGAST-LATCH provided in an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of corner point selection for the FAST algorithm provided in an embodiment of the present invention;

[0050] Figure 3 A flowchart of the FAST algorithm provided in an embodiment of the present invention;

[0051] Figure 4 The binary tree of the AGAST algorithm provided in this embodiment of the invention;

[0052] Figure 5 A flowchart of the AGAST algorithm provided in an embodiment of the present invention;

[0053] Figure 6 This is a schematic diagram of LATCH algorithm descriptor generation provided in an embodiment of the present invention;

[0054] Figure 7 Medical images of the lungs provided in embodiments of the present invention;

[0055] Figure 8 Medical images of the foot joint provided in embodiments of the present invention;

[0056] Figure 9Medical images of the neck provided for embodiments of the present invention;

[0057] Figure 10 Medical images of the sacroiliac joint provided for embodiments of the present invention;

[0058] Figure 11 Abdominal medical images provided for embodiments of the present invention;

[0059] Figure 12 Medical images of the testis provided in embodiments of the present invention;

[0060] Figure 13 The original medical image provided for the embodiments of the present invention;

[0061] Figure 14 The original watermark image provided in the embodiments of the present invention;

[0062] Figure 15 The encrypted watermark image provided in this embodiment of the invention;

[0063] Figure 16 The watermark extracted without interference, as provided in the embodiments of the present invention;

[0064] Figure 17 A medical image with a Gaussian noise interference intensity of 10% provided for an embodiment of the present invention;

[0065] Figure 18 The watermark extracted when the Gaussian noise interference intensity is 10% is provided in the embodiment of the present invention;

[0066] Figure 19 A medical image with a JPEG compression quality of 25% provided for an embodiment of the present invention;

[0067] Figure 20 The watermark extracted when the JPEG compression quality is 25% as provided in this embodiment of the invention;

[0068] Figure 21 The medical image provided in this embodiment of the invention has a filtering window size of [3×3] and is filtered 10 times by median filtering;

[0069] Figure 22 The watermark extracted after median filtering 10 times, with a filtering window size of [3×3] provided for this embodiment of the invention;

[0070] Figure 23 The medical image provided in this embodiment of the invention has a filtering window size of [5×5] and is filtered 15 times by median filtering;

[0071] Figure 24 The watermark extracted after median filtering 15 times, with a filtering window size of [5×5] provided for this embodiment of the invention;

[0072] Figure 25 A medical image rotated 20° counterclockwise, as provided in an embodiment of the present invention;

[0073] Figure 26 The watermark extracted when rotated 20° counterclockwise according to an embodiment of the present invention;

[0074] Figure 27 A medical image rotated 60° clockwise, as provided in an embodiment of the present invention;

[0075] Figure 28 The watermark extracted when rotated 60° clockwise, as provided in an embodiment of the present invention;

[0076] Figure 29 A medical image scaled 1.5 times as provided in an embodiment of the present invention;

[0077] Figure 30 The watermark extracted when scaled by 1.5 times according to an embodiment of the present invention;

[0078] Figure 31 A medical image horizontally shifted 30% to the right, provided as an embodiment of the present invention;

[0079] Figure 32 The watermark extracted when the horizontal position is shifted 30% to the right, as provided in this embodiment of the invention;

[0080] Figure 33 A medical image vertically shifted downwards by 30% as provided in an embodiment of the present invention;

[0081] Figure 34 The watermark extracted when the vertical displacement is 30% as provided in the embodiment of the present invention;

[0082] Figure 35 A medical image with 40% cropped along the X-axis provided for an embodiment of the present invention;

[0083] Figure 36 The watermark extracted when 40% is cut along the X-axis, as provided in this embodiment of the invention;

[0084] Figure 37 A medical image with 40% cropped along the Y-axis provided for an embodiment of the present invention;

[0085] Figure 38 The watermark extracted when the Y-axis is cut by 40% is provided in an embodiment of the present invention. Detailed Implementation

[0086] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0087] This invention provides a robust watermarking method for medical images based on AGAST-LATCH, such as... Figure 1 As shown, it includes the following steps:

[0088] S101. Extract AGAST feature points from medical images using the AGAST algorithm;

[0089] It should be noted that the above medical image is considered as the original high-quality image, for example, a medical image of size 512*512, denoted as I(i,j). I(i,j) represents the pixel grayscale value of the original medical image.

[0090] Specifically, this invention uses the AGAST (Adaptive and Generic Accelerated Segment Test) algorithm to extract AGAST feature points from medical image I(i,j) to obtain AGAST feature point information.

[0091] S102. Use the LATCH algorithm to describe the AGAST feature points of the medical image and generate a binary feature descriptor matrix of the medical image.

[0092] Specifically, for the extracted AGAST feature points, the LATCH (Learned Arrangements of Three Patch Codes) algorithm is used to describe the AGAST feature points with LATCH descriptors to generate the binary feature descriptor matrix A(i,j) of the medical image I(i,j).

[0093] S103. Perform DCT transformation on the binary feature descriptor matrix of the medical image, and obtain the binary feature vector of the medical image by combining it with a hash function.

[0094] Specifically, a Discrete Cosine Transform (DCT) is performed on the binary feature descriptor matrix A(i,j) of the medical image I(i,j), and a hash function is used to obtain the robust binary feature vector V(i,j) of the medical image I(i,j).

[0095] S104. Perform a bitwise XOR operation on the binary feature vector of the medical image and the encrypted scrambled watermark to embed the watermark information into the medical image.

[0096] It should be noted that, to ensure the security and uncrackability of the digital watermark, this invention requires chaotic scrambling encryption of the original watermark before embedding it. Specifically, the encrypted scrambling watermark refers to the watermark obtained after chaotic scrambling encryption of the original watermark. A meaningful binary text image can be selected as the watermark embedded in the medical image I(i,j), denoted as W = {W(i,j)|W(i,j) = 0,1; 1≤i≤M1, 1≤j≤M2}, where M1 and M2 are the length and width dimensions of the original watermark, respectively. W(i,j) represents the pixel grayscale value of the original watermark. After chaotic scrambling encryption of the original watermark W(i,j), the encrypted scrambling watermark BW(i,j) can be obtained.

[0097] It should be noted that by performing a bitwise XOR operation between the binary feature vector V(i,j) of the medical image I(i,j) and the encrypted scrambled watermark BW(i,j), the watermark information can be embedded into the medical image I(i,j).

[0098] In the above-mentioned robust watermarking method for medical images based on AGAST-LATCH provided in this embodiment of the invention, the advantages of AGAST and LATCH algorithms, such as fast calculation speed and strong resistance to geometric attacks, as well as the ergodicity and robustness of DCT transform, are combined. The method extracts and describes features of the medical image, and then uses the encrypted scrambled watermark and the feature vector of the medical image for calculation. This ensures the security and uncrackability of the digital watermark, embedding the watermark without modifying the medical image, thus preventing the digital watermark from being discovered and destroyed by malicious attackers. This effectively solves the defects caused by the modification of the original medical image data by traditional digital watermarking embedding technology, ensuring the quality of the medical image, giving it strong robustness and invisibility, and protecting both the patient's privacy information and the data security of the medical image.

[0099] In a specific implementation, in the robust watermarking method for medical images provided in the embodiments of the present invention, step S101 uses the AGAST algorithm to extract AGAST feature points of the medical image. Specifically, it may include: expanding the gray value difference configuration space of the medical image I(i,j) to construct the AGAST scale space of the medical image I(i,j); using a binary tree to perform corner detection in the AGAST scale space of the medical image I(i,j) to extract the corner points in the medical image I(i,j); and using a non-maximum suppression method to filter the corner points in the medical image I(i,j) to filter out the AGAST feature points of the medical image I(i,j).

[0100] It should be noted that the AGAST algorithm used in this invention is an image local feature extraction algorithm, which is an improvement of the FAST corner detection algorithm. It has a higher feature point extraction rate and better scene adaptability, and performs better in complex scene images.

[0101] like Figure 2 As shown, the FAST algorithm randomly selects a point P in the image, assuming its pixel (brightness) value is Ip. A Bresenham circle is constructed with P as the center and a radius of 3 pixels, covering 16 pixels surrounding P. A threshold t is set. The threshold t can be between 60 and 80; the lower the threshold, the more corner points can be extracted. If the absolute value of the difference between the gray values ​​of N consecutive pixels surrounding P and the gray value of P is not less than t, then this point is identified as a corner point. Generally, N = 12 is considered. The specific process is as follows... Figure 3 As shown.

[0102] As an improvement to the FAST algorithm, the AGAST algorithm provides a more detailed grayscale value difference configuration space, which has better performance for complex images. It expands the original configuration space by using "non-brighter" and "non-darker" values:

[0103]

[0104] In the formula, I represents the brightness of the pixel; the relative position of x and p is denoted as p→x; S p→x S′ represents the state of a pixel from p to x with respect to kernel p. p→x d represents the previous state. s,b, This indicates the grayscale comparison between the current point and the center point; d represents the current point being darker than the center point. 's' indicates relatively dim, 's' indicates similar, and 'b' indicates bright. For brighter light, u represents an unknown state.

[0105] Depend on Figure 3 As shown, the FAST algorithm uses a decision tree to find corners, but the corner construction changes significantly when the external environment changes. To overcome this shortcoming, the AGAST algorithm improves upon the underlying AST of FAST by converting the decision tree into a binary tree, as shown below. Figure 4 As shown, this allows the algorithm to automatically adapt to the external environment, and the detection speed is also significantly improved.

[0106] The flowchart for AGAST algorithm to extract corner points is as follows: Figure 5 As shown, based on the FAST algorithm, the configuration space is expanded, and a binary tree is used for corner detection. After extracting the corners, the non-maximum suppression method is still used to filter the corners to obtain AGAST feature points.

[0107] The non-maximum suppression method involves calculating the AGAST score at the corner (the sum of the absolute values ​​of the grayscale differences between the 16 points in the Bresenham circle and the corner point), retaining the point with the highest score within the corner's neighborhood, and suppressing other points. The scoring formula is as follows:

[0108]

[0109] In the formula, V represents the score and t represents the threshold.

[0110] In a specific implementation, in the robust watermarking method for medical images provided in the embodiments of the present invention, step S102 uses the LATCH algorithm to describe the AGAST feature points of the medical image and generate a binary feature descriptor matrix of the medical image. Specifically, it may include: selecting a detection window W of a set size centered on the AGAST feature points of the medical image I(i,j); selecting three pixel blocks in the detection window W, and taking one pixel block as the main pixel block and the other two pixel blocks as the accompanying pixel blocks; calculating and comparing the F norm between the main pixel block and each accompanying pixel block to obtain the value of the corresponding binary bit of the medical image I(i,j); and generating a binary feature descriptor matrix A(i,j) of the medical image I(i,j) based on the value of the corresponding binary bit of the medical image I(i,j).

[0111] Specifically, the LATCH algorithm is a local image feature descriptor. Currently available binary descriptors all select an image patch of a certain size, i.e., a detection window W, centered on the detected feature point. Sampling is performed within the detection window, selecting T pairs of sampling points. The binary descriptor of that feature point is then determined by the set S of sampling points formed by the T pairs of sampling points, expressed by the formula:

[0112] S={s t} t＝1,2,3...,T ={P t,1 ,P t,2 ]} t＝1,2,...,T ;

[0113] In the formula, s t P represents the t-th sample point pair in the point set. t,1 P t,2 Let t be the coordinates of each pair of points in the image. The index t is associated with both a pair of coordinates in W and the Gaussian smoothing kernel σ. t =(σ t,1 ,σ t,2 ) t＝1,2,3...,T For each sampling pair s t Compare P t,1 and P t,2 After smoothing the grayscale, the value of the corresponding bit in the binary representation is set using the above formula:

[0114]

[0115] In the formula, W(p) t,1 ,σ t,1 ), W(p t,2 ,σ t,2 ) are respectively the detection window W in P t,1 With P t,2 The grayscale values ​​are subjected to a standard deviation of σ. t,1 σ t,2 The value after Gaussian filtering.

[0116] Finally, the binary descriptor b of a feature point w The representation is:

[0117]

[0118] This algorithm, which directly compares the pixel grayscale values ​​of point pairs or the grayscale values ​​after Gaussian smoothing, is easily affected by noise and will lose a lot of feature information in high-frequency regions.

[0119] Therefore, this invention employs the LATCH algorithm, a method based on comparing triplet pixel blocks to determine the value of each bit in the feature descriptor. Considering a sequence of t = 1, 2, 3, ..., T triplets, S is transformed into... like Figure 6 As shown, three pixel blocks P are selected in the detection window W. t,a P t,1 P t,2 This gives the location of the center pixel of a k*k pixel block. Let P... t,a Defined as the main pixel block, P t,1 and P t,2 The similarity is evaluated by calculating the F-norm between the main pixel block and its accompanying pixel block, using the following formula:

[0120]

[0121] Therefore, the binary descriptor of the feature points obtained using the LATCH algorithm can be represented as:

[0122]

[0123] In terms of runtime, the LATCH descriptor maintains the advantage of binary descriptors and is an order of magnitude faster than histogram-based descriptors. In terms of robustness, LATCH outperforms other binary descriptors on most datasets, narrowing the gap with histogram-based descriptors.

[0124] In a specific implementation, in the robust watermarking method for medical images provided in the embodiments of the present invention, step S103 performs DCT transformation on the feature descriptor matrix of the medical image and obtains the binary feature vector of the medical image by combining a hash function. Specifically, it may include: performing DCT transformation on the feature descriptor matrix A(i,j) of the medical image I(i,j) to obtain the coefficient matrix F(i,j) of the medical image I(i,j): F(i,j) = DCT2(A(i,j)); selecting modules of a set size (e.g., 2*16) at the low frequency of the coefficient matrix F(i,j) of the medical image I(i,j) to form a new matrix D(i,j) of the medical image I(i,j); and performing symbol conversion on the new matrix D(i,j) of the medical image I(i,j) by a perceptual hash algorithm to obtain the binary feature vector V(i,j) of the medical image I(i,j).

[0125] Specifically, after selecting 2*16 modules from the low-frequency part of the coefficient matrix F(i,j) of the medical image I(i,j) to form a new matrix D(i,j), the mean of the new matrix D(i,j) is calculated as mean0. Based on the mean0 of the new matrix D(i,j), a hash function is used to perform a sign conversion on D(i,j). If an element in D(i,j) is greater than 0, it is set to 1; otherwise, it is set to 0, generating a 32-bit binary feature vector V(i,j) of the medical image.

[0126] To demonstrate the reliability of medical image feature vectors extracted using the AGAST-LATCH-DCT transform, this invention uses the AGAST-LATCH-DCT transform to... Figures 7-12 Feature extraction was performed on six different medical images. If the correlation coefficient between the extracted feature vectors of different medical images is less than 0.5, it can be considered that the feature vectors of different medical images are almost uncorrelated, and the method of using AGAST-LATCH-DCT transform to extract feature vectors of medical images is truly reliable. The test results are shown in Table 1:

[0127] Table 1. Correlation coefficients among different medical image feature vectors

[0128] Figure 7 1.00 0.35 0.23 0.22 0.26 0.35 Figure 8 0.35 1.00 0.20 0.08 0.29 0.36 Figure 9 0.23 0.20 1.00 0.44 0.12 0.23 Figure 10 0.22 0.08 0.44 1.00 0.45 0.22 Figure 11 0.26 0.29 0.12 0.45 1.00 0.01 Figure 12 0.35 0.36 0.23 0.22 0.01 1.00

[0129] It should be noted that digital watermarking technology has multiple implementation methods, the most common being spatial domain watermarking and frequency domain watermarking. Spatial domain watermarking embeds information into the pixels of the original image. This method requires some modification to the original content during the embedding process so that the information can be embedded and extracted. Frequency domain watermarking embeds information in the frequency domain of the original content. This method does not require modification of the original content, thus making it easier to achieve reversibility.

[0130] The main reason why most medical image watermarking algorithms are currently weak against geometric attacks is that digital watermarks are embedded in pixels or transform coefficients. Slight geometric transformations of medical images often lead to significant changes in pixel or transform coefficient values, making the embedded watermark vulnerable to attack. If feature vectors reflecting the geometric characteristics of medical images can be found, then when the image undergoes geometric transformations, the feature values ​​of the image will not undergo significant abrupt changes. Observation of the LATCH feature matrices of a large number of medical images revealed that when medical images are processed according to the method of this invention, the numerical values ​​of the LATCH feature matrix may change when a common geometric transformation is performed on a medical image, but the curvature of all data curves remains essentially unchanged. Furthermore, considering the characteristic that the image energy of DCT transform is concentrated in the low-frequency region, and according to human visual characteristics (HVS), low-to-mid-frequency signals have a greater impact on human vision and represent the main features of the image. Therefore, this invention selects the data after the AGAST-LATCH-DCT transform of medical images as the visual feature vector.

[0131] Furthermore, in a specific implementation, in the robust watermarking method for medical images provided in the embodiments of the present invention, before performing the bitwise XOR operation between the binary feature vector of the medical image and the encrypted scrambled watermark in step S104, the method may further include: using a Logistic Map algorithm to perform chaotic scrambling encryption on the original watermark in the frequency domain. Specifically, firstly, a chaotic sequence X(j) is generated based on the initial value x0. For example, the initial value of the chaotic coefficient can be set to 0.2, the growth parameter can be set to 4, and the number of iterations can be set to 1024. Then, the values ​​of the chaotic sequence X(j) are arranged in a set order (such as from smallest to largest), and the pixel position space of the original watermark W(i,j) is scrambled according to the position changes of each value in the chaotic sequence X(j) before and after sorting, to obtain the encrypted scrambled watermark BW(i,j).

[0132] In a specific implementation, the robust watermarking method for medical images provided in the embodiments of the present invention may further include, while performing step S104 of embedding watermark information into the medical image, generating a logical key Key(i,j): Save Key(i,j), which will be used later in the watermark extraction process. By applying for Key(i,j) as a key from a third party, ownership and usage rights of the original medical image can be obtained, thereby achieving the purpose of protecting the medical image.

[0133] Next, obtain the binary feature vector V′(i,j) of the medical image to be tested I′(i,j); perform an XOR operation between the binary feature vector V′(i,j) of the medical image to be tested I′(i,j) and the logical key Key(i,j) to extract the encrypted watermark BW′(i,j): This algorithm only requires the logical key Key(i,j) to extract the watermark, without the involvement of the medical image I(i,j), making it a zero-watermark extraction algorithm.

[0134] It should be noted that the method for obtaining the binary feature vector V′(i,j) of the encrypted medical image I′(i,j) under test can be the same as the method for obtaining the binary feature vector V(i,j) of the medical image I(i,j). For details, please refer to the method for obtaining the binary feature vector V(i,j) of the medical image I(i,j), which will not be elaborated here.

[0135] In a specific implementation, the robust watermarking method for medical images provided in the embodiments of the present invention, obtaining the binary feature vector V′(i,j) of the medical image to be tested I′(i,j) may specifically include: extracting AGAST feature points of the medical image to be tested I′(i,j) using the AGAST algorithm; describing the AGAST feature points of the medical image to be tested I′(i,j) using the LATCH algorithm to generate a binary feature descriptor matrix A′(i,j) of the medical image to be tested I′(i,j); performing DCT transformation on the binary feature descriptor matrix A′(i,j) of the medical image to be tested I′(i,j) and combining it with a hash function to obtain the binary feature vector V′(i,j) of the medical image to be tested I′(i,j).

[0136] Specifically, the binary feature descriptor matrix A′(i,j) of the medical image to be tested I′(i,j) is transformed by DCT to obtain the coefficient matrix F′(i,j): F′(i,j)=DCT2(A′(i,j)); 2*16 modules are selected from the coefficient matrix F′(i,j) to form a new matrix D′(i,j), and the mean of matrix D′(i,j) is calculated as mean1. The 32-bit binary feature vector V′(i,j) of the medical image to be tested is obtained through a hash function.

[0137] In specific implementation, the robust watermarking method for medical images provided in the embodiments of the present invention uses the AGAST algorithm to extract AGAST feature points of the medical image I′(i,j) to be tested. Specifically, this may include: expanding the gray value difference configuration space of the medical image I′(i,j) to construct the AGAST scale space of the medical image I′(i,j); using a binary tree to perform corner detection in the AGAST scale space of the medical image I′(i,j) to extract the corner points in the medical image I′(i,j); and using a non-maximum suppression method to filter the corner points in the medical image I′(i,j) to filter out the AGAST feature points of the medical image I′(i,j).

[0138] In a specific implementation, the robust watermarking method for medical images provided in the embodiments of the present invention uses the LATCH algorithm to describe the AGAST feature points of the medical image I′(i,j) to be tested, generating a binary feature descriptor matrix A′(i,j) of the medical image I′(i,j). Specifically, this may include: selecting a detection window of a set size centered on the AGAST feature points of the medical image I′(i,j); selecting three pixel blocks in the detection window, and taking one pixel block as the main pixel block and the other two pixel blocks as accompanying pixel blocks; calculating and comparing the F norm between the main pixel block and each accompanying pixel block to obtain the value of the corresponding binary bit of the medical image I′(i,j); and generating the binary feature descriptor matrix A′(i,j) of the medical image I′(i,j) based on the value of the corresponding binary bit of the medical image I′(i,j).

[0139] In specific implementation, the robust watermarking method for medical images provided in the embodiments of the present invention, after extracting the encrypted watermark, may further include: restoring the pixel position space of the encrypted watermark BW′(i,j) according to the position changes of each value in the chaotic sequence X(j) before and after sorting, to obtain the restored watermark W′(i,j). It should be noted that the method of generating the chaotic sequence X(j) here can be the same as the method of generating the chaotic sequence X(j) during the watermark encryption process.

[0140] The above process can be understood as follows: First, the watermark is encrypted by chaotic scrambling in the frequency domain using the properties of Logistic Map chaotic mapping; then, the medical image is transformed by AGAST-LATCH-DCT to extract the medical image feature vector for watermark embedding, and the feature vector is associated with the binary watermark to obtain a binary logical sequence, which is stored in a third party; then, the medical image to be tested is transformed by AGAST-LATCH-DCT to extract its feature vector, and associated with the binary sequence stored in the third party, thereby restoring and extracting the watermark.

[0141] Finally, the ownership of the medical image I(i,j) and the embedded watermark information can be determined by calculating the normalized cross correlation (NC) values ​​of the original watermark W(i,j) and the restored watermark W′(i,j).

[0142] The invention will be further described below with reference to the accompanying drawings. The experimental test object is a 512×512 abdominal medical image, see... Figure 13 Let I(i,j) represent the image, where 1≤i,j≤512. Select a meaningful binary image as the original watermark image, denoted as W(i,j), see [reference needed]. Figure 14The size of the watermark image here is 32×32.

[0143] First, the original medical image is processed using the AGAST-LATCH algorithm, and the feature matrix is ​​transformed using DCT. Considering the robustness of the method and the capacity of embedding the watermark in one step, 32 coefficients are selected as the feature values ​​of the image, i.e., a 2×16 module. The initial value of the chaos coefficient is set to 0.2, the growth parameter is 4, and the number of iterations is 1024. The original watermark image is then chaotically encrypted, and the encrypted watermark image is shown below. Figure 15 After detecting and restoring the watermark W′(i,j) using the watermark extraction algorithm, the normalized correlation coefficient NC is calculated to determine whether a watermark is embedded. The closer the NC value is to 1, the higher the similarity between the two images, thus assessing the robustness of the algorithm. The PSNR value represents the degree of image distortion; a higher PSNR value indicates less image distortion.

[0144] Figure 16 The image shows the watermark image extracted without interference. NC = 1.00, indicating that the watermark image can be accurately recovered.

[0145] The following experiments will determine the resistance to conventional attacks and geometric attacks of this digital watermarking method.

[0146] First, Gaussian noise was added to the medical image using the `imnoise` function. Table 2 shows the experimental data on the watermark image's resistance to Gaussian noise interference. As can be seen from Table 2, when the Gaussian noise interference intensity reaches 40%, the PSNR value of the interfered image drops to 8.01 dB. At this point, the extracted watermark image has a correlation coefficient NC = 0.82, indicating that the algorithm can still accurately extract the watermark, and the overall data are all around 0.9. This demonstrates that the present invention can give medical images a high degree of resistance to Gaussian noise.

[0147] Figure 17 This is a medical image with a Gaussian noise intensity of 10%, which is visually significantly different from the original abdominal medical image.

[0148] Figure 18 This is the watermarked image extracted when the Gaussian noise intensity is 10%, NC=0.94.

[0149] Table 2. Experimental data on watermark resistance to Gaussian noise interference.

[0150] PSNR (dB) 21.99 17.45 15.36 12.60 11.03 10.07 8.01 NC 0.94 0.88 0.94 0.94 0.9 0.88 0.82

[0151] Next, the abdominal medical image was JPEG compressed using the percentage of image compression quality as a parameter; Table 3 shows the experimental data of watermark resistance to JPEG compression attacks. When the compression quality is only 5%, the image quality is low, but the watermark can still be extracted from the attacked image using this invention, NC=0.56.

[0152] Figure 19 These are medical images with a compression quality of 25%.

[0153] Figure 20 This is the watermark image extracted when the compression quality is 25% and NC=1.00, which can accurately extract the watermark.

[0154] Table 3. Experimental Data on Watermark Anti-JPEG Compression Processing

[0155] PSNR (dB) 26.90 30.58 31.86 33.86 35.14 36.06 36.52 NC 0.56 0.72 0.74 1 1 0.94 1

[0156] The following describes the median filtering process performed on the abdominal medical image. Table 4 shows the experimental data of anti-median filtering processing of the medical image watermark. As can be seen from Table 4, when the median filtering parameter is [7×7] and the number of filtering repetitions is 15, the watermark image can still be recovered well, with NC = 0.66.

[0157] Figure 21 The image is a medical image with a median filter parameter of [3×3] and a filter repetition count of 10. The image is slightly blurred.

[0158] Figure 22 The watermark extracted when the median filtering parameter is [3×3] and the number of filtering repetitions is 10 has an NC = 0.85, indicating that it can recover the watermark of medical images relatively well.

[0159] Figure 23 The image is a medical image with a median filter parameter of [5×5] and a filter repetition count of 15. The image is already significantly blurred.

[0160] Figure 24 The watermark extracted when the median filtering parameter is [5×5] and the number of filtering repetitions is 15 has an NC = 0.76, indicating that it can recover the watermark of medical images relatively well.

[0161] Table 4. Experimental Data on Watermark Anti-Median Filtering Processing

[0162]

[0163] The following table shows the rotation transformation of the abdominal medical image, and the experimental data of watermark anti-rotation transformation are presented in Table 5. Table 5 shows that when the medical image is rotated 60° counterclockwise, NC = 0.63, and the watermark can still be accurately extracted.

[0164] Figure 25 It is a medical image rotated 20° counterclockwise;

[0165] Figure 26 The watermark is extracted when the medical image is rotated 20° counterclockwise. NC = 0.82, indicating that the watermark can be extracted accurately.

[0166] Figure 27 It is a medical image rotated 60° clockwise;

[0167] Figure 28 The watermark is extracted when the medical image is rotated 60° clockwise. NC = 0.6, indicating that the watermark can be extracted accurately.

[0168] Table 5. Experimental Data on Watermark Anti-Rotation Transformation

[0169] PSNR (dB) 13.92 15.35 16.66 16.68 15.33 13.92 NC 0.63 0.82 0.76 0.8 0.74 0.6

[0170] Note: A negative rotation degree indicates that the image rotates clockwise, and a positive rotation degree indicates that the image rotates counterclockwise.

[0171] Next, the abdominal medical image was scaled. Table 6 shows the experimental data of the medical image watermark anti-scaling transformation. As can be seen from Table 6, when the scaling factor is as small as 0.6, the correlation coefficient NC = 0.64, indicating that the watermark can be effectively extracted by the present invention.

[0172] Figure 29 This is a scaled medical image (scaling factor of 1.5);

[0173] Figure 30 The watermark is extracted from the medical image after scaling (scaling factor is 1.5), and NC = 0.6, indicating that the watermark of the medical image can be accurately extracted.

[0174] Table 6. Experimental Data on Watermark Anti-Scaling Transformation

[0175] NC 0.64 0.83 0.74 0.6 0.66

[0176] The following section describes the translation transformation of abdominal medical images. Table 7 shows the experimental data on watermark resistance to translation transformation. As can be seen from Table 7, when the horizontal or vertical translation of the medical image does not exceed 30%, its NC value is higher than 0.80, indicating that the present invention can accurately extract the watermark and has strong resistance to translation transformation.

[0177] Figure 31 This is a medical image that has been horizontally shifted 30% to the right.

[0178] Figure 32 The watermark extracted when the medical image is horizontally shifted 30% to the right has an NC value of 0.85, indicating that the present invention can accurately extract the watermark.

[0179] Figure 33 This is the image after a medical image has been vertically shifted downwards by 30%.

[0180] Figure 34 The watermark extracted when the medical image is vertically shifted downward by 30% has an NC value of 0.82, indicating that the present invention can accurately extract the watermark.

[0181] Table 7. Experimental Data on Watermark Anti-Translation Transformation

[0182]

[0183] The following section describes the shearing transformation of abdominal medical images. Table 8 shows the experimental data of watermark resistance to shearing transformation. As can be seen from Table 8, when the medical image is sheared along the X or Y coordinate axis, the NC value is still greater than 0.7 when the shearing amount reaches 40%, indicating that the watermark can still be accurately extracted. This demonstrates that the present invention has strong resistance to shearing transformation.

[0184] Figure 35 Medical images cropped 40% along the X-axis;

[0185] Figure 36 The watermark extracted when the medical image is sheared by 40% along the X-axis has an NC of 0.82, indicating that the present invention can accurately extract the watermark.

[0186] Figure 37 Learn the image by cropping it 40% along the Y-axis;

[0187] Figure 38 The watermark extracted by cutting the image along the Y-axis by 40% yielded NC = 0.74, indicating that the present invention can accurately extract the watermark.

[0188] Table 8. Experimental Data on Watermark Shear Transformation

[0189]

[0190] The data above shows that the present invention has good robustness, and its effect on Gaussian noise and attacks such as rotation, translation and shearing is particularly outstanding. Furthermore, the embedding of the watermark does not change the content of the original encrypted data, thus achieving the goal of zero-watermark resistance to geometric attacks and conventional attacks on medical images.

[0191] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0192] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0193] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0194] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0195] The robust watermarking method for medical images based on AGAST-LATCH provided by this invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A robust watermarking method for medical images based on AGAST-LATCH, characterized in that, include: The AGAST algorithm is used to extract AGAST feature points from medical images. Specifically, the grayscale value difference configuration space of the medical image is expanded to construct the AGAST scale space of the medical image. A binary tree is used in the AGAST scale space of the medical image to detect corner points and extract them. Non-maximum suppression is used to filter the corner points in the medical image, thereby selecting the AGAST feature points. The LATCH algorithm is used to describe the AGAST feature points of the medical image, generating a binary feature descriptor matrix for the medical image. A detection window of a predetermined size is selected, centered on the AGAST feature points of the medical image. Three pixel blocks are selected within the detection window, with one pixel block designated as the main pixel block and the other two as accompanying pixel blocks. The F-norm between the main pixel block and each of the accompanying pixel blocks is calculated and compared to obtain the value of the corresponding binary bit in the medical image. Based on the value of the corresponding binary bit in the medical image, the binary feature descriptor matrix of the medical image is generated. The binary feature descriptor matrix of the medical image is subjected to DCT transformation, and the binary feature vector of the medical image is obtained by combining it with a hash function; The binary feature vector of the medical image and the encrypted scrambled watermark are XORed bit by bit to embed the watermark information into the medical image.

2. The robust watermarking method for medical images based on AGAST-LATCH according to claim 1, characterized in that, Performing a DCT transform on the feature descriptor matrix of the medical image and combining it with a hash function to obtain the binary feature vector of the medical image includes: The feature descriptor matrix of the medical image is subjected to DCT transformation to obtain the coefficient matrix of the medical image; A new matrix of the medical image is constructed by selecting modules of a set size at the low-frequency part of the coefficient matrix; The new matrix of the medical image is symbolically transformed using a perceptual hash algorithm to obtain the binary feature vector of the medical image.

3. The robust watermarking method for medical images based on AGAST-LATCH according to claim 2, characterized in that, Before performing a bitwise XOR operation on the binary feature vector of the medical image and the encrypted scrambled watermark, the method further includes: Generate chaotic sequences; The values ​​of the chaotic sequence are arranged in a set order, and the pixel position space of the original watermark is scrambled according to the position changes of each value before and after sorting in the chaotic sequence to obtain an encrypted scrambled watermark.

4. The robust watermarking method for medical images based on AGAST-LATCH according to claim 3, characterized in that, In addition to embedding watermark information into the medical image, the method also includes: Generate a logical key; Obtain the binary feature vector of the medical image to be tested; The encrypted watermark is extracted by performing an XOR operation between the binary feature vector of the medical image to be tested and the logical key.

5. The robust watermarking method for medical images based on AGAST-LATCH according to claim 4, characterized in that, Obtain the binary feature vector of the medical image to be tested, including: The AGAST algorithm is used to extract AGAST feature points from the medical image under test; The LATCH algorithm is used to describe the AGAST feature points of the medical image under test, and a binary feature descriptor matrix of the medical image under test is generated. A DCT transformation is performed on the binary feature descriptor matrix of the medical image under test, and the binary feature vector of the medical image under test is obtained by combining it with a hash function.

6. The robust watermarking method for medical images based on AGAST-LATCH according to claim 5, characterized in that, The AGAST algorithm is used to extract AGAST feature points from the medical image under test, including: The grayscale value difference configuration space of the medical image under test is expanded to construct the AGAST scale space of the medical image under test; Corner detection is performed using a binary tree in the AGAST scale space of the medical image under test to extract the corners in the medical image under test. The corner points in the medical image under test are screened using the non-maximum suppression method to identify the AGAST feature points of the medical image under test.

7. The robust watermarking method for medical images based on AGAST-LATCH according to claim 6, characterized in that, The LATCH algorithm is used to describe the AGAST feature points of the medical image under test, generating a binary feature descriptor matrix of the medical image under test, including: The detection window of the set size is selected with the AGAST feature points of the medical image to be tested as the center; Three pixel blocks are selected in the detection window, and one pixel block is designated as the main pixel block, while the other two pixel blocks are designated as the accompanying pixel blocks. Calculate and compare the F-norm between the main pixel block and each of the accompanying pixel blocks to obtain the value of the corresponding binary bit of the medical image under test; Based on the values ​​of the corresponding bits in the binary representation of the medical image under test, a binary feature descriptor matrix of the medical image under test is generated.

8. The robust watermarking method for medical images based on AGAST-LATCH according to claim 7, characterized in that, After extracting the encrypted watermark, the process also includes: The pixel position space of the encrypted watermark is restored according to the position changes of each value before and after sorting in the chaotic sequence to obtain the restored watermark.

Images