Robust zero-watermarking method and system for medical images based on adaptive 2dewt, pht and gabor transform

Abstract

This invention discloses a robust zero-watermarking method and system for medical images based on local minimum adaptive two-dimensional empirical wavelet transform (2DEWT), polar harmonic transform (PHT), and Gabor transform. The method includes the following steps: Step 1, copyright watermark encryption process; Step 2, copyright watermark embedding process; Step 3, copyright watermark extraction process; Step 4, copyright watermark decryption process; Step 5, copyright system usage process. This invention uses 2DEWT in the extraction of medical image features and, based on the characteristics of medical images, designs an adaptive improvement strategy that directly uses the local minimum points of the frequency domain energy curve as sub-band boundaries. The local minimum points are essentially the boundary points of different feature components in the frequency domain of medical images: the boundaries between low-frequency global structures and high-frequency lesion details, and the boundaries of lesion textures in different directions (such as blood vessel orientation and lesion edge angles) all exist in the form of local minimum points. Two-dimensional empirical wavelet transform is performed on medical images to extract low-frequency subbands. Gabor transform is applied to the low-frequency subbands, and PHT is applied to the high-frequency subbands. Image features are obtained through feature fusion, and the copyright watermark is embedded by obtaining the key through XOR with the copyright watermark.

Description

Technical Field

[0001] This invention belongs to the field of information security technology and relates to a robust zero-watermarking method and system for medical images. Specifically, it relates to a robust zero-watermarking method and system for medical images based on adaptive two-dimensional empirical wavelet decomposition (2DEWT), polar harmonic transform (PHT), and Gabor transform. Background Technology

[0002] In the wave of information digitization, the dissemination and sharing of professional data such as medical images and hyperspectral images are becoming increasingly frequent. However, this type of data is easily copied and tampered with, which brings serious risks of intellectual property infringement and data integrity breaches. Digital watermarking technology has become an important means of ensuring data security and copyright protection. As professional images, watermarking technology faces stringent requirements: it must strictly maintain the original pixels from being modified, while also possessing strong robustness and high uniqueness.

[0003] Traditional EWT, due to its fixed basis functions, struggles to effectively adapt to the complex frequencies of medical images. 2DEWT, extended to the image processing domain by Gilles in 2014, demonstrates superior feature extraction capabilities compared to traditional transforms through Fourier spectrum construction and customized basis functions. While PHT possesses rotation invariance, it falls short in capturing local textures; Gabor transform excels at extracting local textures but is sensitive to geometric transformations. A single transform cannot meet the multi-dimensional performance requirements of watermarking.

[0004] To address the aforementioned issues, this paper proposes a zero-watermarking method and system that integrates adaptive 2DEWT, PHT, and Gabor transforms, and designs an edge detection strategy based on local minima according to the characteristics of medical images. By leveraging the complementary advantages of multiple transforms, a watermark feature is constructed that balances global robustness and local uniqueness, adapting to the specific security requirements of medical images. Summary of the Invention

[0005] This invention discloses a robust zero-watermarking method and system for medical images based on adaptive 2DEWT, PHT, and Gabor transforms. Building upon the 2DEWT algorithm, an adaptive strategy based on minimum edge detection is proposed to adapt to the characteristics of medical images. The initial maximum boundary detection is changed to minimum boundary detection, and the algorithm is optimized to more stably extract low-frequency global structures and high-frequency lesion details. The local details from Gabor and the global details from PHT complement each other, improving the uniqueness and anti-attack capability of the zero-watermark. The robustness of this patented method is measured by normalized correlation (NC) and bit error rate (BER).

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A robust zero-watermarking method and system for medical images based on 2DEWT, PHT, and Gabor includes the following steps:

[0008] (1) Copyright watermark encryption process:

[0009] First, the original watermark image Standardization and resizing are performed to ensure compatibility with the host medical image during the embedding stage. Subsequently, chaotic sequences are generated using Tent chaotic mapping, by setting initial parameters (…). ), to obtain a pseudo-random number sequence Then perform binarization. This binary sequence... Performing a bitwise XOR operation with the watermark image performs encryption preprocessing on the watermark image, resulting in an encrypted watermark image. This process effectively enhances the security and resistance to attacks of watermarks.

[0010] (2) Copyright watermark embedding process:

[0011] Step 1: Analyze the host's medical images Perform 2DEWT to obtain a multi-scale, multi-directional set of subbands:

[0012]

[0013] in, Includes low-frequency subband and high-frequency subband The low-frequency subband contains the main structural information, while the high-frequency subband contains the main edge information.

[0014] Step 2: Apply PHT to the high-frequency subband for feature extraction to obtain rotation-invariant feature moment vectors:

[0015]

[0016] Step 3: Apply Gabor transform to the low-frequency subband to extract its orientation and texture response features:

[0017]

[0018] Step 4: Convert the shape feature vector and Gabor vector Perform feature fusion:

[0019]

[0020] This process enhances the stability of features in the frequency and orientation domains, making the extracted features more resistant to noise, compression, and ensemble attacks.

[0021] Step 5: Transfer the watermarked image encrypted with Tent. With fusion features Establish mapping relationship:

[0022]

[0023] And Stored as a zero-watermark feature key for subsequent watermark extraction.

[0024] (3) Copyright watermark extraction process:

[0025] During the detection phase, medical images that may have been tampered with are examined. Repeat the 2DEWT decomposition, PHT feature extraction, and Gabor transform process to obtain a new feature vector:

[0026]

[0027] Based on the original key and mapping relationship XOR to generate encrypted watermark:

[0028]

[0029] (4) Copyright watermark decryption process:

[0030] Using the decryption key The extracted encrypted watermark image Perform the reverse XOR operation:

[0031]

[0032] Thus, the watermark image can be recovered. .

[0033] (5) Copyright system usage process:

[0034] ①System copyright information embedding process:

[0035] Step 1: The system receives the medical image to be protected, the copyright watermark information, and the encryption key parameters provided by the user, and performs format verification on the copyright watermark information and the key parameters.

[0036] Step 2: The system calls the copyright information embedding module, uses the key parameters provided by the user to encrypt the copyright watermark information, and automatically triggers the zero-watermark construction process.

[0037] Step 3: Without modifying the pixel information of the medical image, the system generates a watermark extraction key based on the zero-watermark construction process.

[0038] Step 4: The system establishes a correspondence between the generated watermark extraction key and the key parameters and the medical image to be protected, and stores or registers them securely, thereby completing the embedding and registration of copyright information of the medical image.

[0039] ② Copyright information verification process:

[0040] Step 1: The system receives the medical image to be verified and extracts the watermark and verifies the key parameters provided by the user.

[0041] Step 2: The system calls the copyright information verification module and uses the watermark extraction key to perform feature extraction and XOR processing with the zero watermark key on the medical image to be verified.

[0042] Step 3: The system recovers the encrypted copyright watermark information based on the XOR result, and decrypts the encrypted copyright watermark information using the verification key parameter to obtain the copyright watermark image.

[0043] Step 4: The system performs a consistency check on the recovered copyright watermark image and outputs the copyright verification result to determine the copyright ownership status of the medical image.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] Experimental results show that the watermarking technology based on the optimized 2DEWT algorithm, combined with PHT and Gabor, has good robustness and can be applied to various interference scenarios in practical applications. Attached Figure Description

[0046] Figure 1 This is a flowchart of the watermarking process.

[0047] Figure 2 Original host image: Figure 2 (a) is a CT image of the chest cavity. Figure 2 (b) is a CT image of the lungs. Figure 2 (c) is a chest CT image. Figure 2 (d) is a CT image of the kidney;

[0048] Figure 3 The original watermark image;

[0049] Figure 4 The image encrypted using the Tent chaotic map;

[0050] Figure 5 For 43 medical images used in the test, after nine different attacks, the NC value of the watermarked image was extracted from each image.

[0051] Figure 6For 43 medical images used in the test, after nine different attacks, the BER value of the watermarked image was extracted from each image.

[0052] Figure 7 (a~i) shows the NC values ​​of the watermarked images extracted from 43 tested medical images after being subjected to nine attacks of different degrees, including Gaussian noise, salt-and-pepper noise, median filtering, mean filtering, Gaussian low-pass filtering, JPEG compression, rotation, center cropping, and edge cropping (the horizontal axis represents different attacks, and the vertical axis represents the NC value of the extracted watermarked image). Detailed Implementation

[0053] The technical solution of the present invention will be further described below with reference to the accompanying drawings, but it is not limited thereto. Any modifications or equivalent substitutions to the technical solution of the present invention that do not depart from the spirit and scope of the technical solution of the present invention should be covered within the protection scope of the present invention.

[0054] This invention discloses a robust zero-watermarking method and system for medical images based on local minimum adaptive two-dimensional empirical wavelet transform, polar harmonic transform, and Gabor transform. First, the watermark image is encrypted using Tent chaotic mapping to ensure its security. Next, feature information of the medical image is extracted using 2DEWT, PHT, and Gabor transform, and XORed with the encrypted watermark image to obtain the watermark key. This is the second layer of encryption, ensuring both the security of the watermark information and the prerequisite for information acquisition, further improving the reliability of the method. The method includes the following steps: Step 1, copyright watermark extraction process; Step 2, copyright watermark embedding process; Step 3, copyright watermark extraction process; Step 4, copyright watermark decryption process; Step 5, copyright system usage process. Figure 1 As shown, the specific steps include the following:

[0055] (1) Copyright watermark encryption process:

[0056] First, the original watermark image Standardization and resizing are performed to ensure compatibility with the host medical image during the embedding stage. Subsequently, chaotic sequences are generated using Tent chaotic mapping, by setting initial parameters (…). ), to obtain a pseudo-random number sequence Then perform binarization. This binary sequence... Performing a bitwise XOR operation with the watermark image performs encryption preprocessing on the watermark image, resulting in an encrypted watermark image. This process effectively enhances the security and resistance to attacks of watermarks.

[0057] (2) Copyright watermark embedding process:

[0058] Step 1: Analyze the host's medical images Perform 2DEWT to obtain a multi-scale, multi-directional set of subbands:

[0059]

[0060] in, Includes low-frequency subband and high-frequency subband The low-frequency subband contains the main structural information, while the high-frequency subband contains the main edge information.

[0061] Step 2: Apply PHT to the high-frequency subband for feature extraction to obtain rotation-invariant feature moment vectors:

[0062]

[0063] Step 3: Apply Gabor transform to the low-frequency subband to extract its orientation and texture response features:

[0064]

[0065] Step 4: Convert the shape feature vector and Gabor vector Perform feature fusion:

[0066]

[0067] This process enhances the stability of features in the frequency and orientation domains, making the extracted features more resistant to noise, compression, and ensemble attacks.

[0068] Step 5: Transfer the watermarked image encrypted with Tent. With fusion features Establish mapping relationship:

[0069]

[0070] And Stored as a zero-watermark feature key for subsequent watermark extraction.

[0071] (3) Copyright watermark extraction process:

[0072] During the detection phase, medical images that may have been tampered with are examined. Repeat the 2DEWT decomposition, PHT feature extraction, and Gabor transform process to obtain a new feature vector:

[0073]

[0074] Based on the original key and mapping relationship XOR to generate encrypted watermark:

[0075]

[0076] (4) Copyright watermark decryption process:

[0077] Using the decryption key The extracted encrypted watermark image Perform the reverse XOR operation:

[0078]

[0079] Thus, the watermark image can be recovered. .

[0080] (5) Copyright system usage process:

[0081] ①System copyright information embedding process:

[0082] Step 1: The system receives the medical image to be protected, the copyright watermark information, and the encryption key parameters provided by the user, and performs format verification on the copyright watermark information and the key parameters.

[0083] Step 2: The system calls the copyright information embedding module, uses the key parameters provided by the user to encrypt the copyright watermark information, and automatically triggers the zero-watermark construction process.

[0084] Step 3: Without modifying the pixel information of the medical image, the system generates a watermark extraction key based on the zero-watermark construction process.

[0085] Step 4: The system establishes a correspondence between the generated watermark extraction key and the key parameters and the medical image to be protected, and stores or registers them securely, thereby completing the embedding and registration of copyright information of the medical image.

[0086] ② Copyright information verification process:

[0087] Step 1: The system receives the medical image to be verified and extracts the watermark and verifies the key parameters provided by the user.

[0088] Step 2: The system calls the copyright information verification module and uses the watermark extraction key to perform feature extraction and XOR processing with the zero watermark key on the medical image to be verified.

[0089] Step 3: The system recovers the encrypted copyright watermark information based on the XOR result, and decrypts the encrypted copyright watermark information using the verification key parameter to obtain the copyright watermark image.

[0090] Step 4: The system performs a consistency check on the recovered copyright watermark image and outputs the copyright verification result to determine the copyright ownership status of the medical image.

[0091] The following explanation is based on theoretical foundations and experimental data:

[0092] 1. Two-dimensional empirical wavelet transform (2DEWT)

[0093] Two-dimensional empirical wavelet transform (2DEWT) is an extension of one-dimensional empirical wavelet transform to two-dimensional signals such as images. Its core principle is to construct a wavelet basis that matches the signal's frequency characteristics through data-driven adaptive frequency domain partitioning, achieving accurate extraction of the scale and direction features of two-dimensional signals. This overcomes the limitation of traditional two-dimensional wavelets in adapting to complex and heterogeneous signals. Its principle can be simplified into three core steps:

[0094] Step 1: Fourier Spectrum Analysis and Polar Coordinate Transformation

[0095] First, the two-dimensional signal (such as an image) The signal is converted to the frequency domain using a two-dimensional Fourier transform to obtain the frequency domain signal. ( The frequency is spatial frequency, and it is centered (low-frequency components are moved to the center of the frequency domain) to highlight the frequency distribution characteristics of the signal.

[0096] To facilitate the division of the frequency domain by scale and direction, the frequency domain rectangular coordinates are... Convert to polar coordinates :

[0097]

[0098] in, Radial frequency ( (with the frequency domain center) reflects the scale. The smaller the value, the smoother the global features in the image;

[0099]

[0100] in, Angular frequency, reflecting direction: range This corresponds to the direction of the image texture / edge.

[0101] Step 2: Adaptive Frequency Domain Subband Division

[0102] Radial scale division: Calculate the radial energy curve (for each) Summing all (corresponding frequency domain energy), find the local minimum point of the curve as the radial boundary. ( Starting from a low frequency, (where the maximum radial frequency is), the frequency domain is divided into One radial ring (corresponding to) (Scale).

[0103] Angular direction division: For each radial replacement, calculate the angular energy curve (for each Summing all the elements within the ring. The corresponding energy), also with local minima as the angular boundary. Divide each circumference into Each sector sub-band (corresponding to) (One direction).

[0104] Ultimately, the frequency domain was divided into Each sector is a sub-band, and each sub-band corresponds to a feature at a specific scale and orientation in the image.

[0105] Step 3: Constructing a fixed two-dimensional empirical wavelet basis:

[0106] Based on the divided frequency domain sub-bands, an adaptive empirical wavelet basis is constructed, and the original signal is projected onto the basis to achieve decomposition.

[0107] Basis function construction: Design a frequency domain mask for each sub-band, use a smooth transition function to ensure no spectral leakage between sub-bands, and then convert the frequency domain mask into a spatial empirical wavelet basis through inverse two-dimensional Fourier transform.

[0108] Signal decomposition and XOR: decomposing the original image By performing an inner product with the spatial basis functions, we obtain the empirical wavelet coefficients of each sub-band (the magnitude of the coefficients reflects the intensity of the corresponding scale or direction features). Since the basis functions satisfy the tight frame property, the original signal can be completely XORed with the coefficients through a linear combination of the basis functions, without information loss.

[0109] 2. Polar Harmonic Transform (PHT)

[0110] If using The representative size is The image of size, defined on the unit circle, can be represented as PHT. Kernel function With the number of repetitions A linear combination of .

[0111]

[0112] in, Represents the PHT moment.

[0113] PHT moments are calculated as follows: and The product of.

[0114]

[0115] in, It is a constant. yes . conjugate.

[0116] The kernel function of PHT consists of radial components. and circular components constitute

[0117]

[0118] Kernel functions should satisfy the following orthogonality conditions

[0119]

[0120] in, As the Kronecker function.

[0121] In fact, PHT is a collective term for Polar Complex Exponential Transform (PCET), Polar Cosine Transform (PCT), and Polar Sine Transform (PST). Different orthogonal moments have different radial components. The radial components of PCET, PCT, and PST are respectively... , , .

[0122] The above definitions are all in the continuous domain, while digital images in practical applications are represented discretely. In this case, PCET can be calculated as...

[0123]

[0124] in, , , , Using PCET moments and its kernel function XOR graph function, i.e.

[0125]

[0126] The calculation of discrete domain PCT and PST moments and their corresponding image reconstruction formulas are similar to the calculation of PCET moments.

[0127] 3. Gabor Transform

[0128] The Gabor transform is an extension of the Short-Time Fourier Transform (STFT) that introduces a Gaussian window function to provide time-frequency localization analysis. It combines time and frequency information, making it particularly suitable for analyzing non-stationary signals. In image processing, the Gabor transform is commonly used for tasks such as texture analysis, edge detection, and face recognition.

[0129] Frequency domain response of Gabor filters:

[0130]

[0131] in, Indicates the time shift parameter. Indicates frequency parameters, Representing the Gaussian window function:

[0132]

[0133] in, Indicates the window width.

[0134] 4. Tent Chaos Mapping

[0135] A chaotic system refers to a seemingly random and irregular motion within a deterministic system. Its behavior is characterized by uncertainty, non-repeatability, and unpredictability, which is the phenomenon of chaos.

[0136] The tent map, in mathematics, refers to a piecewise linear mapping, named for its function graph resembling a tent. In addition, it is a two-dimensional chaotic mapping, widely used in chaotic encryption systems, and frequently employed in the generation of chaotic spreading codes, the construction of chaotic encryption systems, and the implementation of chaotic optimization algorithms.

[0137] The definition of tent mapping is as follows:

[0138]

[0139] in, The tent map is a chaotic map within its parameter range, exhibiting a uniform distribution function and good correlation. It is also worth noting that the tent map and the logistic map are topologically conjugate maps. Within the acceptable range, the system is in a chaotic state. Especially At that time, the system exhibits a short-cycle state, so it is generally not taken. When using this mapping, it is important to note the initial system values. Cannot be combined with system parameters If they are the same, otherwise it will evolve into a periodic system, which is no longer a chaotic system.

[0140] Example:

[0141] The experimental dataset consisted of multiple medical images randomly selected from MedPix's public dataset, including, for example, the attached images. Figure 2 As shown. The watermark image size is... As attached Figure 3 The encrypted image is as follows: Figure 4As shown, the image has undergone significant changes, greatly improving security. From a robustness perspective, performance tests were conducted on the watermarking scheme proposed in this invention. The robustness and stability of the scheme were measured by applying various attack methods, including filtering attacks, noise attacks, compression attacks, and geometric attacks.

[0142] The robustness of the method was verified by subjecting the watermarked image to different attacks and detecting the similarity between the extracted watermark and the original watermark. Normalized cross-correlation (NC) and bit error rate (BER) were used to measure robustness. These attacks were categorized as Gaussian noise, salt-and-pepper noise, JPEG compression, median filtering, mean filtering, Gaussian low-pass filtering, rotation, and cropping. Table 1 shows the magnitude of normalized cross-correlation (NC) after the attacks.

[0143] Table 1. Robustness Test

[0144] attack NC BER Gaussian noise (0.1) 0.9975 0.0025 Salt and pepper noise (0.1) 0.9982 0.0017 Median filtering (9) 0.9974 0.0025 Mean filtering (9) 0.9995 0.0005 Gaussian low-pass filter (9) 0.9995 0.0005 JPEG compression (20) 0.9998 0.0002 Rotation (25) 0.9913 0.0087 Center clipping (0.1) 0.9962 0.0038 Edge trimming (0.1) 0.9976 0.0024

[0145] From Table 1 and Figure 5 , Figure 6 The experimental results show that the proposed watermarking method exhibits high stability and robustness on various medical images. Table 1 presents the average NC values ​​of multiple test images, reflecting the consistency of the method's overall performance. Figure 5 and Figure 6 The trends of NC and BER values ​​for each image under different attack conditions are shown. Except for a few images that performed slightly lower under extreme attacks, the indicators of the remaining images remained at a good level, fully demonstrating the robustness of the method under various attack scenarios. Furthermore, Figures 7(a-i) show the changes in NC for multiple medical images under different attack intensities, indicating that the method can effectively maintain high robustness under various attack types such as noise, compression, and filtering. In summary, the experimental data fully verify the universality and robustness of the method on diverse medical images, demonstrating its good application potential in the field of medical image copyright protection.

Claims

1. A robust zero-watermarking method and system for medical images based on two-dimensional empirical wavelet transform (2DEWT), polar harmonic transform (PHT), and Gabor transform, characterized in that... The method includes the following steps: First, obtain the image to be verified for copyright. (1) Copyright watermark extraction process Step 1: Verify the copyright image Perform 2DEWT to obtain a multi-scale, multi-directional set of subbands: Step 2: High-frequency subband Using PHT for feature extraction, rotation-invariant feature moment vectors are obtained: Step 3: Low-frequency sub-band Using Gabor transform, its orientation and texture response features are extracted: Step 4: Convert the shape feature vector and Gabor vector Perform feature fusion: This process enhances the stability of features in the frequency and orientation domains, making the extracted features more resistant to noise, compression, and ensemble attacks. Step 5: Transfer the zero-watermark feature key With fusion features Establish mapping relationship: Obtain encrypted watermark image . (2) Copyright watermark verification process: Step 1: Using the decryption key The extracted encrypted watermark image Perform the reverse XOR operation: Step 2: Decrypt the watermark image The image is matched with the original watermarked image in the database; after matching, the copyright of the image to be verified is determined to be legitimate.

2. The computer-implemented copyright protection method according to claim 1, characterized in that, The decryption key is obtained through the following method: Chaotic sequences are generated using Tent chaotic mapping by setting initial parameters ( ), to obtain a pseudo-random number sequence Then perform binarization. This binary sequence... Performing a bitwise XOR operation with the watermark image performs encryption preprocessing on the watermark image, resulting in an encrypted watermark image. This process effectively enhances the security and resistance to attacks of watermarks.

3. The computer-implemented copyright protection method according to claim 2, characterized in that, The key It is obtained through the following method: Step 1: Analyze the host's medical images Perform 2DEWT to obtain a multi-scale, multi-directional set of subbands: Step 2: Apply PHT to the high-frequency subband for feature extraction to obtain rotation-invariant feature moment vectors: Step 3: Apply Gabor transform to the low-frequency subband to extract its orientation and texture response features: Step 4: Convert the shape feature vector and Gabor vector Perform feature fusion: This process enhances the stability of features in the frequency and orientation domains, making the extracted features more resistant to noise, compression, and ensemble attacks. Step 5: Transfer the watermarked image encrypted with Tent. With fusion features Establish mapping relationship: And Stored as a zero-watermark feature key for subsequent watermark extraction.

4. The robust zero-watermarking method and system for medical images based on 2DEWT, PHT, and Gabor as described in claim 1, characterized in that... The 2DEWT provides an adaptive strategy for medical images: using a strategy based on local minima for edge detection.

5. A copyright protection device, characterized in that, The device includes a processor and a memory storing instructions, which, when executed by the processor, cause the copyright protection device to perform a computer-implemented copyright protection method as described in any one of claims 1-4.

6. A non-volatile storage medium, characterized in that, It stores instructions that, when executed, cause the computer to perform the computer-implemented copyright protection method according to any one of claims 1-4.

Images