Method for hiding regional reversible information of medical image

Abstract

The invention discloses a medical image regionalization reversible information hiding method, and aims to solve the problems of low embedding capacity, obvious visual quality degradation, insufficient reversibility, poor regional adaptability and the like in the existing medical image information hiding technology. A carrier image is generated through self-adaptive neighborhood interpolation, the image comprises seed pixels and non-seed pixels, region segmentation of the carrier image is carried out by combining an OTSU segmentation threshold, information embedding is carried out on the non-seed pixels of a region of interest through a DPD histogram technology, and the region of interest is obtained. And embedding the residual information to be embedded into non-seed pixels of the non-interested area by adopting a prediction error extension method, and finally realizing a high-capacity, high-fidelity and strong-reversibility regionalized medical image information hiding method so as to cope with diversified medical safety application scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image processing and information security, and in particular relates to a regionalized reversible information hiding method for medical images. Background Art

[0002] With the widespread use of medical images in clinical diagnosis, telemedicine, and electronic medical records, data security and privacy issues are becoming increasingly prominent. In particular, unauthorized access and tampering during image transmission and storage can severely impact diagnostic results. Consequently, researchers have turned their attention to reversible information hiding, which, due to its reversible nature, has become a research hotspot. Reversible information hiding not only accurately extracts secret information but also perfectly restores the underlying information. This holds significant practical significance for fields requiring high-quality images, such as medicine, the military, and law.

[0003] Research on reversible information hiding in the medical field has garnered significant attention. When medical data is stored or shared, sensitive data such as patient personal data (e.g., name, age, address, biometrics, medical insurance number, social security card number, etc.), electronic medical records (e.g., sensor parameters, diagnostic information), and medical electronic signatures are vulnerable to leakage. This can have serious consequences for patients and hospitals, and even have a significant negative impact on society. Given the aforementioned situation, research on reversible information hiding in medical images holds enormous potential for development. This approach uses digital medical images as carrier information, hiding secret information within the carrier information using a reversible embedding algorithm. Digital medical images are composed of multiple pixels, and spatial resolution, often referred to as the distance between adjacent pixels, is the distance between adjacent pixels. Based on the pixel grayscale, digital medical images can be categorized as binary, grayscale, or color images. Based on the imaging method, common medical images can be categorized as computerized tomography (CT), magnetic resonance imaging (MRI), X-ray computed tomography (X-ray), and chest X-ray film (CXR). The human eye often has a certain degree of redundancy in image processing and a certain degree of error in its perception of spatial resolution. Reversible information hiding algorithms based on medical images exploit this characteristic.

[0004] Typical reversible information hiding algorithms include difference expansion (DE), histogram shift (HS), and prediction error expansion (PEE). Compared to the first two algorithms, the prediction error expansion algorithm can better exploit the local correlation between the pixels of the carrier image, making fewer modifications to it, and reducing the distortion of the encrypted image at the same embedding rate. Traditional interpolation techniques, such as diamond mean interpolation, are used to predict the interpolated pixels in the carrier image. Although this scheme has good image quality, it is computationally intensive and not conducive to processing large amounts of information. Traditional histogram techniques, such as using the results of the ROI and non-ROI division to select the information embedding location, rather than directly applying stretched histogram shift embedding, have limited flexibility in the information embedding method itself.

[0005] Existing reversible information hiding algorithms for medical images are based on the characteristics of medical images and are combined with other image processing techniques and reversible information hiding technologies. Large amounts of patient privacy and diagnostic information require a certain embedding capacity, reversible information hiding requires reversibility and imperceptibility, and practical medical applications require real-time and security. Therefore, when designing reversible information hiding algorithms for medical images, it is important to consider the following three aspects: the balance between embedding capacity and image quality, the balance between algorithm complexity and security, and meeting the actual needs of medical diagnosis (storage and transmission). In summary, a regionalized medical image information hiding method with high capacity, high fidelity, and strong reversibility is urgently needed to address diverse medical security application scenarios. Summary of the Invention

[0006] The present invention provides a regionalized reversible information hiding method for medical images, aiming to solve the problems of low embedding capacity, significant visual quality degradation, insufficient reversibility and poor regional adaptability in existing medical image information hiding technologies. The method integrates interpolation technology, image segmentation technology and histogram technology, generates a carrier image through adaptive neighborhood interpolation, and the image includes seed pixels and non-seed pixels. The carrier image is regionally segmented by combining the maximum between-class variance method (OTSU) segmentation threshold, and information is embedded in the non-seed pixels of the region of interest using the pixel difference (DPD) histogram technology. The remaining information to be embedded is embedded in the non-seed pixels of the non-region of interest using the prediction error expansion method. Ultimately, a regionalized medical image information hiding method with high capacity, high fidelity and strong reversibility is realized to cope with diverse medical security application scenarios.

[0007] The regionalized reversible information hiding method for medical images of the present invention is as follows:

[0008] 1. Preprocess medical images (including MRI images and CT images) to obtain standardized two-dimensional grayscale images (M×N). Medical image preprocessing includes image denoising, spatial registration, grayscale normalization, and boundary enhancement.

[0009] 2. Adopt the adaptive neighbor interpolation algorithm (AIA) to interpolate and expand the two-dimensional grayscale image to generate a carrier image containing seed pixels and non-seed pixels. The seed pixels remain in their original state for subsequent image restoration, and the non-seed pixels are used for information embedding. The formula of the adaptive neighbor interpolation algorithm is as follows;

[0010] I(2i,2j)=A(i,j)

[0011]

[0012] Where: A(i,j) represents the pixel value of the i-th row and j-th column of the medical image, I(i,j) represents the pixel value of the i-th row and j-th column of the carrier image, i = 1, 2, 3...; j = 1, 2, 3...;

[0013] 3. Use the maximum inter-class variance method to determine the segmentation threshold, divide the interpolated intermediate carrier image into regions of interest, regions of no interest, and square regions, and identify seed pixels and non-seed pixels in regions of interest and regions of no interest;

[0014] 4. Analyze the least significant bit of the square area, extract the lowest bit of the pixel and the lowest vacant bit of the pixel in the area, combine the lowest bit of the pixel with the information to be hidden, and obtain the binary information to be embedded after encryption;

[0015] The information to be hidden includes patient information and diagnosis information. Patient information includes identity information and condition information. Diagnosis information includes doctor information and condition diagnosis information.

[0016] 5. Non-seed pixels in the region of interest are embedded with the information to be embedded using the pixel difference (DPD) histogram technique

[0017] (1) performing raster scanning on the carrier image to obtain the grayscale values ​​of all non-seed pixels in the region of interest of the carrier image, the non-seed pixel payload, the grayscale value of the last non-seed pixel with the highest frequency, the maximum value of the non-seed pixels, and the minimum value of the non-seed pixels; constructing a grayscale histogram with the grayscale value of the non-seed pixels as the horizontal axis and the frequency of the grayscale value of the non-seed pixels as the vertical axis;

[0018] Calculate the DPD values ​​of the three adjacent pixel groups (A, B, C) in the grayscale histogram using the formula d = |CB| - |BA|. Find the DPD value with the highest frequency among the DPD values ​​and record it as Ppeak.

[0019] Based on the mean and standard deviation of all DPD values, the highest value θ of the DPD value interval is determined by the following formula high and the lowest value θl ow , when Ppeak>θ highWhen θ low <Ppeak≤θ high When Ppeak≤θ low When , the position of the adjacent three-pixel group corresponding to Ppeak in the region of interest is divided into a low-frequency area;

[0020] θ high =μ ROI +σ ROI θl ow =μ ROI -0.5σ ROI ;

[0021] (2) A histogram is constructed with the grayscale value of the non-seed pixel corresponding to the DPD value divided into the intermediate frequency region as the horizontal axis and the frequency of the grayscale value of the non-seed pixel as the vertical axis. The histogram is stretched using the following formula to obtain the stretched histogram pixel value and the peak point value is counted;

[0022]

[0023] Where: H(x,y) represents the pixel value of the pixel point (x,y) in the histogram; H min 、H max Indicates the minimum grayscale value and maximum pixel value of non-seed pixels in the histogram; Lmin and Lmax indicate the minimum and maximum pixel values ​​of non-seed pixels in the stretched histogram; round indicates the use of rounding rule;

[0024] (3) Find the pixel value corresponding to the peak point with missing adjacent positions in the stretched histogram. If there is no corresponding pixel value in the grayscale histogram in step (1), the peak point will not be used. The pixel points corresponding to the pixel values ​​obtained in the stretched histogram are recorded as a set {S1, S2, ... Sn}. The histogram shift method (refer to the method in the literature Ren Fang, Liu Yuge, Zhang Xing, et al. Reversible information hiding scheme based on interpolation and histogramshift for medical images [J]. Multimedia Tools and Applications. 2023, 82 (18): 1-27) is used to embed the hidden information at the position of the pixel points in the set corresponding to the position in the grayscale histogram in step (1), completing the embedding of the hidden information in the mid-frequency region; at the same time, the information to be embedded is embedded in the high-frequency region to obtain a dense region of interest (non-seed pixels and seed pixels containing embedded information);

[0025] 6. Due to the limited image embedding capacity, the remaining information to be embedded is embedded into the non-seed pixels of the non-interest region using the prediction error expansion method, while the seed pixels remain unchanged, thus obtaining the encrypted non-interest region. In order to correctly extract the hidden information and restore the carrier image, the location map and payload of the non-interest region are also saved.

[0026] When embedding information in non-interested areas, the prediction error expansion method embeds the remaining information to be embedded into non-seed pixels by using the difference between the predicted pixel value and the actual pixel value, while ensuring that the original value of the seed pixel is preserved, and using a position map to record the embedding position to prevent pixel value overflow;

[0027] 7. The non-seed pixel payload of step 5, the grayscale value of the last non-seed pixel with the highest frequency, the maximum value of the non-seed pixel, the minimum value of the non-seed pixel, the position map of the non-interest area in step 6, and the payload are embedded into the pixels in the lowest vacant position of the pixels in the square area using the LSB method to obtain the secret square area; finally, a hidden image containing the reversible information of the secret area of ​​interest, the secret area of ​​non-interest, and the secret square area is obtained.

[0028] The present invention discloses a method for recovering the seed pixels of the region of interest and the non-region of interest in the hidden image of reversible information without any change during embedding, and the region can be recovered by adopting the same interpolation method as step 2 and the same segmentation method as step 3; the encrypted region of interest is recovered by decompressing the encrypted region of interest in step 7 through the improved compression algorithm RLE; the information of the encrypted region of interest is extracted by performing prediction error expansion extraction on the non-seed pixels of the encrypted non-region of interest to obtain embedded information, and the information of the encrypted non-region of interest is recovered; the information extracted from the encrypted region of interest and the encrypted non-region of interest is encrypted information to be embedded, and the hidden information and the lowest bit of the region of interest are obtained by decryption; the lowest bit of the region of interest obtained after decryption is embedded into the vacant lowest bit of the region of interest by adopting the LSB method, and the region of interest is recovered; the recovered region of interest, the region of interest and the non-region of interest constitute a carrier image, and the carrier image is recovered by the inverse process of the adaptive interpolation algorithm to obtain the initial medical image.

[0029] The advantages of the present invention are:

[0030] High embedding capacity and quality: The DPD histogram method effectively avoids the pixel distortion problem caused by traditional histogram stretching methods, ensuring that the image still has good subjective visibility and objective performance after embedding;

[0031] Reversibility: The introduction of seed pixels makes the image reconstruction process simpler, reduces the overall complexity of the algorithm, and achieves lossless restoration.

[0032] Regional adaptability: Through precise ROI / NROI segmentation, a differentiated information embedding strategy is implemented, which not only ensures the information hiding requirements of key diagnostic areas, but also takes into account the visual quality of non-critical areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a schematic diagram of a standardized two-dimensional grayscale image after preprocessing of multimodal medical image data, including (a) Abdominal CT, 256x256; (b) Brain MRI, 180x180; (c) Pancreas MRI, 256x256; (d) COVID-19 CT, 720x541; (e) COVID-19 CXR, 256x256; (f) Lung CT, 888x495; (g) Liver MRI, 256x256; (h) Hand X-ray, 406x512; (i) Leg X-ray, 406x512; (j) Brain CT, 180x180;

[0034] Figure 2Schematic diagram of the histogram and stretched histogram of the present invention. Figure a is the original histogram of the image, and Figure b is the stretched histogram. Figure c is the original histogram with density, and Figure d is the original stretched histogram with density.

[0035] Figure 3 Figure 1 is a schematic diagram of the embedding process. Figure a shows the embedding position selection process using the stretched histogram SH, and Figure b shows the histogram H information embedding process.

[0036] Figure 4 Therefore Figure 1 (c) Pancreas MRI is the encrypted image of the carrier image at the embedding rates of 0.3bpp, 0.5bpp, and 1bpp using the four algorithms. Figures a, e, and i represent the Yang algorithm, Figures b, f, and j represent the Gao algorithm, Figures c, g, and k represent the Ren algorithm, and Figures d, h, and l represent the encrypted images of the proposed algorithm at different embedding rates.

[0037] Figure 5 Therefore Figure 1 (j) Brain CT, 180x180, is the encrypted image of the four algorithms when the embedding rate of the carrier image is 0.3bpp, 0.5bpp, and 1bpp. Figures a, e, and i represent the Yang algorithm, Figures b, f, and j represent the Gao algorithm, Figures c, g, and k represent the Ren algorithm, and Figures d, h, and l represent the encrypted images of the algorithm in this paper at different embedding rates (only the embedded rate of encrypted image i is 0.7bpp, and the embedded rates of encrypted images j, k, and l are all 1bpp). DETAILED DESCRIPTION

[0038] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Unless otherwise specified, the methods in the embodiments are all conventional methods.

[0039] Example 1: The method for regionalized reversible information hiding in medical images is as follows:

[0040] The proposed algorithm was simulated and evaluated on a grayscale image dataset of a multimodal medical imaging dataset using a Windows 10 operating system with an Intel(R) Core(TM) i5-7200U processor (2.50GHz main frequency, 2.70GHz maximum turbo frequency) and the PyTorch 2.0 and Python 3.9 development environments using the PyCharm compiler.

[0041] 1. The dataset of this embodiment comes from the open source project linhandev / dataset website. Different multimodal medical image data are selected and image denoising, spatial registration, grayscale normalization, and boundary enhancement are performed to obtain 30 standardized two-dimensional grayscale images ( Figure 1 );

[0042] 2. Adopt the adaptive neighbor interpolation algorithm (AIA) to interpolate and expand the two-dimensional grayscale image to generate a carrier image containing seed pixels and non-seed pixels. The seed pixels remain in their original state for subsequent image restoration, and the non-seed pixels are used for information embedding. The formula of the adaptive neighbor interpolation algorithm is as follows;

[0043] I(2i,2j)=A(i,j)

[0044]

[0045] Where: A(i,j) represents the pixel value of the i-th row and j-th column of the medical image, I(i,j) represents the pixel value of the i-th row and j-th column of the carrier image, i = 1, 2, 3...; j = 1, 2, 3...;

[0046] 3. Use the maximum inter-class variance method to determine the segmentation threshold, divide the interpolated intermediate carrier image into regions of interest, regions of no interest, and square regions, and identify seed pixels and non-seed pixels in regions of interest and regions of no interest;

[0047] 4. Perform least significant bit analysis on the square area to extract the lowest bit of the pixels and the lowest vacant bit of the pixels in the area. Combine the lowest bit of the pixels with the information to be hidden and encrypt it to obtain the binary information to be embedded.

[0048] The information to be hidden includes patient information and diagnosis information. Patient information includes identity information and condition information. Diagnosis information includes doctor information and condition diagnosis information.

[0049] 5. Non-seed pixels in the region of interest are embedded with the information to be embedded using the pixel difference (DPD) histogram technique

[0050] (1) performing raster scanning on the carrier image to obtain the grayscale values ​​of all non-seed pixels in the region of interest of the carrier image, the non-seed pixel payload, the grayscale value of the last non-seed pixel with the highest frequency, the maximum value of the non-seed pixels, and the minimum value of the non-seed pixels; constructing a grayscale histogram with the grayscale value of the non-seed pixels as the horizontal axis and the frequency of the grayscale value of the non-seed pixels as the vertical axis;

[0051] Calculate the DPD values ​​of the three adjacent pixel groups (A, B, C) in the grayscale histogram using the formula d = |CB| - |BA|. Find the DPD value with the highest frequency among the DPD values ​​and record it as Ppeak.

[0052] Based on the mean and standard deviation of all DPD values, the highest value θ of the DPD value interval is determined by the following formula high and the lowest value θl ow , when Ppeak>θ high When θ low <Ppeak≤θ high When Ppeak≤θ low When , the position of the adjacent three-pixel group corresponding to Ppeak in the region of interest is divided into a low-frequency area;

[0053] θ high =μ ROI +σ ROI θl ow =μ ROI -0.5σ ROI ;

[0054] (2) A histogram is constructed with the grayscale value of the non-seed pixel corresponding to the DPD value divided into the intermediate frequency region as the horizontal axis and the frequency of the grayscale value of the non-seed pixel as the vertical axis. The histogram is stretched using the following formula to obtain the stretched histogram pixel value and the peak point value is counted;

[0055]

[0056] Where: H(x,y) represents the pixel value of the pixel point (x,y) in the histogram; H min 、H max Indicates the minimum grayscale value and maximum pixel value of non-seed pixels in the histogram; Lmin and Lmax indicate the minimum and maximum pixel values ​​of non-seed pixels in the stretched histogram; round indicates the use of rounding rule;

[0057] In medical images, the pixel value range of non-seed pixels in the stretched histogram is usually set to [0, 255], so Lmin is usually 0 and Lmax is 255;

[0058] (3) Find the pixel value corresponding to the peak point with missing neighboring positions in the stretched histogram. If there is no corresponding pixel value in the grayscale histogram in step (1), the peak point is not used. The pixel points corresponding to the pixel values ​​obtained in the stretched histogram are recorded as a set {S1, S2, ... Sn}. The histogram shift method is used to embed the information to be hidden at the position of the pixel points in the set corresponding to the position in the grayscale histogram in step (1), completing the embedding of the information to be hidden in the intermediate frequency area; at the same time, the information to be embedded is embedded in the high frequency area to obtain a dense region of interest (non-seed pixels and seed pixels containing embedded information). The histogram shift steps are as follows:

[0059] A. Split the stretched histogram in half. Based on the position of the information to be hidden and the stretched pixel value in the stretched histogram, shift the corresponding pixel value in the original histogram to embed the information to be hidden. The specific shifting rules are as follows:

[0060] When the information to be hidden is '1' and the stretched pixel value is in the left half of the stretched histogram, the pixel value at the corresponding position in the original histogram is shifted right by one position; when the information to be hidden is '1' and the stretched pixel value is in the right half of the stretched histogram, the pixel value at the corresponding position in the original histogram is shifted left by one position; when the information to be hidden is '0', the pixel value at the corresponding position in the original histogram remains unchanged;

[0061]

[0062] In the formula, k1 is the pixel value of the original histogram peak point, k2 is the pixel value of the stretched histogram peak point, k' is the pixel value after embedding the hidden information, bi represents the information bit of the hidden information, which is 0 / 1, SH(k2+1) and SH(k2-1) represent the left and right neighboring pixel values ​​of the peak point k2 in the stretched histogram, respectively.

[0063] B. Repeat step A until there are no pixel values ​​that meet the conditions.

[0064] Figure 2 ab、 Figure 3It shows that the pixel value range is stretched from [3,8] to [1,10] for information embedding. In the first round, the original histogram has a peak point k1∈(3,4,7,8), and the stretched histogram has a peak point k2(1,3,8,10). In the stretched histogram, the peak point SH(3)=20 is found, which belongs to [1,4] and the right adjacent position is vacant. The embedding position is determined to be the pixel point with a pixel value of 4 in the original histogram, and 20 bits of information are embedded in the corresponding point of the original histogram H. Assuming that the information to be embedded '0' and '1' appear with equal probability, H(4)=H(5)=10, then SH(3)=SH(4)=10. In the second round, the peak point of the stretched histogram is SH(8)=16, which belongs to [7,10] and the left adjacent position is vacant, and 16 bits of information are embedded in H(7). H(7)=(H 6)=8, then SH(8)=SH(7)=8. In the third round, the peak point of the stretched histogram is SH(1)=10, which belongs to [1,4] and the right adjacent bit is vacant, so 10 bits of information are embedded in H(3). H(3)=H(4)=5, so SH(1)=SH(2)=5. In the fourth round, SH(3) has no vacant right adjacent bit, and SH(4) and SH(7) have no corresponding bits in the original histogram and are not used. The peak point of the stretched histogram is SH(10)=8, which belongs to [7,10] and the left adjacent bit is vacant, so 8 bits of information are embedded in H(8). H(8)=H(7)=4, so SH(10)=SH(9)=4; all optional bits are filled, and the peak point of the last round of the stretched histogram is saved. A maximum of 54 bits of information are embedded in the optional bits; through the above process, we can get the following: Figure 2 (c) shows the original dense histogram, such as Figure 2 (d) shows the stretched histogram of the density element.

[0065] 6. Due to the limited image embedding capacity, the remaining information to be embedded is embedded into the non-seed pixels of the non-interest region using the prediction error expansion method, while the seed pixels remain unchanged, thus obtaining the encrypted non-interest region. In order to correctly extract the hidden information and restore the carrier image, the location map and payload of the non-interest region are also saved.

[0066] When embedding information in non-interested areas, the prediction error expansion method embeds the remaining information to be embedded into non-seed pixels by using the difference between the predicted pixel value and the actual pixel value, while ensuring that the original value of the seed pixel is preserved, and using a position map to record the embedding position to prevent pixel value overflow;

[0067] 7. The non-seed pixel payload of step 5, the grayscale value of the last non-seed pixel with the highest frequency, the maximum value of the non-seed pixel, the minimum value of the non-seed pixel, the position map of the non-interest area in step 6, and the payload are embedded into the pixels in the lowest vacant position of the pixels in the square area using the LSB method to obtain the secret square area; finally, a hidden image containing the reversible information of the secret area of ​​interest, the secret area of ​​non-interest, and the secret square area is obtained.

[0068] 8. Data extraction and image recovery:

[0069] The seed pixels of the region of interest and the non-region of interest in the hidden image of reversible information do not undergo any changes during embedding, and the region can be restored by using the same interpolation method as step 2 and the same segmentation method as step 3; the secret square region in step 7 is decompressed by the improved compression algorithm RLE to restore the square region; according to the inverse process of the embedding method in step 5, the secret region of interest information is extracted; the prediction error expansion extraction is performed on the non-seed pixels of the secret non-region of interest to obtain the embedded information, and the secret non-region of interest information extraction is completed; the information extracted from the secret region of interest and the secret non-region of interest is the encrypted information to be embedded, and the hidden information and the lowest bit of the square region are decrypted to obtain the hidden information and the lowest bit of the square region. The lowest bit of the decrypted square region is embedded into the vacant lowest bit of the square region using the LSB method to restore the square region. The restored square region, region of interest and non-region of interest constitute the carrier image, and the carrier image is restored by the inverse process of the adaptive interpolation algorithm to obtain the initial medical image.

[0070] In order to evaluate the performance of the method of the present invention, the method disclosed in the literature Gao G, Tong S, Xia Z, et al. Reversible data hiding with automatic contrast enhancement for medical images [J]. Signal Processing. 2021, 178: 107817 was used as a comparative experiment. This paper also uses the medical image regionalization reversible information hiding technology to achieve regionalized medical image information with high capacity, high fidelity and strong reversibility, and has good performance.

[0071] In order to objectively evaluate the quality of encrypted images, this paper selects the peak signal-to-noise ratio (PSNR) as an important indicator, and other indicators such as cross-correlation coefficient (NC), cross entropy (C-EN), structural similarity (SSIM), Gini coefficient (G) and visual fidelity (VIF) as references, and compares these indicators with other algorithms;

[0072] The PSNR calculation formula is as follows:

[0073]

[0074] Where: MSE represents mean square error, I represents the input image, I' represents the image to be processed, the input image and the image to be processed are given sizes M×N, M represents the height of the image, N represents the width of the image, i and j are the row index and column index of the pixel, respectively, used to traverse each pixel in the image.

[0075] Table 1 According to Figure 4 According to Table 2 Figure 5 The objective index test results of pancreatic MRI and brain CT images at embedding rates of 0.3bpp, 0.5bpp and 1bpp (0.7bpp for brain CT images) are respectively presented. Data analysis shows that the invention in this paper maintains a PSNR value above 50dB under all embedding rate conditions, and as the embedding capacity increases, the PSNR gap with the comparison algorithm tends to widen, which fully verifies that the proposed scheme can still maintain excellent image quality under high load conditions. Further observation shows that the SSIM, NC and VIF indicators obtained by the invention in this paper are close to the ideal value of 0.999, indicating that the encrypted image has a high structural similarity and visual fidelity with the original image. At the same time, the cross entropy index between the carrier image and the encrypted image is significantly lower than that of the other three comparison algorithms, which confirms the advantage of the proposed algorithm in maintaining the statistical characteristics of the image from the perspective of information theory. To ensure the universality of the experimental conclusions, the study conducted extended experiments using an additional 30 medical images of different modalities from a medical image database. The average performance metrics summarized in Table 3 show that, in high-embedding scenarios, the proposed algorithm achieves an 8%-12% improvement in PSNR compared to the comparison algorithm. These experimental results, from both subjective and objective perspectives, confirm that the proposed algorithm achieves secure information hiding while fully meeting the stringent image quality requirements of medical diagnosis, providing reliable technical support for clinical applications.

[0076] Table 1 Performance evaluation of three schemes for Pancreas MRI encrypted images

[0077]

[0078]

[0079] Table 2 Performance evaluation of three schemes for Brain CT encrypted images

[0080]

[0081] Table 3 Average performance evaluation results of three schemes for 30 medical images containing encrypted images

[0082]

[0083]

Claims

1. A method for hiding reversible information in regionalized medical images, characterized in that: Here are the steps: (1) Preprocess the medical image to obtain a standardized two-dimensional grayscale image; (2) Adopting an adaptive neighboring pixel interpolation algorithm to interpolate and expand the two-dimensional grayscale image to generate a carrier image containing seed pixels and non-seed pixels; (3) Using the maximum inter-class variance method to determine the segmentation threshold, the interpolated intermediate carrier image is divided into regions of interest, regions of no interest, and square regions, and the seed pixels and non-seed pixels in the regions of interest and regions of no interest are identified; (4) Performing least significant bit analysis on the square area, extracting the lowest bit of the pixels and the lowest vacant bit of the pixels in the area, combining the lowest bit of the pixels with the information to be hidden, and obtaining the binary information to be embedded after encryption; (5) performing raster scanning on the carrier image to obtain the grayscale values ​​of all non-seed pixels in the region of interest of the carrier image, the non-seed pixel payload, the grayscale value of the last non-seed pixel with the highest frequency, the maximum value of the non-seed pixels, and the minimum value of the non-seed pixels; constructing a grayscale histogram with the grayscale value of the non-seed pixels as the horizontal axis and the frequency of the grayscale value of the non-seed pixels as the vertical axis; Calculate the DPD values ​​of the adjacent three-pixel group in the grayscale histogram using the formula d = |CB| - |BA|, and find the DPD value with the highest frequency among the DPD values, which is recorded as Ppeak; Based on the mean and standard deviation of all DPD values, the highest value θ of the DPD value interval is determined by the following formula high and the lowest value θl ow , when Ppeak>θ high When θ low <Ppeak≤θ high When Ppeak≤θ low When , the position of the adjacent three-pixel group corresponding to Ppeak in the region of interest is divided into a low-frequency area; i high =μ ROI +s ROI θl ow =μ ROI -0.5σ ROI ; (6) A histogram is constructed with the grayscale value of the non-seed pixel corresponding to the DPD value divided into the intermediate frequency region as the horizontal axis and the frequency of the grayscale value of the non-seed pixel as the vertical axis. The histogram is stretched using the following formula to obtain the stretched histogram pixel value and the peak point value is counted; Where: H(x,y) represents the pixel value of the pixel point (x,y) in the histogram; H min 、H max Indicates the minimum grayscale value and maximum pixel value of non-seed pixels in the histogram; Lmin and Lmax indicate the minimum pixel value and maximum pixel value of non-seed pixels in the stretched histogram; Round means using rounding rule; (7) Find the pixel value corresponding to the peak point of the missing neighbor in the stretched histogram. If there is no corresponding pixel value in the grayscale histogram in step (6), the peak point is not used. The pixel points corresponding to the pixel values ​​obtained in the stretched histogram are recorded as a set {S1, S2, ... Sn}. The histogram shift method is used to embed the information to be hidden at the position of the pixel points in the set corresponding to the position in the grayscale histogram in step (6), completing the embedding of the information to be hidden in the mid-frequency area; at the same time, the information to be embedded is embedded in the high-frequency area to obtain a dense region of interest; (8) Due to the limitation of image embedding capacity, the remaining information to be embedded is embedded into the non-seed pixels of the non-interest region using the prediction error expansion method, and the seed pixels remain unchanged to obtain the encrypted non-interest region; in order to correctly extract the hidden information and restore the carrier image, the location map and payload of the non-interest region are saved at the same time; (9) The non-seed pixel payload of step (5), the grayscale value of the last non-seed pixel with the highest frequency, the maximum value of the non-seed pixel, the minimum value of the non-seed pixel, the position map of the non-interest area of ​​step (8), and the payload are embedded into the pixels in the lowest vacant position of the pixels in the square area using the LSB method to obtain the secret square area; finally, a hidden image containing the reversible information of the secret interest area, the secret non-interest area, and the secret square area is obtained.

2. The method for hiding regionalized reversible information in medical images according to claim 1, characterized in that: The information to be hidden includes patient information and diagnosis information. Patient information includes identity information and condition information, and diagnosis information includes doctor information and condition diagnosis information.

3. The method for hiding regionalized reversible information in medical images according to claim 1, characterized in that: The seed pixels of the interest region and the non-interest region in the hidden image of the reversible information do not undergo any changes during the embedding process, and the region can be restored by using the same interpolation method as step (2) and the same segmentation method as step (3); the encrypted square region in step (9) is decompressed by the improved compression algorithm RLE to restore the square region; the inverse process of the embedding method in steps (5)-(7) is used to complete the extraction of the information of the encrypted interest region; the prediction error expansion extraction is performed on the non-seed pixels of the encrypted non-interest region to obtain the embedded information, and the extraction of the information of the encrypted non-interest region is completed; the information extracted from the encrypted interest region and the encrypted non-interest region is the encrypted information to be embedded, and the hidden information and the lowest bit of the square region are obtained by decryption. The lowest bit of the square region obtained after decryption is embedded into the lowest bit vacancy of the square region using the LSB method, and the square region can be restored. The restored square region, the interest region and the non-interest region constitute the carrier image, and the carrier image is restored by the inverse process of the adaptive interpolation algorithm to obtain the initial medical image.

4. The method for hiding regionalized reversible information in medical images according to claim 1, characterized in that: When embedding information in non-interest areas, the prediction error expansion method embeds the remaining information to be embedded into non-seed pixels by calculating the difference between the predicted pixel value and the actual pixel value, while ensuring that the original value of the seed pixel is retained and using a position map to record the embedding position to prevent pixel value overflow.

5. The method for hiding regionalized reversible information in medical images according to claim 1, characterized in that: Medical image preprocessing includes image denoising, spatial registration, grayscale normalization, and boundary enhancement.

Images