A medical image data-oriented blind watermark embedding and tracing method

Abstract

The application discloses a kind of medical image data-oriented blind watermark embedding and tracing method, comprising the following steps: S1, original medical image is embedded watermark operation;S11, original medical image is parsed, and the pixel data in original medical image is extracted to obtain gray matrix;S12, according to gray matrix, the diagnosis area in original medical image is segmented, and safe embedding area and forbidden embedding area are obtained;S13, construct trace information tuple;S14, trace information vector is encrypted operation;S15, the transformation domain conversion of safe embedding area is carried out;S16, selected coefficient and watermark bit are quantized index modulation;S17, safe embedding area inverse transformation and image reconstruction are carried out, and the medical image embedded with watermark is generated;S2, the medical image embedded with watermark is traced operation;S21, the medical image embedded with watermark is extracted operation;S22, the extracted watermark is decoded and path restoration operation.

Description

Technical Field

[0001] This invention relates to the field of medical imaging, and more particularly to a blind watermark embedding and tracing method for medical image data. Background Technology

[0002] In the process of sharing and collaborative research of medical image data (such as DICOM format images from CT, MRI, and X-rays), preventing the secondary dissemination and leakage of data by authorized parties is a key challenge. While existing technologies offer general digital watermarking solutions, their direct application to medical image data suffers from the following significant drawbacks: Risk of compromising medical diagnostic value: In pursuit of robustness or capacity, general watermarking algorithms may embed watermarks in critical diagnostic areas (such as lesion edges and minute textures), introducing noise or artifacts, affecting the biometric characteristics of images and the accuracy of doctors' diagnoses, which is unacceptable in clinical practice.

[0003] Ignoring DICOM standards and medical workflows: Existing solutions are often incompatible with the DICOM standard. Embedding watermarks may damage key medical tags (such as patient information and scanning parameters) in the DICOM header file, causing the image to be unable to be correctly recognized and displayed by the PACS system.

[0004] Robustness is not optimized for medical image processing: Medical images undergo unique processing in clinical and research settings, such as window width / level adjustment, edge enhancement, measurement annotation, lossy compression (e.g., JPEG 2000), and normalization and resampling for AI training. General-purpose watermarking algorithms are not robust enough to withstand these medical-specific operations.

[0005] The limited dimensions of traceability information make it difficult to reconstruct the complete flow path: Simply embedding user IDs is insufficient to build a complete view of the data lifecycle, and the lack of association with the data authorization context (such as time and purpose) makes it difficult to quickly reconstruct the image sharing chain in the event of a leak.

[0006] Therefore, there is an urgent need in this field for a data traceability solution specifically designed for medical image data, which can achieve accurate and robust data traceability while ensuring diagnostic safety and DICOM compatibility. Summary of the Invention

[0007] The purpose of this invention is to address the above-mentioned problems by providing a blind watermark embedding and tracing method for medical image data that enables accurate data traceability while ensuring diagnostic security and DICOM compatibility.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows: A blind watermarking embedding and source tracing method for medical image data includes the following steps: S1. Perform watermark embedding operation on the original medical image; S11. Analyze the original medical image and extract the pixel data to obtain the grayscale matrix. ; S12, Based on the grayscale matrix The diagnostic region in the original medical image is segmented to obtain safe embedding regions and prohibited embedding regions; S13. Construct traceability information tuples; S14. Encrypt the traceability information vector; S15. Perform transform domain transformation on the secure embedding region; S16. Select coefficients and watermark bits and perform quantization index modulation; S17. Perform inverse transformation of the secure embedding region and image reconstruction to generate a medical image with watermark. S2. Perform source tracing operations on watermarked medical images; S21. Perform watermark extraction on the medical image with embedded watermark; S22. Decode and restore the path of the extracted watermark.

[0009] Furthermore, step S11 specifically includes the following steps: S121, Based on the grayscale matrix Generate a binary mask M, where, Represents pixels This is a prohibited embedding area. , Represents pixels Belongs to the secure embedded area ; S122. Automatically identify lesion regions using a pre-trained semantic segmentation model. Its formula is: ; S123, Regarding the lesion area A morphological dilation operation is performed to expand the protective boundary and generate the final mask. The formula is as follows: ; in, It is a structural element. This indicates the expansion operation.

[0010] Furthermore, in step S13, the traceability information tuple is: ; in, A unique identifier representing the sharing entity. Indicates the authorization timestamp. Indicates the data usage encoding. The hash value representing the watermark information of the parent image.

[0011] Furthermore, step S14 specifically includes the following steps: S141. Convert the source tracing information vector into a byte stream. ; S142. Use encryption algorithm byte stream Encryption is performed to obtain ciphertext. Its formula is: ; S143, Regarding the ciphertext Application of error correction coding Generate watermark sequence Its formula is: ; Furthermore, step S15 specifically includes the following steps: S151, For the secure embedded area Perform an L-level discrete wavelet transform to generate DWT subbands; the formula is as follows: ; S152. Select an intermediate frequency sub-band in the DWT sub-band as the watermark embedding area; S111, Based on the embedded key Generate a pseudo-random sequence, and pseudo-randomly select N carrier coefficients from the selected mid-frequency subband to form a carrier coefficient set. .

[0012] Furthermore, step S16 specifically includes the following steps: S161. Based on the HVS model, calculate the selected carrier coefficient. variance of the local region centered The calculation formula is as follows: ; in, and It is a preset constant that controls the range of step size; S162. For the selected carrier coefficient and the watermark location to be embedded The quantization modulation function is: ; in, It is the carrier coefficient after embedding the watermark.

[0013] Furthermore, step S71 specifically includes the following steps: S171, The carrier coefficient after embedding the watermark Put it back into the DWT sub-tape; S172. Perform inverse discrete wavelet transform on the modified DWT subbands to reconstruct the secure embedding region. Its formula is: ; S173, Embed the reconstructed security area With unchanged prohibited embedding regions Merge to obtain a pixel matrix with watermark. .

[0014] S174, watermarked pixel matrix Package the original medical image with the original image to generate a watermarked medical image.

[0015] Furthermore, step S21 specifically includes the following steps: S211. Analyze the watermarked medical image and extract pixel data to obtain the prohibited embedding region and the safe embedding region; S212. Perform multi-level discrete wavelet transform on the secure embedding region, based on the embedding key. The generated pseudo-random sequence is used to select the set of carrier coefficients that perfectly match the embedding stage position. ; S213, For each carrier coefficient to be detected The watermark position is determined based on how close its value is to the quantization range. Its formula is: ; in, The center point of the quantization interval corresponding to watermark bit 0 that is closest to the coefficient of the carrier to be detected. This is the center point of the quantization interval corresponding to watermark bit 1; S214, All watermark positions By concatenating them in sequence, a watermark sequence is obtained. This enables the extraction of watermarks.

[0016] Furthermore, step S11 specifically includes the following steps: S221, Using error correction decoding algorithm For watermark sequence Correction was performed to recover the ciphertext. ; S222. Perform the inverse encryption operation using the decryption key. , ciphertext Decrypt into byte stream It is then parsed into traceability information tuples according to a preset format. ; S223, Source tracing information tuple Match the authorization logs with the source tracing database to locate the authorization record corresponding to the current leak node; S224, Check the traceability information tuple In field, if the If the field is not empty, recursively query the source database for the next higher level authorization record that matches the hash value, tracing upwards step by step; if the... If the field is empty, it means that the medical image has been traced back to its source. S225. Construct a directed tracing chain from the current leak node back to the source of the medical image through recursive queries. Its formula is: ; in, This is the current leak node. ( ) represents the intermediate authorization node. This is the original node for generating medical images.

[0017] S226. If there is no hash value association, the authorization records of the current leaked node and the original medical image generation node are integrated; if there is a multi-level hash value association, all authorized node information obtained by recursive query is integrated; a medical image data flow path diagram is generated based on the integration result and a visual traceability report is output.

[0018] Furthermore, in step S26, the content of the visual traceability report includes the identity of the authorizing party, the authorization time, the purpose of use, and the propagation relationship of each authorized node.

[0019] Compared with the prior art, the advantages and positive effects of this invention are: 1. Diagnostic safety and seamless integration with clinical practice: This invention fundamentally guarantees the diagnostic value of medical images through precise segmentation of key diagnostic regions and mask dilation. It can completely preserve the DICOM header file and ensure 100% compatibility with existing workflows such as PACS.

[0020] 2. High robustness for medical image processing: The present invention provides multi-resolution robustness in the selection of the DWT domain. Its adaptive quantization step size based on local variance enables the watermark to adapt to the image texture, remain invisible in smooth areas, and enhance robustness in textured areas, effectively resisting medical image-specific processing such as window width and window level adjustment and JPEG 2000 compression.

[0021] 3. Precise multi-level traceability capability: The parent hash value Hparent introduced in the watermark encoding of this invention is the key innovation. It makes each authorized distribution a link in the traceability chain, which can restore the complex multi-level shared path and realize precise traceability from the leak node to all parties involved.

[0022] 4. Blind extraction and strong security: The watermark extraction of this invention does not require the original image, making it highly practical. The embedding position and content of the watermark are encrypted and pseudo-randomized, effectively preventing detection and removal by unauthorized parties. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a system flow logic diagram of the present invention; Figure 2 This is a flowchart illustrating the process of embedding watermarks. Figure 3 This is a flowchart illustrating the process of tracing the origin of medical images. Detailed Implementation

[0025] 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, any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art to all other embodiments obtained without creative effort should be included within the protection scope of the present invention.

[0026] This invention discloses a blind watermark embedding and source tracing method for medical image data, such as... Figure 1 As shown, it constructs a system framework consisting of an "embedding end" and a "tracing end". The embedding end is integrated into the medical data sharing gateway or authorized client and is responsible for embedding an invisible traceability watermark into the DICOM image before it is distributed. The traceability end is an analysis platform responsible for extracting the watermark from the leaked image and reconstructing the circulation path.

[0027] The core technical process of this system revolves around two pillars: "preferential embedding of diagnostic security areas" and "traceability of multi-chain associations". The former ensures the invisibility and robustness of the watermark by diagnosing key area segmentation and DWT domain adaptive embedding algorithm; the latter achieves accurate positioning of leaked images and reconstruction of propagation paths by designing watermark encoding that includes spatiotemporal context and associated fingerprints and linking with background logs.

[0028] This invention can be deployed in hospital medical image archiving and communication systems (PACS), data anonymization systems, or research data management platforms. The embedded end, as a microservice or software module, is integrated into the image data export interface. When users (such as doctors or researchers) download DICOM images through an authorized approval process, the system automatically invokes the embedding process described in the core details. The embedding process is transparent to the user, outputting a DICOM-compliant image copy with a robust watermark. The traceability end acts as an independent security management platform. Upon discovering a suspected leaked DICOM image, security personnel upload the file through this platform. The platform automatically executes watermark extraction and decoding algorithms, and performs matching and correlation analysis with the backend authorization log database, ultimately generating a visualized image flow path report indicating which stages the image may have been leaked through. Through these methods, the system provides traceable and verifiable technical guarantees for the secure sharing and lifecycle management of medical image data.

[0029] The blind watermarking embedding and tracing method for medical image data mainly consists of two parts: 1. Embed watermarks into the original medical images; 2. Trace the source of watermarked medical images; like Figure 2 As shown, the watermark embedding operation on the original medical image includes the following steps: Step 1: DICOM Image Input and Parsing; (1) Input: Raw DICOM image .

[0030] (2) Parse the DICOM header file and extract pixel data to obtain the grayscale matrix. And read the existing ROI (Region of Interest) labels (if they exist).

[0031] Step 2: Diagnose key region segmentation; Generate a binary mask M, where Represents pixels Diagnostic critical areas that are prohibited from being embedded , This indicates that it belongs to the secure embedded area. .

[0032] (1) Through a pre-trained semantic segmentation model (e.g., U-Net) automatically identifies lesion areas. : ; (2) Perform morphological dilation on the segmented region to expand the protective boundary and generate the final mask: ; in, It is a structural element (such as a circular core). This represents the expansion operation. Ultimately, (Right now (The complement).

[0033] Step 3: Construct an enhanced source tracing information vector; Define traceability information For a tuple: ; in, A unique identifier representing the sharing entity. Indicates the authorization timestamp. Indicates the data usage encoding. A hash value representing the watermark information of the parent image (can be empty when shared for the first time), used to build a chain relationship.

[0034] Step 4: Encryption and error correction coding; (1) Sequence converted to byte stream .

[0035] (2) Use encryption algorithms Encryption is performed using algorithms based on chaotic systems (such as encryption algorithms based on chaotic systems) to obtain ciphertext. : ; (3) To Application of error correction coding (e.g., Reed-Solomon code) to generate the final watermark sequence. : ; Step 5: Transform domain transformation and coefficient selection; (1) For the security embedded area Perform L-level Discrete Wavelet Transform (DWT): ; (2) Selecting the intermediate frequency subband (For example Watermarks are embedded in the data to balance robustness and invisibility.

[0036] (3) Based on the embedded key Generate a pseudo-random sequence, and pseudo-randomly select N coefficients from the selected sub-band to form the carrier coefficient set. .

[0037] Step 6: Perceptually adaptive quantization index modulation; (1) Adaptive quantization step size: Based on the HVS model, the quantization step size is... Coefficient The local background complexity is proportional to the computational complexity. variance of the local region centered .

[0038]

[0039] in, and It is a preset constant that controls the range of step size.

[0040] (2) QIM embedding: for the selected coefficients and the watermark location to be embedded The quantization modulation function is: ; in, These are the coefficients after the watermark is embedded. This formula quantizes the coefficient values ​​to different interval centers, representing 0 and 1 respectively.

[0041] Step 7: Inverse Transform and Image Reconstruction; (1) Modify the coefficients Put it back into the original DWT sub-tape.

[0042] (2) Perform inverse discrete wavelet transform on the modified wavelet subband to reconstruct the secure embedding region. : ; (3) With unchanged prohibited embedding regions Merge to obtain a pixel matrix with watermark. .

[0043] (4) Packaged with the original DICOM header file, the final watermarked DICOM image is generated. .

[0044] like Figure 3 As shown, the process of tracing the source of watermarked medical images includes the following steps: Step 1: Blind watermark extraction; (1) Preprocessing and coefficient extraction: DICOM image to be detected Perform the same steps 1 (DICOM image parsing) and 5 (transform domain transformation and coefficient selection) as the embedding process, i.e., first parse the DICOM header file and extract pixel data, then perform multi-level DWT transformation on the secure embedding region, based on the embedding key. The generated pseudo-random sequence is used to select the set of carrier coefficients that perfectly match the embedding stage position. This provides a foundation for subsequent watermark extraction.

[0045] (2) QIM watermark extraction: For each carrier coefficient to be detected The watermark position is determined based on how close its value is to the quantization range. : ; in, Distance coefficient The nearest center point of the quantization interval corresponding to watermark bit 0. The center point of the quantization interval corresponding to watermark bit 1 is used; by comparing the distance between the coefficient and the two types of center points, the watermark information is restored bit by bit.

[0046] (3) Watermark sequence integration: integrate all extracted watermark positions By concatenating them in sequence, a complete sequence of extracted watermark bits is obtained. Complete the blind watermark extraction process.

[0047] Step 2: Decoding and path restoration; (1) Error correction decoding and decryption: the extracted watermark sequence Perform the two inverse operations sequentially: ① Call the error correction and decoding algorithm It corrects bit errors caused by image compression, format conversion, and other operations to recover the ciphertext sequence. ; ② Perform the inverse encryption operation using the decryption key. Decrypt the ciphertext into a byte stream It is then parsed into traceability information tuples according to a preset format. This completes the initial reconstruction of the source information.

[0048] (2) Chain-based tracing and authorization record matching: ①Restore the source information Accurately match the authorization logs in the traceability database to locate the authorization record (including authorizer ID, authorization time, purpose of use, etc.) corresponding to the current leak node. ② Check the traceability information Field: If the field is not empty (i.e., there is a parent watermark hash association), then recursively query the database for the parent authorization record that matches the hash value, tracing upwards step by step; if If empty, it means that the original source of the image has been traced.

[0049] ③ By recursively querying, a directed tracing chain is constructed that traces back from the leak node to the source of the image. : ; in, This is the current leak node. ( ) represents the intermediate authorization node. This is the original node generated from the image.

[0050] (3) Path integration and visualization output: If there is no parent watermark hash association, the authorization records of the current leak node and the source node are directly integrated; if there is a multi-level association, all node information obtained by recursive query is integrated to generate a complete medical image data flow path diagram. The final output is a visualized traceability report, which clearly shows the complete life cycle of the image from its original generation to the leak, including the identity of the authorizing party, authorization time, purpose of use and propagation relationship of each node, providing intuitive and verifiable technical basis for the determination of responsibility and investigation of the leak incident.

[0051] The present invention has the following beneficial effects: 1. Seamless Integration of Diagnostic Safety and Clinical Application: This invention fundamentally safeguards the diagnostic value of DICOM images through precise segmentation of critical diagnostic regions and mask dilation. It fully preserves the DICOM header file, ensuring 100% compatibility with existing workflows such as PACS.

[0052] 2. High robustness for medical image processing: The selection of the DWT domain in this invention provides multi-resolution robustness. The adaptive quantization step size based on local variance enables the watermark to adapt to image texture, remaining invisible in smooth areas and enhancing robustness in textured areas, effectively resisting medical image-specific processing such as window width and level adjustments and JPEG 2000 compression.

[0053] 3. Precise multi-level traceability capability: This invention introduces a parent hash value in the watermark encoding. This is the key innovation, which makes each authorized distribution a link in the traceability chain, enabling the reconstruction of complex multi-level sharing paths and achieving precise traceability from the point of leakage to all parties involved.

[0054] 4. Blind extraction and strong security: The watermark extraction operation in this invention does not require the original image, making it highly practical. The embedding position and content of the watermark are encrypted and pseudo-randomized, effectively preventing detection and removal by unauthorized parties.

Claims

1. A method for blind watermark embedding and tracing of medical image data, characterized in that: Includes the following steps: S1. Perform watermark embedding operation on the original medical image; S11. Analyze the original medical image and extract the pixel data to obtain the grayscale matrix. ; S12, Based on the grayscale matrix The diagnostic region in the original medical image is segmented to obtain safe embedding regions and prohibited embedding regions; S13. Construct traceability information tuples; S14. Encrypt the traceability information vector; S15. Perform transform domain transformation on the secure embedding region; S16. Select coefficients and watermark bits and perform quantization index modulation; S17. Perform inverse transformation of the secure embedding region and image reconstruction to generate a medical image with watermark. S2. Perform source tracing operations on watermarked medical images; S21. Perform watermark extraction on the medical image with embedded watermark; S22. Decode and restore the path of the extracted watermark.

2. The blind watermarking embedding and tracing method for medical image data as described in claim 1, characterized in that: Step S11 specifically includes the following steps: S121, Based on the grayscale matrix Generate a binary mask M, where, Represents pixels This is a prohibited embedding area. , Represents pixels Belongs to the secure embedded area ; S122. Automatically identify lesion regions using a pre-trained semantic segmentation model. Its formula is: ； S123, Regarding the lesion area A morphological dilation operation is performed to expand the protective boundary and generate the final mask. The formula is as follows: ； in, It is a structural element. This indicates the expansion operation.

3. The blind watermarking embedding and tracing method for medical image data as described in claim 2, characterized in that: In step S13, the source tracing information tuple is: ； in, A unique identifier representing the sharing entity. Indicates the authorization timestamp. Indicates the data usage encoding. The hash value representing the watermark information of the parent image.

4. The blind watermarking embedding and tracing method for medical image data as described in claim 3, characterized in that: Step S14 specifically includes the following steps: S141. Convert the source tracing information vector into a byte stream. ; S142. Use encryption algorithm byte stream Encryption is performed to obtain ciphertext. Its formula is: ； S143, Regarding the ciphertext Application of error correction coding Generate watermark sequence Its formula is: 。 5. The blind watermarking embedding and tracing method for medical image data as described in claim 4, characterized in that: Step S15 specifically includes the following steps: S151, For the secure embedded area Perform an L-level discrete wavelet transform to generate DWT subbands; the formula is as follows: ； S152. Select an intermediate frequency sub-band in the DWT sub-band as the watermark embedding area; S111, Based on the embedded key Generate a pseudo-random sequence, and pseudo-randomly select N carrier coefficients from the selected mid-frequency subband to form a carrier coefficient set. .

6. The blind watermarking embedding and tracing method for medical image data as described in claim 5, characterized in that: Step S16 specifically includes the following steps: S161. Based on the HVS model, calculate the selected carrier coefficient. variance of the local region centered The calculation formula is as follows: ； in, and It is a preset constant that controls the range of step size; S162. For the selected carrier coefficient and the watermark location to be embedded The quantization modulation function is: ； in, It is the carrier coefficient after embedding the watermark.

7. The blind watermarking embedding and tracing method for medical image data as described in claim 6, characterized in that: Step S71 specifically includes the following steps: S171, The carrier coefficient after embedding the watermark Put it back into the DWT sub-tape; S172. Perform inverse discrete wavelet transform on the modified DWT subbands to reconstruct the secure embedding region. Its formula is: ； S173, Embed the reconstructed security area With unchanged prohibited embedding regions Merge to obtain a pixel matrix with watermark. ; S174, watermarked pixel matrix Package the original medical image with the original image to generate a watermarked medical image.

8. The blind watermarking embedding and tracing method for medical image data as described in claim 7, characterized in that: Step S21 specifically includes the following steps: S211. Analyze the watermarked medical image and extract pixel data to obtain the prohibited embedding region and the safe embedding region; S212. Perform multi-level discrete wavelet transform on the secure embedding region, based on the embedding key. The generated pseudo-random sequence is used to select the set of carrier coefficients that perfectly match the embedding stage position. ; S213, For each carrier coefficient to be detected The watermark position is determined based on how close its value is to the quantization range. Its formula is: ； in, The center point of the quantization interval corresponding to watermark bit 0 that is closest to the coefficient of the carrier to be detected. This is the center point of the quantization interval corresponding to watermark bit 1; S214, All watermark positions By concatenating them in sequence, a watermark sequence is obtained. This enables the extraction of watermarks.

9. The blind watermarking embedding and tracing method for medical image data as described in claim 8, characterized in that: Step S11 specifically includes the following steps: S221, Using error correction decoding algorithm For watermark sequence Correction was performed to recover the ciphertext. ; S222. Perform the inverse encryption operation using the decryption key. , ciphertext Decrypt into byte stream It is then parsed into traceability information tuples according to a preset format. ; S223, Source tracing information tuple Match the authorization logs with the source tracing database to locate the authorization record corresponding to the current leak node; S224, Check the traceability information tuple In field, if the If the field is not empty, recursively query the source database for the next higher level authorization record that matches the hash value, tracing upwards step by step; if the... If the field is empty, it means that the medical image has been traced back to its source. S225. Construct a directed tracing chain from the current leak node back to the source of the medical image through recursive queries. Its formula is: ； in, This is the current leak node. ( ) represents the intermediate authorization node. This is the original generation node for medical images; S226. If there is no hash value association, the authorization records of the current leaked node and the original medical image generation node are integrated; if there is a multi-level hash value association, all authorized node information obtained by recursive query is integrated; a medical image data flow path diagram is generated based on the integration result and a visual traceability report is output.

10. The blind watermarking embedding and tracing method for medical image data as described in claim 9, characterized in that: In step S26, the content of the visual traceability report includes the identity of the authorizing party, the authorization time, the purpose of use, and the propagation relationship of each authorized node.

Images