Watermark embedding and extracting method for multi-format electronic chart data

Abstract

The invention provides a watermark embedding and extracting method for multi-format electronic chart data, and the watermark embedding process comprises the steps: carrying out the preprocessing of the electronic chart data, extracting visual features through a convolutional neural network, carrying out the dimension reduction of the visual features through principal component analysis, and embedding a watermark through discrete cosine transform. According to the method, the defects in the aspects of theft prevention and safety management under various electronic chart data formats can be overcome, the copyright and safety of the electronic chart data can be effectively protected, and the method is suitable for various electronic chart data formats.

Description

Technical Field

[0001] The present invention specifically relates to the technical field of watermarking of electronic nautical chart data, and more specifically refers to a watermark embedding and extraction method for multi-format electronic nautical chart data. Background Art

[0002] As a core tool for navigation safety, nautical charts provide key information such as marine topography, waterways, obstacles, and aids to navigation, and play an irreplaceable role in ship navigation, route planning, and maritime traffic management. With the rapid development of digital technology, the popularity of electronic nautical charts has also brought a series of challenges to copyright protection. Electronic nautical charts are easy to copy and disseminate, and unauthorized use and piracy pose a serious threat to the intellectual property rights and commercial interests of nautical chart data.

[0003] As an effective means of digital content protection, watermark technology has been widely used in multimedia fields such as images, audio, and video by embedding invisible information in the original data to identify the copyright owner or usage conditions of the data. However, the diversity of electronic charts, including vector data, image data, document data, and tile data, as well as the special requirements of different formats for watermark technology, make it difficult for existing watermark technology to fully adapt to all electronic chart data formats, and cannot fully meet the needs of electronic chart anti-theft and security management.

[0004] The challenge of electronic chart data theft prevention and security management is how to provide a method that does not affect data quality, is highly robust, can effectively trace the source to the responsible unit, and can adapt to a variety of electronic chart data formats. This requires that watermark technology should not only ensure that the watermark is not destroyed during data transmission and processing, and does not affect the accuracy of the chart data, but also be able to be embedded in chart data of different formats.

[0005] Therefore, it is particularly urgent to develop an anti-theft and security management method for electronic chart data that does not affect data quality, can adapt to a variety of chart data formats, and can effectively protect the copyright of chart data. This watermarking method needs to be highly adaptable and robust to ensure that watermark information can be effectively embedded and extracted in various electronic chart data formats, thereby protecting the copyright and security of chart data. Summary of the invention

[0006] 1. Technical issues

[0007] The purpose of the present invention is to provide a watermark embedding and extraction method for multi-format electronic nautical chart data, so as to solve the problem that the prior art cannot perform anti-theft and security management on multiple electronic nautical chart data formats.

[0008] 2. Technical content

[0009] In order to solve the above technical problems, the technical solution of the present invention is: a watermark embedding and extraction method for multi-format electronic nautical chart data, comprising the following steps:

[0010] Step S1: pre-processing the electronic nautical chart that needs to be embedded with a watermark, and taking different pre-processing measures for different types of electronic nautical chart data;

[0011] Step S2: using the pre-trained convolutional neural network (CNN) model to process the pre-processed electronic chart through a forward propagation method to extract visual feature vectors;

[0012] Step S3: Applying the principal component analysis method to perform dimensionality reduction processing on the extracted high-dimensional visual feature vector, and converting the high-dimensional feature vector into a low-dimensional feature vector of fixed length;

[0013] Step S4: Select a watermark string with copyright information from a preset database, and encode the string into a watermark bit code through a hash function;

[0014] Step S5: In the frequency domain of the feature vector, a discrete cosine transform (DCT) is used to select a number of discontinuous positions in the low-frequency region, and the watermark bit code is embedded in these positions;

[0015] Step S6: Use the inverse CNN to reconstruct the feature vector embedded with the watermark into the electronic chart data and convert it into the original format using the inverse processing method.

[0016] Preferably, in step S1, for vector data, rasterization technology is used to convert it into image format, for image data, standardization and normalization are performed, for document data, image elements are separated and converted into raster images, and for tile data, standardization and normalization are performed.

[0017] Preferably, in step S2, a pre-trained CNN model is selected as a feature extractor, the electronic chart data processed in step 1 is input into the CNN model, a deeper intermediate layer in the model is selected as a feature extraction layer, and a high-dimensional visual feature vector X output by the selected layer is obtained through forward propagation.

[0018] Preferably, in step S3, the extracted feature vector X is reduced in dimension using principal component analysis and converted into a feature vector of fixed length. First, the feature vector X is standardized so that the mean of each feature is 0 and the variance is 1, and the formula is:

[0019]

[0020] Among them, μ is the mean of each feature, and σ is the standard deviation of each feature.

[0021] Preferably, in step S3, the normalized feature vector X is calculated. std The covariance matrix ∑ of is:

[0022]

[0023] Here, n is the number of samples.

[0024] Preferably, in step S3, the covariance matrix ∑ is subjected to eigenvalue decomposition to obtain the eigenvalue λ i and the corresponding eigenvector v i , according to the size of the eigenvalue, select the eigenvectors v1, v2, ..., v corresponding to the first k largest eigenvalues k These eigenvectors form a new coordinate system. Finally, the eigenvector X std Projecting into the new coordinate system, we get the reduced-dimensional feature vector Y, where k is the selected fixed length.

[0025] Preferably, in step S4, a pre-stored watermark string is extracted from the database, the watermark string including key information such as the management ID, organization name, and control level of the electronic nautical chart, and the watermark string is converted into a binary bit code w through a hash function so as to be embedded in the feature vector of the electronic nautical chart.

[0026] Preferably, in step S5, the feature vector Y is subjected to discrete cosine transform to obtain its representation Y in the frequency domain. D , manually select the i-th to i+k-1-th positions as embedding positions. Define the embedding strength factor α, and embed the watermark bit code w into the i-th to i+k-1-th positions:

[0027] Y' D [i+j]=Y D [i+j]+α×w[j]

[0028] Among them, Y' D is the frequency domain coefficient after embedding the watermark, and w[j] is the jth bit of the watermark bit code.

[0029] Preferably, in step S5, Y' D Perform inverse discrete cosine transform to obtain the feature vector Y' after embedding the watermark.

[0030] Preferably, in step S6, the feature vector embedded with the watermark is restored to the electronic chart data using an inverse CNN. The inverse CNN is a decoder model whose structure matches the encoder part of the CNN used for feature extraction in step 2. The operation of each layer can be expressed as:

[0031] Z i+1 =ReLU(Wi *Z i +b i )

[0032] Among them, Z i and W i They represent the output feature map and bias vector of the i-th layer respectively, * represents the convolution operation, and ReLU represents the activation function.

[0033] Preferably, in step S6, an image can be reconstructed from the feature vector Y' after embedding the watermark through reverse CNN. For vector data, the image format is converted back to vector data using the reverse rasterization technology, and for image data, the reverse normalization process is performed to restore the original size and pixel value, and for document data, the image elements are reassembled back to the document format, and for tile data, the reverse normalization process is performed to restore the original size and pixel value.

[0034] Another aspect of the present invention provides a watermark extraction method for multi-format electronic nautical chart data. The watermark extraction process is the inverse process of the embedding process. The specific steps are as follows:

[0035] Step S7: pre-processing the electronic nautical chart from which the watermark needs to be extracted, and taking different pre-processing measures for different types of electronic nautical chart data;

[0036] Step S8: using the pre-trained CNN model, processing the pre-processed electronic chart through the forward propagation method to extract the visual feature vector;

[0037] Step S9: Applying the principal component analysis method to perform dimensionality reduction processing on the extracted high-dimensional visual feature vector, and converting the high-dimensional feature vector into a low-dimensional feature vector of fixed length;

[0038] Step S10: In the frequency domain of the feature vector, use DCT to extract the watermark bit code at the selected position in the low-frequency region;

[0039] Step S11: Decode the extracted watermark bit code into a watermark string through an inverse hash function, and compare it with the original watermark string in the database.

[0040] Preferably, in step S7, the vector data embedded with the watermark is converted into an image format using rasterization technology, and the image data embedded with the watermark is standardized and normalized. For the document data embedded with the watermark, the image elements are separated and converted into a raster image, and the tile data embedded with the watermark is standardized and normalized.

[0041] Preferably, in step S8, a pre-trained CNN model is selected as a feature extractor, and the watermarked electronic chart data processed in step 1 is input into the CNN model, and the CNN model should be consistent with the CNN model in step 1 of the watermark embedding process. The same intermediate layer as in step 2 of the watermark embedding process is selected as the feature extraction layer, and the high-dimensional visual feature vector X' output by the selected layer is obtained through forward propagation.

[0042] Preferably, in step S9, the extracted feature vector X' is reduced in dimension using principal component analysis and converted into a feature vector of fixed length. The processing steps are consistent with step 3 of the embedding process, and a feature vector Y' containing a watermark after dimension reduction is obtained.

[0043] Preferably, in step S10, the feature vector Y' containing the watermark is subjected to discrete cosine transform to obtain its representation Y' in the frequency domain. D , locate the embedding position in step five of the watermark embedding process and extract the watermark information w.

[0044] Preferably, in step S11, the encoding scheme in step 4 of the watermark embedding process is used to reconstruct the extracted watermark information w into the original watermark string using an inverse hash function. The database is queried to retrieve matching pre-stored watermark strings. Key information such as the management ID, organization name, and control level of the electronic chart is obtained. Thus, the copyright ownership and the responsible unit of the electronic chart are confirmed.

[0045] Preferably, the classification and format of the electronic chart data include: chart data, in the format of S57 or 000; vector data, in the format of Shp, Dwg; image data, in the format of Tif, Jpg, Png; document data, in the format of Pdf, Docx, Xlsx; tile data, in the format of vector tile, raster tile.

[0046] 3. Technical Effect

[0047] Compared with the prior art, the advantages of the present invention are as follows: the present invention provides a watermark embedding and extraction method for multi-format electronic nautical chart data, which can effectively process and embed watermarks in different types of electronic nautical chart data, including vector data, image data, document data and tile data. This compatibility can well realize the copyright protection and security management of electronic nautical charts. At the same time, by using the CNN model to extract visual feature vectors and combining principal component analysis for dimensionality reduction processing, the present invention can embed watermark information into the low-frequency area of ​​the feature vector without affecting the visual quality of the electronic nautical chart, thereby improving the concealment of the watermark. During the watermark embedding process, DCT is used to select discontinuous positions in the frequency domain of the feature vector to embed the watermark bit code. This method enhances the watermark's resistance to common image processing operations and ensures the integrity and recoverability of the watermark information. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The present invention is a watermark embedding flow chart of a watermark embedding and extraction method for multi-format electronic nautical chart data.

[0049] Figure 2 The present invention is a watermark extraction flow chart of a watermark embedding and extraction method for multi-format electronic nautical chart data.

[0050] Figure 3 This is an experimental data diagram of a watermark embedding and extraction method for multi-format electronic nautical chart data of the present invention. Specific embodiments

[0051] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0052] This embodiment selects an electronic chart data in an image format as experimental data, such as Figure 3 shown.

[0053] (I) Watermark embedding process Figure 1 As shown, a watermark embedding and extraction method for multi-format electronic chart data is provided, and the specific steps are as follows:

[0054] Step S1: pre-processing the electronic nautical chart that needs to be embedded with a watermark, and taking different pre-processing measures for different types of electronic nautical chart data;

[0055] Step S2: using the pre-trained CNN model, processing the pre-processed electronic chart through the forward propagation method to extract the visual feature vector;

[0056] Step S3: Applying the principal component analysis method to perform dimensionality reduction processing on the extracted high-dimensional visual feature vector, and converting the high-dimensional feature vector into a low-dimensional feature vector of fixed length;

[0057] Step S4: Select a watermark string with copyright information from a preset database, and encode the string into a watermark bit code through a hash function;

[0058] Step S5: In the frequency domain of the feature vector, use DCT to select several discontinuous positions in the low-frequency region and embed the watermark bit code into these positions;

[0059] Step S6: Use the inverse CNN to reconstruct the feature vector embedded with the watermark into the electronic chart data and convert it into the original format using the inverse processing method.

[0060] The classification and format of the electronic chart data include: chart data, format is S57 or 000; vector data, format is Shp, Dwg; image data, format is Tif, Jpg, Png; document data, format is Pdf, Docx, Xlsx; tile data, format is vector tile, raster tile.

[0061] It should be noted that this embodiment uses a JPEG format electronic chart data with a resolution of 1024x1024 pixels. After normalization, the pixel values ​​of the image are adjusted to the range of [0,1], with a mean of 0.5 and a standard deviation of 0.2.

[0062] The VGG16 model pre-trained on the ImageNet dataset is used as a feature extractor, and the electronic chart data processed in step S1 is input into the VGG16 model. The 10th layer is selected as the feature extraction layer. After forward propagation, a vector X containing 512 features is obtained, and the dimension of each feature is 32×32.

[0063] The principal component analysis method is used to reduce the dimension of the extracted feature vector X and convert it into a feature vector of fixed length. First, the feature vector X is standardized so that the mean of each feature is 0 and the variance is 1. The formula is:

[0064]

[0065] Among them, μ is the mean of each feature, and σ is the standard deviation of each feature.

[0066] Then calculate the normalized eigenvector X std The covariance matrix ∑ of is:

[0067]

[0068] Here, n is the number of samples.

[0069] Perform eigenvalue decomposition on the covariance matrix ∑ again to obtain the eigenvalue λ i and the corresponding eigenvector v i , according to the size of the eigenvalue, select the eigenvectors υ1, υ2, …, υ corresponding to the first 128 largest eigenvalues 128 These eigenvectors form a new coordinate system. Finally, the eigenvector X std Projecting into the new coordinate system, we get the reduced-dimensional feature vector Y, where 128 is the selected fixed length.

[0070] Extract a watermark string containing the management ID "CHART12345", the organization name "NAVY", and the control level "SECRET" from the database. Use the MD5 hash function to encode the string into a 128-bit watermark bit code w.

[0071] Perform discrete cosine transform on the feature vector Y to obtain its representation Y in the frequency domain D , manually select the 10th to 137th bits as the embedding position. Define the embedding strength factor α as 0.15, embed the watermark bit code w into the 10th to 137th bits, and get the frequency domain coefficient Y' after embedding the watermark D :

[0072] Y' D [i+j]=Y D [i+j]+α×w[j]

[0073] Where w[j] is the jth bit of the watermark code.

[0074] Y' D Perform an inverse discrete cosine transform to obtain the feature vector Y' after embedding the watermark. Step 6: Use the inverse CNN to reconstruct the image from the feature vector Y' embedded with the watermark. The inverse CNN is a decoder model whose structure matches the encoder part of the VGG16 model used for feature extraction in step 2. The operation of each layer can be expressed as:

[0075] Z i+1 =ReLU(W i *Z i +b i )

[0076] Among them, Z i and W i They represent the output feature map and bias vector of the i-th layer respectively, * represents the convolution operation, and ReLU represents the activation function.

[0077] Finally, the inverse normalization process is performed to restore the electronic chart data to its original size and pixel value.

[0078] (ii) The watermark extraction process is the reverse process of the embedding process, such as Figure 2 As shown, a watermark extraction method for multi-format electronic chart data is provided, and the specific steps are as follows:

[0079] Step S7: pre-processing the electronic nautical chart from which the watermark needs to be extracted, and taking different pre-processing measures for different types of electronic nautical chart data;

[0080] Step S8: using the pre-trained CNN model, processing the pre-processed electronic chart through the forward propagation method to extract the visual feature vector;

[0081] Step S9: Applying the principal component analysis method to perform dimensionality reduction processing on the extracted high-dimensional visual feature vector, and converting the high-dimensional feature vector into a low-dimensional feature vector of fixed length;

[0082] Step S10: In the frequency domain of the feature vector, use DCT to extract the watermark bit code at the selected position in the low-frequency region;

[0083] Step S11: Decode the extracted watermark bit code into a watermark string through an inverse hash function, and compare it with the original watermark string in the database.

[0084] It should be noted that the extraction step uses the JPEG format electronic chart data that has been embedded with a watermark during the embedding process. After standardization, the pixel values ​​of the image are adjusted to the range of [0,1], with a mean of 0.5 and a standard deviation of 0.2.

[0085] The same pre-trained VGG16 model as in the embedding process is used, and the same 10th layer is selected as the feature extraction layer to process the standardized image. After forward propagation, a vector X' containing 512 features containing the watermark is obtained, and the dimension of each feature is 32×32.

[0086] The feature vector X' is processed by the same principal component analysis as in the embedding process to obtain the feature vector Y' containing the watermark after dimensionality reduction. The feature vector Y' containing the watermark is subjected to discrete cosine transform to obtain its representation Y' in the frequency domain D , locate the embedding position of the 10th to 137th bits in the watermark embedding process and extract the watermark information w.

[0087] The extracted watermark information w is reconstructed into the original watermark string using the inverse hash function. By querying the database, the result matching the pre-stored watermark string is retrieved, and it is confirmed that the management ID of the electronic chart is "CHART12345", the organization name is "NAVY", and the control level is "SECRET".

[0088] The present invention and its embodiments are described above. Such description is not restrictive. What is shown in the accompanying drawings is only one of the embodiments of the present invention. In short, if ordinary technicians in the field are inspired by it and do not deviate from the purpose of the invention, they can creatively design structural methods and embodiments similar to the technical solution, which should all fall within the scope of protection of the invention.

Claims

1. A watermark embedding method for multi-format electronic chart data, characterized in that: The following steps are involved: S1: Preprocess the electronic nautical charts that need to be embedded with watermarks, and take different preprocessing measures for different types of electronic nautical chart data; S2: Use the pre-trained convolutional neural network model to process the pre-processed electronic chart through the forward propagation method to extract the visual feature vector; S3: Apply the principal component analysis method to reduce the dimensionality of the extracted high-dimensional visual feature vector and convert the high-dimensional feature vector into a low-dimensional feature vector of fixed length; S4: Select a watermark string with copyright information from a preset database, and encode the string into a watermark bit code through a hash function; S5: In the frequency domain of the feature vector, a number of discontinuous positions are selected in the low-frequency region using discrete cosine transform, and the watermark bit code is embedded in these positions; S6: The feature vector embedded with watermark is reconstructed into the electronic chart data using the inverse CNN and converted into the original format using the inverse processing method.

2. The watermark embedding method for multi-format electronic chart data according to claim 1 is characterized in that: In step S1, for vector data, rasterization technology is used to convert it into image format, for image data, standardization and normalization processing is performed, for document data, image elements are separated and converted into raster images, and for tile data, standardization and normalization processing is performed.

3. The watermark embedding method for multi-format electronic nautical chart data according to claim 1 is characterized by: In step S2, the pre-trained CNN model is selected as the feature extractor, the electronic chart data processed in step S1 is input into the CNN model, a deeper intermediate layer in the model is selected as the feature extraction layer, and the high-dimensional visual feature vector X output by the selected layer is obtained through forward propagation.

4. The watermark embedding method for multi-format electronic chart data according to claim 3 is characterized by: In step S3, the extracted feature vector X is reduced in dimension using principal component analysis and converted into a feature vector of fixed length. The feature vector X is first standardized so that the mean of each feature is 0 and the variance is 1. The formula is: Among them, μ is the mean of each feature, and σ is the standard deviation of each feature; In step S3, the normalized feature vector X is calculated. std The covariance matrix ∑ of is: Where n is the number of samples; In step S3, the covariance matrix ∑ is subjected to eigenvalue decomposition to obtain the eigenvalue λ i and the corresponding eigenvector v i , according to the size of the eigenvalue, select the eigenvectors v1, v2, ..., v corresponding to the first k largest eigenvalues k These eigenvectors form a new coordinate system. Finally, the eigenvector X std Projecting into the new coordinate system, we get the reduced-dimensional feature vector Y, where k is the selected fixed length.

5. The watermark embedding method for multi-format electronic chart data according to claim 1 is characterized in that: In step S4, a pre-stored watermark string is extracted from the database, the watermark string including the management ID, organization name, and control level key information of the electronic nautical chart. The watermark string is converted into a binary bit code w through a hash function so as to be embedded in the feature vector of the electronic nautical chart.

6. The watermark embedding method for multi-format electronic chart data according to claim 4 is characterized by: In step S5, the feature vector Y is subjected to discrete cosine transform to obtain its representation Y in the frequency domain. D , manually select the i-th to i+k-1-th positions as the embedding positions, define the embedding strength factor α, and embed the watermark bit code w into the i-th to i+k-1-th positions: Y' D [i+j]=Y D [i+j]+α×w[j] Among them, Y' D is the frequency domain coefficient after embedding the watermark, and w[j] is the jth bit of the watermark bit code; In step S5, Y' D Perform inverse discrete cosine transform to obtain the feature vector Y' after embedding the watermark.

7. The watermark embedding method for multi-format electronic chart data according to claim 6 is characterized by: In step S6, the feature vector embedded with the watermark is restored to the electronic chart data using the reverse CNN. The reverse CNN is a decoder model whose structure matches the encoder part of the CNN used for feature extraction in step 2. The operation of each layer can be expressed as: WITH i+1 =ReLU(W i *WITH i +b i ) Among them, Z i and W i They represent the output feature map and bias vector of the i-th layer respectively, * represents the convolution operation, and ReLU represents the activation function; In step S6, the image can be reconstructed from the feature vector Y' after the watermark is embedded through the reverse CNN. For vector data, the image format is converted back to vector data using the inverse rasterization technology. For image data, the inverse normalization process is performed to restore the original size and pixel values. For document data, the image elements are reassembled back to the document format. For tile data, the inverse normalization process is performed to restore the original size and pixel values.

8. A watermark extraction method for multi-format electronic chart data, characterized in that: The following steps are involved: S7: pre-processing the electronic nautical chart from which the watermark needs to be extracted, and taking different pre-processing measures for different types of electronic nautical chart data; S8: Use the pre-trained convolutional neural network model to process the pre-processed electronic chart through the forward propagation method to extract the visual feature vector; S9: Apply the principal component analysis method to reduce the dimensionality of the extracted high-dimensional visual feature vector and convert the high-dimensional feature vector into a low-dimensional feature vector of fixed length; S10: in the frequency domain of the feature vector, using discrete cosine transform to extract the watermark bit code at a selected position in the low frequency region; S11: Decode the extracted watermark bit code into a watermark string through an inverse hash function, and compare it with the original watermark string in the database.

9. The watermark extraction method for multi-format electronic chart data according to claim 8, characterized in that: In step S7, the vector data embedded with the watermark is converted into an image format using a rasterization technique, the image data embedded with the watermark is standardized and normalized, the document data embedded with the watermark is separated and converted into a raster image, and the tile data embedded with the watermark is standardized and normalized; In step S8, the pre-trained CNN model is selected as the feature extractor, and the watermarked electronic chart data processed in step 1 is input into the CNN model. The CNN model should be consistent with the CNN model in step 1 of the watermark embedding process. The same intermediate layer as that in step 2 of the watermark embedding process is selected as the feature extraction layer, and the high-dimensional visual feature vector X' output by the selected layer is obtained through forward propagation; In step S9, the extracted feature vector X' is reduced in dimension using the principal component analysis method and converted into a feature vector of fixed length. The processing steps are consistent with step 3 of the embedding process to obtain a feature vector Y' containing a watermark after dimension reduction. In step S10, the feature vector Y' containing the watermark is subjected to discrete cosine transform to obtain its representation Y' in the frequency domain. D , locate the embedding position in step 5 of the watermark embedding process and extract the watermark information w; In step S11, the encoding scheme in step 4 of the watermark embedding process is used, and the extracted watermark information w is reconstructed into the original watermark string using an inverse hash function. The database is queried to retrieve and match the pre-stored watermark string to obtain the key information such as the management ID, organization name, and control level of the electronic nautical chart, thereby confirming the copyright ownership and responsible unit of the electronic nautical chart.

10. The watermark embedding method for multi-format electronic chart data according to claim 1, characterized in that: The classification and format of the electronic chart data include: chart data, format is S57 or 000; vector data, format is Shp, Dwg; image data, format is Tif, Jpg, Png; document data, format is Pdf, Docx, Xlsx; tile data, format is vector tile, raster tile.

Images