A GF-2 image watermarking method using improved SIFT feature points

Abstract

The present invention discloses a GF‑2 image watermarking method using improved SIFT feature points, including: a screening method for SIFT feature points of GF‑2 images and a GF‑2 image copyright protection watermarking method; the feature point screening method is: extracting SIFT feature points from the original GF‑2 image, using the Mean Shift method to perform feature point clustering optimization, using the screening results as new SIFT feature points and determining feature areas; the GF‑2 image copyright protection watermarking method is: constructing SIFT feature areas with key points as watermark embedding domains. When embedding the watermark, the feature area is decomposed into sub-bands using NSCT, a low-frequency sub-band is selected and singular value decomposition is performed, and an additive watermark embedding method is used to embed the singular values ​​of the watermark image into the above-mentioned sub-band singular values. Experiments show that the method extracts a moderate number of feature points in the GF‑2 image, has high efficiency, and has high robustness to common watermark geometric attacks, and can be used for GF‑2 image digital watermark copyright protection.

Description

Technical Field

[0001] The invention relates to the technical field of cartography and geographic information system, and relates to a GF-2 image watermarking method using improved SIFT feature points. Background Art

[0002] In recent years, with the rapid development of remote sensing science and technology, high-resolution remote sensing images have become an important technical means for earth observation, and remote sensing is developing towards high spatial resolution, temporal resolution, spectral resolution and radiation resolution. As the first satellite in the earth observation system to break through the sub-meter level, the successful launch and use of the Gaofen-2 (GF-2) satellite marks the localization of high-resolution images and reduces the dependence on foreign high-resolution data. GF-2 images have the characteristics of high resolution, high precision and wide application fields, and play an important role in land and resources utilization, land planning, agriculture and ocean applications. High-resolution remote sensing images are widely used to give full play to the value of data. However, data leakage, piracy and illegal dissemination make it difficult to ensure the safe use of data, and even threaten national defense construction and national security. Digital watermarking, as a new technology in the field of information security, provides technical guarantee for the safe use of data.

[0003] Since a GF-2 image often cannot achieve full coverage of the study area, in this case, in order to meet the needs of remote sensing mapping and theoretical research, it is usually necessary to crop and splice the image. In addition, image processing can also cause slight geometric distortion of the image data, such as translation, rotation, and deformation. Geometric attacks can cause the synchronization information of the watermark to be destroyed, making the watermark difficult to detect and extract. Therefore, the high-resolution image watermarking method needs to solve the problem of resisting geometric attacks.

[0004] Solutions to this type of problem can be divided into three categories: (1) embedding the watermark into an invariant domain; (2) embedding a mapping template that reflects the geometric transformation parameters of the image; (3) extracting invariant feature points in the image and embedding the watermark into an invariant feature region constructed with the feature points as the center. Among them, the digital watermarking method based on feature points is a commonly used method for watermark embedding and detection. Scale-Invariant Feature Transform (SIFT), SIFT is an image feature point extraction operator based on scale space. The feature points extracted by this operator have the characteristics of large amount of information and high matching accuracy. At the same time, it is also invariant to operations such as image scaling, rotation, translation and brightness change, and is widely used in anti-geometric attacks. In order to enhance the robustness of the method, this method selects the SIFT feature point extraction method, constructs the feature region with feature points, and embeds the watermark in the invariant domain of the feature region. Compared with the spatial domain watermarking method, the transform domain watermarking method has better imperceptibility and robustness. For example, the Discrete Wavelet Transformation (DWT) can decompose the image into low-frequency information and high-frequency information, and is widely used in robust watermarking methods. However, since DWT cannot effectively represent the contour and edge information of the image, there are certain limitations when DWT is applied to remote sensing image watermarking technology. The Non-Subsampled Contourlet Transform (NSCT) can effectively represent the edge structure of two-dimensional and multi-dimensional signals, making up for the shortcomings of DWT in terms of directional representation and multi-resolution characteristics. At the same time, compared with the Contourlet transform, NSCT can avoid spectral aliasing and better express the content of image data. Applying it to the GF-2 image watermarking technology is expected to obtain better watermark embedding and detection effects.

[0005] The present invention proposes a method based on SIFT, which adapts the SIFT feature point watermark method to the GF-2 image characteristics and realizes the GF-2 image anti-geometric watermark method. Summary of the invention

[0006] In view of this, this paper combines the idea of ​​digital watermarking based on feature points with the features of GF-2 images, and proposes a GF-2 image watermarking method using improved SIFT feature points. This method first extracts SIFT feature points of GF-2 images, then uses the Mean Shift clustering method to filter and screen the feature points, and finally uses the singular value decomposition (SVD) to embed the watermark information into the optimized feature area. Considering the stability and geometric invariance of SIFT feature points in the feature area, and the multi-directionality, multi-resolution and approximate invariance of NSCT, this method combines the excellent properties of the two and applies them to digital watermarking technology, achieving a new breakthrough in the anti-geometric aspect of GF-2 image copyright protection.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A GF-2 image watermarking method using improved SIFT feature points, comprising:

[0009] 1. The steps for screening SIFT feature points are as follows:

[0010] S1: Use SIFT feature point extraction operator to extract GF-2 image feature points;

[0011] S2: The extracted GF-2 image feature point coordinates are used as training samples and clustered using the Mean Shift method;

[0012] S3: Calculate the average SIFT descriptor for key points in the same cluster;

[0013] S4: After clustering SIFT feature points, after determining their position (x1, y1), direction, and scale S, a circular area is constructed with R as the radius;

[0014] S5: The distance D of the feature points will be screened once;

[0015] 2. The steps of generating, embedding and detecting GF-2 image copyright protection watermark are as follows:

[0016] S6: Select the Arnold transformation method to perform scrambling processing on the copyright watermark;

[0017] S7: Read the original data. Select the GF-2 image Q0, Q0{g(i,j),0≤i≤M,0≤j≤N}, g(i,j) is the image pixel value, and the size is M×N;

[0018] S8: Read binary watermark information. The watermark information is W, and the length of the watermark information is Z×L;

[0019] S9: Use NSCT to extract the low-frequency subband L0 of the feature area, perform SVD processing, and obtain its singular value λ0;

[0020] S10: Extract the singular value λ of the watermark information W using SVD w (Save U W With V W is the key K3) and is embedded in the singular values ​​of the low-frequency subband L1;

[0021] S11: The low frequency subband singular value λ of the singular value of the embedded watermark information L′ Perform inverse SVD to obtain L1;

[0022] S12: Perform inverse NSCT transformation on the low-frequency subband to restore it to the selected SIFT feature area, and obtain the final watermarked image Q1;

[0023] S13: End.

[0024] It can be seen from the above technical scheme that, compared with the prior art, the present invention discloses a GF-2 image watermarking method using improved SIFT feature points. This scheme has strong robustness against common geometric attacks while maintaining image quality, and solves the problem that the GF-2 image copyright protection watermarking method is not good at resisting geometric attacks. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0026] Figure 1 A GF-2 image watermarking method process using improved SIFT feature points provided by the present invention;

[0027] Figure 2 NSCT principle diagram provided by the present invention;

[0028] Figure 3 The extracted watermark image provided by the present invention;

[0029] Figure 4 The present invention provides a watermarked GF-2 image. DETAILED DESCRIPTION

[0030] The following will be combined with the 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 described embodiments 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 creative work are within the scope of protection of the present invention.

[0031] See attached Figure 1 The embodiment of the present invention discloses a GF-2 image watermarking method using improved SIFT feature points. The method involved in the present invention is further described below:

[0032] S1: Use SIFT feature point extraction operator to extract GF-2 image feature points;

[0033] S2: The extracted GF-2 image feature point coordinates are used as training samples and clustered using the Mean Shift method;

[0034] After determining the position, scale, direction and other information of the extracted feature points, the SIFT operator can construct an n×128-dimensional feature descriptor, where n is the number of feature points extracted in the image. Since not all feature points extracted by the SIFT method are suitable for watermark information embedding, and the feature areas overlap, the present method is invented to filter and screen feature points.

[0035] S3: Calculate the average SIFT descriptor for the key points in the same cluster, the calculation formula is:

[0036]

[0037] where d k is the SIFT descriptor in the cluster group, is the average SIFT descriptor in cluster i, j is the 128-dimensional index in the descriptor, and θ h is the set of feature points in the cluster, |θ h | is the cardinality of key points with similar positions.

[0038] S4: After clustering SIFT feature points, after determining their position (x1, y1), direction, and scale S, a circular area is constructed with R as the radius. The formula is as follows:

[0039] (xx i ) 2 +(yy i ) 2 =(Rs) 2 (2)

[0040] S5: The distance D of the feature points will be screened once, and the formula is as follows:

[0041]

[0042] Where M and N are the length and width of the image, and k is a constant;

[0043] 2. The steps of generating, embedding and detecting GF-2 image copyright protection watermark are as follows:

[0044] S6: Select the Arnold transformation method to perform scrambling processing on the copyright watermark;

[0045] In view of the interpretation requirements of GF-2 images, the spatial correlation cannot be affected when embedding watermarks, so the watermark image needs to be scrambled first. The present invention first selects the Arnold transform method for scrambling processing, and destroys the correlation of the watermark image by disrupting the pixel position and grayscale in its spatial domain, and then embeds it into the low-frequency subband of the GF-2 image to encrypt the original watermark and improve the security of the embedded watermark. The Arnold transform is defined as follows:

[0046]

[0047] Where (x, y) is the pixel coordinates in the original watermark image, and (x', y') is the pixel coordinates after transformation. The Arnold transformation period is T. If T / 2 transformations are performed, the watermark image has the largest scrambling degree and the strongest robustness; after T transformations, the original watermark image is obtained again. This method performs transformations for K times, and K is saved as a key for watermark extraction.

[0048] S7: Read the original data. Select the GF-2 image Q0, Q0{g(i,j),0≤i≤M,0≤j≤N}, g(i,j) is the image pixel value, and the size is M×N;

[0049] S8: Read binary watermark information as W, the length of the watermark information is Z×L;

[0050] S9: Use NSCT to extract the low-frequency subband L0 of the feature area, perform SVD processing, and obtain its singular value λ0;

[0051] The non-subsampled Contourlet transform (NSCT) can effectively represent the edge structure of two-dimensional and multi-dimensional signals, making up for the shortcomings of DWT in terms of directional representation and multi-resolution characteristics. At the same time, compared with the Contourlet transform, NSCT can avoid spectrum aliasing and better express the content of image data. Applying it to the GF-2 image watermarking technology is expected to achieve better watermark embedding and detection effects.

[0052] S10: Extract the singular value λ of the watermark information W using SVD L′ (Save U W With V W is the key K3), and according to the formula ( 5 ) is embedded in the singular value λ0 of the low-frequency subband L1,

[0053] λ L′ =λ0+αλ W (5) where α is the watermark strength;

[0054] S11: The low frequency subband singular value λ of the singular value of the embedded watermark information L′ Perform inverse SVD to obtain L1;

[0055] The singular values ​​obtained by SVD can reflect the intrinsic properties of the image, satisfying the characteristics of non-negativity, uniqueness, stability to perturbations, and invariance to matrix transposition. When the image undergoes conventional image operations, the singular values ​​will not change much due to their strong stability.

[0056] S12: Perform inverse NSCT transformation on the low-frequency subband to restore it to the selected SIFT feature area, and obtain the final watermarked image Q1;

[0057] S13: End.

[0058] This method uses Mean Shift clustering to screen and filter the feature points extracted from the GF-2 image, constructs a rectangular feature area and uses it as the watermark embedding position. Combining the good detail retention and directionality of NSCT on the image, and the excellent stability of SVD on the image, a geometric watermarking method for GF-2 images is proposed. This method has the following characteristics: (1) Clustering the SIFT feature points of the image to achieve the purpose of feature point screening and filtering, eliminating the overlapping phenomenon of feature areas, making the feature point watermarking method feasible for GF-2 images; (2) Embedding the watermark information in the feature area NSCT domain, well retaining the image detail information, and improving the efficiency of watermark detection and extraction.

[0059] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0060] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A GF-2 image watermarking method using improved SIFT feature points, comprising: SIFT feature point screening and GF-2 image copyright protection watermarking method; The steps for SIFT feature point screening are as follows: S1: Use SIFT feature point extraction operator to extract GF-2 image feature points; S2: The coordinate values ​​of the extracted GF-2 image feature points are used as training samples and clustered using the MeanShift method; S3: Calculate the average SIFT descriptor for key points in the same cluster; S4: After clustering SIFT feature points, after determining their position (x1, y1), direction, and scale S, a circular area is constructed with R as the radius; S5: The distance D of the feature points will be screened once; The steps of generating, embedding and detecting the copyright protection watermark of GF-2 images are as follows: S6: Select the Arnold transformation method to perform scrambling processing on the copyright watermark; S7: read the original GF-2 image Q0, Q0{g(i, j), 0≤i≤M, 0≤j≤N}, g(i, j) is the image pixel value, the size is M×N; S8: Read binary watermark information as W, the length of the watermark information is Z×L; S9: Use NSCT to extract the low-frequency subband L0 of the feature area, perform SVD processing, and obtain its singular value λ0; S10: Extract the singular value λ of the watermark information W using SVD w , the left singular matrix U W With the right singular matrix V W , save U W With V W , as the key K3, and is embedded in the singular value λ0 of the low-frequency subband L1 according to the following formula, where λ L , is the singular value of the low-frequency subband after the watermark information has been embedded, and α is the watermark strength; l L′ =λ0+αλ w S11: The low frequency subband singular value λ of the singular value of the embedded watermark information L′ Perform inverse SVD to obtain L1; S12: Perform inverse NSCT transformation on the low-frequency subband to restore it to the selected SIFT feature area, and obtain the final watermarked image Q1; S13: End.

2. The GF-2 image watermarking method using improved SIFT feature points according to claim 1, characterized in that In steps S1 to S5, the Mean Shift method is used to screen and optimize the SIFT feature points of the GF-2 image.

3. According to the GF-2 image watermarking method using improved SIFT feature points as described in claim 1 or 2, the watermark generation method is in step S6, and also includes executing steps S9 to S12 based on the obtained feature area, thereby completing the copyright protection of the GF-2 image.

Images