Single-frame video non-blind watermark embedding and detecting method based on DWT-SVD (Discrete Wavelet Transform-Singular Value Decomposition)

Abstract

The invention discloses a single-frame video non-blind watermark embedding and detecting method based on DWT-SVD (Discrete Wavelet Transform-Singular Value Decomposition). The watermark embedding method comprises the following steps: S1, converting an original video frame into a YCbCr color format, and performing secondary discrete wavelet decomposition on a Y component to obtain a low-frequency component LL; s2, dividing the LL into a plurality of isometric matrixes LLi, performing singular value decomposition on each matrix LLi, and arranging singular values in a descending order to obtain a singular value diagonal matrix S; s3, modifying S to obtain S * according to watermark information to be embedded, performing inverse SVD decomposition to obtain LLi *, and merging to obtain LLi *; and S4, replacing the original low-frequency component LL with the watermark-embedded low-frequency component LL *, performing secondary discrete wavelet inverse transformation to obtain Y *, replacing the original Y component, converting into an original color format, and outputting the watermark-embedded video frame. According to the method, the defects of poor watermark robustness and concealment of the existing single-frame video are overcome, and the detection accuracy is further improved.

Description

Technical Field

[0001] This application relates to the field of video watermark embedding, and specifically relates to a single-frame video non-blind watermark embedding and detection method based on DWT-SVD. Background Art

[0002] Digital watermark technology is to embed imperceptible copyright marks in carriers such as pictures, videos, and audios. With the increasing prominence of digital video copyright security issues, digital watermark technology has become an effective technical means for piracy tracing and copyright protection. However, with the rapid development of technologies such as short videos and AIGC, the existing long-video blind watermarks that require several minutes of samples to be accurately detected can no longer meet the new requirements, and there is an urgent need for a video digital watermark embedding and detection technical method that can be detected with a single frame. Currently, most single-frame video watermark technologies are difficult to ensure concealment and robustness in order to ensure a high embedding speed. In addition, since the samples to be detected are often tampered with by means such as cropping, occlusion, and compression, in order to improve the detection efficiency and accuracy, it is necessary to further enhance the robustness of the watermark and optimize the watermark detection method. Summary of the Invention

[0003] In order to solve at least one of the above technical problems, this application provides a single-frame video non-blind watermark embedding and detection method based on DWT-SVD.

[0004] A single-frame video non-blind watermark embedding and detection method based on DWT-SVD provided by this application, the watermark embedding method includes: S1: Convert the original video frame to the YCbCr color format, perform a two-level discrete wavelet decomposition on the Y component to obtain the low-frequency component LL; S2: Divide LL into several equal-sized matrices LL i , where the number of divided matrices should be greater than or equal to the number of bits of the watermark information, and the number of rows and columns of each matrix should be greater than or equal to 2. Perform singular value decomposition on each matrix LL i , and arrange the singular values in descending order to obtain the singular value diagonal matrix S; S3: According to the watermark information to be embedded, modify S to obtain S * , perform inverse SVD decomposition to obtain LL i * , and then merge to obtain LL * ; S4: Replace the original low-frequency component LL with the low-frequency component LL* embedded with the watermark, perform an inverse two-level discrete wavelet transform to obtain Y * , replace the original Y component, convert to the original color format, and output the video frame embedded with the watermark. The watermark detection method includes: S1': Use the SURF algorithm to extract and match feature points of the original video frame and the video frame to be detected, and correct the video frame to be detected according to the matching result; S2': Convert the corrected video frame to the YCbCr color format, and perform operations on the Y component Y *Perform a two-level discrete wavelet decomposition to obtain the low-frequency component LL * ; S3': Divide LL * into several matrices of equal size according to the segmentation method during embedding as LL i * , and for each matrix LL i * perform a singular value decomposition and arrange the singular values in descending order to obtain the singular value diagonal matrix S * ; S4': Decode the watermark information by comparing the difference between S* and S.

[0005] Furthermore, the specific method for modifying S according to the watermark information to be embedded as described in the watermark embedding method S3 is as follows: When modifying S, only S(2,2) is modified, specifically:

[0006]

[0007] where to ensure a certain robustness, m≥5 is taken.

[0008] Furthermore, the value of m is; m = 10.

[0009] Furthermore, the specific method for decoding the watermark information as described in the watermark detection method S4' is: If S * (2,2)>S(2,2), the watermark information is 1, otherwise the watermark information is 0.

[0010] The beneficial technical effects of the present invention are:

[0011] 1. The present invention embeds watermark information by modifying S(2,2), improving the robustness and invisibility of the single-frame video watermark.

[0012] 2. The present invention first uses the SURF algorithm to perform matching correction on the frame to be detected during detection, further improving the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is the watermark embedding flowchart of a single-frame video non-blind watermark embedding and detection method based on DWT-SVD of the present invention;

[0014] Figure 2 is the watermark extraction flowchart of a single-frame video non-blind watermark embedding and detection method based on DWT-SVD of the present invention;

[0015] Figure 3 (a) is the video frame image to which the watermark is to be embedded, Figure 3 (b) is the video frame image after the watermark is embedded, Figure 3 (c) is the video frame image to be detected after being attacked;

[0016] Figure 4 (a) is the matching result of SURF algorithm feature points, Figure 4 (b) is the corrected video frame image to be detected. Specific implementation mode

[0017] The present invention will be further described in detail below with reference to the accompanying drawings.

[0018] Refer to Figure 1 and Figure 2 , an embodiment of the present invention discloses a single-frame video non-blind watermark embedding and detection method based on DWT-SVD. The watermark embedding method includes: S1: Convert the original video frame into the YCbCr color format, perform a two-level discrete wavelet decomposition on the Y component to obtain the low-frequency component LL; S2: Divide LL into several equal-sized matrices LL i , where the number of divided matrices should be greater than or equal to the number of bits of the watermark information, and the number of rows and columns of each matrix should be greater than or equal to 2. Perform singular value decomposition on each matrix LL i , and arrange the singular values in descending order to obtain the singular value diagonal matrix S; S3: According to the watermark information to be embedded, modify S to obtain S * , perform inverse SVD decomposition to obtain LL i * , and then merge to obtain LL * ; S4: Replace the original low-frequency component LL with the low-frequency component LL* embedded with the watermark, perform inverse two-level discrete wavelet transform to obtain Y * , replace the original Y component, convert it to the original color format, and output the video frame embedded with the watermark. The watermark detection method includes: S1: Use the SURF algorithm to extract and match the feature points of the original video frame and the video frame to be detected, and correct the video frame to be detected according to the matching result; S2: Convert the corrected video frame into the YCbCr color format, perform a two-level discrete wavelet decomposition on the Y component Y * to obtain the low-frequency component LL * ; S3: Divide LL * into several equal-sized matrices LL i * according to the segmentation method during embedding, perform singular value decomposition on each matrix LL i * , and arrange the singular values in descending order to obtain the singular value diagonal matrix S * ; S4: Decode the watermark information by comparing the difference between S* and S.

[0019] The watermark embedding steps are as follows:

[0020] In this embodiment, the watermark information to be embedded is 64 bits in total, and the video frame to be embedded with the watermark is as Figure 3As shown in (a), with a size of 512×512, it is converted to the YCbCr color format, and the Y component is subjected to a two-level discrete wavelet decomposition to obtain the low-frequency component LL;

[0021] LL is divided into 1764 3×3 matrices, and the (mod(i - 1, 64)+1)-th bit of watermark information is embedded in the i-th matrix LL i in the following specific way: For each matrix LL i perform singular value decomposition, LL i = U×S×V T , and arrange the singular values in descending order to obtain a 3×3 singular value diagonal matrix S;

[0022] The specific way to modify S is as follows:

[0023]

[0024] To ensure a certain robustness, generally the value of m is taken as: m≥5. In this embodiment, m = 10;

[0025] In this embodiment, when modifying S, only S(2,2) is modified. Because through experiments, it is found that when the modification amplitude is large, especially in darker regions, modifying S(1,1) has a greater impact on the overall visual effect, while modifying S(2,2) with the same amplitude results in finer granularity and a smaller impact on the visual effect. In addition, since S(2,2) is "more hidden", when the modification amplitude is small, modifying S(2,2) has better robustness. Therefore, considering both concealment and robustness, here only S(2,2) is selected for modification instead of changing all values including S(1,1).

[0026] Perform inverse SVD decomposition, When all 1764 3×3 matrices are updated, they are merged to obtain LL * ;

[0027] Replace the original low-frequency component LL with the watermark-embedded low-frequency component LL*, perform inverse two-level discrete wavelet transform to obtain Y * , replace the original Y component, convert it to the original color format, and output the video frame with the embedded watermark, as Figure 3 shown in (b).

[0028] The watermark detection steps are as follows:

[0029] Perform text occlusion on the video frame image with the embedded watermark and rotate it 5° clockwise. The video frame image to be detected is as Figure 3 shown in (c). Use the SURF algorithm to extract and match the feature points of the original video frame and the video frame to be detected, and correct the video frame to be detected according to the matching result. The feature point matching result is asFigure 4 As shown in (a), the corrected video frame image to be detected is as Figure 4 shown in (b);

[0030] Convert the corrected video frame to the YCbCr color format, and perform a two-level discrete wavelet decomposition on the Y component Y * to obtain the low-frequency component LL * ;

[0031] Divide LL * into 1764 3×3 matrices LL i * in the same segmentation manner as during embedding. For each matrix LL i * perform singular value decomposition and arrange the singular values in descending order to obtain the singular value diagonal matrix S * . If S * (2,2)>S(2,2), the watermark information w(i) is 1; otherwise, the watermark information w(i) is 0;

[0032] Reconstruct the decoded watermark information w into a 27×64 matrix, and take the mode for each column to obtain the 64-bit watermark detection result.

[0033] The above is the preferred embodiment of this application. It does not limit the protection scope of this application accordingly. Therefore, all equivalent changes made according to the method and principle of this application should be covered within the protection scope of this application.

Claims

1. A single-frame video non-blind watermark embedding and detection method based on DWT-SVD, characterized in that The watermark embedding method includes: S1: Convert the original video frame into the YCbCr color format, perform a two-level discrete wavelet decomposition on the Y component to obtain the low-frequency component LL; S2: Divide LL into several matrices of equal size, LL i , where the number of divided matrices should be greater than or equal to the number of bits of the watermark information, and the number of rows and columns of each matrix should be greater than or equal to 2. For each matrix LL i perform singular value decomposition and arrange the singular values in descending order to obtain the singular value diagonal matrix S; S3: Modify the singular value diagonal matrix S according to the watermark information to be embedded to obtain S * , perform inverse SVD decomposition to obtain LL i * , and then merge to obtain LL * ; S4: Replace the original low-frequency component LL with the low-frequency component LL embedded with the watermark, and perform the inverse two-level discrete wavelet transform to obtain Y * Replace the original low-frequency component LL, perform the inverse two-level discrete wavelet transform to obtain Y * , replace the original Y component, convert it to the original color format, and output the video frame with the embedded watermark; The watermark detection method includes: S1': Use the SURF algorithm to extract and match feature points from the original video frame and the video frame to be detected, and correct the video frame to be detected according to the matching result to obtain the corrected video frame; S2’: Convert the corrected video frame into the YCbCr color format, and perform a two-level discrete wavelet decomposition on the Y component Y * to obtain the low-frequency component LL * ; S3’: Divide LL * into several matrices of equal size according to the segmentation method during embedding, LL i * , for each matrix LL i * perform singular value decomposition and arrange the singular values in descending order to obtain the singular value diagonal matrix S * ; S4’: Compare S * and decode the watermark information from the difference between S and S.

2. According to the single-frame video non-blind watermark embedding and detection method based on DWT-SVD described in claim 1, characterized in that In the watermark embedding method, the specific method of modifying S according to the watermark information to be embedded is as follows: When modifying S, only S(2,2) is modified, specifically:

3. According to the single-frame video non-blind watermark embedding method based on DWT-SVD described in claim 2, characterized in that To ensure a certain robustness, the value of m is taken as: m≥5.

4. According to the single-frame video non-blind watermark embedding method based on DWT-SVD described in claim 3, characterized in that The value of m is; m = 10.

5. A single-frame video non-blind watermark embedding and detection method based on DWT-SVD according to claim 1, characterized in that, The specific method for decoding the watermark information described in the watermark detection method S4’ is as follows: If S * (2,2)>S(2,2), the watermark information is 1; otherwise, the watermark information is 0.

Images