Non-blind single-frame video watermark adding and detecting method

Abstract

The invention discloses a non-blind single-frame video watermark adding and detecting method, which comprises the following steps of: decoding original video frames frame by frame to obtain a video frame sequence, and performing discrete wavelet decomposition on each independent video frame in the video frame sequence to obtain a low-frequency component LL; performing histogram analysis on the low-frequency component LL, taking a position where half of pixels are accumulated as a boundary, performing (0, 1) quantization to generate a template LLT, setting the pixels on the left side of the boundary to be 0, setting the pixels on the right side of the boundary to be 1, analyzing the template LLT, finding a center reversible point as a watermark embedding position, and embedding watermark information into the LL to generate LLM; and replacing the original low-frequency component LL with the LLM low-frequency component embedded with the watermark, performing discrete wavelet inverse transformation, and outputting a complete video frame added with the watermark. According to the method, the defect that a complete watermark can be added only through multi-frame combination in the existing video watermark is overcome, and a reliable guarantee is provided for copyright confirmation and pirate traceability.

Description

Technical Field

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

[0002] With the popularization of broadband, it has become increasingly common to distribute and obtain film and television works through the network. However, due to the openness and sharing nature of the network, it is difficult to control video information during transmission, resulting in an increasingly rampant piracy phenomenon. In the production, processing, storage, playback, and transmission of digital video products, illegal copying can be easily carried out. Digital watermark technology is a new technology developed in recent years. It embeds digital watermarks into the original data of works through certain algorithms without affecting the use and appreciation of the works, and usually cannot be detected by users. The publisher stores copyright information, serial numbers, etc. in the digital watermark and distributes it along with the copy. When it is necessary to authenticate the copy, by extracting the watermark in the work copy, the information stored by the publisher in the watermark can be obtained to identify the copyright information of the work. It has natural advantages in meeting the security requirements of media content.

[0003] How to integrate digital watermark technology into video streams is a current technical difficulty.

[0004] Existing video watermarks require multiple frames to be combined to add a complete watermark. For example, the DCI stipulates that a complete watermark can be detected in a 15-minute video.

[0005] How to add watermarks to each frame in a video and detect them is an unprecedented challenge. Summary of the Invention

[0006] To at least solve one of the above technical problems, this application provides a non-blind single-frame video watermark addition and detection method.

[0007] A non-blind single-frame video watermark addition method provided by this application includes: S1: Decode the original video frames frame by frame to obtain a video frame sequence, perform discrete wavelet decomposition on each individual video frame in the video frame sequence to obtain the low-frequency component LL; S2: Perform histogram analysis on the low-frequency component LL, take the position where the cumulative number of pixels is half as the dividing line, perform (0, 1) quantization to generate a template LLT, set the pixels on the left side of the dividing line to 0, and set the pixels on the right side of the dividing line to 1, so that the template LLT only contains 0 and 1 values; S3: Analyze the template LLT, find the central flip point as the watermark embedding position, and embed the watermark information into LL to generate LLM; S4: Replace the original low-frequency component LL with the low-frequency component LLM embedded with the watermark, perform inverse discrete wavelet transform, and output the complete video frame with the watermark added.

[0008] Further, the central flip point described in step S3 is specifically: the central point of the nine-square grid in the central flip mode.

[0009] Further, embedding the watermark information into LL in step S3 is specifically:

[0010] If the watermark information is 1, then LLM(i,j) = LL(i,j) + m; (m >= 1);

[0011] If the watermark information is 0, then LLM(i,j) = LL(i,j) - m; (m >= 1);

[0012] Where i, j are the coordinates of the central point of the nine-square grid in the central flip mode.

[0013] Further, the discrete wavelet decomposition in step S1 is a first-level discrete wavelet decomposition or a second-level or higher discrete wavelet decomposition.

[0014] Further, the inverse discrete wavelet transform in step S4 is a first-level inverse discrete wavelet transform or a second-level or higher inverse discrete wavelet transform.

[0015] A non-blind single-frame video watermark detection method provided by the present application includes: S5: Decoding the original video frames frame by frame to obtain a first video frame sequence, performing discrete wavelet decomposition on each individual video frame in the first video frame sequence to obtain a low-frequency component LL; decoding the video frames with watermarks frame by frame to obtain a second video frame sequence, performing discrete wavelet decomposition on each individual video frame in the second video frame sequence to obtain a low-frequency component LLM; S6: Performing histogram analysis on the low-frequency component LL, taking the position where the cumulative number of pixels is half as the dividing line, performing (0, 1) quantization to generate a template LLT, setting the pixels on the left side of the dividing line to 0 and the pixels on the right side of the dividing line to 1, so that the template LLT only contains 0 and 1 values; S7: Analyzing the template LLT to find the central flip point as the position for watermark embedding; S8: Decoding the watermark information by comparing the difference between the central flip points of LL and LLM.

[0016] Further, the central flip point described in step S7 is specifically: the central point of the nine-square grid in the central flip mode.

[0017] Further, in step S8, comparing the difference between the central flip points of LL and LLM to decode the watermark information is specifically: comparing the difference between the central points of the nine-square grids that conform to the central flip mode of LL and LLM to decode the watermark information:

[0018] LLM(i,j) - LL(i,j) > 0; then the watermark information is 1,

[0019] LLM(i,j) - LL(i,j) < 0; then the watermark information is 0,

[0020] where \(i\) and \(j\) are the coordinates of the center point of the nine - grid of the center - flippable pattern.

[0021] Further, in the step S5, the discrete wavelet decomposition is a first - level discrete wavelet decomposition or a discrete wavelet decomposition of two levels or more.

[0022] The beneficial technical effects of the present invention are as follows:

[0023] 1. The present invention adopts the center - flippable point watermark technology, which can effectively eliminate the influence brought by attacks such as zooming in, zooming out, rotating, flipping, mirroring, transcoding, and pirated recording by cameras, providing a reliable guarantee for detecting watermarks and confirming copyright.

[0024] 2. Since the present invention adopts a single - frame watermark, it provides a good guarantee for single - frame copyright confirmation. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is the flowchart of watermark embedding of a non - blind single - frame video watermark adding and detecting method of the present invention;

[0026] Figure 2 is the flowchart of watermark extraction of a non - blind single - frame video watermark adding and detecting method of the present invention;

[0027] Figure 3 is the schematic diagram of wavelet decomposition of a non - blind single - frame video watermark adding and detecting method of the present invention;

[0028] Figure 4 is the analysis of the low - frequency component LL histogram and the quantization flowchart of (0, 1) of a non - blind single - frame video watermark adding and detecting method of the present invention;

[0029] Figure 5 is the 32 patterns of center - flippable points of a non - blind single - frame video watermark adding and detecting method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0030] The following further describes the present invention in detail with reference to the accompanying drawings.

[0031] Refer to Figure 1 and Figure 2, an embodiment of the present invention discloses a method for adding non-blind single-frame video watermark, including: S1: Decode the original video frames frame by frame to obtain a video frame sequence, perform a first-level discrete wavelet decomposition on each individual video frame to obtain the low-frequency component LL for the next step. Without loss of generality, to enhance the robustness of the watermark, a discrete wavelet decomposition of two levels or more can be performed. The higher the level, the stronger the robustness, but the watermark capacity will decrease rapidly; S2: Perform a histogram analysis on the low-frequency component LL obtained in the previous step, take the position where the cumulative number of pixels is half as the dividing line, perform (0, 1) quantization to generate the template LLT, set the pixels on the left side of the dividing line to 0, and set the pixels on the right side of the dividing line to 1, so that the template LLT only contains 0 and 1 values; S3: Analyze the template LLT, find the central invertible point as the watermark embedding position, and embed the watermark information into LL to generate LLM; S4: Replace the original low-frequency component with the low-frequency component LLM with the embedded watermark, perform an inverse first-level discrete wavelet transform, and output the complete video frame with the added watermark. An embodiment of the present invention also discloses a method for detecting non-blind single-frame video watermark, including S5: Decode the original video frames frame by frame to obtain a video frame sequence, perform a first-level discrete wavelet decomposition on each individual video frame to obtain the low-frequency component LL for the next step, decode the video frames with watermarks frame by frame to obtain a video frame sequence, perform a first-level discrete wavelet decomposition on each individual video frame to obtain the low-frequency component LLM for the next step; S6: Perform a histogram analysis on the low-frequency component LL obtained in the previous step, take the position where the cumulative number of pixels is half as the dividing line, perform (0, 1) quantization to generate the template LLT, set the pixels on the left side of the dividing line to 0, and set the pixels on the right side of the dividing line to 1, so that the template LLT only contains 0 and 1 values; S7: Analyze the template LLT, find the central invertible point as the position where the watermark is embedded; S8: Decode the watermark information by comparing the difference between the central invertible points of LL and LLM.

[0032] The steps of adding the watermark are as follows:

[0033] In this embodiment, for example Figure 3 , to enhance the robustness of the video watermark, we select the low-frequency component of the video frame to add the watermark. Perform a first-level discrete wavelet decomposition on each individual video frame to obtain the low-frequency component LL for adding the watermark. Without loss of generality, to enhance the robustness of the watermark, a discrete wavelet decomposition of two levels or more can be performed. The higher the level, the stronger the robustness, but the watermark capacity will decrease rapidly. Those skilled in the art can set it according to actual needs.

[0034] Next is the histogram analysis and (0, 1) quantization of the obtained low-frequency component LL of the video frame, for example Figure 4 .

[0035] To prevent the abnormal distribution of the template LLT(0,1) after (0,1) quantization, we first perform a histogram analysis on the low-frequency component LL of the video frame. Take the position where the cumulative number of pixels is half as the dividing line, perform (0,1) quantization to generate the template LLT. Pixels on the left side of the dividing line are set to 0, and pixels on the right side of the dividing line are set to 1, so that the template LLT only contains 0 and 1 values.

[0036] Then, perform the operation of embedding watermark information. In this embodiment, analyze the template LLT to find the central invertible point as the watermark embedding position.

[0037] Next, analyze and count the number and position coordinates of the nine-square grids in the template LLT that conform to the central invertible pattern. The central invertible pattern set is as Figure 5 shown. Among these 32 patterns, flipping the 0-1 value of the center point of the nine-square grid does not affect the vision. In actual use, a subset of less than 32 patterns can be selected according to the key. Note that when counting the nine-square grids, non-overlapping counting is performed, that is, the step size of the row and column values is 3. Suppose the watermark length is 32 bits. If the number of nine-square grids that conform to the central invertible pattern is not less than 32, then select to repeatedly embed the watermark value at these points; if it is less than 32, end the program and return an error code;

[0038] Embed the watermark information into LL:

[0039] If the watermark information is 1, then LLM(i,j) = LL(i,j) + m; (m >= 1);

[0040] If the watermark information is 0, then LLM(i,j) = LL(i,j) - m; (m >= 1);

[0041] Where i,j are the coordinates of the center point of the nine-square grid of the central invertible pattern.

[0042] Then replace the original low-frequency component with the low-frequency component LLM embedded with the watermark, perform the first-level inverse discrete wavelet transform, and output the complete video frame with the watermark added.

[0043] The watermark detection steps are as follows:

[0044] Decode the original video frame frame by frame to obtain the video frame sequence, perform the first-level discrete wavelet decomposition on each individual video frame, and obtain the low-frequency component LL for the next step;

[0045] Decode the video frame containing the watermark frame by frame to obtain the video frame sequence, perform the first-level discrete wavelet decomposition on each individual video frame, and obtain the low-frequency component LLM for the next step;

[0046] Next, perform histogram analysis on the low-frequency component LL obtained in the previous step. Take the position where the cumulative number of pixels is half as the dividing line, and perform (0, 1) quantization to generate the template LLT. Pixels to the left of the dividing line are set to 0, and pixels to the right of the dividing line are set to 1, so that the template LLT only contains 0 and 1 values;

[0047] Then analyze the template LLT to find the central invertible point as the position for watermark embedding. Analyze and count the number and position coordinates of the nine-square grids in the template LLT that conform to the central invertible pattern. The set of central invertible patterns is as Figure 5 shown. Among these 32 patterns, flipping the 0-1 value at the center of the nine-square grid does not affect the vision. In actual use, a subset of less than 32 patterns can be selected according to the key. Note that when counting the nine-square grids, non-overlapping counting is performed, that is, the step size of the row and column values is 3. The center point of the nine-square grid that conforms to the central invertible pattern is the watermark embedding position.

[0048] Finally, decode the watermark information by comparing the difference between the center points of the nine-square grids of the central invertible patterns of LL and LLM,

[0049] For example, if LLM(i, j) - LL(i, j) > 0; then the watermark information is 1,

[0050] For example, if LLM(i, j) - LL(i, j) < 0; then the watermark information is 0,

[0051] where i, j are the coordinates of the center point of the nine-square grid of the central invertible pattern.

[0052] In this embodiment, for example Figure 5 , for the selection of 32 central invertible points, a specific pattern is selected according to the key for addition and detection. Those skilled in the art can select a suitable key according to needs, which will not be elaborated here.

[0053] The above is a 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 non-blind single-frame video watermarking addition method, characterized in that Including: S1: Decode the original video frames frame by frame to obtain a video frame sequence, perform discrete wavelet decomposition on each individual video frame in the video frame sequence to obtain the low-frequency component LL; S2: Perform histogram analysis on the low-frequency component LL, take the position where the cumulative number of pixels is half as the dividing line, perform (0, 1) quantization to generate the template LLT, set the pixels on the left side of the dividing line to 0, and set the pixels on the right side of the dividing line to 1, so that the template LLT only contains 0 and 1 values; S3: Analyze the template LLT to find the central flippable point as the watermark embedding position, and embed the watermark information into LL to generate LLM; S4: Replace the original low-frequency component LL with the low-frequency component LLM with the embedded watermark, perform inverse discrete wavelet transform, and output the complete video frame with the added watermark.

2. The non-blind single-frame video watermark addition method according to claim 1, wherein The central flippable point described in step S3 is specifically: the center point of the nine-square grid in the central flip mode.

3. The non-blind single-frame video watermark addition method according to claim 2, characterized in that, The embedding of the watermark information into LL in step S3 is specifically: If the watermark information is 1, then LLM(i, j) = LL(i, j) + m; (m >= 1); If the watermark information is 0, then LLM(i, j) = LL(i, j) - m; (m >= 1); Where i and j are the coordinates of the center point of the nine-square grid in the central flip mode.

4. The non-blind single-frame video watermark addition method according to claim 1 or 2, characterized in that The discrete wavelet decomposition in step S1 is a first-level discrete wavelet decomposition or a second-level or higher-level discrete wavelet decomposition.

5. The non-blind single-frame video watermark addition method according to claim 1 or 2, characterized in that The inverse discrete wavelet transform in step S4 is a first-level inverse discrete wavelet transform or a second-level or higher-level inverse discrete wavelet transform.

6. A non-blind single-frame video watermark detection method, characterized in that: S5: Decode the original video frames frame by frame to obtain a first video frame sequence, perform discrete wavelet decomposition on each individual video frame in the first video frame sequence to obtain the low-frequency component LL; Decode the video frames with watermarks frame by frame to obtain a second video frame sequence, perform discrete wavelet decomposition on each individual video frame in the second video frame sequence to obtain the low-frequency component LLM; S6: Perform histogram analysis on the low-frequency component LL, take the position where the cumulative number of pixels is half as the dividing line, perform (0, 1) quantization to generate the template LLT, set the pixels on the left side of the dividing line to 0, and set the pixels on the right side of the dividing line to 1, so that the template LLT only contains 0 and 1 values; S7: Analyze the template LLT to find the central flippable point as the watermark embedding position; S8: Decode the watermark information by comparing the difference between the central flippable points of LL and LLM.

7. The non-blind single-frame video watermark detection method according to claim 6, characterized in that The central flippable point described in step S7 is specifically: the center point of the nine-square grid in the central flip mode.

8. The non-blind single-frame video watermark detection method according to claim 7, characterized in that The decoding of the watermark information by comparing the difference between the central flippable points of LL and LLM in step S8 is specifically: Decode the watermark information by the difference between the center points of the nine-square grids that conform to the center-flippable pattern of LL and LLM: If LLM(i,j) - LL(i,j) > 0, the watermark information is 1. If LLM(i,j) - LL(i,j) < 0, the watermark information is 0. Where i and j are the coordinates of the center point of the nine-square grid in the center-flippable pattern.

9. The non-blind single-frame video watermark detection method according to any one of claims 6-8, characterized in that the discrete wavelet decomposition in step S5 is a first-level discrete wavelet decomposition or a second-level or higher discrete wavelet decomposition.

Images