CN114998081B - Video blind watermark embedding and extracting method based on H.265/HEVC - Google Patents

Abstract

The invention relates to a video blind watermark embedding and extracting method based on H.265/HEVC, which comprises the following steps of; encrypting the watermark image with copyright information; the pixel value of the encrypted watermark image is used as watermark information to be embedded into an I frame image and a P frame image of HEVC in a binary form; in the HEVC video decoding stage, watermark information embedded into an I frame image and a P frame image is subjected to blind extraction; the extracted watermark information is decrypted to restore the watermark image, so as to achieve the purpose of copyright protection. The blind watermark embedding and extracting method provided by the invention can effectively limit the increase of the video code rate, realize the blind extraction of the watermark, and ensure the video visual quality while obtaining higher data embedding capacity.

Description

A video blind watermark embedding and extraction method based on H.265/HEVC

Technical Field

The present invention belongs to the technical field of multimedia information security and relates to a video blind watermark embedding and extraction method based on H.265/HEVC.

Background Art

Video watermarking is mainly to achieve the purpose of copyright protection by embedding information into the carrier file. Its characteristics include imperceptibility, robustness and high embedding capacity. The embedding points of video watermarks are generally selected in the spatial domain (pixel domain), transform domain and compression domain. However, video applications generally have real-time requirements. Since the computational complexity of embedding watermarks in the spatial domain and transform domain is high, most algorithms choose to embed in the compression domain. At present, a large number of video watermarking schemes have been proposed, especially for the most widely used video compression standard H.264/AVC.

However, as people increasingly pursue higher video clarity experience, H.264/AVC is difficult to provide ideal compression efficiency when processing ultra-high resolution videos. In response to the above problems, the new generation of video compression standard H.265/HEVC was developed and released by the Joint Collaboration Group on Video Coding, and is gradually becoming a new hot research direction during its rapid development. Although the watermark algorithm based on H.265/HEVC has made some progress, there are still some shortcomings:

(1) In video watermarking, blind extraction means that the watermark information can be successfully extracted at the video decoding end without relying on additional information. Among the existing digital watermarking schemes based on HEVC, most of them need to save the embedding position in a local file during the data embedding process, and then use the position map to correctly extract the watermark when extracting data at the decoding end. However, in actual application scenarios, it is difficult to transmit the position map file securely and completely from the sender to the receiver, so these schemes are not very practical.

(2) There is no good balance between the imperceptibility and embedding capacity of the watermark algorithm. The video embedded with the watermark should not affect the quality of the original video, but as the data embedding capacity increases, many existing algorithms will have a greater impact on imperceptibility. In addition, the existing algorithms only embed the watermark information into a single frame type of the video, thus wasting a lot of embedding space and reducing the embedding efficiency.

Summary of the invention

In order to solve the above technical problems, the present invention provides a video blind watermark embedding and extraction method based on H.265/HEVC, which can effectively limit the growth of video bit rate, realize blind extraction of watermarks, and ensure video visual quality while achieving higher data embedding capacity.

The present invention provides a video blind watermark embedding and extraction method based on H.265/HEVC, comprising the following steps:

Step 1: Encrypt the watermark image with copyright information;

Step 2: The pixel values of the encrypted watermark image are embedded into the HEVC I frame image in binary form as watermark information;

Step 3: The pixel values of the encrypted watermark image are embedded into the HEVC P frame image in binary form as watermark information;

Step 4: In the HEVC video decoding stage, the watermark information embedded in the I frame image is blindly extracted;

Step 5: In the HEVC video decoding stage, the watermark information embedded in the P frame image is blindly extracted;

Step 6: Decrypt the extracted watermark information and restore it to the watermark image.

In the video blind watermark embedding and extraction method based on H.265/HEVC of the present invention, step 1 is specifically as follows:

The encryption method used in the watermark image encryption processing is the Fibonacci transformation encryption algorithm, the Hilbert curve transformation encryption algorithm, the affine transformation encryption algorithm or the magic square transformation encryption algorithm; the pixel value of the encrypted watermark image will be embedded into the video in binary form as the watermark information.

In the video blind watermark embedding and extraction method based on H.265/HEVC of the present invention, step 2 is specifically as follows:

Step 2.1: First, find the 4×4 luminance DST coefficient block of the I frame, and then calculate the PNNZ value of this coefficient block according to the corresponding coefficients in the second row and the second column of the luminance DST coefficient block. The PNNZ value represents the number of non-zero coefficients in the specified position.

Step 2.2: If the PNNZ value of the current coefficient block is less than a preset threshold N _th , N _th =1-7, then the coefficient block abandons embedding watermark information, otherwise the current 4×4 luminance DST coefficient block is dequantized and inversely transformed to obtain the first reconstructed residual matrix R ^r , as shown in formula (1);

Step 2.3: Calculate the average value of the eight pixel values around the pixel point _x'33 in the first reconstructed residual matrix ^Rr according to formula (2):

Step 2.4: Design the watermark embedding matrix W as shown in formula (3):

Where, δ is the embedding strength;

Step 2.5: Use _wk to represent the binary bit of the embedded watermark information. When the binary bit of the watermark information _wk = 1, determine Is it true? If R _th > 1, the watermark information is embedded and the next step is executed; otherwise, the embedding strength in the watermark embedding matrix W is set according to the following formula:

Among them, Qstep is the equivalent quantization step size;

Step 2.6: Modify the original DST coefficient matrix using the obtained watermark embedding matrix W according to the following formula;

X _W =X+W (5)

Among them, X is the original DST coefficient matrix, X _W represents the modified DST coefficient matrix, which realizes the embedding of watermark;

Step 2.7: When the binary bit w _k of the watermark information is 0, determine Is it true? If so, the watermark information is embedded and the next step is executed; otherwise, the embedding strength in the watermark embedding matrix W is set according to the following formula:

Among them, Qstep is the equivalent quantization step size;

Step 2.8: Modify the original coefficient matrix using the watermark embedding matrix W obtained in the previous step according to the following formula;

_XW ＝XW (7)

Among them, X is the original DST coefficient matrix, _XW represents the modified DST coefficient matrix, which realizes the embedding of watermark.

Step 2.9: The current 4×4 luminance DST coefficient block has been processed, and steps 2.1 to 2.8 are repeated until all 4×4 luminance DST coefficient blocks have been processed.

In the video blind watermark embedding and extraction method based on H.265/HEVC of the present invention, step 3 is specifically as follows:

Step 3.1: Get the 4×4 luminance DCT coefficient block of each P frame. In order to realize the blind extraction of watermark, use the coefficients _A1 and _A2 in the DCT coefficient matrix A as flags. If the number of non-zero values of _A1 and _A2 is greater than or equal to a threshold _Tth , _Tth is 1 or 2, then proceed to the next step of embedding, otherwise the block abandons the embedded data;

Step 3.2: Select the (1,7,3) matrix coding scheme, use the 3rd to 9th coefficients A ₃ ~A ₉ in the DCT coefficient matrix as carrier signals, and use their least significant bits to construct the carrier vector S _1×7 :

S _1×7 = (v ₁ , v ₂ , v ₃ , v ₄ , v ₅ , v ₆ , v ₇ ) (9)

Among them, v ₁ ~v ₇ are the least significant bits of A ₃ ~A ₉ respectively;

Step 3.3: Select the binary Hamming code matrix H _3×7 :

Then, the dependency vector D _3×1 is calculated from H _3×7 and S _1×7 according to equation (11);

Among them, S _1×c represents a 1-row and c-column carrier vector, H _m×c represents an m-row and c-column Hamming code matrix, D _m×1 represents an m-row and 1-column dependency vector, and mod represents a modulo operation;

Step 3.4: Select 3 bits of watermark information in order to form the information vector E _3×1 = (m ₁ , m ₂ , m ₃ ), then calculate the position P to be modified in the carrier vector S _1×7 according to formula (12), and add 1 to the coefficients of the corresponding positions in A ₃ to A ₉ of the DCT coefficient matrix A according to the modified position to complete the watermark embedding;

Among them, bin2dec() represents the binary to decimal function. represents an XOR operation; if position P is zero, it means that the carrier vector does not change.

Step 3.6: The current 4×4 luminance DCT coefficient block is processed, and steps 2.2.1 to 2.2.3 are repeated until all 4×4 luminance DCT coefficient blocks are processed.

In the video blind watermark embedding and extraction method based on H.265/HEVC of the present invention, step 4 is specifically as follows:

Step 4.1: At the HEVC video decoding end, for each 4×4 luma DST coefficient block of an I frame, calculate the PNNZ value of this coefficient block according to the corresponding coefficients in the second row and the second column of the luma DST coefficient block.

Step 4.2: When the PNNZ value of this coefficient block is less than the preset threshold _Nth , the information extraction is abandoned; otherwise, the coefficient block is dequantized and inversely transformed to obtain the second reconstructed residual matrix ^RW :

Step 4.3: Calculate the second reconstruction residual matrix R ^W The average value of the surrounding 8 pixels

Step 4.4: Extract the embedded watermark information according to formula (14), _wk represents the extracted binary information bit:

Step 4.5: Save the extracted watermark information and repeat steps 4.1 to 4.4 until all 4×4 luminance DST coefficient blocks are processed.

In the video blind watermark embedding and extraction method based on H.265/HEVC of the present invention, step 5 is specifically as follows:

Step 5.1: During the HEVC video decoding process, obtain the 4×4 luminance DCT coefficient block of each P frame. If the modified DCT coefficient matrix The coefficients in and If the number of non-zero values of is greater than or equal to the threshold value T _th , the next step of extraction is performed, otherwise the watermark extraction for this block is abandoned;

Step 5.2: Perform the inverse process of matrix encoding, using the modified DCT coefficient matrix The 3rd to 9th coefficients As the carrier signal, The least significant bit of constructs the carrier vector S' _1×7 :

S' _1×7 = (v' ₁ ,v' ₂ ,v' ₃ ,v' ₄ ,v' ₅ ,v' ₆ ,v' ₇ ) (17)

Step 5.3: By multiplying the Hamming code matrix H _3×7 in formula (10) by the transpose of S' _1×7 , the extracted watermark information M _3×1 can be obtained:

M _3×1 ＝H _3×7 S' _1×7 (18)

Step 5.4: Save the extracted watermark information and repeat steps 5.1 to 5.3 until the processing of all 4×4 luminance DCT coefficient blocks is completed.

The H.265/HEVC-based video blind watermark embedding and extraction method of the present invention also includes evaluating the video quality after the watermark information is inserted through the peak signal-to-noise ratio PSNR and the structural similarity index SSIM.

The video blind watermark embedding and extraction method based on H.265/HEVC of the present invention has at least the following beneficial effects:

1. Converting one-dimensional information into two-dimensional information improves the robustness of the watermark, and encryption processing improves security. Generally, the copyright information of the video author is in the form of a line of text or a special number. When the information is embedded in the video in this way, once it is attacked maliciously, the recognizability of the information will be sharply reduced. If the information is embedded in a two-dimensional form such as an image, even if the extracted image pixels are changed, shifted, or even missing some blocks, it is still possible to identify the watermark information.

2. The decoding end can successfully extract the watermark information without relying on additional information. The present invention uses the non-zero number of some coefficients in the coefficient matrix to determine whether the current block is embedded with a watermark, thereby achieving true blind watermark extraction at the receiving end, thereby enhancing the practicability of the method.

3. The video quality and data embedding capacity after watermark embedding are guaranteed. In the video I frame, the present invention uses an information embedding algorithm without intra-frame prediction error propagation by analyzing the characteristics of 35 intra-frame prediction modes of HEVC. By modifying the 4×4 size brightness DST coefficient block, the final embedding effect may only cause a change in one pixel in the spatial domain, which effectively guarantees the quality of the video. In the video P frame, since most P frames use inter-frame prediction, even if the coefficients are directly modified, it will not cause intra-frame prediction error propagation, so the present invention chooses to use matrix coding to increase the embedding capacity.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a method for embedding and extracting blind video watermarks based on H.265/HEVC of the present invention;

FIG2 is an original watermark image with copyright information used in this embodiment;

Figure 3 is the original watermark image after encryption;

FIG4 is a flowchart of embedding a 4×4 luma DST coefficient block watermark in an I frame of HEVC;

FIG5 is a diagram showing the calculation of the PNNZ value position of a 4×4 luma DST coefficient block in an I frame of HEVC;

Figure 6(a) is a comparison diagram of the original video frame and the watermarked video frame at the 6th frame of the test video RaceHorses_416x240_30;

Figure 6(b) is a comparison diagram of the original video frame and the watermarked video frame at frame 65 of the test video RaceHorses_832x480_30;

Figure 6(c) is a comparison diagram of the original video frame and the watermarked video frame in the first frame of the test video FourPeople_1280x720_60;

Figure 6(d) is a comparison diagram of the original video frame and the watermarked video frame at frame 65 of the test video BasketballDrive_1920x1080_50.

DETAILED DESCRIPTION

In order to more clearly explain the purpose, technical methods and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

The present invention proposes a video blind watermark embedding and extraction method based on H.265/HEVC, which provides a copyright protection solution for the new generation video compression standard H.265/HEVC by efficiently embedding information into the video compression domain. As shown in Figure 1, the specific implementation of the method of the present invention includes the following steps:

Step 1: Encrypt the watermark image with copyright information to enhance security. In this embodiment, the watermark image is a 64×64 grayscale image, as shown in Figure 2. The effect of the encrypted watermark image is shown in Figure 3. Specifically, the encryption method used in this step can be a Fibonacci transformation encryption algorithm, a Hilbert curve transformation encryption algorithm, an affine transformation encryption algorithm, or a magic square transformation encryption algorithm. The pixel value of the encrypted watermark image will be embedded in the video in binary form as the watermark information.

In the HEVC encoding stage, different embedding methods are designed for I frames and P frames. For ease of description, the methods for I frames and P frames are introduced separately, but it should be noted that in general, I frames and P frames of a video appear alternately, that is, the embedding steps described below selectively execute steps 2 and 3 according to different frame types.

Step 2: The pixel value of the encrypted watermark image is embedded into the HEVC I frame image in binary form as watermark information. As shown in FIG4 , step 2 is specifically as follows;

Step 2.1: First find the 4×4 luminance DST coefficient block of the I frame, and then calculate the PNNZ value of this coefficient block based on the corresponding coefficients in the 2nd row and 2nd column of the luminance DST coefficient block, that is, the light gray block in Figure 5. The PNNZ value represents the number of non-zero coefficients in the specified position.

Step 2.2: If the PNNZ of the current coefficient block is less than a preset threshold N _th , N _th =1-7, then the coefficient block abandons embedding watermark information, otherwise the current 4×4 luminance DST coefficient block is dequantized and inversely transformed to obtain the first reconstructed residual matrix R ^r , as shown in formula (1);

Step 2.3: Calculate the average value of the eight pixel values around the pixel point _x'33 in the first reconstructed residual matrix ^Rr according to formula (2):

Step 2.4: Design the watermark embedding matrix W as shown in formula (3):

Where, δ is the embedding strength;

Step 2.5: Use _wk to represent the binary bit of the embedded watermark information. When the binary bit of the watermark information _wk = 1, determine Is it true, R _th > 1, if it is true, the watermark information is embedded and the next step is executed; otherwise, the embedding strength in the watermark embedding matrix W is set according to the following formula:

Among them, Qstep is the equivalent quantization step size;

Step 2.6: Modify the original DST coefficient matrix using the obtained watermark embedding matrix W according to the following formula;

X _W =X+W (5)

Among them, X is the original DST coefficient matrix, X _W represents the modified DST coefficient matrix, which realizes the embedding of watermark;

Step 2.7: When the binary bit w _k of the watermark information is 0, determine Is it true? If so, the watermark information is embedded and the next step is executed; otherwise, the embedding strength in the watermark embedding matrix W is set according to the following formula:

Among them, Qstep is the equivalent quantization step size;

Step 2.8: Modify the original coefficient matrix using the watermark embedding matrix W obtained in the previous step according to the following formula;

_XW ＝XW (7)

Where X is the original DST coefficient matrix, _XW represents the modified DST coefficient matrix. By modifying the original coefficient matrix, x' ₃₃ and The relationship between the two is used to realize the embedding of watermark.

Step 2.9: The current 4×4 luminance DST coefficient block has been processed, and steps 2.1 to 2.8 are repeated until all 4×4 luminance DST coefficient blocks have been processed.

There are two important threshold values in step 2, _Nth and _Rth ; these two values are set to balance the imperceptibility and embedding capacity of the data embedding algorithm. In this embodiment, _Nth = 6 and _Rth = 3 are set respectively. According to the different values of the binary bit _wk of the embedded watermark information, the embedding strength δ in the embedding matrix W is set, and the original DST coefficient matrix is modified by the embedding matrix W to achieve the embedding of watermark information.

Step 3: The pixel value of the encrypted watermark image is embedded into the HEVC P frame image in binary form as watermark information. Step 3 is specifically as follows;

Step 3.1: Get the 4×4 luminance DCT coefficient block of each P frame. In order to realize the blind extraction of watermark, use the coefficients _A1 and _A2 in the DCT coefficient matrix A as flags. If the number of non-zero values of _A1 and _A2 is greater than or equal to a threshold _Tth , _Tth is 1 or 2, then proceed to the next step of embedding, otherwise the block abandons the embedded data;

Step 3.2: Select the (1,7,3) matrix coding scheme, use the 3rd to 9th coefficients A ₃ ~A ₉ in the DCT coefficient matrix as carrier signals, and use their least significant bits to construct the carrier vector S _1×7 :

S _1×7 = (v ₁ , v ₂ , v ₃ , v ₄ , v ₅ , v ₆ , v ₇ ) (9)

Among them, v ₁ ~v ₇ are the least significant bits of A ₃ ~A ₉ respectively;

Step 3.3: Select the binary Hamming code matrix H _3×7 :

Then, the dependency vector D _3×1 is calculated from H _3×7 and S _1×7 according to equation (11);

Among them, S _1×c represents a 1-row and c-column carrier vector, H _m×c represents an m-row and c-column Hamming code matrix, D _m×1 represents an m-row and 1-column dependency vector, and mod represents a modulo operation;

Step 3.4: Select 3 bits of watermark information in order to form the information vector E _3×1 = (m ₁ , m ₂ , m ₃ ), then calculate the position P to be modified in the carrier vector S _1×7 according to formula (12), and add 1 to the coefficients of the corresponding positions in A ₃ to A ₉ of the DCT coefficient matrix A according to the modified position to complete the watermark embedding;

Among them, bin2dec() represents the binary to decimal function. represents an XOR operation; if position P is zero, it means that the carrier vector does not change.

Step 3.6: The current 4×4 luminance DCT coefficient block is processed, and steps 2.2.1 to 2.2.3 are repeated until all 4×4 luminance DCT coefficient blocks are processed.

Step 4: In the HEVC video decoding stage, the watermark information embedded in the I frame image is blindly extracted. Step 4 is as follows:

Step 4.1: At the HEVC video decoding end, for each 4×4 luma DST coefficient block of an I frame, calculate the PNNZ value of this coefficient block according to the corresponding coefficients in the second row and the second column of the luma DST coefficient block.

Step 4.2: When the PNNZ value of this coefficient block is less than the preset threshold _Nth , the information extraction is abandoned; otherwise, the coefficient block is dequantized and inversely transformed to obtain the second reconstructed residual matrix ^RW :

Step 4.3: Calculate the second reconstruction residual matrix R ^W The average value of the surrounding 8 pixels x ^W ;

Step 4.4: Extract the embedded watermark information according to formula (14), _wk represents the extracted binary information bit:

Step 4.5: Save the extracted watermark information and repeat steps 4.1 to 4.4 until all 4×4 luminance DST coefficient blocks are processed.

Step 5: In the HEVC video decoding stage, the watermark information embedded in the P frame image is blindly extracted. Step 5 is as follows:

Step 5.1: During the HEVC video decoding process, obtain the 4×4 luminance DCT coefficient block of each P frame. If the modified DCT coefficient matrix The coefficients in and If the number of non-zero values of is greater than or equal to the threshold value T _th , the next step of extraction is performed, otherwise the watermark extraction for this block is abandoned;

Step 5.2: Perform the inverse process of matrix encoding, using the modified DCT coefficient matrix The 3rd to 9th coefficients As the carrier signal, The least significant bit of constructs the carrier vector S' _1×7 :

S' _1×7 = (v' ₁ ,v' ₂ ,v' ₃ ,v' ₄ ,v' ₅ ,v' ₆ ,v' ₇ ) (17)

Step 5.3: By multiplying the Hamming code matrix H _3×7 in formula (10) by the transpose of S' _1×7 , the extracted watermark information M _3×1 can be obtained:

M _3×1 ＝H _3×7 S' ^T _1×7 (18)

Step 5.4: Save the extracted watermark information and repeat steps 5.1 to 5.3 until the processing of all 4×4 luminance DCT coefficient blocks is completed.

Step 6: The extracted watermark information is restored to the watermark image through decryption to complete the extraction of the watermark image.

Experimental results and analysis

1. Video bitrate growth analysis

In general, embedding data in a video will inevitably lead to a certain degree of bit rate expansion. However, the embedding algorithm should not significantly affect the encoding and decoding process of the video and reduce the user experience. The present invention tests the influence of the watermarking method on the video bit rate by conducting experiments on a variety of video sequences. The experimental results are shown in Table 1. In Table 1, the composition of the video sequence name is as follows: Taking RaceHorses_416x240_30 as an example, RaceHorses represents the sequence name, 416×240 represents the video resolution, and 30 represents the frame rate of the video. QP represents the quantization parameter during video compression. The basic unit of the bit rate is kbps. It can be seen from the experimental results that the average bit rate expansion of all test sequences is only 0.2033%. It can be seen that the watermarking method proposed in the present invention has little effect on the expansion of the video bit rate.

Table 1 Video bit rate expansion introduced by watermark

2. Video Subjective Quality Analysis

Subjective quality mainly refers to whether human vision can detect the presence of watermarks after watermarks are embedded in the video. Figure 6 (a)-Figure 6 (d) are comparisons of the original and watermarked video frames of the test sequence. In the figure, the left side is the video frame decoded after the original video is compressed, and the right side is the video frame compressed and decoded after the watermark is added. The subjective quality analysis experimental results are all obtained under the condition of QP = 28. Figure 6 (a) is a comparison of the original video frame and the watermarked video frame of the 6th frame of the test video RaceHorses_416x240_30, Figure 6 (b) is a comparison of the original video frame and the watermarked video frame of the 65th frame of the test video RaceHorses_832x480_30, Figure 6 (c) is a comparison of the original video frame and the watermarked video frame of the 1st frame of the test video FourPeople_1280x720_60, and Figure 6 (d) is a comparison of the original video frame and the watermarked video frame of the 65th frame of the test video BasketballDrive_1920x1080_50. The experimental results show that it is difficult for the human eye to perceive the presence of watermarks in videos. This watermarking method does not have a significant impact on the subjective quality of the video, ensuring the imperceptibility of the watermarking algorithm.

3. Objective video quality analysis

The subjective quality evaluation system of the video often requires a lot of manpower and time costs, and will be affected by the subjective consciousness of the tester, while the objective quality evaluation system uses algorithms and visual models to complete the quality evaluation of the reconstructed image on the computer. The following uses multiple objective evaluation indicators to analyze the impact of the watermark algorithm on video quality from multiple dimensions.

(1) PSNR

Peak Signal to Noise Ratio (PSNR) is a classic full-reference image metric used to measure the difference between the original image and the distorted image. The PSNR definition is shown in formula (19):

In formula (19), MaxErr is the absolute value of the maximum value of the corresponding image color component (in this case is set to 1), MSE is the root mean square error, which is defined as shown in formula (20):

In formula (20), m and n correspond to the length and width of the image, respectively, and I(i, j) and K(i, j) correspond to the original image and the distorted image, respectively. In this experiment, the value range of PSNR is 0 to 100. The larger the value, the smaller the distortion of the video image. When the value is between 30 and 40, the distortion of the image is difficult to be detected. Table 2 shows the impact of the watermark algorithm on the PSNR value of the video. Among them, the calculation result of the PSNR value is the average value of the video sequence on the YUV component. The reference image is the video image before compression, and the original video is the video compressed and decoded by HEVC when the watermark is not embedded. It can be seen from Table 2 that the PSNR value of the algorithm of the present invention is only slightly reduced compared with the original video. The experimental results show that the video watermark algorithm of the present invention has good imperceptibility.

Table 2 The influence of the watermark algorithm of the present invention on the video PSNR value

(2)SSIM

The structural similarity index SSIM (Structure Similarity) evaluates image distortion from three components: brightness similarity, contrast similarity, and structural similarity. The algorithm uses a window function W _i,j (i,j = 0, 1, 2, ..., N) to perform convolution on the image, and is defined as shown in formula (21):

It should be noted that in equation (21), the image U uses negative indices that extend beyond the image area, and here the nearest edge pixels are extended to the desired position. Then, using the convolution operation in equation (22), the SSIM of the image pixels is calculated in equation (23):

In formula (23), C ₁ = 0.01 ² , C ₂ = 0.03 ² . The SSIM value range is from 0 to 1. The larger the value, the smaller the image distortion. SSIM = 1 means that the two images are the same. Table 3 shows the influence of the watermark algorithm on the SSIM value of the video image in the experiment. From the experimental results in Table 3, it can be seen that the watermark algorithm of the present invention has little influence on the SSIM value, indicating that the watermark only causes slight distortion to the video image.

Table 3 The impact of the watermark algorithm of the present invention on the video SSIM value

The platform used for testing in this experiment is: Intel(R) Core(TM) i5-10400 CPU@2.90GHz; memory (RAM): 16GB; operating system: Windows 10 64-bit; algorithm implementation language: C++ language.

The above series of analyses show that the blind watermarking method based on H.265/HEVC proposed in the present invention can effectively limit the growth of video bit rate, realize blind extraction of watermarks, and ensure video visual quality while achieving higher data embedding capacity. Experiments have proved that the watermarking method of the present invention is suitable for application scenarios such as video copyright protection and has high practical value.

The above description is only a preferred embodiment of the present invention and is not intended to limit the concept of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (3)

1. A video blind watermark embedding and extraction method based on H.265/HEVC, characterized in that it comprises the following steps: Step 1: Encrypt the watermark image with copyright information; Step 2: The pixel values of the encrypted watermark image are embedded into the HEVC I frame image in binary form as watermark information; Step 3: The pixel values of the encrypted watermark image are embedded into the HEVC P frame image in binary form as watermark information; Step 4: In the HEVC video decoding stage, the watermark information embedded in the I frame image is blindly extracted; Step 5: In the HEVC video decoding stage, the watermark information embedded in the P frame image is blindly extracted; Step 6: Decrypt the extracted watermark information and restore it to the watermark image; Step 2 is specifically as follows: Step 2.1: First, find the 4×4 luminance DST coefficient block of the I frame, and then calculate the PNNZ value of this coefficient block according to the corresponding coefficients in the second row and the second column of the luminance DST coefficient block. The PNNZ value represents the number of non-zero coefficients in the specified position. Step 2.2: If the PNNZ value of the current coefficient block is less than a preset threshold N _th , N _th =1-7, then the coefficient block abandons embedding watermark information, otherwise the current 4×4 luminance DST coefficient block is dequantized and inversely transformed to obtain the first reconstructed residual matrix R ^r , as shown in formula (1); Step 2.3: Calculate the average value of the eight pixel values around the pixel point _x'33 in the first reconstructed residual matrix ^Rr according to formula (2): Step 2.4: Design the watermark embedding matrix W as shown in formula (3): Where, δ is the embedding strength; Step 2.5: Use _wk to represent the binary bit of the embedded watermark information. When the binary bit of the watermark information _wk = 1, determine Is it true, R _th > 1, if it is true, the watermark information is embedded and the next step is executed; otherwise, the embedding strength in the watermark embedding matrix W is set according to the following formula: Among them, Qstep is the equivalent quantization step size; Step 2.6: Modify the original DST coefficient matrix using the obtained watermark embedding matrix W according to the following formula; X _W =X+W (5) Among them, X is the original DST coefficient matrix, X _W represents the modified DST coefficient matrix, which realizes the embedding of watermark; Step 2.7: When the binary bit w _k of the watermark information is 0, determine Is it true? If so, the watermark information is embedded and the next step is executed; otherwise, the embedding strength in the watermark embedding matrix W is set according to the following formula: Among them, Qstep is the equivalent quantization step size; Step 2.8: Modify the original coefficient matrix using the watermark embedding matrix W obtained in the previous step according to the following formula; _XW ＝XW (7) Among them, X is the original DST coefficient matrix, X _W represents the modified DST coefficient matrix, which realizes the embedding of watermark; Step 2.9: The current 4×4 luminance DST coefficient block is processed, and steps 2.1 to 2.8 are repeated until all 4×4 luminance DST coefficient blocks are processed; Step 3 is specifically as follows: Step 3.1: Get the 4×4 luminance DCT coefficient block of each P frame. In order to realize the blind extraction of watermark, use the coefficients _A1 and _A2 in the DCT coefficient matrix A as flags. If the number of non-zero values of _A1 and _A2 is greater than or equal to a threshold _Tth , _Tth is 1 or 2, then proceed to the next step of embedding, otherwise the block abandons the embedded data; Step 3.2: Select the (1,7,3) matrix coding scheme, use the 3rd to 9th coefficients A ₃ ~A ₉ in the DCT coefficient matrix as carrier signals, and use their least significant bits to construct the carrier vector S _1×7 : S _1×7 = (v ₁ , v ₂ , v ₃ , v ₄ , v ₅ , v ₆ , v ₇ ) (9) Among them, v ₁ :v ₇ are the least significant bits of A ₃ to A ₉ respectively; Step 3.3: Select the binary Hamming code matrix H _3×7 : Then, the dependency vector D _3×1 is calculated from H _3×7 and S _1×7 according to equation (11); Among them, S _1×c represents a 1-row and c-column carrier vector, H _m×c represents an m-row and c-column Hamming code matrix, D _m×1 represents an m-row and 1-column dependency vector, and mod represents a modulo operation; Step 3.4: Select 3 bits of watermark information in order to form the information vector E _3×1 = (m ₁ , m ₂ , m ₃ ), then calculate the position P to be modified in the carrier vector S _1×7 according to formula (12), and add 1 to the coefficients of the corresponding positions in A ₃ to A ₉ of the DCT coefficient matrix A according to the modified position to complete the watermark embedding; Among them, bin2dec() represents the binary to decimal function. represents an XOR operation; if position P is zero, it means that the carrier vector does not change; Step 3.5: The current 4×4 luminance DCT coefficient block is processed, and steps 3.1 to 3.4 are repeated until all 4×4 luminance DCT coefficient blocks are processed; Step 4 is specifically as follows; Step 4.1: At the HEVC video decoding end, for each 4×4 luma DST coefficient block of an I frame, calculate the PNNZ value of this coefficient block according to the corresponding coefficients in the second row and the second column of the luma DST coefficient block; Step 4.2: When the PNNZ value of this coefficient block is less than the preset threshold _Nth , the information extraction is abandoned; otherwise, the coefficient block is dequantized and inversely transformed to obtain the second reconstructed residual matrix ^RW : Step 4.3: Calculate the second reconstruction residual matrix R ^W The average value of the surrounding 8 pixels Step 4.4: Extract the embedded watermark information according to formula (14), _wk represents the extracted binary information bit: Step 4.5: Save the extracted watermark information and repeat steps 4.1 to 4.4 until all 4×4 luminance DST coefficient blocks are processed; Step 5 is specifically as follows; Step 5.1: During the HEVC video decoding process, obtain the 4×4 luminance DCT coefficient block of each P frame. If the modified DCT coefficient matrix The coefficients in and If the number of non-zero values of is greater than or equal to the threshold value T _th , the next step of extraction is performed, otherwise the watermark extraction for this block is abandoned; Step 5.2: Perform the inverse process of matrix coding, using the 3rd to 9th coefficients in the modified DCT coefficient matrix A As the carrier signal, The least significant bit of constructs the carrier vector S' _1×7 : Step 5.3: By multiplying the Hamming code matrix H _3×7 in formula (10) by the transpose of S' _1×7 , the extracted watermark information M _3×1 can be obtained: M _3×1 ＝H _3×7 S' _1×7 (18) Step 5.4: Save the extracted watermark information and repeat steps 5.1 to 5.3 until the processing of all 4×4 luminance DCT coefficient blocks is completed. 2. The method for embedding and extracting blind video watermarks based on H.265/HEVC according to claim 1, wherein step 1 specifically comprises: The encryption method used in the watermark image encryption processing is the Fibonacci transformation encryption algorithm, the Hilbert curve transformation encryption algorithm, the affine transformation encryption algorithm or the magic square transformation encryption algorithm; the pixel value of the encrypted watermark image will be embedded into the video in binary form as the watermark information. 3. The video blind watermark embedding and extraction method based on H.265/HEVC as described in claim 1 is characterized in that it also includes evaluating the video quality after the watermark information is inserted through the peak signal-to-noise ratio PSNR and the structural similarity index SSIM.