CN112132731A - DWT-SVD domain adaptive robust watermarking algorithm adopting preset PSNR - Google Patents

Abstract

The invention belongs to the technical field of image information processing, and in particular relates to a DWT-SVD domain adaptive robust watermarking algorithm using a preset PSNR. The method utilizes the attack resistance of discrete wavelet transform and the stability of singular value decomposition, embeds watermark by modifying the size relationship between the elements in the first column of the left singular matrix of the image, and modifies the right singular matrix to achieve quality compensation. This method is different from other algorithms that use fixed embedding parameters or repeated experiments to obtain embedding parameters. The embedding strength of the watermark depends on the embedding parameters. The present invention not only establishes the correlation between the embedding parameters, the carrier image and the watermark information, but also does not need to sacrifice the algorithm's performance. Time complexity, pixel overflow and correction after the embedding is completed enhances the reliability of the algorithm.

Description

DWT-SVD Domain Adaptive Robust Watermarking Algorithm Using Preset PSNR

technical field

The invention belongs to the technical field of image information processing, and in particular relates to a DWT-SVD domain adaptive robust watermarking algorithm using a preset PSNR.

Background technique

The opening of the network brings convenience to the circulation of multimedia, and also makes the intellectual property protection of various content information face unprecedented challenges. Among all kinds of electronic information protection methods, electronic watermarks with the characteristics of concealment and security have attracted the attention of scholars. Electronic watermark refers to the secret embedding of electronic information called watermark in carrier information such as images, videos, audios, etc. The extractor shows the ownership and integrity of the carrier by extracting the accuracy of the watermark.

Traditional image watermarks are divided into two categories according to their functions: robust watermarks and fragile watermarks, which are used to protect image copyright and image content integrity respectively. Among them, the robust watermark refers to the watermark embedded by modifying the carrier content, and the recognizable watermark information can still be extracted after the image is subjected to various attacks. Embedding the modification of carrier information will inevitably lead to the degradation of image quality. Therefore, there are three main indicators of robust watermarking performance: capacity, robustness, and visibility. These three indicators trade off one another. When the watermark capacity is determined, in order to pursue robustness, the watermarking method needs to sacrifice the image quality after the watermark is embedded, and vice versa. Trying to make the information load of the carrier, the quality of the watermark image and the integrity of the watermark extracted after being attacked to achieve the optimum at the same time, has always been the purpose of the robust watermarking method design.

The watermark embedding process can be divided into spatial domain watermarking and frequency domain watermarking according to the embedding domain. Spatial watermarking directly modifies the spatial pixel embedding watermark. Although the computational complexity is low, the image processing interferes greatly with the pixels, resulting in a decrease in robustness. Taking advantage of the energy-gathering ability, multi-resolution ability, and embodying the time-domain or frequency-domain features of the frequency-domain transformation, the frequency-domain-based algorithm has stronger robustness and image quality, so it is widely used in watermarking algorithms. Among the frequency domain algorithms, DCT, DWT, RDWT, and FrFT are common transformation methods. Multiple transformations make the algorithm combine the advantages of different transformations, and finally achieve the purpose of improving the performance of the algorithm, which has become a research hotspot of watermarking algorithms in recent years.

In addition to the design of the embedding extraction process, the parameter selection method is also important in the robust watermark design. More and more algorithms use artificial intelligence techniques such as evolutionary algorithms and neural networks to select parameters. Compared with the watermarking algorithm that used fixed parameters in the past, using artificial intelligence technology to select parameters can establish a connection between parameters, watermarks, and carriers, such as particle swarm optimization (PSO), firefly algorithm (FA), artificial bee colony optimization ( ABC) select embedding parameters, which can balance the watermark image quality and robust embedding strength, but these techniques cannot ensure that the watermark image reaches the preset quality.

SUMMARY OF THE INVENTION

Aiming at the defects and problems of the current SVD-based robust watermarking algorithms, the present invention provides a DWT-SVD domain adaptive robust watermarking algorithm that does not rely on experimental feedback results and can ensure the quality of watermarked images using a preset PSNR.

The scheme adopted by the present invention to solve the technical problem is: a DWT-SVD domain adaptive robust watermarking algorithm that adopts preset PSNR, and the algorithm comprises the following steps:

Step 1. Select an embedding block: set the carrier image as A∈R _M×N , M and N are even numbers, the watermark to be embedded W={w _r |1≤r≤m}, m is the watermark length; the carrier image is divided into For non-overlapping blocks, calculate the entropy value of the block and calculate the weighted average sum of the entropy values of the block and adjacent blocks, then sort the entropy values, and select the block with the same number of watermark bits as the embedding block.

时的嵌入块和当w_r＝0且

时的嵌入块两类，两类嵌入块的序号分别记为序列S₁与S₂，根据载体图像DWT低频域的每个分块的奇异值、左奇异向量第一列第二、第三个元素的差值，计算S₁与S₂子序列水印嵌入前后图像像素的平方误差Ls₁和Ls₂，根据载体图像、水印和预设PSNR值，通过公式

计算得到自适应量化步长t。Step 2. Determine the adaptive quantization step size: divide the embedded block into when w _r =1 and

Embedding block when and when _wr = 0 and

There are two types of embedding blocks, the serial numbers of the two types of embedding blocks are respectively recorded as sequences S ₁ and S ₂ . The difference value of the elements, calculate the squared errors Ls ₁ and Ls ₂ of the image pixels before and after the watermark embedding of the S ₁ and S ₂ subsequences, according to the carrier image, the watermark and the preset PSNR value, through the formula

The adaptive quantization step size t is obtained by calculation.

Step 3: Embedding watermarks: Perform first-level Haar wavelet transform on the carrier image, divide the low-frequency subbands into non-overlapping blocks, perform SVD decomposition on the embedded blocks, and decompose the first column of the left singular vector, the second and third The difference between elements is quantized index modulation to embed the watermark;

Step 4. Construct a watermark image: perform reverse SVD decomposition and reverse Haar wavelet transformation on the modified embedded block to obtain a watermark image;

Step 5: Extract the watermark: perform SVD decomposition on the low-frequency subband of the watermark image that may be modified, and compare the magnitude relationship of the second and third elements of the first column of the singular vector to extract the watermark.

The above-mentioned DWT-SVD domain adaptive robust watermarking algorithm using the preset PSNR, step 1 includes the following steps:

为向上取整符号，分块数量为

(1) Divide the carrier image into non-overlapping blocks, each block contains 8 × 8 pixels, and record the block set as

For the round-up symbol, the number of blocks is

与

分别表示分块的视觉熵与边缘熵，计算分块与相邻块的加权平均和：(2) Calculate the entropy value of the blocks l _{i, j} in turn

and

Represent the visual entropy and edge entropy of the block, respectively, and calculate the weighted average sum of the block and adjacent blocks:

where max and min represent the maximum and minimum values of the variables, respectively;

S将作为提取过程的边信息。(3) Sort E _i,j from small to large, select the first m blocks as the embedded block, and record the sequence number set of the embedded blocks as the sequence

S will be used as side information for the extraction process.

The above-mentioned DWT-SVD domain adaptive robust watermarking algorithm using the preset PSNR, the calculation process of t in step 2 is:

做水印嵌入后记为

与

的SVD分解形式可表示为：for embedded blocks

After doing the watermark embedding, it is recorded as

and

The SVD decomposition form of can be expressed as:

Modified embedding blocks fall into two categories:

时的嵌入块；1) When _wr = 1 and

Embedding block when;

时的嵌入块。2) When _wr = 0 and

Embedding block.

不同，其具体定义为：In this embodiment, the sequence numbers of the two types of embedding blocks are respectively recorded as sequences S ₁ and S ₂ , and S ₁ and S ₂ are subsequences of S and do not intersect with each other. The first and second types of embedding blocks correspond to

different, which is specifically defined as:

Then the difference between the embedded blocks before and after embedding is:

After knowing the modification range of the elements of the embedded block, the corresponding relationship is further established between the difference of the embedded block and the squared error of the LL subband coefficients after the first-level Haar wavelet transform, and the squared error of the LL subband coefficients is calculated:

It is also known that the square error of the LL subband coefficients of the Haar wavelet transform and the image pixels satisfies

The calculation formula of known PSNR is:

where MAX _A is the maximum value of the elements of matrix A.

Available

即

根据确定的图像、水印和预设PSNR值，通过公式计算得到自适应量化步长t。which is:

which is

According to the determined image, watermark and preset PSNR value, the adaptive quantization step size t is obtained by formula calculation.

The above-mentioned DWT-SVD domain adaptive robust watermarking algorithm using the preset PSNR, step 3 includes the following steps:

个分块，其中的嵌入块集合记作

(1) Perform first-level Haar wavelet transform on the carrier image, and divide the LL subbands into non-overlapping 4×4 blocks to obtain

blocks, where the set of embedded blocks is denoted as

做SVD分解，左奇异向量

的第一列向量记为

其中第二、三个元素之间的差值记为

(2) For the embedded block

Do SVD decomposition, left singular vector

The first column vector of is denoted as

The difference between the second and third elements is recorded as

的差值

量化以嵌入水印w_r，第二、三个元素的相应修改方式为：(3) For the embedded block

difference

Quantization is used to embed the watermark w _r , and the corresponding modification of the second and third elements is:

if w _r = 1

if w _r = 0

Among them, t is the adaptive quantization step size of the quantization index modulation strategy.

The above-mentioned DWT-SVD domain adaptive robust watermarking algorithm using the preset PSNR, step 5 includes the following steps:

(1) After the first-level Haar wavelet transformation is performed on the watermark image, the LL subband is divided into blocks, and the block for extracting the watermark is determined as the extraction block according to the set S of the embedded block sequence numbers;

第二、三元素的差值为

其中s_r∈S；(2) Perform SVD decomposition on each extraction block, and record the first column vector of the left singular vector U ^* as

The difference between the second and third elements is

where s _r ∈ S;

最终合成完整水印。(3) Extract 1-bit watermark from each extraction block in turn, and the extraction method is as follows:

Finally, a complete watermark is synthesized.

Beneficial effects of the present invention: the present invention obtains the relationship between the PSNR value, the quantization step size, and image features, and in the case of determining the carrier image and the watermark, calculates the optimal quantization step size according to the preset PSNR value, and adjusts the image quality. The corresponding relationship with the quantization step size is clarified, and the quantization step size that ensures the image quality reaches the preset value can be obtained without repeated embedding and extraction process. DWT conversion enhances the watermark algorithm’s ability to resist noise, compression and other common image processing, while SVD decomposition Increases its ability to resist geometric attacks; the stability of singular vector element relationships is exploited in the proposed algorithm, and further enhancing the robustness can not only ensure that the actual watermarked image reaches the preset image quality, but also does not need to sacrifice the time complexity of the algorithm ; The pixel overflow and correction after embedding also enhances the reliability of the algorithm. Its invisibility and robustness are superior to other similar algorithms, and it has practical value in applications such as copyright protection.

Description of drawings

FIG. 1 is a flow chart of the watermarking algorithm based on the adaptive quantization step size of the present invention.

FIG. 2 is a flow chart of the embedding process of the present invention.

FIG. 3 is a list of images and watermarks of the present invention.

FIG. 4 shows the correspondence between the parameter t mean value of the test image, the preset PSNR mean value, and the actual PSNR mean value.

Figure 5 shows the test images after embedding the watermark and their embedding parameters t and the actual PSNR under the premise that the preset PSNR is 40db.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

Embodiment 1: This embodiment provides a DWT-SVD domain adaptive robust watermarking algorithm using a preset PSNR. The watermarking algorithm mainly includes the following contents, and the flow is shown in FIG. 1 .

Step 1: Let the carrier image be a 512×512 grayscale image Lena, and the watermark to be embedded is a 32×32 binary image, denoted as W={ _wr |1≤r≤1024}. The specific embedding steps are as follows:

Step1: Divide the carrier image into non-overlapping blocks, each block contains 8×8 pixels; record the block set as L={l _i,j| |1≤i≤64,1≤j≤64 }, the number of blocks is 4096.

与

分别表示分块的视觉熵与边缘熵；计算分块与相邻块的熵值的加权平均和：Step2: Calculate the entropy value of the blocks l _{i, j} in turn

and

Represent the visual entropy and edge entropy of the block, respectively; calculate the weighted average sum of the entropy values of the block and adjacent blocks:

where max and min represent the maximum and minimum values of the variable, respectively.

Step3: Sort E _i,j from small to large, select the first 32×32 blocks as the embedding blocks, and the sequence number set of the embedding blocks is recorded as the sequence S={s _r |1≤r≤1024, 1≤s _r ≤4096}, S as the side information of the extraction process.

分别表示图像A经过一级哈尔小波变换后的LL、LH、HL、HH子带系数。Assume

respectively represent the LL, LH, HL, and HH subband coefficients of the image A after the first-level Haar wavelet transform.

原空域像素记为a_k,l,1≤k≤M,1≤l≤N，水印图像像素记为a_k,l'，它们之间的关系见图2。The low-frequency subband coefficients after watermark embedding are written as

The original air domain pixels are denoted as _ak,l , 1≤k≤M, 1≤l≤N, and the watermark image pixels are denoted as _ak,l ', and the relationship between them is shown in Figure 2.

Then the squared error of the LL subband coefficients and the image pixels satisfies the following equation:

It shows that there is an equivalent relationship between the squared error of image pixels before and after watermark embedding and the squared error of the Haar wavelet transform low-frequency subband coefficients.

Step4: Perform a first-level Haar wavelet transform on the carrier image, and divide the LL sub-band into non-overlapping 4×4 sub-blocks to obtain 4096 sub-blocks, in which the set of embedded blocks is denoted as

做SVD分解，左奇异向量

的第一列向量记为

其中第二、三个元素之间的差值记为

Step5: Embedding the block

Do SVD decomposition, left singular vector

The first column vector of is denoted as

The difference between the second and third elements is recorded as

的差值

量化以嵌入第r位水印w_r，第二、三个元素的相应修改方式如下：Step6: Embedding the block

difference

Quantized to embed the rth bit watermark w _r , the corresponding modifications of the second and third elements are as follows:

if w _r = 1

if w _r = 0

In equations (3) to (6), t is the adaptive quantization step size of the quantization index modulation strategy, and the quantization step size t can be determined by the formula

二分法快速求出，其中量化步长t具体选取方式如下。

The bisection method can be quickly obtained, and the specific selection method of the quantization step size t is as follows.

做水印嵌入后记为

与

的SVD分解形式可表示为：for embedded blocks

After doing the watermark embedding, it is recorded as

and

The SVD decomposition form of can be expressed as:

Modified embedding blocks can be divided into two categories:

时的嵌入块；1) When _wr = 1 and

Embedding block when;

时的嵌入块。2) When _wr = 0 and

Embedding block.

不同，其具体定义为：In this embodiment, the sequence numbers of the two types of embedding blocks are respectively recorded as sequences S ₁ and S ₂ , and S ₁ and S ₂ are subsequences of S and do not intersect with each other. The first and second types of embedding blocks correspond to

different, which is specifically defined as:

According to equations (7) to (9), the difference between the embedded blocks before and after embedding is:

After knowing the modification range of the elements of the embedded block, the corresponding relationship is further established between the difference of the embedded block and the square error of the LL subband coefficients after the first-level Haar wavelet transform, that is, the square of the LL subband coefficients is calculated according to formula (10). error:

which is:

It is illustrated that there is a relationship between the squared error of the low-frequency subband coefficients and the modification magnitude of the embedding block.

The calculation formula of known PSNR is:

where MAX _A is the maximum value of the elements of matrix A.

Available

which is:

which is

It can be seen that the left side of the equal sign is related to the maximum singular value of the embedded block, the difference between the two elements of the singular vector in the first column, the watermark and the quantization step size t. When the image, watermark and PSNR values are determined, the quantization step size of equation (16) is satisfied. t can be quickly found by the dichotomy method. The quantization step size selection strategy combines the PSNR calculation formula and the embedding method of the watermarking algorithm based on the robustness of singular vectors, and obtains the corresponding relationship between the quantization step size t and the preset image quality, so that the guaranteed image can be obtained without repeated embedding and extraction process. The quantization step size at which the quality reaches the preset value.

Step7: Perform reverse SVD decomposition on the modified embedding block.

Step 8: Repeat Step 5 to Step 7 for all embedded blocks; replace the original LL subband after the blocks are merged, and then perform reverse Haar wavelet transformation to obtain a watermarked image.

2. Watermark extraction

Step1: First perform first-level Haar wavelet transformation on the watermark image, and then divide the LL sub-band into blocks, and select the block for extracting the watermark as the extraction block according to the set of embedded block serial numbers S;

第二、三元素的差值为

其中s_r∈S。Step2: Perform SVD decomposition on each extraction block, and record the first column vector of the left singular vector U ^* as

The difference between the second and third elements is

where s _r ∈ S.

Step 3: Extract a 1-bit watermark from each extraction block in turn, the extraction method is shown in formula (18), and finally a complete watermark is synthesized.

Embodiment 2: In order to verify the effect of the algorithm of the present invention, the present embodiment will proceed from the two perspectives of watermark image quality and watermark robustness, and combine the method of the present invention with three algorithms that are also based on SVD and have the same watermark capacity. One, two, Three are compared; the parameters of the four algorithms are shown in Table 1.

Table 1. Summary of parameters of the four algorithms

表示按位异或运算符。Select 10 pairs of 512×512 classic test images in Figure 3 as carrier images, and 32×32, 48×32, 48×48, 64×64 binary images as watermarks to be embedded. In the quantitative analysis of the experimental results in this paper, PSNR and bit error rate (BER) are respectively used as the measurement mechanism of watermark image quality and algorithm robustness.

Represents the bitwise XOR operator.

(1) Invisibility

Minimizing the impact on the original image is one of the goals pursued by the invisible watermarking algorithm. Different from other algorithms that balance the relationship between image quality and robustness by embedding parameters, the watermarking algorithm of the present invention uses the preset PSNR value as a parameter, and inversely derives the embedding parameter, the parameter t mean of the test image, the preset PSNR mean value and the actual PSNR mean value. The corresponding relationship between them is shown in Figure 4.

It can be seen from Figure 4 that when the preset PSNR is 30, 35, 40, 45, and 50, the mean value of the embedded parameters of the ten test images and the mean value of the actual PSNR basically coincide, which proves that this method can ensure that the obtained watermark image quality reaches default value.

Generally speaking, to measure the difference between the original image and the watermarked image, subjective and objective methods are used. Under the premise that the preset PSNR is 40db, the test images after embedding the watermark and their embedding parameters t and the actual PSNR are the premise that the preset PSNR is equal to 40db, the six pairs of watermark images, their embedding parameters t and the actual PSNR obtained As shown in Figure 5.

As can be seen from Figure 5, the algorithm ensures that the PSNR of the watermark image is not less than 40db, which not only proves the effectiveness of the adaptive parameter selection strategy of the present invention, but also illustrates the watermark image quality performance of the algorithm of the present invention from both subjective and objective aspects. meet application requirements. At the same time, the correlation between PSNR and watermark length is shown. Theoretically, when the watermark length is less than 64×64, the watermark image quality of this watermark method can be guaranteed.

In order to verify this conclusion, this embodiment is verified by experiments on the Lena image with watermarks of different sizes, and the results are shown in Table 2.

Table 2 Comparison of PSNR between fixed embedding parameters of different watermark sizes and the method of the present invention (Lena image)

It can be seen from Table 2 that when the size of the watermark is 32×32, 32×48, 48×48, 64×64, the PSNR of the watermark image can reach the preset 40db, while the method using fixed embedding parameters Can't do this.

(2) Robustness

Robustness is another indicator to measure the watermarking algorithm. The present invention uses BER as an index to measure the robustness of the watermark. The smaller the BER value, the higher the similarity between the extracted watermark and the original watermark, that is, the higher the robustness of the watermark. In order to verify the robustness of the algorithm, the present invention selects 13 representative attack methods, which include not only common image processing methods such as compression, filtering, and noise, but also geometric attacks such as scaling and rotation operations. shown in Table 3.

Table 3 Robustness of the algorithm of the present invention to different attacks

As shown in Table 3, when the 32×32 watermark is embedded in the Luk Fook test image, it can be seen from the BER results of extracting the watermark after 13 kinds of attacks. From the robustness of different images, Peppers, Man images In the face of image processing such as jpeg compression and median filtering with low quality factor, the anti-attack ability is weak, which is related to the fact that the algorithm preferentially selects the block with small entropy value as the embedded block and the smooth area of the image itself is relatively small. But in general, the algorithm has a certain robustness against common attacks, especially for histogram equalization, contrast enhancement, reduction, and rotation attacks, the algorithm shows excellent robustness.

In this embodiment, the robustness of the four watermarking methods is compared under the premise that the PSNR values of the four watermarking methods are all set to be about 41db. Tables 4 and 5 show the PSNR values of the four watermarking methods obtained in this paper for the images Lena and Peppers respectively when the watermark size is 32×32, and the watermark ber values after facing 13 kinds of attacks.

Table 4 Robustness comparison between the algorithm of the present invention and similar algorithms (Lena image)

Table 5 Robustness comparison between the algorithm of the present invention and similar algorithms (Peppers image)

In Table 4 and Table 5, although the algorithm of the present invention presets PSNR of 41db for both test images, the obtained t values are different (0.041 and 0.045), which shows the necessity of adaptive embedding parameters. From the comparison results, although the four algorithms embed the watermark by changing the relationship between the two elements, the two comparison elements in the second algorithm are taken from the largest singular value of the two matrices composed of the DCT intermediate frequency coefficients, which are not as high as The similarity between the elements in the first column of the block singular matrix is strong, so it can be seen from Tables 4 and 5 that under the same embedding capacity, Algorithm 1 and Algorithm 3 have excellent robustness in the face of most attacks, especially noise attacks. in Algorithm 2. The algorithm of Algorithm 1 embeds a watermark in the RIDWT domain. Although the watermark can resist continuous 90-degree rotation and row-column flipping attacks, the RIDWT transformation includes a pixel position replacement step, which destroys the similarity of adjacent pixel values of the image. Not resistant to JPEG compression, median filter, mean filter, size reduction, Gaussian filter operations. The comparison shows the shortcomings of the third algorithm: the embedding parameters cannot be dynamically selected according to different carriers, which leads to the situation that the watermark image quality is unstable when the algorithm is applied to different images. In addition, the algorithm proposed by the present invention is superior to the third algorithm in the optimization of entropy block selection, which is reflected in the face of jpeg compression.

(3) Running time

In order to verify the advantages of the quantization step size selection strategy proposed in this paper in terms of time complexity, this section uses the ACO-based quantization step size selection strategy and the proposed adaptive quantization step size selection in the watermarking algorithm based on singular vector robustness. strategies and compare their runtimes. The main frequency of the experiment will be 2.90GHz, the memory will be 8GB, and the software environment will be Microsoft Windows 10 Ultimate and MATLAB 2018. Table 6 is a comparison of the average running time of the two selection strategies for the test images Lena and Peppers. In the selection strategy based on ACO, this paper sets the number of iterations of ACO to 50 and the ant colony size to 10; the results are shown in the table 6 shown.

Table 6 Comparison of running time of two quantization step selection strategies

As can be seen from Table 6, the execution time of the two selection strategies is proportional to the watermark size, but the execution time of the adaptive quantization step selection strategy of the present invention is far less than the selection strategy based on ACO, which shows that the selection strategy proposed in this paper is in the It has obvious advantages in time complexity.

The above are only preferred embodiments of the present invention and do not limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principle scope of the present invention shall be included in the protection scope of the present invention. Inside.

Claims (5)

1. a DWT-SVD domain adaptive robust watermarking algorithm that adopts preset PSNR, is characterized in that: comprise the following steps: Step 1. Select an embedding block: set the carrier image as A∈R _M×N , M and N are even numbers, the watermark to be embedded W={w _r |1≤r≤m}, m is the watermark length; the carrier image is divided into For non-overlapping blocks, calculate the entropy value of the block and calculate the weighted average sum of the entropy values of the block and adjacent blocks, then sort the entropy values, and select the blocks with the same number of watermark bits as the embedded block. The set of serial numbers is denoted as sequence S; Step 2. Determine the adaptive quantization step size: divide the embedded block into when w _r =1 and

Embedding block when and when _wr = 0 and

There are two types of embedding blocks, the serial numbers of the two types of embedding blocks are respectively recorded as sequences S ₁ and S ₂ . The difference value of the elements, calculate the squared errors Ls ₁ and Ls ₂ of the image pixels before and after the watermark embedding of the S ₁ and S ₂ subsequences, according to the carrier image, the watermark and the preset PSNR value, through the formula

Calculate the adaptive quantization step size t; Step 3: Embedding watermarks: Perform first-level Haar wavelet transform on the carrier image, divide the low-frequency subbands into non-overlapping blocks, perform SVD decomposition on the embedded blocks, and decompose the first column of the left singular vector, the second and third The difference between elements is quantized index modulation to embed the watermark; Step 4. Construct a watermark image: perform reverse SVD decomposition and reverse Haar wavelet transformation on the modified embedded block to obtain a watermark image; Step 5: Extract the watermark: perform SVD decomposition on the low-frequency subband of the watermark image that may be modified, and compare the magnitude relationship of the second and third elements of the first column of the singular vector to extract the watermark. 2. the DWT-SVD domain adaptive robust watermarking algorithm that adopts preset PSNR according to claim 1, is characterized in that: step 1 comprises the following steps: (1) Divide the carrier image into non-overlapping blocks, each block contains 8 × 8 pixels, and record the block set as

For the round-up symbol, the number of blocks is

(2) Calculate the entropy value of the blocks l _{i, j} in turn

and

Represent the visual entropy and edge entropy of the block, respectively, and calculate the weighted average sum of the block and adjacent blocks:

where max and min represent the maximum and minimum values of the variables, respectively; (3) Sort E _i,j from small to large, select the first m blocks as the embedded block, and record the sequence number set of the embedded blocks as the sequence

S will be used as side information for the extraction process. 3. the DWT-SVD domain adaptive robust watermarking algorithm that adopts preset PSNR according to claim 1, is characterized in that: in step 2, the calculation process of t is: for embedded blocks

After doing the watermark embedding, it is recorded as

and

The SVD decomposition form of can be expressed as:

The modified embedding block is divided into when _wr = 1 and

Embedding block when and when _wr = 0 and

When there are two types of embedding blocks, the serial numbers of the second types of embedding blocks are respectively recorded as sequences S ₁ and S ₂ , S ₁ and S ₂ are subsequences of S and do not cross each other. The first and second types of embedding blocks correspond to

different, which is specifically defined as:

Then the difference between the embedded blocks before and after embedding is

After knowing the modification range of the elements of the embedded block, the corresponding relationship is further established between the difference of the embedded block and the squared error of the LL subband coefficients after the first-level Haar wavelet transform, and the squared error of the LL subband coefficients is calculated:

It is also known that the square error of the LL subband coefficients of the Haar wavelet transform and the image pixels satisfies

but:

The calculation formula of known PSNR is:

where MAX _A is the maximum value of the elements of matrix A. Available

which is:

which is

According to the determined image, watermark and PSNR values, the adaptive quantization step size t is obtained by formula calculation. 4. the DWT-SVD domain adaptive robust watermarking algorithm adopting preset PSNR according to claim 1, is characterized in that: step 3 comprises the following steps: (1) Perform first-level Haar wavelet transform on the carrier image, and divide the LL subbands into non-overlapping 4×4 blocks to obtain

blocks, where the set of embedded blocks is denoted as

(2) For the embedded block

Do SVD decomposition, left singular vector

The first column vector of is denoted as

The difference between the second and third elements is recorded as

(3) For the embedded block

difference

Quantization is used to embed the watermark w _r , and the corresponding modification of the second and third elements is: if w _r = 1

if w _r = 0

Among them, t is the adaptive quantization step size of the quantization index modulation strategy. 5. the DWT-SVD domain adaptive robust watermarking algorithm that adopts preset PSNR according to claim 1, is characterized in that: step 5 comprises the following steps: (1) After the first-level Haar wavelet transformation is performed on the watermark image, the LL subband is divided into blocks, and the block for extracting the watermark is determined as the extraction block according to the set S of the embedded block sequence numbers; (2) Perform SVD decomposition on each extraction block, and record the first column vector of the left singular vector U ^* as

The difference between the second and third elements is

where s _r ∈ S; (3) Extract 1-bit watermark from each extraction block in turn, and the extraction method is as follows:

Finally, a complete watermark is synthesized.

Images