WO2014071714A1

WO2014071714A1 - Watermark embedding and extracting method

Info

Publication number: WO2014071714A1
Application number: PCT/CN2013/072468
Authority: WO
Inventors: 谭铁牛; 董晶; 王再冉
Original assignee: 中国科学院自动化研究所
Priority date: 2012-11-06
Filing date: 2013-03-12
Publication date: 2014-05-15
Also published as: CN102890814A; CN102890814B

Abstract

A watermark embedding method, comprising the steps of: conducting wavelet decomposition on a to-be-protected image, and dividing a low-frequency wavelet coefficient into non-overlapping blocks; conducting singular value decomposition on each block; randomly selecting two image blocks satisfying the condition to form a block pair; modifying the maximum singular value of the block pair to embed a watermark; conducting inverse singular value decomposition on each sub-block, and then conducting wavelet reconstruction on the wavelet coefficient. The present invention embeds a watermark at the most stable portion of an image, thus increasing the watermark robustness. The normalized distance between two image blocks is defined to select an image block pair for watermark embedding, thus reducing the amount of image modification, and ensuring the transparency of the image having the watermark embedded.

Description

Watermark embedding and extraction method

The present invention is a method of embedding and extracting a watermark into a digital image, which has high robustness and can be used for copyright protection of digital images. Background technique

With the popularization and rapid development of information technology, digital multimedia content has been copied and distributed in an uncontrolled manner, making the copyright protection of digital products more and more important. In order to solve this problem, digital watermarking technology came into being and gained wide attention. The digital watermarking technology directly embeds some identification information into the digital media, but does not affect the use value of the original carrier, and is not easy to be detected and modified again, but can be extracted and recognized by the embedded party to identify the copyright information of the multimedia. After nearly a decade of development, digital watermarking technology has achieved certain results.

Digital watermarking technology mainly focuses on the following aspects: First, transparency, that is, the image quality change after embedding watermark information is imperceptible; Second, robustness, that is, after a variety of unintentional or even intentional signal processing, the number The watermark still maintains partial integrity and can be extracted correctly. Possible signal processing includes channel noise, filtering, digital/analog and analog-to-digital conversion, resampling, clipping, displacement, scaling, and lossy compression. The third is the watermark capacity, which can embed watermarks in multimedia. The maximum amount of information.

The present invention is a watermarking technique for digital image copyright protection. Current digital image watermarking techniques fall into two main categories: the first is to embed the watermark directly in the spatial domain; the second is to embed the watermark into the transform domain. For the first method, the image after embedding the watermark has good transparency, but the robustness is poor, and it can be used for image integrity authentication or tamper detection, and is not suitable for copyright protection of images. For the second method, the image is converted to the transform domain and then watermarked. Most of the current algorithms are embedded in the transform domain. The embedding of the watermark in the transform domain is more robust and more suitable for copyright protection. Currently commonly used transform techniques are singular value decomposition, Fourier transform, DCT transform, wavelet transform, Contourlet transform, and the like.

At present, digital watermarking techniques based on singular value decomposition are mainly divided into the following categories: The embedding of the same value vector matrix and the embedding based on the singular value matrix. Watermark embedding based on singular value vector matrix may change the orthogonality of the matrix during the embedding process, thus affecting the quality of the image after embedding the watermark and the robustness of the watermark. The embedding based on the singular value matrix is mainly to modify the singular value of the image to embed the watermark. Liu Ruizhen published in the "Electronic Journal" article "Digital Image Watermarking Method Based on Singular Value Decomposition" in 2001. It proposes a watermarking method to protect the image copyright by adding the singular value of the watermark image to the singular value of the image to be protected. However, since the extraction process requires a singular vector matrix of the watermark image, it causes a large false detection rate.

Hsien-Chu Wu, uen-Jie Jang, and Yu-Chi Liu, in "A robust watermarking scheme based on singular value decomposition and quantization technique" 2010 International Computer Symposium (IC S) - proposed to block the low-frequency wavelet coefficients of images. Singular value decomposition, which quantizes the maximum singular value of each block to embed a watermark. However, this method is less robust to some common attacks, such as JPEG compression and so on. In response to the above problems, the present invention proposes a secure, robust digital watermarking method based on wavelet transform and singular value decomposition. The feature of the present invention is that after wavelet decomposition, the low frequency coefficients are first divided, and then each block is subjected to singular value decomposition. Two blocks satisfying the qualification condition are selected according to the pseudo-randomness of the key, and the watermark is embedded by modifying the relationship between the maximum singular values of the two blocks. Summary of the invention

The object of the present invention is to provide a digital watermarking method with strong robustness, transparency and security to solve the copyright protection problem of current digital image products.

According to an aspect of the invention, a method of embedding a watermark, comprising the steps of:

Perform wavelet decomposition on the protected image, and divide the low-frequency wavelet coefficients into blocks that do not overlap each other; perform singular value decomposition on each block;

Randomly selecting two image blocks that satisfy the condition to form a block pair;

Embedding a watermark by modifying the maximum singular value of the block pair;

The singular value inverse decomposition is performed for each sub-block, and then the wavelet coefficients are wavelet reconstructed. According to another aspect of the present invention, a method for extracting a watermark includes the steps of: performing discrete wavelet decomposition on a detected image, dividing the low frequency wavelet coefficients into non-overlapping blocks; performing singular value decomposition on each block; Calculating the position of each block pair embedded in the watermark according to the largest singular array of the original image and the threshold interval;

Obtaining the maximum singular value of the block pair according to the position of each block pair, and calculating the normalized distance of the two blocks;

The watermark is extracted by comparing the normalized distance of each block pair with a detection threshold.

The present invention embeds a watermark in the most stable portion of the image, thereby enhancing the robustness of the watermark. By defining the normalized distance between the two image blocks to select the image block pair for watermark embedding, the amount of modification of the image is reduced, thereby ensuring the transparency of the image after embedding the watermark. DRAWINGS

Figure 1 is a flow chart of the present invention, including a watermark embedding process and a watermark extraction process.

Figure 2 is a flow chart of the present invention randomly selecting image block pairs from the largest singular matrix M. Figure 3a shows an original image taken in an embodiment of the invention with an image size of 1500 X 2000. Figure 3b shows an image after embedding a watermark in the embodiment of the present invention, wherein the number of wavelet decomposition layers is 2, the block size is 8 X 8, the threshold t/il = 0.02, t/i2 = 0.045, and the embedding strength t/i3 =

0.02, ί/ι4 = 0.02.

Figure 3c shows an original watermark image used in the embodiment of the present invention, the watermark size being 32 X 32. Figure 3d shows the watermark image extracted after the mean filtering of the image shown in Figure 3b, with a window size of 3 X 3 .

Figure 3e shows the watermark image extracted after JPEG compression of the image shown in Figure 3b, with compression factor Q=20.

Figure 3f shows the watermark image extracted after adding Gaussian noise (mean 0, variance 0.01) to the image shown in Figure 3b.

Figure 3g shows a watermark image extracted after adding salt and pepper noise (probability density of 0.01) to the image shown in Figure 3b.

Figure 3h is a watermark image extracted after scaling the image shown in Figure 3b (reduced to 250 X 250 and then to 1500 X 2000). detailed description

The invention provides a high robust watermark based on discrete wavelet transform and singular value decomposition The method is robust, resistant to a variety of digital image processing and attacks, and can be used for copyright protection of digital images. The methods of the present invention will be described in detail below with reference to the accompanying drawings.

FIG. 1(a) is a main flowchart of a method for embedding a watermark according to an embodiment of the present invention. Step S11: Perform L-layer discrete wavelet decomposition on the original image, extract its low-frequency wavelet coefficients, and divide the low-frequency wavelet coefficients into non-overlapping blocks, each block has a size of mxm, and then perform singular value decomposition on each sub-block. .

Step S12: The maximum singular value of each block is formed into an array M of maximum singular values, and two blocks satisfying the embedding condition are randomly selected to form a block pair.

Step S13: Modify the maximum singular value of the selected block pair to embed the watermark. The specific modification method is as follows - if '0, embed, then: Mi = (Mi + Mj)/2, Mj = (Mi + Mj)/2;

If '1' is embedded and Mi≥, Bay (J: Μ[ = Μ; X (1 + th3), Mj' = Mj X (1 - t/i4) _;

If '嵌入 and < Mj are embedded, IJ: Μ[ = Mi X (1 - th3), Mj = Mj X (1 + tM).

Where M is the maximum singular value of the modified image block pair, th3, t/i4 is the modification strength, and 0 ≤ t/i3 ≤ l, 0 ≤ t/i4 ≤ 1. When '0' is embedded, the maximum singular values of the two blocks are equal, so that the normalized distance between the two blocks becomes 0; when 'Γ is embedded, the larger maximum singular value is increased, and the smaller is reduced. The largest singular value, thus widening the normalized distance between the two blocks. The present invention embeds a watermark by quantizing the normalized distance between two blocks.

Step S14: Perform singular value inverse decomposition on each sub-block, and finally perform wavelet reconstruction on the wavelet coefficients to obtain an image after embedding the watermark.

FIG. 1(b) is a main flowchart of a method for extracting a watermark according to an embodiment of the present invention. Step S21: Perform L-layer discrete wavelet decomposition on the image to be detected, extract low-frequency wavelet coefficients, and divide the low-frequency wavelet coefficients into non-overlapping blocks, each block has a size of mxm, and then perform singular value decomposition on each sub-block. .

Step S22: Combining the maximum singular value of each block into a maximum singular value array M', and rooting The position of each block pair embedded in the watermark is calculated from the largest singular array M of the original image and the threshold interval [t/il, t/i2].

Step S23: Obtain the maximum singular value M^, M; of the block pair according to the position of each block pair, and calculate the normalized distance ΰ'(ΐ, ') of the two blocks.

Step S24: The watermark is extracted by comparing the normalized distance ΰ' of each block pair with the detection threshold. The specific comparison method is as follows:

If D'(ij')> t/ι, the watermark bit is extracted as;

If D' i, ) ≤ th, the watermark bit is extracted as '0'.

Finally, the extracted watermark bits are combined to obtain complete watermark information.

The method of randomly selecting two blocks forming a block pair satisfying the embedding condition according to the array M described in step S12 is as shown in FIG. 2:

Step S121: Calculate the normalized distance DG' of each image block and any other block. Step S122: Statistics are satisfied for each image block -

D j, k) e [thl,th2]

The number of blocks N _fc , where [t/il, t/i2] is a predetermined threshold interval. If N _fc > 0, the block fc is a block that satisfies the embedding condition.

Step S123: randomly select a block from the block satisfying the embedding condition according to the key fceyl, and record it as .

Step S124: According to the key fcey2, the normalized distance from the threshold is in the threshold interval

[t / il, t / i2 ] blocks randomly select a block, referred to _as; '. This forms a block pair (, ).

Step S125: Mark the block as marked as used. The used image block cannot be selected to be embedded in the watermark again.

The embodiment will be described with reference to the image in Fig. 3.

First, the image (Fig. 3a, image size is 1500X 2000) is subjected to discrete wavelet decomposition (decomposition layer number L = 2), and its low frequency wavelet coefficients are extracted for segmentation (block size is 8 X 8), and each block is performed. Singular value decomposition.

Then, from the array M consisting of the largest singular values of each sub-block, according to the method shown in Fig. 2, two image blocks satisfying the embedding condition are randomly selected to form a block pair. Where thl = 0.02, t/i2 = 0.045.

Second, modify the maximum singular value of the selected block pair to embed the watermark information, such as the watermark image Figure 3c, where the modification strength is: th3 = 0.02, ί/ι4 = 0.02.

Finally, each image block is subjected to singular value inverse decomposition and wavelet reconstruction to obtain an image after watermarking. As shown in Fig. 3b, it can be seen that the watermarked image produced by the present invention has no obvious visual difference from the original image.

Mean filtering is a common image processing method. Figure 3d is a watermark image extracted after mean filtering. The window size is 3 X 3 .

JPEG compression is a compression mode that is often taken when an image is transmitted. Figure 3e is an extracted watermark image after JPEG compression with a compression factor of 20. Figure 3f is a watermark image obtained by adding Gaussian noise to the image after the watermark is embedded, and the noise variance is 0.01. Fig. 3g is a watermark image obtained by adding salt and pepper noise to the image after the watermark is printed, and the noise variance is 0. 01. Fig. 3h is a watermark image after the watermarked image is reduced by 0.5 times and then enlarged to the original size.

It can be seen from the above that after the image works are subjected to common transformation (noise, JPEG compression, filtering, scaling), the watermark image indicating the copyright can still be extracted.

The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art familiar with the art may make changes or substitutions within the technical scope of the present invention. The transformations and substitutions made are intended to be included within the scope of the invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

Claim

What is claimed is: 1. A method of embedding a watermark, comprising the steps of:

The singular value inverse decomposition is performed for each sub-block, and then the wavelet coefficients are wavelet reconstructed.

2. The method according to claim 1, wherein said randomly selecting two image block forming block pairs satisfying the condition comprises:

Calculate the normalized distance between each image block and any other block

Count all the conditions for each image block -

D (j, k e [thl, th2]

The number of blocks N _fc , where [t/il, t/i2] is a predetermined threshold interval. If N _k > 0, the block 1J block c is a block that satisfies the embedding condition;

According to the key fceyl, a block is randomly selected from the block satisfying the embedding condition;

According to the key fcey2, a block is randomly selected from the block whose normalized distance is in the threshold interval [t/il, t/i2];

The block ΰ ' is marked as 巳, and the used image block cannot be selected to be embedded in the watermark again.

3. The method of claim 1 wherein said embedding a watermark by modifying a maximum singular value of a pair of blocks comprises:

When 0 ' is embedded, the maximum singular value of the selected block pair is modified so that the normalized distance between the two blocks is 0; when embedding, the maximum singular value of the two blocks is modified to make the normalized distance between the two blocks Increase.

4. A method for extracting a watermark, comprising the steps of: performing discrete wavelet decomposition on a detected image, dividing the low frequency wavelet coefficients into non-overlapping blocks; performing singular value decomposition on each block;

Calculating the position of each block pair embedded in the watermark based on the largest singular array of the original image and the threshold interval;

Get the maximum singular value of the block pair according to the position of each block pair, and calculate the normalization of the two blocks Distance

5. The method according to claim 4, wherein the comparing comprises: if the normalized distance is greater than the detection threshold, extracting the watermark to 1;

If the normalized distance is less than the detection threshold, the extracted watermark is zero.