JP7142256B2 - Digital watermarking device and method - Google Patents

Description

The present invention is an electronic watermarking method for embedding information in digital image data for the purpose of copyright protection and anti-counterfeiting. It is about.

In today's digital information society, information duplication has made it possible for many people to share information, and society has developed significantly. However, this convenience has led to incidents such as copyright infringement due to the illegal duplication and distribution of other people's works. In the case of images, the high image quality of digital cameras and printers in recent years has made it possible to easily obtain exact reproductions of original images. It is a result of encouraging vicious criminal acts called acts.

In such a situation, invisible digital watermark technology embeds other information, such as author information, in image information to protect copyright or warn against illegal duplication. has developed. For example, for the purpose of copyright protection, this digital watermarking technology not only embeds the copyright owner, URL, contact information, terms of use, etc. , and in case of unauthorized use, it is possible to trace it. Also, in the case of illegal copying, a warning is given, and at the same time, copying can be stopped in the copying machine, or a black image can be output. In this way, electronic watermarking has begun to be widely used as a security measure for managing and protecting copyrights and monitoring unauthorized copying.

The electronic watermark as a security measure must be robust and cannot be easily lost or removed. This is because if watermark information can be removed or erased by a simple operation, unauthorized use can escape tracking and surveillance.

Electronic watermarks as digital data do not deteriorate or disappear as long as they are copied as electronic data. However, when an image is edited or processed by conversion such as enlargement, reduction, or rotation, or when image compression such as JPEG is performed, the embedded information is likely to be lost. Cropping, which cuts out a part of the image, destroys the embedded watermark information. In order to avoid such a situation and to protect copyrights and prevent counterfeiting, a robust digital watermarking technology that can withstand such editing and processing is required.

Furthermore, it is assumed that a third party removes the watermark information for unauthorized use (third party attack). By analyzing and estimating the embedding algorithm with various methods and images, the watermark information is removed from the embedded image and restored to the original image, making unauthorized use possible. Therefore, watermarking algorithms need to defend against such attacks.

Therefore, the digital watermarking method must be robust against image editing/processing and attacks from third parties. On the other hand, it is necessary to increase the strength (gain) of embedding for robustness, but in general, a large gain causes deterioration in image quality, and artifacts become visible. Image quality is important for commercial Internet distribution of stock photos, etc. Invisible digital watermarks that do not cause image quality deterioration are effective as a measure to protect copyrights, and the quality of watermarked images is close to the original image. is requested.

Normally, there is a trade-off relationship between such durability and image quality. That is, if the resistance is strengthened, the image quality deteriorates, and if the deterioration of the image quality is reduced, the resistance is weakened. Digital watermarking for copyright protection requires algorithms and methods that achieve both image quality and robustness.

As a digital watermarking method that satisfies such a request, there is a method of embedding watermark information in frequency space. This is a method of frequency transforming an image and embedding information in a specific frequency band. Since the embedded information is distributed over the entire image in real space, it is visually inconspicuous.

As an electronic watermarking method in frequency space, Non-Patent Document 1 proposes a method of DCT transforming an image (discrete cosine transform) and embedding watermark information in an intermediate frequency band in DCT space. The embedded information is dispersed and diluted throughout the image in real space. If the gain of the embedding strength is increased, the artifact is strongly expressed, and the shape of the artifact changes depending on the embedding position. If the embedded position is a plurality of intermediate frequency points radially from the position of the DC component (zero frequency), oblique streaks and long streaks of low-frequency noise are visible in the real space, and the image quality is degraded.

On the other hand, in the digital watermark method of embedding after Wavelet transformation, an image is DWT (Discrete Wavelet Transform) transformed and a watermark is embedded in the Wavelet component. When gain is increased, embedded patterns appear as artifacts on the inverse transformed image. The reason for this is that the Wavelet transform is a spectral intensity transform that preserves the spatial position of the image, so it appears as it is at the embedding position without being distributed over the entire image in real space. Those embedded in the LL component appear stronger than those embedded in the HH component. In addition, in multi-layered transformations, the higher the layer is embedded, the stronger the artifact becomes. For this reason, methods such as embedding and dispersing them in edge portions are taken, but they are highly dependent on the image. If the gain is reduced, there is no problem in terms of image quality, but it becomes a serious problem when high tolerance is required.

Therefore, a method of embedding a green noise pattern in image data (hereinafter referred to as a green noise diffusion method) disclosed in Patent Literature 1 was previously proposed as a high-resilience, high-quality digital watermarking method. The green noise pattern has its spectrum confined to a specific band f_(max) to f_min, and its main frequency can be set to a frequency band that is difficult to recognize by human visual characteristics. For this reason, when used for digital watermarking, when the embedded image is observed at a distance of clear vision, the dot pattern exhibits low responsiveness and is difficult to recognize. For this reason, even if the embedding strength (gain) is increased to make editing and processing tougher, artifacts are not felt, and high durability can be maintained while maintaining high image quality.

In the green noise diffusion method, an embedding operation is performed by dividing an image into blocks and superimposing a green noise dot pattern corresponding to embedding bits on the image data. For watermark extraction, FFT (Fast Fourier Transform) is applied to the embedded image on a block-by-block basis, and embedded information is extracted from the spectrum distribution. It is a so-called asymmetric type in which the embedding is performed in real space and the extraction is performed in frequency space, and since the green noise pattern is a pattern of random dots, it is difficult to erase or tamper with the watermark itself.

However, it is assumed that the watermark will be removed by a so-called collusion attack in which the watermark pattern is specified from the original image, the pattern of the flat portion, a plurality of images, or the like. The embedded pattern can be extracted by taking the difference between the embedded image and the original image. Even if the original image is not available, the watermark pattern is partially extracted block by block from the image embedded in the flat part of the image (e.g. skyscape, seascape, etc.), and the entire embedding pattern is estimated from many blocks. It is possible to In order to avoid this problem and improve security, it is necessary to further improve resistance to these attacks.

JP 2018-182471

Tadashi Mizumoto, Kikoo Matsui, ``Study of print capture resistance of digital watermark using DCT'', Journal of Institute of Electronics, Information and Communication Engineers A, Vol J85-A, No.4, pp.451-459 (2002)

The challenge is to obtain a digital watermarking method that is resistant to image editing and processing, has little deterioration in image quality even when strongly embedded, and exhibits strong resistance to attacks from third parties, in which watermark information is not removed. be. A further challenge is that authors and purchasers can remove watermarks if desired. Another challenge is to create a secure system that prevents the damage from spreading over a wide area in the event of key loss.

In order to solve the above problems, the present invention performs digital watermark embedding by dividing an image into blocks and superimposing and embedding a random cluster dot pattern exhibiting green noise characteristics on the original image for each block.

The image data is divided into blocks, and in each block, the bit information (0 or 1) of the watermark to be embedded is selected from two basic dot patterns p0 and q0 that exhibit green noise characteristics and exhibit different spectral shapes, and their inverse patterns. and selecting an embedding pattern from the selection list, multiplying the pattern by a gain and superimposing it on the image data to generate a watermark-embedded image,
The watermark information is extracted by dividing the watermarked image data into blocks and extracting the embedded bit information from the shape of the Fourier spectrum pattern of the blocks.

Here, the inverted dot pattern is generated from the two basic dot patterns p0 and q0,
p1,q1: upside down dot pattern,
p2,q2: left-right reversed dot pattern,
p3,q3: vertical and horizontal reversed dot pattern,
p4,q4: negative dot pattern of p0,q0,
p5,q5: negative dot pattern of p1,q1,
p6,q6: negative dot pattern of p2,q2,
p7,q7: negative dot pattern of p3,q3,
All or part of 8 dot patterns pi, qi (i = 0, 1, …, 7) including 7 inversion patterns each,
The selection list indicates the selection order of the pi, qi dot patterns.

The basic dot patterns p0 and q0 consist of k patterns each having the same spectral shape but different dot patterns. part is a dot pattern for embedding,
The selection list indicates the selection order of those dot patterns.

Such a selection list is obtained by randomly generating the selection order of the plurality of dot patterns to be used by a random number generator.

Furthermore, such a digital watermark can remove the embedded watermark by a key composed of i basic dot patterns pi and qi, gain, embedded information and a selection list.

If the basic dot pattern qi is generated using the basic dot pattern pi, the digital watermark is embedded using a key composed of i basic dot patterns pi and gain, embedding information and a selection list. You can remove watermarks that have been

The present invention enables robust watermark embedding against attacks from third parties. Since there are a plurality of watermark-embedded dot patterns, including reversed dot patterns, it is difficult to estimate the basic dot pattern from the difference from the original image and the flat portion of the image.

Also, when a plurality of basic dot patterns are used, the number of dot patterns increases when the inverted patterns are included. Moreover, since the selection list is random, there is no regularity. This makes dot pattern estimation even more difficult.

On the other hand, the key for elimination normally requires all dot patterns, but since the reverse dot pattern is generated from the basic dot pattern, the key information is the basic dot patterns p0, q0, gain, embedded information and a selection list, and a small amount of keys to remove.

Furthermore, when the basic dot pattern q0 is generated from p0, the key information is only p0, gain, embedding information, and selection list, and the key to be removed becomes even smaller.

With such a key, the watermark can be removed from the watermarked image to restore the original image (original image). By changing the key for each distributed image, it is useless for other images. It is also possible to embed new secondary copyright information as an electronic watermark using the key.

1 is a diagram of a watermark embedding and extraction apparatus of the present invention; A diagram showing the distribution of watermarked images over the Internet, Basic dot pattern for embedding: (a) 32 × 32 with spectral characteristics of vertically elongated elliptical ring, (b) rotated by 90 degrees, (c) 64 × 64 vertically elongated elliptical ring and (d) rotated by 90 degrees. The figure showing a basic dot pattern and an inversion dot pattern. Embedded image with different gain and partially enlarged view, (a) gain=0.0625, (b) 0.125, (c) 0.25. FIG. 4 is a diagram showing a processing flow of watermark embedding; FIG. 10 is a diagram showing the flow of automatic gain adjustment processing when embedding a watermark; FIG. 4 is a diagram showing a processing flow of watermark extraction; A diagram of a mask pattern for watermark extraction. FIG. 4 is a diagram showing mask processing for watermark extraction; FIG. 10 is a diagram showing improvement in reliability by mask processing for watermark extraction; Watermark embedding, extraction, and removal: (a) is a watermarked image (gain=0.1875), (b) is a spectral image for watermark extraction, and (c) is a watermark-removed image. Diagram representing different dot patterns of the same spectrum. FIG. 4 is a diagram showing two basic dot patterns and an inverted dot pattern;

FIG. 1 is a configuration diagram of a digital watermarking system for information embedding and extraction according to the present invention. In the copyright holder's computer 3, copyrighted image data is stored in a data memory 7 such as a hard disk. The image data is image-processed by the image processing program in the program memory 6 using the CPU 11, ROM 4, RAM 5, etc., and is displayed on the monitor 8. FIG. A scanner 1 and a printer 2 are connected to the computer 3 , and the processed image can be output from the printer and the image can be read from the scanner 1 . Such image processing increases the computational load when handling images of a large size. Therefore, in some cases, a processing board such as a GPU is included for performing parallel processing or increasing the speed.

In order to embed confidential information and copyright information in the image, character information such as the author's name, date, URL, etc. is input from the keyboard 9 . The number of characters that can be embedded varies depending on the image size and block size. When the block size is 512 pixels x 512 pixels and the block size is 64 pixels x 64 pixels, the maximum size is 8 bytes, and when the block size is 32 pixels x 32 pixels, the maximum size is 32 bytes. This is 8 characters and 32 characters, respectively, if limited to ASCII code.

FIG. 2 shows a processing procedure for Internet distribution in one operation mode. The image embedded with the watermark information is distributed via the Internet 12 from the computer of the distributor or copyright holder 16 (13). The prospective purchaser 17 who sees it informs the distributor of his purchase desire (14). After a predetermined procedure, the distributor distributes the private key for watermark embedding and removal (15). Based on the contract with the distributor, the purchaser can remove the electronic watermark embedded in the sent image and edit the image.

FIG. 2 shows a processing procedure for Internet distribution in one operation mode. The image embedded with the watermark information is distributed via the Internet 12 from the computer of the distributor or copyright holder 16 (13). The prospective purchaser 17 who sees it informs the distributor of his purchase desire (14). After a predetermined procedure, the distributor distributes the private key for watermark embedding and removal (15). Based on the contract with the distributor, the purchaser can remove the electronic watermark embedded in the sent image and edit the image.

A third party who sees the image can confirm that the image has the copyright by using watermark extraction software. The watermark extracting software is not confidential and can be freely downloaded, for example, from the homepages of distributors and copyright holders, and made widely available for anyone to use. Third parties can know the owner of the image, copyright information, contact information, etc. by this, so it can also prevent unauthorized use and warn.

On the other hand, copyright holders and secondary copyright holders can track and monitor image data as copyrighted works by using the watermark. If a third party uses the image illegally, the copyright holder or secondary copyright holder will read the watermark information from the image using monitoring software, confirm that it is their own work, and that the third party will use it without permission. If so, it can be detected.

A key corresponds to each embedded image on a one-to-one basis. For this reason, the copyright holder uses a different key for each image to prevent abuse. This key cannot be used to remove watermarks on other images. Even if the key is lost or the key is obtained through a collusion attack by a third party, the damage will be limited to one image, and the watermark cannot be removed from the other images. At this time, the watermark reading software can be used in common even if the keys are different. Also. Since keys can be issued almost infinitely, the probability of having the same key is extremely low.

A detailed description will be given below along with examples.
Here, first, the dot profile creation algorithm showing the green noise characteristics used in the present invention will be explained.
Let R×R (R=2^m) be the desired dither matrix size, and g (0≤g≤1: g = 1 is all black, g = 0 is all white). do. Assuming that the blackening rate is g and the dot profile at the point (x, y) is p(x, y), the intermediate density (g=1/2) dot profile is obtained as follows.
(STEP 1) First, (R^2)/2 random dots (white noise in the initial state) are generated by a pseudo-random number generator, and let p(x,y) be generated. At this time, the dot profile in the initial state can be changed by changing the SEED value of the pseudo-random number generator. A two-dimensional Fourier transform is performed on this dot profile to obtain P(fx,fy).
(STEP 2) Multiply P(fx,fy) by filter D(fx,fy) to obtain new P'(fx,fy). Here, D(fx,fy) is a filter whose radial frequency fr has a value in the range of fmin to fmax.
(STEP 3) Perform an inverse Fourier transform on P'(fx,fy) to obtain a multivalued point profile p'(x,y).
(STEP 4) Obtain an error function e(x, y)=p'(x, y)-p(x, y), and perform white and black inversion in descending order of error at each pixel position.
(STEP 5) Repeat the above operation until the error is within the allowable amount, and finally obtain a dot profile p0 of g=1/2.

Here, setting of the radial frequencies fmax and fmin will be described. The frequency due to the average dot spacing at g=1/2 at fmax and fmin of filter D(u,v) is
fo=√g・fn=√(1/2)・fn
and fmax and fmin are based on fo,
a≡(fmax fo)/fn
b≡(fmin fo)/fn
Define the parameters (a,b) as Here, fn indicates the Nyquist frequency. By changing (a,b), dot patterns with different cluster sizes can be obtained.

As an example, find the dot profile when R=32, (a,b)=(-1/16,-5/16), and ellipticity 1.5. Two types of patterns corresponding to embedding bits (0, 1) are required for watermark embedding. Therefore, p0 is a vertically long ellipse, and q0 is a horizontally long ellipse pattern. The dot pattern p0 and spectral characteristics are shown in FIG. 3(a). An ellipticity of 1.5 means that the filter D(fx,fy) is enlarged 1.3 times in the y-axis direction. The imaginary part is 0 because the spectral distribution is symmetrical with respect to the x and y axes. . Figure (b) shows the dot pattern q0. This is generated individually, but in order to reduce the memory capacity of the key described later, p0 rotated by 90 degrees is substituted. The spectrum is similarly rotated by 90°. (c) and (d) of the figure show p0 and q0 when R=64. The spectral shape becomes clearer.

Next, an algorithm for embedding watermarks in image data will be described. In embedding watermark information, color image data is converted into Y, Cb, and Cr, and watermark information is embedded in luminance Y. Embedding in blue (yellow when printing) is ideal because it is the least visually recognizable, but this is because it is difficult to accurately separate colors from printed images and monitor images read by cameras and scanners. .

Subsequently, the image data is divided into blocks of R pixels×R pixels. It should be the same size as the dot pattern size. A block size of 64 pixels×64 pixels is normally used, but if a large amount of information is desired to be embedded, a block size of 32 pixels×32 pixels enables embedding of four times as much information.

Watermark information is embedded block by block in real space. Let i(x,y) be the image data, gain be the embedding strength, and w(x,y) be the watermarked image.
When padding bit = 0: w(x,y)=i(x,y)+gain・p0(x,y) (1)
When padding bit = 1: w(x,y)=i(x,y)+gain・q0(x,y) (2)
becomes. As for the watermark information, a character string is converted into a bit string in advance, and the watermark information is sequentially embedded in block units. Here, p0 and q0 are binary values of (1,0), but are set to (1/2,-1/2) to preserve the average luminance.

Since p0 and q0 are random dots, the boundaries between them are inconspicuous. The dot pattern, which exhibits green noise characteristics, is also used as an FM screen in printing. Dispersive dots are excellent in uniformity, so they are visually uniform and do not give the impression of graininess.

If only p0 and q0 are embedded in the image data, there is a risk that the embedded pattern will be reproduced by a third party analyzing the pattern. Since the patterns of p0 and q0 are repeatedly embedded, the original pattern can be estimated by observing them closely. Especially in images with a uniform background, such as the sky or the sea, the dot pattern is easy to identify, and it is possible to obtain an estimated pattern based on the information obtained from each block.

For this reason, the present invention uses a large number of embedding patterns to prevent easy estimation. In FIG. 4, by generating reversed dot patterns from one green noise dot pattern p0, q0, each pattern is multiplied into eight patterns.
Assuming that P0 and q0 are basic dot patterns, the following inverted dot patterns are generated from p0 and q0 as follows.
p1,q1: upside down dot pattern,
p2,q2: left-right reversed dot pattern,
p3,q3: vertical and horizontal reversed dot pattern,
p4,q4: p0,q0 negative/positive inverted dot pattern,
p5,q5: p1,q1 negative/positive inverted dot pattern,
p6,q6: p2,q2 negative/positive inverted dot pattern,
p7,q7: p3,q3 negative/positive inverted dot pattern,
FIG. 4 shows inverted dot patterns pi, qi (i=0, 1, . . . , 7) for 32×32. All or part of this dot pattern is used to embed watermark information in the image.

The order of embedding dot patterns is determined by a selection list L for embedding watermark information. The selection list L is, for example, a list for selecting which dot pattern number to use, and is determined by a random number before embedding. Now, if all pi,qi (i=0,1,2,...,7) are used, random numbers from 0 to 7 are generated by the number of blocks by the random number generator. For example, we obtain the following selection list L.
L={0,7,5,2,4,7,1,5,6,1,6,4,2,7,6,0,3,1,6,0,3,2,7,5 ,...} (3)
Using such a selection list, the number i of the dot pattern to be used in block order is determined. That is, if the block number is n,
i=L(n)
and pi and qi are determined. For example, p0,q0 if i=0, and p7,q7 if i=7. Using this pi,qi, the embedding is done for each block as
When padding bit = 0: w(x,y)=i(x,y)+gain・pi(x,y) (1')
When padding bit=1: w(x,y)=i(x,y)+gain・qi(x,y) (2')
becomes.

Since embedding is performed in units of blocks, the image size can be any size. The amount of data that can be embedded in an image (maximum number of bits) N is given by
N=int(W/R)×int(H/R) (4)
is given by Here, int() represents rounding down after the decimal point. The number of characters that can be embedded is int (N/8) characters for ASCII characters. Normally, images for which copyright protection is desired are images of full HD size (1920 pixels x 1080 pixels) or larger. For full HD images, N = 480 bits = 60 bytes, which is sufficient for embedding copyright information. . However, if a small size image requires a sufficient amount of information, embedding it with R = 32, for example, in the case of 512 pixels x 512 pixels, N = 256 bits = 32 bytes of information can be embedded. be.

Embedding images can be made more robust by increasing the gain. However, increasing the gain makes the dot pattern visible. FIG. 5 shows an image embedded with a different gain and its partially enlarged image. (a) of the same figure is embedded with gain=0.0625, and the embedded information is hardly visible. However, watermark extraction is error prone. On the other hand, (c) of the same figure is embedded with gain=0.25, and the watermark extraction is almost 100%, but the dot structure becomes noticeable. As a result of the experiment, it was found that if the gain is 0.125 to 0.25, the green noise dot pattern is inconspicuous due to the low-pass effect due to the visual characteristics, and it is durable and practical.

FIG. 6 shows a watermark embedding processing flow. It is assumed that the dot pattern to be embedded already exists. First, a watermark information bit string wm(i,j) for embedding and a selection list L(n) are prepared (20). Here, i represents the i-th bit of the character to be embedded, and j represents the j-th bit, which are converted in advance to a one-dimensional bit string. Subsequently, the image data is divided into R×R blocks (21), and dot patterns are synthesized for each block. We start with the first block n=1 (22). Next, the selected dot pattern number i is obtained from the selection list L(n) from the block number n, and the embedded dot patterns pi and qi are determined (23). For embedding, it is determined whether the embedding bit is 0 or 1 (24). If it is 1, qi is embedded (25), and if it is 0, pi is embedded (26). Subsequently, it is determined whether or not the block has ended (27), and if not, the block number is incremented by 1 (28), and the process returns to (23). The process ends when all blocks have been filled.

The embedded watermark information may not be extracted because the embedded gain is small. Such happens when the spatial frequency of the corresponding block of the original image is high. FIG. 7 shows the flow of automatic gain adjustment. First, watermark information wm(i,j) is embedded in all blocks of the image with gain=g (30). After embedding all the blocks, the watermark is read (32). Here, the extracted watermark information wm'(i,j) and the watermark information wm(i,j) before embedding are compared to determine whether they match. At the same time, it is determined whether or not the reliability wmrel'(i,j) of extracted bits, which will be described later, is greater than or equal to a certain threshold (33). If the watermark information does not match, or if the reliability is below the threshold value even if the watermark information matches, a constant value Δg is added to the gain (34), and the watermark is embedded again. Since the reliability can be quantified, it is possible to automatically embed watermarks with the optimum watermark strength as described above. Even if the watermark information matches, if the reliability is less than the threshold, it may not be possible to read the watermark.

Next, watermark extraction will be described.
Extraction of watermark information is performed by correcting the size and inclination of the read image, and then obtaining a spectrum image for each block. By Fourier transforming equation (2),
F{w(x,y)}=F{i(x,y)} + gain・F{pi(x,y) or qi(x,y)} (5)
becomes. where F{ } represents the Fourier transform. Normally, F{i(x,y)} is the spectrum of the original image, localized near 0 frequency, and F{pi(x,y) or qi(x,y)} is the ring-shaped spectrum surrounding it. becomes. Therefore, there is little overlap between the two, and it is easy to separate the two.

FIG. 8 shows the processing flow of watermark extraction. First, the input image is divided into R×R blocks (41), and FFT is performed on the first block (42). The resulting spectral image is then subjected to histogram equalization (43). This is for improving the contrast of the spectral pattern and for standardization in the quantification of the degree of reliability, which will be described later. Subsequently, mask processing (44), which will be described later, is performed to obtain watermark information wm'(i,j) and reliability wmrel'(i,j). It is determined whether or not all blocks have been completed (46), and if not completed, the process goes to the next block and the same processing is performed again.

FIG. 9 shows a mask pattern as a discriminator for extracting watermark information by pattern recognition from the spectrum distribution after Fourier transform. It is a vertical and horizontal elliptical ring-shaped mask that corresponds to the spectral characteristics of the dot pattern for embedding. Assume that the black area in the figure has a value of 1, and the white area has a value of 0. This mask is superimposed on the watermarked spectrum, and whether the bit is 0 or 1 is determined from the difference in the integrated luminance value of the overlapped portion. That is, in FIG. 10, the following integrated luminance value outputs Q0 and Q1 are obtained for the block spectrum pattern W(x,y) (40).
Q0=(1/Z)ΣM0(x,y)・W(x,y)
Q1=(1/Z)ΣM1(x,y)・W(x,y)(6)
From such output,
When Q0>Q1, extraction bit = wm'(x,y)=0
When Q1>Q0, extraction bit = wm'(x,y)=1
becomes (41). At the same time, the reliability is obtained as follows (42).
wmrel'(i,j)=|Q0-Q1|(7)
The higher the reliability, the higher the reliability of the extracted bit.

FIG. 11 explains how to improve the accuracy and reliability of watermark information. The size of the image gives the maximum number of embedding bits N from equation (4). Assuming that the number of bits to be embedded is n and n<N, duplicate bit strings can be embedded. In the extraction of watermark information in Fig. 8, continuous watermark information wm' and reliability wmrel' up to N were obtained regardless of the number of bits (without considering character string repetition). Watermark information up to n is obtained by selecting a highly reliable character string from this continuous character string.

First, when the s-th character number (0<s≤N) of the embedded character string and t is the bit number (0≤t≤7), the remainder k modulo the number of characters n is k = smod n (50), when the (s,t)-th reliability wmrel' is wmrel'(s,t)>wmrel(k,t), the reliability is high, so
wm(k,t)=wm'(s,t)
wmrel(k,t)=wmrel'(s,t) (8)
(52) to replace the watermark information. This is done for all bits and all blocks until s reaches N, that is, until the number of embedding bits N ends.

By re-dividing and extracting embedded blocks with low reliability in this way, extraction with high reliability is possible. Since all of the blocks divided into four represent the same spectral characteristics, if even one high reliability is extracted, the entire block is upgraded to that reliability. In this way, highly reliable watermark extraction is possible for the printed image.

Next, watermark removal will be described.
By knowing the embedding pattern, the watermark information can be removed to obtain the original image. That is, from equations (1') and (2'),
When padding bit = 0: i(x,y)= w(x,y)-gain・pi(x,y) (9)
When padding bit = 1: i(x,y)= w(x,y)-gain qi(x,y) (10)
and knowing the embedding patterns pi(x,y), qi(x,y) and gain, the embedding information and the selection list, we can return to the current image. These pieces of information can be obtained by receiving a "private key". In order to completely restore the original image, it is necessary to limit the dynamic range in advance to avoid overflow and underflow of the embedded image.

FIG. 12 is obtained by embedding 8 characters "20181118" in an image of 256 pixels×256 pixels using the dot pattern of FIG. (a) shows a diagram when embedded with gain=0.1875, and (b) shows a spectrum for extraction, and the embedded watermark information could be accurately extracted. (c) is an image obtained by removing the watermark from the embedded image of (a) using a key, and the original image is reproduced.

Only the basic dot patterns p0 and q0 may be included in the secret key, and the inverted dot pattern can be generated from these basic dot patterns. Therefore, when R = 64, the key size is 4096 bits (512 bytes) for both p0 and q0, plus the embedded information and selection list, which is proportional to the number of characters. However, in general, it is about several tens of bytes to hundreds of bytes, and the capacity of the dot pattern is almost the capacity of the key.

If the basic dot pattern of q0 is a pattern obtained by rotating the basic dot pattern of p0 by 90°, the key will be the basic dot pattern p0, gain, embedded information and selection list, and the key capacity will be even smaller. .

Next, the attack resistance of this method will be explained. Attacks on digital watermarking include signal enhancement (sharpness adjustment), noise addition, filtering (linear, nonlinear), lossy compression (JPEG, MPEG), and transformation (rotation, scaling). These attacks are attacks on luminance information, and embedding by a set of dispersed dots like our method can be maintained as long as there is no attack in the direction of resolution (for example, a strong low-pass filter), and is generally robust. I can say. Tolerance can be further enhanced by increasing gain.

Next, the possibility that a third party fraudulently detects and removes the embedded pattern will be described. In order to specify the embedding pattern, it is generally assumed that the watermark pattern is specified from an original image in which no watermark is embedded or a plurality of embedded images. If only the basic dot pattern is used, the same pattern is repeated many times because these are used repeatedly. By connecting the partial patterns obtained from each block, the dot pattern of the original block size can be estimated. .

In order to avoid this, it is conceivable to have many basic dot patterns. This is because using a plurality of dot patterns makes it difficult to identify the pattern. However, the capacity of the key increases and becomes difficult to handle. Therefore, by generating an inverted pattern from a basic dot pattern, the number of patterns can be increased and an increase in key size can be prevented.

For loss or theft of a key, by using different basic dot patterns for each image, the same key cannot remove the watermark of another image. FIG. 13 shows different dot patterns with the same spectral characteristics. These are obtained by changing the seed value of the initial random number when generating the green noise pattern. The spectral distribution is the same, the extraction software is common, but the dot profile is different. In the case of R=64, only (_4096^)C_2048 different patterns are possible, and the probability of having the same pattern is extremely low, so security is high.
The copyright holder provides each purchaser with a dot pattern obtained from different random numbers as a secret key. Since this key is different for each purchaser, it is not possible to remove the watermark from another purchaser's image.
As described above, the present invention provides a highly resistant digital watermark that eliminates attacks from third parties.

Next, a method of increasing the number of basic dot patterns and obtaining a digital watermark with higher attack resistance is shown. FIG. 14 uses two basic dot patterns (pattern 1 and pattern 2) as basic dot patterns.
As shown below, 8 patterns p0 to p7 including 7 inversion patterns from pattern 1 are obtained. Eight patterns p8 to p15 are obtained from pattern 2 in the same way.
p0, p8: basic dot pattern
p1, p9: upside down dot pattern,
p2, p10: left-right reversed dot pattern,
p3, p11: vertical and horizontal reversed dot pattern,
p4, p12: p0, p8 patterns of negative/positive inverted dot patterns,
p5, p13: p1, p9 negative/positive reversed dot pattern,
p6, p14: p2, p10 negative/positive reversed dot pattern,
p7, p15: p3, p11 negative/positive reversed dot pattern,
A reverse dot pattern is similarly generated from q0 and q8, which have spectra obtained by rotating p0 and p8 by 90°.

When using all patterns, the selection list L selects which of the 16 patterns p0 to p15 and q0 to q15 to use. Let it occur for a few minutes. For example, we obtain the following selection list L.
L={4,10,5,12,9,7,14,15,6,11,6,4,2,7,6,8,13,1,6,0,3,12,7,… } (3)
Using such a selection list, the number i of the dot pattern to be used in block order is determined. That is, if the block number is n,
i=L(n)
and pi and qi are determined. For example, if i=4, p4,q4, if i=10, p10,q10. Using this pi,qi, the embedding is done for each block as
When padding bit = 0: w(x,y)=i(x,y)+gain・pi(x,y) (1')
When padding bit=1: w(x,y)=i(x,y)+gain・qi(x,y) (2')
becomes.

Although the case where there are two basic dot patterns p0 and q0 has been described above, it is assumed that the number of basic dot patterns p0 and q0 is further increased to include k patterns, respectively. Since a maximum of seven reversed patterns are generated for each basic dot pattern, the total number of basic dot patterns and reversed dot patterns is a maximum of 8×k, and all or part of them are used as dot patterns for embedding.
The selection list indicates the selection order of those dot patterns.

As in the first embodiment, the watermark is extracted by first dividing the watermarked image into R×R blocks, performing FFT on each block, performing masking, and obtaining watermark information and reliability. Even if the number of dot patterns is increased and embedded, the extraction software does not need to be changed.

Increasing the number of dot patterns is effective in avoiding attacks from third parties. However, the size of the key increases and becomes difficult to handle. In the present invention, in order to reduce the capacity of the key, a reverse pattern is generated from the basic dot pattern, and only the basic dot pattern information is used as the key information. For this reason, the keys are compact and easy to handle, such as distribution and storage.

Furthermore, by generating a basic dot pattern q0 with a spectral shape rotated by 90° from the dot pattern of p0, the key size is further reduced. However, since reducing the number of basic dot patterns makes it easier to be attacked, it is necessary to select the optimum technique according to the level of resistance.

The digital watermarking method of the present invention has been described above, but since this method embeds a dot pattern that exhibits green noise characteristics, it is possible to embed strongly with little visual discomfort. Highly resistant to attacks from humans. In addition, since it is possible to restore the original image by key information, it is effective not only for copyright protection but also for various applications by linking printed images and electronic data.

In addition, since the watermark is embedded in the real space in units of blocks, there is no restriction on the size of the image such as a large image, and processing can be performed at a high speed equivalent to that of the dither method. Therefore, it is possible to embed moving images in real time, and it is expected to be applied to surveillance cameras and drive recorders.

The present watermark embedding and extraction method is algorithm-friendly. All confidentiality is encapsulated in the dot pattern, which is stored as the private key. Even if the key is lost or stolen, if a different dot pattern is used for each image, since there is, in principle, one key for one image, another image can be watermarked using that key. cannot be removed. For this reason, security is extremely high, and there are no problems with disclosure of algorithms, standardization, and disclosure of extraction software.

1 is scanner, 2 is printer, 3 is computer system, 4 is ROM, 5 is RAM, 6 is program memory, 7 is data memory, 8 is monitor, 9 is keyboard, 10 is communication function, 11 is CPU, 12 is Internet, 13 image data distribution, 14 purchase request notification, 15 secret key delivery, 16 copyright holder's computer, and 17 purchaser's computer.

Claims (6)

In the digital watermarking method for embedding watermark information in digital image data, the image data is divided into blocks, and in each block, two basic dot patterns p0 and q0 exhibiting green noise characteristics and exhibiting different spectral shapes, and their inverse patterns are used. bit information (0 or 1) of the watermark to be embedded and an embedding pattern are selected from the selection list, and the pattern is multiplied by gain and superimposed on the image data to generate a watermark-embedded image;
An electronic watermarking method characterized by extracting watermark information by dividing watermarked image data into blocks and extracting embedded bit information from the shape of the Fourier spectrum pattern of the block. The reverse pattern is generated from the two basic dot patterns p0 and q0,
p1,q1: upside down dot pattern,
p2,q2: left-right reversed dot pattern,
p3,q3: vertical and horizontal reversed dot pattern,
p4,q4: p0,q0 negative/positive inverted dot pattern,
p5,q5: p1,q1 negative/positive inverted dot pattern,
p6,q6: p2,q2 negative/positive inverted dot pattern,
p7,q7: p3,q3 negative/positive inverted dot pattern,
All or part of 8 dot patterns pi, qi (i = 0, 1, ..., 7) including 7 inversion patterns each,
2. The digital watermarking method according to claim 1, wherein said selection list indicates the selection order of pi, qi dot patterns. The basic dot patterns p0 and q0 are each composed of k patterns having the same spectral shape but different dot patterns, and all or part of a total of 8×k dot patterns including the inverted dot pattern. is the dot pattern for embedding, and
2. The digital watermarking method according to claim 1, wherein the selection list indicates the selection order of those dot patterns. 4. The digital watermarking method according to claim 3, wherein the selection list is obtained by randomly generating the selection order of the plurality of dot patterns to be used by a random number generator. The watermark embedded in the image embedded with the digital watermark can be removed by a key composed of i basic dot patterns pi and qi, gain, embedding information, and a selection list. Item 5. The digital watermarking method according to item 4. The digital watermark is characterized in that the basic dot pattern qi is generated from the basic dot pattern pi, and the key is composed of i basic dot patterns pi, gain, embedded information, and a selection list. The digital watermarking method according to claim 4.

Images