JP2020184663A - High resistance digital watermarking method - Google Patents

Abstract

To obtain a robust digital watermarking method that maintains high image quality and resistance to editing and processing, makes it difficult to decipher a pattern by collusion attack from a third party.SOLUTION: In a green noise diffusion method in which image data is divided into multiple blocks and watermarks are embedded in different patterns that indicate green noise characteristics corresponding to the embedded information for each block, the phase of the pattern is shifted for each block, thereby making pattern decoding by collusion attacks difficult and improving attack resistance. Shift information is locked at the same time as the pattern information and is used at the time of removal.SELECTED DRAWING: Figure 11

Description

The present invention is a digital watermarking method that embeds watermark information such as copyright information in digital image data. In particular, invisible and reversible electrons that show high resistance to attacks from third parties even if the algorithm is disclosed. It is about the watermarking method.

In the current digital information society, information can be easily duplicated, and many people can share information, and the society has greatly developed. However, its convenience has also caused incidents such as copyright infringement by illegally copying and distributing the copyrighted works of others. With the recent improvement in the image quality of digital cameras and printers, it has become easier to obtain duplicates of images that are exactly the same as the original images, and not only infringing copies but also forgery of banknotes and securities. This has resulted in the promotion of malicious criminal acts called acts.

Under such circumstances, invisible digital watermarking technology that embeds other information, such as author information, in image information to protect copyrights and warn against illegal copying. Has developed. This digital watermarking technology, for example, for the purpose of copyright protection, not only draws the user's attention by embedding the copyright holder, URL, contact information, terms of use, etc. in the copyrighted work of the document or image. , In case of unauthorized use, it is possible to track it. In this way, digital watermarks are beginning to be widely used as security measures such as copyright management / protection and monitoring of unauthorized copying.

The watermark as a security measure must be robust and highly resistant and cannot be easily erased or removed. This is because if the watermark information can be removed or erased by editing or processing the image or by a simple operation, it can be illegally used to escape the eyes of tracking and monitoring.

Usually, in order to enhance the resistance, it is necessary to embed a digital watermark in a deep layer of image data. Therefore, the strength of embedding (gain) is increased for embedding. This has the adverse effect of increasing damage to the original image and deteriorating the image quality. That is, image quality and tolerance are in a trade-off relationship with each other, and it is usually difficult to satisfy both.

As a digital watermarking method that simultaneously satisfies image quality and resistance, we have proposed a method of embedding the green noise pattern shown in Patent Documents 1 to 6 and Non-Patent Documents 1 and 2 in image data (hereinafter referred to as a green noise diffusion method). It was. The spectrum of the green noise pattern is confined in a specific band fmax to fmin, and its main frequency can be set to a frequency band that is difficult to recognize by human visual characteristics. Therefore, when used for digital watermarking, when the embedded image is observed at a clear vision distance, the dot pattern shows low responsiveness and the artifact is difficult to recognize. In addition, since it has high response sensitivity when writing or reading with a printer or scanner, embedded information can be retained even when printed. In order to increase the reliability, even if the embedding strength (gain) is increased, it is difficult for the artifacts to be visually recognized, and the image quality is maintained.

The green noise diffusion method is as robust as the spectral diffusion method in editing and processing, but it must also be robust against the removal of watermark information for unauthorized use by a third party. By a third party analyzing, estimating, and attacking the embedding algorithm with various methods, various images, and various information (hereinafter referred to as collusion attack), the watermark information is removed from the embedded image and returned to the original image. , Allows unauthorized use. In the algorithm-public digital watermark, it is easy to reverse-analyze the signal processing by knowing the algorithm, and therefore it is vulnerable to attack. Embedded parameters are usually held in the form of "keys". This key usually uses various parameters such as the position information of the pixel to be embedded, the strength (gain information), the random number string, and the password of the hash function, and the watermark cannot be removed without the key information. A collusion attack is to estimate the information of this key.

In the usage form as steganography, confidential information is secretly embedded in the image, and watermark extraction cannot be performed without a key. Therefore, the algorithm is private. However, the digital watermarking method for the purpose of copyright protection requires that any user can know the existence and conditions of copyright information, and it is necessary for anyone to easily extract the watermark. .. In addition, in order to spread widely, it is necessary to standardize as a method unified in the industry, and it is desirable that the algorithm can be published.

At this time, in order to ensure safety even if the algorithm is disclosed, it is usually considered that an asymmetric digital watermarking method is desirable. This is because the symmetric type extracts by the reverse procedure of embedding the watermark, so that the embedding position and method are easily discovered and attacked.
The green noise diffusion method is an asymmetric type in which embedding is performed in real space and extraction is performed in spectral space, and the green noise pattern is a pattern in which random dots are diffused, so it is generally difficult to completely remove it in the same manner as spread spectrum. .. Even if the algorithm is made public, the various embedded parameters are passed to the recipient in the form of a "private key" as secret information. Confidentiality is centralized in this key to ensure security, and normally the watermark cannot be removed without this key.

Furthermore, since the parameters such as the embedding pattern are changed for each image, the key information is different for each image. Therefore, even if the key is lost or stolen, the watermark cannot be removed by manipulating other images with the key.

On top of that, it is necessary to take measures against the above-mentioned collusion attack. In a collusion attack, the key information is calculated by taking the difference from the original image without a key, decoding the embedding parameters and patterns from the image with many flat parts, multiple images, embedding characteristics, image correlation, etc. .. Therefore, it is necessary to take measures against such attacks. In such a collusion attack from a third party, it is becoming possible to estimate key information by learning various patterns using machine learning using AI technology in recent years. In particular, in the block-type digital watermarking method, since the watermark information is embedded for each block, a large number of learning data for the number of blocks are collected from one image, so that the embedding parameters can be easily estimated.

Therefore, Patent Document 5 proposes a method in which the number of green noise patterns to be embedded is increased by using an inversion pattern, and the selection of a plurality of patterns is included in the key information as a random number string for the bit information (0, 1) to be embedded. did. The plurality of patterns consisted of a spatial inversion pattern of the original pattern and an inversion pattern of luminance data (negative / positive pattern), and eight inversion patterns were generated for one green noise pattern. This increases the types and number of patterns, reduces the risk of repeated occurrences of the same pattern, and improves safety.

However, since there is no pattern phase (shift amount of spatial position information) only in the inverted pattern and the phase is always the same, it is possible to identify the pattern from a lot of pattern embedding information, and it is still completely safe. Absent.

JP 2018-182471 JP 2018-201134 JP 2018-207247 Japanese Patent Application 2017-117460 Japanese Patent Application 2018-216362 Japanese Patent Application No. 2019-000178

N.Kawamura; "Digital Halftoning by Clustered Dither Screen Masks Representing Green-noise Characteristics", Proceeding of the 1st ICAI 2015, T109-3, pp.708-711 (2015). N.Kawamura; "Digital Halftoning by Stochastic Dither Screen Mask Representing Green to Blue Noise Characteristics", Trans. On IEVC, Vol.5, No.1, pp.34-41 (2017).

Therefore, it is an issue to maintain high image quality and resistance to editing / processing, and to obtain a digital watermarking method sufficiently strong against collusion attacks. That is, in the algorithm open type digital watermarking method, it is a problem to obtain a digital watermarking method showing strong resistance in which a third party cannot determine the embedded pattern by analyzing the embedded pattern without a key.

In order to solve the above problems, the present invention is a digital watermarking method in which an image is divided into blocks and watermark information is embedded for each block. When an embedded pattern showing green noise characteristics is superimposed and embedded on the original image, the pattern is embedded for each block. The phase of is shifted or fluctuated and embedded.

That is, in the digital watermarking method in which the image data I (x, y) is divided into R × R blocks and different patterns corresponding to the watermark information bits are embedded in each block, the embedded patterns are a plurality of different patterns exhibiting green noise characteristics. Create an embedded image by embedding with a pattern P'i (x, y) in which the phase of the embedded pattern is shifted by a different amount in the x and y directions for each block with Pi (x, y).
To extract the watermark information, the image data with the embedded watermark is divided into blocks, and the embedded information is read out by extracting the embedded bit information from the shape of the Fourier spectrum pattern of the block.
Watermark removal can be performed by using the embedding pattern P'i (x, y) and embedding strength and phase shift information to remove the watermark and return it to the original image. ..

The phase-shifted pattern P'i (x, y) is based on the basic pattern Pi (x, y) of R × R, and only (x0, y0) (however, 0 ≦ x0, y0 <R). It ’s a shift,
P'i (x, y) = Pi ((x + x0) mod R, (y + y0) mod R) (where mod is the modulo operation)
The image W (x, y) after embedding is calculated by multiplying the embedding strength (gain) and adding it to the image data.
W (x, y) = I (x, y) + gain ・ P'i (x, y)
The shift amount (x0, y0) varies from block to block, and the shift amount is stored as a phase shift list L (n) (n is a block number).
The key consists of the basic pattern Pi (x, y), the embedding strength gain, and the phase shift list L (n).
Watermark removal uses such a key,
I (x, y) = W (x, y) -gain · P'i (x, y)
The removal is performed by the operation. By doing so, it is possible to make it difficult to estimate the pattern from the collusion attack.

The embedded pattern Pi (x, y) is a green noise pattern generated from different initial values for each image, and the phase shift list L (m) is composed of different shift columns for each image, and these are incorporated. You cannot remove the watermark of other images with the key.

According to the present invention, it is possible to obtain a secure digital watermark that is highly resistant to editing and is resistant to collusion attacks from a third party, is robust, and cannot be removed without a key.

In addition, the original image can be restored with the key. This saves the trouble of having the image purchaser distribute the original image directly from the distributor, and since the original image is not distributed, there is no concern about leakage or theft, and it is safe in terms of security. Highly sex. Furthermore, the purchaser can use the key to embed the copyright information as the second author in the edited image.

The key is only the basic pattern Pi (x, y), gain (embedding strength), and phase shift list L (n), and the size of the key can be reduced and the key is easy to handle.

In addition, the basic pattern Pi (x, y) included in the key uses a different green noise pattern for each image, and even if the key is lost or stolen, the watermark cannot be removed from other images using this key. Therefore, the safety is extremely high.

The embedded image is secured by the shift pattern against collusion attacks that estimate the embedded pattern from the embedded image. Therefore, even if the algorithm is made public, the embedding pattern cannot be estimated from the algorithm, and it does not promote the attack.

Diagram of the device for embedding and extracting digital watermarks of the present invention, Diagram showing the distribution of watermarked images on the Internet, Diagram showing the flow of dot pattern generation for embedding, Diagram showing the rotation direction and embedded bits of the spectrum of the dot pattern, The figure which shows the shape of the filter at the time of generating a dot pattern for embedding Diagram showing the generated dot pattern and its spectrum and embedding information, Diagram showing the watermark embedding process flow, Diagram showing the watermark extraction process flow, Diagram showing the pattern identification processing flow, Diagram showing the mask pattern, Diagram showing the phase shift pattern, Diagram showing a phase shift pattern due to random number generation, In the figure showing the basic pattern, the phase shift pattern and their spectra, (a) shows the case of R = 64, and (b) shows the case of R = 32. An image when embedded in a 256 x 256 image, (a) is the original image, (b) is the embedded image, and (c) is the image showing the spectrum.

FIG. 1 is a block diagram of a digital watermarking system for embedding and extracting information of the present invention. In the computer 3 of the copyright holder, image data having the copyright 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, and the like, and displayed on the monitor 8. 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. Since such image processing increases the calculation load when handling a large-sized image, a processing board for performing parallel processing or speeding up such as GPU may be included.

In order to embed copyright information in an image, character information such as the author's name, date, and URL is input from the keyboard 9, but the number of characters to be embedded must be at least 16 characters. This requires 16 bytes (128 bits) to be embedded with ASCII code characters.

FIG. 2 shows one operation mode, and shows the processing procedure of Internet distribution. The image in which the watermark information is embedded is distributed and exhibited on the Internet 12 from the computer or server of the distributor or the copyright holder 16 (13). Upon seeing this, the purchase applicant 17 informs the distributor of the purchase request (14). After the prescribed procedure, the distributor distributes the private key for embedding and removing the watermark (15). Based on the contract with the distributor, the purchaser can remove the digital watermark embedded in the sent image and edit the image.

The purchaser can embed the further edited and processed image data as a secondary author by using this private key to embed information such as his / her own copyright information and URL. Images with these watermark information embedded are made public and distributed to the market for various uses.

A third party who sees the image can confirm that the image is copyrighted by the watermark extraction software. Watermark extraction software has no confidentiality. For example, it can be freely downloaded from the homepages of distributors and copyright holders, made publicly available, and can be used by anyone. This allows a third party to know the owner, copyright information, contact information, etc. of the image, which also prevents and warns of unauthorized use.

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 it illegally, the copyright holder or secondary copyright holder will read the watermark information from the image with monitoring software, and it will be his own copyrighted work, and the third party will use it without permission. If so, it can be detected.

There is a one-to-one correspondence between the keys for each image to be embedded. For this reason, copyright holders use different keys for each image to prevent misuse. This key cannot be used to remove watermarks in other images. Even if the key is lost or the key is acquired by a collusion attack by a third party, the damage is only the corresponding 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 the key can be issued almost infinitely, the probability of having the same key is extremely low.

Hereinafter, details will be described with reference to Examples.
Here, first, the generation of a dot pattern showing the green noise characteristics used in the present invention will be described with reference to FIG.
Now, let's assume that the matrix size is R pixel x R pixel (R = 2 ^ m where ^ is a power) and a black dot of 1 pixel size is embedded in it. Let the area ratio of black dots be g (0 ≤ g ≤ 1: g = 1 is all black, g = 0 is all white). Let the dot pattern at the point (x, y) be p (x, y), and obtain the dot pattern showing the green noise characteristic of g = 1/2 as follows.
(Step 1) First, a random dot p (x, y) is generated by a pseudo-random number generator (20). At this time, the initial state can be changed by changing the SEED value of the pseudo-random number generator.
(Step 2) A spectrum P (u, v) is obtained by performing a two-dimensional Fourier transform on p (x, y) (21). Here, (x, y) represents the spatial coordinates, and (u, v) represents the spatial frequency coordinates. When g = 0.5, (R ^ 2) / 2 random dots (white noise in the initial state) are generated and set to p (x, y).
(Step 3) A filter D (u, v) is applied to P (u, v) to obtain a new P'(u, v) (22). That is,
P'(u, v) = P (u, v) ・ D (u, v)
(Step 4) Inverse Fourier transform is performed on P'(u, v) to obtain a multi-valued point profile p'(x, y) (23).
(Step 5) Find the error function e (x, y) = p'(x, y) -p (x, y) (24), and invert the same number of white and black in descending order of error at each pixel position (25). ).
(Step 6) The operations from step 2 to step 5 are repeated until the error is within the permissible amount, and finally a dot pattern of g = 1/2 is obtained.

Here, the pattern setting of the filter D (u, v) will be described. Now, suppose that D (u, v) is a circular ring pattern, its maximum radial frequency is fmax, and its minimum radial frequency is fmin. The frequency f0 due to the average dot spacing at g = 1/2 is
f0 ＝ √g ・ fn ＝ √ (1/2) ・ fn
Given by, fmax and fmin are based on f0
a≡ (fmax --f0) / fn
b≡ (fmin --f0) / fn
The parameters (a, b) are defined as. Here, fn indicates the Nyquist frequency. By changing such (a, b), dot patterns having different cluster sizes can be obtained according to the visual characteristics.

Watermark embedding shall be performed in a 4-value pattern. For this purpose, four types of patterns corresponding to embedded 2-bit (00,01,10,11) are required. Therefore, the pattern of the filter D (u, v) is an elliptical ring pattern, which is four patterns rotated by 45 °. FIG. 4 schematically shows a filter pattern, where the minimum value on the short axis of the elliptical ring is fmin and the maximum value is fmax, the minimum value on the long axis is β · fmin, and the maximum value is β · fmax. Here, β represents a magnification. An upright filter pattern corresponding to 2-bit (00), a 45 ° rotation of it to (01), a 90 ° rotation to bit (10), and a 135 ° rotation to (11). It shall correspond to each. The four watermark patterns obtained in this way are expressed as Pi (x, y) (i = 0, ..., 3).

FIG. 5 shows the shapes of the four filters (filter 0, filter 1, filter 2, and filter 3). In the figure, the black area represents 0 and the white area represents 1. FIG. 6 shows a pattern and a spectrum created through the above steps using this filter D (u, v). Each pattern has a size of R × R, and pattern 0 corresponds to 2-bit embedded information (00), pattern 1 corresponds to (01), pattern 2 corresponds to (10), and pattern 3 corresponds to (11).

Next, a method of embedding a watermark in image data will be described. The watermark information is embedded in the blue (B) data of the color image data consisting of red (R), green (G), and blue (B), or becomes a luminance and color difference signal (Y, Cb, Cr). It is conceivable to convert it into a signal and embed it in the luminance (Y) data. When handling watermark information only in electronic data, it is ideal to embed it in blue (B) because it is most difficult to visually recognize. When considering applications such as reading printed images and monitor images with a camera or scanner, it is difficult to accurately color-separate blue (B) data (yellow when printing), so use brightness (Y) data. Is good.

Subsequently, the processing procedure will be described along the flow of FIG. First, the embedded information is converted into a bit string and used as wm (20). For example, if the ASCII 16 characters whose information to be embedded is "0123456789abcdef" are expanded into a bit string, the 128-bit string becomes wm = "0011000000110001 ...". Let wm (i) be the i-th bit from the first bit.

Next, the image data is divided into blocks of R pixels × R pixels (21). This should be the same size as the dot pattern. As the block size, 64 pixels x 64 pixels can be used for high-precision extraction, but if you want to embed a lot of information, if you use 32 pixels x 32 pixels as the block size, you can embed information four times as large.

The embedding work is done in block units. As an embedding in the nth block (23), first, it starts from n = 1 (22). Two consecutive bits of the embedded watermark information are extracted from the beginning, and this is set as b = (wm (i), wm (i + 1)).

The pattern to be embedded is selected according to the value of b (24). If b = (00), embed using pattern 0 (25a). Similarly, when b = (01), embed using pattern 1 (25b). When b = (10), embed using pattern 2 (25c). If b = (11), embed using pattern 3 (25d).

The watermark information is embedded in the real space (pixel space). Now, if the image data is I (x, y), the embedding strength is gain, and the image data with the watermark embedded is W (x, y),
W (x, y) = I (x, y) + gain ・ Pi (x, y) (i = 0,1,2,3) ------ (1)
However,
When b = (00), Pi (x, y) is pattern 0
When b = (01), Pi (x, y) is pattern 1
When b = (10), Pi (x, y) is pattern 2
When b = (11), Pi (x, y) is pattern 3
Will be. Pi is a binary value of (1,0), but it is set to (1/2, -1 / 2) to preserve the average brightness. As a result of embedding, the pixel data may exceed the dynamic range and overflow or underflow may occur. Therefore, the image I'(x, y) in which the following linear transformation is performed on the image data in advance may be performed.
I'(x, y) = (1-gain) ・ I (x, y) + gain / 2 ----- (2)
This process is indispensable when the watermark information described later is completely removed, but it may be omitted when the removal is not required or when the frequency of overflow or underflow is low.
It is determined whether all blocks are finished (26), and if not, the next block is moved to (27).

Embed all the embedded bit strings wm (i) as described above. If the number of embedded blocks is larger than the number of embedded bits, the reliability can be improved by repeatedly embedding. In this case, a delimiter marker indicating the beginning of the character string is required.
In the embedded image, the block boundary is inconspicuous because Pi (x, y) is a random dot. The dot pattern showing green noise characteristics is also used as an FM screen in printing, and since it is a dispersed dot and has excellent uniformity, it is visually uniform and does not give a feeling of graininess.

Subsequently, the extraction of the watermark information will be described along with the processing flow of FIG. The extraction work is divided into R × R blocks (30) as in the case of embedding, and is performed in block units. As the one to be extracted from the nth block, it starts from n = 1 (31). An R × R image is taken out from the nth block, FFT (Fast Fourier Transform) is performed, and the image is converted into spectral information Wn (32). Subsequently, normalization is performed by histogram equalization (33). This is because the brightness is different for each block, so that the contrast is improved to improve the identification accuracy, and the reliability described later is evaluated equally in each block. Subsequently, the pattern described later is identified (40), and 2-bit data of the embedded information is extracted from the identification pattern (34). It is determined whether or not all the blocks are completed (35), and if not, n is set to n + 1 (36) and the next block is moved to. When finished, all bits are taken out and the original character information is restored (37).

FIG. 9 shows a processing flow of the pattern identification processing (40). The spectrum information Wn in the block is matched with the prepared mask pattern, and its output Qi is obtained (41). As shown in FIG. 10, the mask pattern is the same as the filter pattern, and four masks 0 (M0), mask 1 (M1), mask 2 (M2), and mask 3 (M3) rotated every 45 ° are prepared. There is.
Returning to FIG. 9, the output Qi of matching with Wn is calculated by setting the i-th mask as Mi.
Qi = (1 / Zi) ΣMi ・ Wn (i = 0,1,2,3) -------- (3)
However, Zi = ΣMi
Is required (42). However, Σ is performed for each pixel in the block (in this case, it corresponds to the spatial frequency). As a result, the outputs (Q0, Q1, Q2, Q3) corresponding to each mask can be obtained. The largest of these four outputs is Qmax
When Qmax = Q0, b = 00
When Qmax = Q1, b = 01
When Qmax = Q2, b = 10
When Qmax = Q3, b = 11
The 2-bit watermark information b is obtained (43 and 44a, 44b, 44c, 44d).

Next, if the output next to Qmax is Qnext, the reliability rel is
rel = Qmax-Qnext ------------ (4)
It is represented by. That is, when Qnext is a value very close to Qmax, the probability of erroneous judgment is high, and the larger the difference between Qmax and Qnext, the higher the reliability. This difference can be used as an index of reliability.
As described above, the 2-bit embedded information b and the reliability rel can be obtained.

Watermark removal is possible by subtracting the embedding. That is, from equation (1),
I (x, y) = W (x, y) -gain ・ Pi (x, y) (i = 0,1,2,3) ------------- (5)
By multiplying the embedded pattern by gain from the watermark embedded image, it is possible to completely return to the original image. However, if linear dynamic range correction is performed to avoid overflow or underflow during embedding, it is necessary to perform reverse correction.
Watermark removal is possible as described above, but embedding patterns (pattern 0, pattern 1, pattern 2, pattern 3), gain, and embedding information are required. This information is given and received as a "private key".

For example, when the block size is 64 × 64, the capacity of the private key is 16 Kbits (2 Kbytes) for 4 patterns from pattern 0 to pattern 3, and gain and the capacity of embedded information are added. In the case of 32 x 32, 4 patterns from pattern 0 to pattern 3 are 4K bits (512 bytes). Since gain and embedded information are usually 100 bytes or less, most of the key capacity is occupied by embedded pattern information.

It is easier to handle if the capacity of the key is as small as possible. In order to reduce the key capacity, pattern 2 and pattern 3 can be generated from pattern 0 and pattern 1, respectively. That is, pattern 2 adopts pattern 0 rotated by 90 °, and pattern 3 adopts pattern 1 rotated by 90 °. Since the rotation of the dot pattern becomes the rotation of the spectrum as it is, there is no change in other processing.

When a key is lost or stolen, it is possible to prevent the watermark of another image from being removed with the same key by using a different dot pattern for each image. These are obtained by changing the seed value of the initial random number when the green noise pattern is generated. The spectral distributions are the same and the extraction software is common, but the dot profiles are different. When R = 64, the number of different patterns for one dot pattern is 4096 C 2048, and when multiple dot patterns are used, the product is the product. Therefore, the probability of the same pattern is extremely low, so the safety is low. high. Here, C represents combination. The copyright holder provides the 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 the images of other purchasers.

Next, the method of the phase shift pattern will be described.
Since Pi (x, y) is a green noise pattern, there are no boundaries. Therefore, the phase-shifted pattern also has an inconspicuous feature. Therefore, the pattern Pi'(x, y) whose phase of this pattern is shifted is generated.
Now, let P'i (x, y) be the phase of the basic pattern Pi (x, y) of R × R shifted by x0 in the x direction and y0 in the y direction. However, it is assumed that the shift amount is 0 ≤ x0, y0 <R. P'i (x, y) is expressed by the following equation.
P'i (x, y) = Pi ((x + x0) mod R, (y + y0) mod R) ---------- (6)
Where mod R represents the remainder modulo R. That is, when the values of x + x0 and y + y0 exceed R, it returns to the beginning and a phase-shifted pattern is obtained.

Using the phase-shifted embedding pattern P'i (x, y), embedding is performed as follows, as in the case without phase shift.
W (x, y) = I (x, y) + gain ・ P'i (x, y) ------- (7)
However, P'i (x, y) is shown in Eq. (6).

Figure 11 shows the phase-shifted embedding pattern P'i (x, y), which is based on the pattern (x0.y0) = (0,0) and is R / 4, R in the x, y directions. / 2 This is the pattern when the phase is shifted. When R = 32, the shift amount s has 16 patterns in all combinations, of which (0,0), (8,8), (16,16), (24,24), (0,24). ), (0,16), (16,8), and (24,0) are shown. In actual operation, the shift amount s is determined by the random number generator so that the shift amount is not known. Figure 12 shows a random number generated from 0 to (W-1), and the shift amount s has 1024 patterns, of which (22,25), (23,17), (6, Only eight patterns from s0 to s7 of (17), (28,13), (30,8), (9,14), (19,14), and (30,0) are shown.

All of these patterns show the same spectral characteristics. FIG. 13 shows the patterns with and without the phase shift and their spectral distributions. As an example, the shift value (x0, y0) = (R / 2, R / 2) is shown for the case of R = 64 (32,32) and the case of R = 32 (16,16). Even if it is shifted, the boundary of the pattern is not known, and its spectral shape does not change. Since watermark extraction does not depend on the amount of phase shift, the extraction software does not need to be changed or modified.

FIG. 14 shows a diagram when embedded in an image. Image 1 is a 256 × 256 LENNA image embedded with gain = 0.1875 for each 32 × 32 block, and is embedded with a 4-value green noise pattern using a random number phase shift pattern. There are a total of 64 blocks with 16 characters (ASCII) embedded in them. Figure (b) shows the embedded image, and (c) shows the 16 characters embedded in the spectrum image correctly extracted. Image 2 is embedded in a uniform image having a brightness value of 192, and is intentionally embedded with a large value of gain = 0.5 so that the embedded pattern can be identified to emphasize the pattern. As can be seen from these figures, the pattern boundaries are inconspicuous. This is because the green noise pattern is a random pattern and has no boundaries. Therefore, no matter how the values of x0 and y0 are changed, the boundary is not noticeable.

Since the amount of phase change can be R in 1 pixel increments in both x and y directions, a total of R ^ 2 is possible. If the block size is 32x32, 1024 patterns are possible. For 64x64
4096 patterns are possible. However, since a small shift amount is vulnerable to an attack, it is preferable to reduce the number of states and configure a large shift amount.

The phase shift amount (x0, y0) of this P'i (x, y) is changed for each block, and is stored as a phase shift list L (n) (n is a block number) and stored in the key. The removal of the embedded watermark is from the embedded pattern P'i (x, y), embedded strength gain, and phase shift list L (n).
I (x, y) = W (x, y) -gain ・ P'i (x, y) -------- (8)
As, it can be removed. In order to completely restore the original image, an integer type operation that takes rounding error into consideration is required.

Next, the effect of the phase shift will be described. In a collusion attack from a third party, it is conceivable to estimate the embedding pattern by synthesizing the images for each block. That is, the embedded image pattern Ci (x, y) is collected for each block corresponding to Pi (x, y) from the set of N embedded patterns according to the embedded information. Then, when the embedding pattern Ci (x, y) is added to obtain the average value, if the phase shift is not performed,
Ci (x, y) = (1 / N) ΣWi (x, y)
= (1 / N) ΣIi (x, y) + gain ・ (1 / N) ΣPi (x, y) ---- (8)
Will be. Here, N is the number of blocks corresponding to Pi (x, y), ΣWi (x, y) is the addition of image data with an embedded watermark, and ΣIi (x, y) is the addition of image data divided into blocks. Is shown. If N is made infinite here,
(1 / N) ΣIi (x, y) → const (DC component of image) ------ (9)
(1 / N) ΣPi (x, y) → Pi (x, y) ----- (10)
Therefore, Ci (x, y) is
Ci (x, y) = gain ・ Pi (x, y) ＋ const ----- (11)
And Pi (x, y) is extracted.
On the other hand, in the embedded image with phase shift, the phase of P'i (x, y) changes randomly.
(1 / N) ΣP'i (x, y) → const ------- (12)
Therefore, the embedded pattern Pi (x, y) cannot be extracted.

N must be large for Eq. (12) to hold. Now, assuming that the image size is W × H, the number of blocks of R × R is [W / R] × [H / R]. Here, [・] represents truncation to a decimal number or less. Assuming that this is embedded with the embedding pattern Pi (x, y), the maximum number of embedding patterns Pi (x, y) is [W / R] × [H / R]. For example, if an image of 256 × 256 pixels is embedded in a block of 32 × 32 size, the maximum number of occurrences of Pi (x, y) is 64. When Pi (x, y) has 4 values, there are 4 patterns, so the average number of occurrences is 16. Therefore, if there are 64 or more phase shift patterns P'i (x, y) per pattern, the probability that the same pattern will appear in the image is extremely low.

In order to support various image sizes, 256 phase shift patterns are prepared for one embedded pattern. That is, 256 patterns with x0 and y0 given the phase shift amounts of W / 16, 2W / 16, 3W / 16, ..., 15W / 16, respectively, are generated, numbered from 0 to 255, and added to the phase shift list. Make a list in embedded order. In a 512 × 512 size image, there are 256 blocks in a 32 × 32 block division, and the capacity of the phase shift list may be 256 bytes. The data in this phase shift list is repeatedly used for larger images.

In the block size of 32 × 32, the capacity of patterns 0 and 1 is 1024 bits (128 bytes), the phase shift list is 256 bytes, and the gain is several bytes, for a total of about 520 bytes. The embedded character information can be obtained at the time of extraction, so it is not necessary to include it in the key. The phase shift list L can be further reduced by reducing the number of patterns. Furthermore, when using a fixed random number sequence, the phase list can be omitted. In this case, only the start position of the fixed random number needs to be specified, so only a few bytes are required.

Since the embedded information amount V is four values when the image size is W × H, it is given by V = 2 × [W / R] × [H / R] bits. For example, if an image of 540 x 530 is embedded in a block with R = 64, V = 128 bits, and 16 characters can be embedded with the ASCCI character code. If you embed it in a 32x32 block, you can embed 64 characters. When using it for copyright protection, at least 16 characters are required, so if you embed it in a block size of 64 x 64, the image size is 512 x 512 or more, and if it is a block of 32 x 32, it is 256 x 256. Image size or larger is required.

As described above, in this method, a digital watermark that is strong against collusion attacks from a third party is obtained by giving a phase shift to the embedding pattern. This method becomes even more robust when used in combination with the inversion pattern introduced in Patent Document 5.
In addition, since the green noise diffusion method is originally highly resistant to various editing and processing, it can be said that it is tough against all kinds of attacks because it is also resistant to collusion attacks by this method.

The digital watermarking method of the present invention has been described above. However, since this method embeds a dot pattern showing green noise characteristics that are difficult to see, there is little deterioration in visual image quality and it is resistant to attacks such as image editing and processing. Tolerant. The reason is that this method is a spectral diffusion type watermark method. This is because white noise is superimposed in normal spectrum diffusion, but in the present invention, green noise is superimposed, so that the image can be easily separated from the spectrum.

On top of that, by using the phase shift pattern, the embedded dots are spatially diffused, the pattern is extremely difficult to decipher, and it shows strong resistance to collusion attacks. Therefore, it is highly resistant to all attacks and highly safe. Due to these characteristics, it is possible to embed a strong watermark in the usage form of copyright protection, and it is possible to detect illegal use, copying, falsification, etc. in various environments.

Since the watermark is embedded in the real space in block units, there is no limit to the image size such as a large image, and it can be processed as fast as the dither method. Therefore, it is possible to embed moving images in real time, and it is possible to use drive recorders and surveillance cameras.

In addition, since this method is completely reversible, it can also be used for tampering detection. That is, the hash value of the image data is embedded in the watermark as signature information. When tampering is detected, after removing the watermark with the key and completely returning to the original image, the hash value of the image is obtained, and when both match, it is proved that there was no tampering.
This tampering detection method is limited to a method in which the digital watermarking technology is completely reversible. This is because the Hash value changes when the image data changes as a result of embedding. Decoding the key by a collusion attack completely overturns such a system, and this method is also effective in preventing it.

1 is a scanner, 2 is a printer, 3 is a 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 The Internet, 13 is the distribution of image data, 14 is the notification of purchase request, 15 is the sending of the private key, 16 is the computer of the copyright holder, and 17 is the computer of the purchaser.

Claims (2)

In the digital watermarking method in which image data I (x, y) is divided into R × R blocks and different patterns corresponding to watermark information bits are embedded in each block.
The embedding pattern is a plurality of different patterns Pi (x, y) showing green noise characteristics, and the embedding pattern is embedded in the pattern P'i (x, y) in which the phase of the embedding pattern is shifted in the x and y directions by different amounts for each block. By creating an embedded image,
To extract the watermark information, the image data with the embedded watermark is divided into blocks, and the embedded information is read out by extracting the embedded bit information from the shape of the Fourier spectrum pattern of the block.
Watermark removal is a reversible digital watermarking method that can remove the watermark and return it to the original image using the embedding pattern P'i (x, y) and the embedding strength and phase shift information. The phase-shifted pattern P'i (x, y) is shifted by (x0, y0) (where 0 ≤ x0, y0 <R) based on the basic pattern Pi (x, y) of R × R. I let you
P'i (x, y) = Pi ((x + x0) mod R, (y + y0) mod R) (where mod R is the remainder modulo R)
The image W (x, y) after embedding is calculated by multiplying the embedding strength (gain) and adding it to the image data.
W (x, y) = I (x, y) + gain ・ P'i (x, y)
The shift amount (x0, y0) varies from block to block, and the shift amount is stored as a shift list L (n) (n is a block number).
The key consists of the basic pattern Pi (x, y), the embedding strength gain, and the shift list L (n).
Watermark removal uses such a key,
I (x, y) = W (x, y) -gain · P'i (x, y)
The digital watermarking method according to claim 1, wherein the digital watermarking method is performed by an operation.