JP2019195148A - Digital watermark apparatus and method - Google Patents

Abstract

To provide a method which can perform real-time embedding in each frame of a moving image by a digital watermark method capable of simultaneously performing strong embedding and detection of a tampered position.SOLUTION: A main information pattern and a sub information pattern for tamper detection are multiplexed so that the spectrum distributions of both of the patterns do not overlap each other, and an embedding operation is performed in a real space using a pattern obtained by synthesizing main information and sub information. A tamper detection position can be identified with the sub information, and embedding in a real space realizes embedding into a moving image with reduced calculation load.SELECTED DRAWING: Figure 5

Description

The present invention relates to an invisible digital watermark that embeds information in a document image, and more particularly to a digital watermark apparatus and method that can embed multiplexed information.

The digitization and development of networks in the information society has made it possible to replicate information easily, allowing many people to share information regardless of distance, and societies have greatly developed. However, its convenience has exacerbated various security problems such as copyright infringement, unauthorized information leakage, and information falsification. In recent years, high-quality image equipment has made it easy to obtain copies that are the same size as the original, and not only violating copyright infringement, but also malicious crimes such as counterfeiting of banknotes and securities. It is the result that encourages the act.

As a security countermeasure technique, a digital watermark technique has attracted attention. Unlike encryption and DRM (Digital Right Management), because other information is embedded directly in the document and image information, it is difficult to remove, and it is possible to detect unauthorized use, counterfeiting / tampering, and prevent and track unauthorized leakage. To do. In particular, invisible digital watermarks are embedded in information so that they cannot be recognized by the human eye, and the range of use is expanded because image quality changes and deterioration are not recognized.

Various methods of digital watermarking techniques have been proposed according to their usage. In order to maintain high resistance to various attacks such as editing, processing, conversion, and compression, a robust embedding algorithm is required. In order to prevent the watermark from being lost even when output to a printer or the like, a stronger watermark method is required.

On the other hand, in the detection of tampering and the like, it is required to specify the tampering position. Therefore, a weak and fragile (Fragile) embedding method is used in order to make it possible to specify the position by deleting the watermark information due to tampering.

In this way, because of the conflicting embedding method, the dedicated device or software must be changed depending on the application and purpose, and the operation and management become complicated, which is a factor that hinders its use and spread.
Therefore, a method has been proposed in which both are extracted and detected by the same method or algorithm without changing the method.

Non-Patent Document 1 proposes a digital watermark technique that can implement watermark embedding and tamper detection with the same technique. In this method, image data is converted into a frequency space, for example, wavelet conversion, embedded in an LL region that is a low frequency region, and watermark data and tampering detection are performed according to a rule at the time of extraction. Although it is an effective method for having both functions without requiring an original image, special hardware is required to support moving images such as drive recorders and surveillance cameras due to the complexity of the embedding operation. In addition, it is easy to be attacked from the outside by releasing the algorithm.

Watanabe, Madoka Hasegawa, Shigeo Kato: “Digital Watermarking System that Realizes Copyright Protection and Tampering Detection”, Journal of the Institute of Image Electronics Engineers of Japan, Vol.33, No.5, pp.756-764 (2004)

To solve this problem, a method that can simultaneously embed information firmly and identify a falsification position with one method, and that can embed at high speed without requiring special hardware, has disclosed the algorithm. However, it is a problem to obtain a technique that can maintain safety.

Therefore, the present invention is a multiplexed digital watermarking method that can detect the position when watermark information is embedded and tampered with a single digital watermarking device or technique, and does not require any special hardware. Provide a possible way. Embedding has means for multiplexing two types of information, main information and sub information, and embedding them simultaneously. The embedding of main information uses a green noise pattern, embedding highly resistant to attacks, and the embedding of sub information uses a pattern showing high spatial frequency characteristics, enabling embedding that is fragile and easy to remove.

By embedding these two pieces of watermark information, it is possible to embed a digital watermark that is highly resistant to attacks and to detect a falsification position against data falsification. Also, it is possible to increase the embedded information by embedding different information separately in both the main information and the sub information. In addition, the embedding can be performed at a very high speed because it is only necessary to write a pattern created in advance in the real space (pixel space) to the image data. This makes it possible to write to a movie without any special hardware.

As another effect, the sub information can be used as a synchronization signal of the main information. Since sub-information can be detected even for scaled images compared to the main information, the synchronization signal is detected from the enlarged / reduced watermarked image, and block synchronization at the time of main information extraction is performed with high accuracy. Can be done.

Furthermore, it is possible to remove the watermark if necessary. For this purpose, a key for removal is required. Using this key, the administrator can embed new information.

Even if the key is lost or stolen, the watermark cannot be removed from another image with the stolen key. For this reason, the security is high, and even if the algorithm is disclosed, the security is concentrated on this key, so it does not promote the attack.

In addition, by embedding information such as “copy prohibited” in the sub information, when copying the output image to paper, the sub information is visualized and used as a so-called copy-forgery-inhibited pattern that gives a warning at duplication. Is also possible.

Diagram of the digital watermark embedding and extracting device of the present invention Diagram showing the delivery of watermarked images over the Internet A diagram showing the main and sub-patterns and their spectra A diagram representing the spectrum of the combined pattern of the main and sub patterns (a) Main pattern, (b) Sub-pattern, (c) Composite pattern and each spectrum The figure showing the main information mask pattern and sub information mask pattern of watermark extraction A figure representing a logo or pattern as sub information The figure showing the image processing flow for detecting main information The figure showing the image processing flow for detecting sub information Image with embedded main / sub information An image from which watermark extraction has been performed, (a) an image from which main information has been extracted, (b) an image from which sub information has been extracted A watermark extracted from a tampered image, (a) a tampered image, (b) a sub-information extracted image When used as a synchronization pattern, (a) an image in which the synchronization pattern is embedded, (b) an image from which sub-information is extracted, (c) an image obtained by enlarging the image in (a) by a factor of 1.5, (d) ( Image obtained by extracting the sub information of c) This figure shows the evaluation results of JPEG resistance, (a) the image before compression, (b) the image when compressed with Q = 50, (c) the image when compressed with Q = 30, and main information of each , Sub-information extraction results When used as a background pattern, (a) is electronic data with embedded main and sub information, and (b) is an image representing the background pattern with the sub information visualized.

FIG. 1 is a block diagram of a digital watermark system for embedding and extracting information according to the present invention. The copyright holder's computer 3 stores copyrighted image data in a data memory 7 such as a hard disk. The image data is subjected to image processing by the image processing program in the program memory 6 using the CPU 11, ROM 4, RAM 5, etc. and displayed on the monitor 8. A scanner 1 and a printer 2 are connected to the computer 3, and processed images are output from the printer, and images can be read from the scanner 1. In such image processing, when a large-size image is handled, the calculation load increases, so there are cases where a processing board for performing parallel processing or speeding up a GPU or the like is included.

In order to embed confidential information or copyright information in an image, for example, the name of the author or character information such as URL is entered from the keyboard 9. The number of characters that can be embedded differs depending on the image size and block size. When the pixel size is 512 pixels × 512 pixels and the block size is 64 × 64, the maximum size is 8 bytes, and when the block size is 32 × 32, the maximum size is 32 bytes. This is 8 characters and 32 characters, respectively, for the ASCII code. This is not necessarily a sufficient amount. In addition to characters, image information such as icons, logos, and patterns may be required. Therefore, the amount of embedded information can be increased, and various uses and usage forms can be achieved by multiplexing and embedding both. Since such information is embedded in an image and can be made difficult to be recognized by human eyes as an invisible digital watermark, the range of use of the information is expanded.

FIG. 2 shows an Internet distribution processing procedure in one operation mode. The image in which the watermark information is embedded is distributed over the Internet 12 from the computer of the distributor or copyright holder 16 (13). The purchase applicant 17 who sees it notifies the distributor of the purchase request (14). After a predetermined procedure, the distributor distributes a secret key for embedding and removing the watermark (15). Based on the contract with the distributor, the purchaser can remove the digital watermark in which the sent image is embedded and edit the image.

The purchaser can embed the edited image data as a secondary author by using his secret key and information such as his own copyright information and URL. Images embedded with such watermark information are distributed in the market and used in various ways.

For illegal copying and usage monitoring, watermark extraction software can read watermark information from an image and confirm that it is a copyrighted work. This watermark extraction software has no secrecy, and can be freely released from the homepages of distributors and copyright holders, for example, and can be used by anyone. As a result, a third party can know the owner of the image, copyright information, contact information, and the like, which also prevents or warns unauthorized use.

If a third party has tampered with the image, it is possible to identify the tampered part. The owner of the work can be determined from the main information, and the presence or absence of falsification and the position can be specified from the sub information, so that fraud can be detected.

The secret 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 so as not to be abused. For this reason, the watermark of other images cannot be removed using this key. It is also resistant to collusion attacks as described below. At this time, the watermark reading software can be used in common even if the keys are different, as described above. Also. Since the key can be issued almost infinitely, the probability of the same key is extremely low.

Hereinafter, it demonstrates along an Example.
The green noise pattern used for embedding the main information of this digital watermark is generated by the following method.
Step1: First, R ^ 2/2 random dots are placed in the R × R area as an initial value and set to p (x, y). The initial value indicates white noise characteristics.
Step2: Next, two-dimensional Fourier transform of the dot pattern is performed to obtain P (fx, fy).
Step3: For P (fx, fy), the frequency fx, fy is
fx, min ≦ fx ≦ fx, max
fy, min ≦ fy ≦ fy, max (1)
A new spectral characteristic P ′ (fx, fy) is obtained by multiplying by the filter D (fx, fy) limited to the band of.
Step4: Perform inverse Fourier transform on P '(fx, fy) to obtain a multi-valued dot pattern p' (x, y).
Step5: Error function: e (x, y) = p '(x, y) -p (x, y) (2)
And reverse black and white in descending order of error.
Repeat Steps 2 to 5 to finish when the error function falls below a certain value and obtain the final pattern. Two types of anisotropic patterns are created as watermark embedding patterns. For the bit “0”, the filter D0 (fx, fy) has an elliptical ring distribution as shown below.
(fx ^ 2) / (fx, max ^ 2) + (fy ^ 2) / (fy, max ^ 2) ≦ 1,
(fx ^ 2) / (fx, min ^ 2) + (fy ^ 2) / (fy, min ^ 2) ≦ 1 (3)
The dot pattern thus obtained is defined as p0 (x, y), and the dot pattern rotated by 90 ° is defined as p1 (x, y). An example is shown in the main pattern of FIG.
(a) represents pattern 1, that is, p1 (x, y). The spectrum is elliptical. (b) represents p0 (x, y).

Here, in Step 1, the dot pattern obtained is different by making the initial random number different. That is, it is possible to change the dot pattern by changing the SEED (seed) value for random number generation with a pseudo random number generator. However, the spectral shape does not change. Such a dot profile is used as a secret key described later.

The sub information pattern is generated as follows.
The sub information pattern m (x, y) is set so that its spectrum distribution is distributed in a high frequency region that does not overlap with the spectrum of the main information pattern. For this reason, select the pattern whose spectrum is in the band above fmax. As an example, the sub-pattern shown in FIG. 3 uses a checkered pattern every other pixel. Figure 4 shows the spectral characteristics when both main and sub information are embedded. The spectrum is a bright spot spectrum at the Nyquist frequency and does not overlap with the spectrum of green noise.

Next, watermark data embedding will be described.
To embed watermark data, a pattern composed of main and sub information is combined into image data in block units.
Now, the image data consisting of L pixels xM pixels is assumed to be i (x, y) (0≤i (x, y) <1) and divided into blocks of R pixels x R pixels. R is a power of 2, and the larger the value, the higher the accuracy of the spectrum distribution, but the less information is embedded. Therefore, 32 and 64 are optimal. Let wm∈ {0,1} be the character string converted from the embedded character string.
When embedded bit wm is 0 → pi (x, y) = p0 (x, y)
When embedded bit wm is 1 → pi (x, y) = p1 (x, y) (4)
And
The number of bits that can be embedded is int (L / R) · int (M / R). Where int represents integerization. Since the image is divided into blocks, the image size may be any size.

The sub information is embedded as follows.
The sub information is a binary image of a logo or pattern. The embedding area is M (x, y), the embedding pattern is m (x, y), and the combined pattern pi '(x, y) of the main and sub information is
When inside the embedded area (M (x, y) = 0) pi '(x, y) = m (x, y)
When outside the embedded area (M (x, y) = 1) pi '(x, y) = pi (x, y) (5)
Created as This means that the sub information pattern is inserted into the main information pattern.
Such pi ′ (x, y) is embedded for each block as follows.
w (x, y) = i (x, y) + gain ・ pi '(x, y) (6)
Here, gain represents the embedding strength, and gain << 1 is necessary to obtain an invisible digital watermark. FIG. 5 shows each pattern and its spectrum. (a) is the main information pattern (main pattern) and its spectrum distribution, (b) is the sub information pattern (sub pattern) and its spectrum distribution, (c) is the combined pattern of both main information and sub information and its spectrum. Show the distribution. Since both spectral distributions have no or little overlap, they can be separated and extracted at the time of extraction.

Such watermark embedding is performed at high speed. That is, since p0 (x, y) and p1 (x, y) have already been determined, pi ′ (x, y) is obtained from equations (4) and (4),
The operation of w (x, y) = i (x, y) + gain · pi ′ (x, y) need only be performed in units of blocks. The calculations of equations (4) and (5) can be further speeded up by obtaining possible patterns in advance without performing each calculation for each block. For example, if the number of sub-information patterns is n, if the patterns of p0 (x, y) and p1 (x, y) are obtained in advance for each pattern, select from a total of 2n patterns. Only the calculation of equation (6) needs to be performed.

In general, in the digital watermark method, it is common to embed by performing various transformations such as orthogonal transformation, DCT, and Wavelet transformation for embedding. For this reason, in the embedding process, real space (pixel space) data is temporarily converted into a frequency space, embedded in the frequency space, and then returned to the real space again. It takes time for this conversion process. On the other hand, in this method, embedding is only required in real space, and simple calculation is sufficient, so it is excellent in high speed. This can be compared to dithering when binarizing an image, and can be done at high speed with simple processing. Therefore, it is possible to embed in real time in each frame of a moving image such as a drive recorder or a surveillance camera.

Removal of watermark data is performed as follows.
Converting equation (6),
i (x, y) = w (x, y) -gain ・ pi '(x, y) (7)
Thus, if pi ′ (x, y) and gain are known, it is possible to remove the watermark information. At the time of removal, information on p0 (x, y), p1 (x, y), m (x, y), M (x, y), gain, and embedded bit string is necessary. However, p1 (x, y) is not necessary when it is created by rotating from p0 (x, y). Also, m (x, y) is unnecessary if it is limited to the checkered pattern. Also, since the embedded bit string is obtained at the time of extraction, this is not necessary, and eventually p0 (x, y), M (x, y) and gain are sufficient. Such information is used as a secret key. Also, since M (x, y) and gain are not very secret, they do not need to be included in the key and may be obtained by other methods.
As described above, this secret key can change the dot pattern by changing the initial value (SEED) of the pseudo random number generator when generating the p0 (x, y) pattern. However, there is no need to change the extraction mask. For this reason, by using a pattern generated from different SEED values for each image or for each user, it is possible to suppress damage from removal of watermarks of other images even when lost or stolen.
In order to completely restore the original image when removing the watermark, it is necessary to adjust the dynamic range at the time of embedding in order to avoid the overflow of the embedded data and the flow underflow.

Such dynamic range conversion performs the following linear conversion on image data.
i '(x, y) = (1-gain) i (x, y) + gain / 2
By applying this conversion to the entire image, no overflow or underflow occurs. After the watermark is removed, the original image is obtained by performing inverse transformation.
Alternatively, the maximum value and the minimum value of the pixel are obtained from the image histogram, and conversion is not necessary if overflow and underflow do not occur. If the embedding gain is extremely small and it is not necessary to completely restore the embedding gain, this process may be skipped.

After removing the watermark data, if necessary, a new digital watermark can be added to the edited image using the key. If the key is lost, a new key can be issued and new watermark information can be entered. Since the number of keys can be issued almost infinitely as (4096) C (2048), duplication is probabilistically low. Here, (•) C (•) represents a combination.
The sub information may use the pattern M (x, y) in the key or create a new one. In any case, after completely returning to the original picture, watermark embedding can be freely made, and the right as a secondary copyright holder can be claimed.

Next, watermark information is extracted from the watermarked image as follows.
The main information is extracted by dividing the image into blocks and identifying the embedded bits in the spectrum space.
First, the spectrum is obtained by FFT (Fast Fourier Transform) for each block, and then p0 (x, y) and p1 (x, y) are discriminated and estimated from the shape of the spectrum image W (fx, fy) by a discriminator. Do it. As a discriminating method, it is obtained from the difference information of the luminance value by the discriminator using D0 at the time of dot pattern generation shown in FIG. 6 and D1 rotated by 90 ° as a mask. That is, the following Q0 and Q1 are obtained for each block.
Q0 = (1 / Z0) ΣD0 (fx, fy) ・ W (fx, fy)
Q1 = (1 / Z1) ΣD1 (fx, fy) ・ W (fx, fy) (8)
However, Zi = Σi (fx, fy).
From the above equation, the embedded bit (wm) is
When Q0-Q1> Th, wm = 0
When Q1-Q0> Th wm = 1 (9)
Otherwise wm gets indeterminate. | Q0-Q1 | represents the reliability, and Th represents the identification threshold. This Th makes it possible to control the reliability of the obtained watermark information.

Subsequently, the sub information is extracted as follows.
The sub information is extracted in the real space. First, the embedded image is convolved with an L × L checkered pattern filter. The larger L is, the higher the accuracy of extraction, but the edge of the extraction region becomes dull. L = 3 or 5th is optimal. As such mask patterns, 3 × 3 checkered pattern classifiers are shown in M0 and M1 in FIG. Using this mask pattern,
U0 = w (x, y) * M0
U1 = w (x, y) * M1 (10)
Ask for. Here, * represents convolution. The output Mout at each pixel is α | U0-U1 | using the coefficient α, and the binarization threshold Th1
When α | U0-U1 |> Th1, Mout (x, y) = 1
Otherwise Mout (x, y) = 0 (11)
Is output, showing a strong response in the checkered pattern area. NCC is an evaluation quantity that represents the image quality of the extracted pattern.
NCC = (ΣΣM (x, y) ・ Mout (x, y)) / (ΣΣM (x, y) ^ 2) (12)
Here, M (x, y) represents a pattern without deterioration, and ΣΣ represents all pixels in the block.

The procedure of embedding main information and sub information is performed as follows.
Color image data is converted into Y, Cb, and Cr, and watermark information is embedded in luminance Y. In the case of using only electronic data, a B (blue) signal may be used. The block size R is 64 × 64. The pattern p0 (x, y) is
(fmax, fmin) = (f0-1, f0-5)
However, the ellipticity was set to 1.5 with f0 = √ (1/2) · R / 2. p1 (x, y) is rotated 90 °.

FIG. 7 shows an example of a pattern indicating the sub-information embedding area M (x, y). The sub-information pattern is a pattern representing a binary logo or graphic, and the inside of the closed region is embedded with a checkered pattern of m (x, y). Prepare the required number of patterns of block size R × R. Create a composite pattern of the sub-information and the patterns p0 and p1 corresponding to the embedding bits 0 and 1 of the main information and embed them in the image. Because the main information is a green noise pattern, it is a cluster dot with good dispersibility, and it is embedded with gain << 1, so it is difficult to recognize visually and maintains high image quality. Furthermore, since the characteristics of the green noise pattern are sufficiently reproduced in the image output of a printer or the like, it is possible to extract the watermark from the printed matter. On the other hand, the sub information is not visually recognized because it is in the high frequency range of the Nyquist frequency. For this reason, image quality degradation does not occur. In addition, it is fragile because of high frequency.

Next, watermark information extraction will be described. Watermark information is extracted separately for main information and sub information.
FIG. 8 shows an image processing flow for extracting the watermark information of the main information. The embedded image data 21 is corrected when there is enlargement / reduction, rotation, or distortion of the image to obtain a corrected image. Thereafter, FFT (Fast Fourier Transform) is performed on a block basis, and (23) the frequency space (spectrum space) is converted. Since the FFT output is performed in units of blocks and the luminance changes, tone correction is performed by the histogram equalization 24 in order to improve identification accuracy. Subsequently, pattern identification 25 is performed, bit information is converted into a character code 26, and finally embedded character information is extracted (27).

FIG. 9 shows an image processing flow for sub-information extraction. The image data 21 may be subjected to histogram equalization for all images (32). Subsequently, the extraction pattern is displayed by the convolution 33 using the masks M0 and M1 (34). Equalization over the entire surface is effective when the background is white like document data.

FIG. 10 shows an embedded image of 1024 pixels × 535 pixels. Because it is possible to embed up to 16 bytes of alphanumeric characters with a block size of 64 x 64, 16 characters "Copy Prohibited!" Are embedded as main information. In addition, the sub-information is obtained by embedding two graphic patterns and eight patterns composed of patterned characters such as “COPY prohibited”. These eight patterns are embedded repeatedly. Embedding was performed with gain = 0.1875 to ensure durability during printing. Despite the relatively large gain, embedded images are less likely to be perceived by the naked eye.

FIG. 11A shows a result of watermark extraction of main information. The characters “Copy Prohibited!” Were extracted with high reliability. FIG. 4B shows that 8 patterns embedded with sub-information extraction are clearly extracted with high NCC values.

FIG. 12 (a) shows a falsified image in which the number of money is rewritten. Fig. 2 (b) shows the sub information extracted, and it can be seen that the background of the falsification position disappears and the falsification location can be identified.

FIG. 13 shows another embodiment, and shows an example in which a synchronization header signal is embedded in the sub information.
In general, a block-type electronic watermark requires synchronization of the size and position of a block at the time of embedding and a block at the time of extraction. Therefore, since the scaled image has a different block size and position, it is necessary to estimate the original image size by some method. Embedding the synchronization header signal in the sub information makes it possible to estimate the size of the original image.
Since the sub-information is subjected to watermark extraction processing in real space, it is not affected by block position fluctuations and is more resistant to scaling than the main information.
FIG. 6A shows an image with embedded watermark (1024 pixels × 640 pixels), and FIG. 5B shows extracted sub-information. (c) enlarges (a) 1.5 times to 1536 pixels × 960 pixels. Since the original image size is not known in the watermark extraction, it is not correctly extracted even if the watermark extraction of the main information is performed as it is. (d) is an extract of sub information, which is extracted by embedding a cross-shaped pattern as sub information in the first bit (MSB) of an embedded character bit of the main information. Since it is 8 blocks apart in the horizontal direction, the original image size can be estimated by measuring this.

That is, the distance of the cross pattern is measured, and the distance is set as d. The synchronization pattern is embedded every 8 blocks, and assuming that the block size is 64, the number of pixels for 8 blocks is separated by 64 × 8 = 512 pixels, making it easy to scale the image Can be calculated. Since the horizontal size of the enlarged image in (c) is 16 blocks wide, it can be seen that the horizontal size of this image is 64 × 16 = 1024 pixels. Since the original size is known, the image data is reduced to the original size. The reduction method is, for example, a method such as Bi-Linear or Bi-Cubic, and it is desirable to prevent the dot pattern from disappearing.
Since the dot pattern of the main information is clustered, it is resistant to various editing processes. The main information is not accurately extracted in the enlarged image because the block size and the phase are different, and if only this is corrected, the main information can be extracted with high reliability. Accordingly, if watermark extraction of main information is performed using an image whose image has been corrected to the original size, the main information is extracted with a high degree of reliability with the synchronization of the blocks.

Next, resistance to attacks will be described.
The main information is highly resistant to various attacks. This is because most attacks are attacks on the brightness of the image, and the green noise pattern is less affected by changes in brightness because of the spectral characteristics in the state of a set of dispersed dots.
FIG. 14 shows an examination of resistance to JPEG compression, and main information is correctly extracted up to a quality factor Q = 20 of JPEG. On the other hand, the sub information can be estimated to some extent by the human recognition ability, but is difficult to estimate below Q = 50. This is because it is easy to disappear because of its high spatial frequency. Therefore, it can be said that the sub information is a fragile embedding.

For the enlarged / reduced image, the sub-information can be extracted for the enlarged / reduced image of 0.8 to 1.5 times as shown in FIG.
Since the main information is subjected to FFT processing in units of blocks, if the block size and position are different, a detection error occurs and a correct result cannot be obtained. On the other hand, the sub information is output by performing the convolution process on the entire image using the detection mask, so that no error due to the block occurs. Therefore, as described above, it is possible to detect the synchronization signal pattern of the scaled image by detecting the synchronization signal with the sub information and know the original image size. By obtaining an image corrected to this image size, the embedded information of the main information can be obtained accurately, and the synchronization problem is solved.

It is convenient to use this synchronization pattern in combination with the alteration detection pattern. For example, by setting the number of sub-information embedding patterns to eight, and embedding a synchronization pattern at the first and an alteration detection pattern at the second to eighth positions, both functions can be satisfied. A pattern whose position can be specified as a synchronization pattern, such as a cross-shaped pattern, is preferred, while a mesh-like pattern or the like is suitable as a tamper detection pattern.

Further, the sub information can be used as a background pattern for printing.
The embedded image is output by a printer or the like. At this time, it is assumed that the printer can print out the original faithfully. Note that depending on the printer driver, checkered patterns may not be reproduced because halftoning by dithering or error diffusion may be performed without permission. If you try to make a duplicate of a copy of the print image, be sure to read it with a scanner. At this time, a fine pattern such as a checkered pattern disappears from the response characteristic of the reading, and a copy-forgery-inhibited pattern is formed on the reproduction in combination with the nonlinear characteristic of the printer. Such a copy-forgery-inhibited pattern has an effect of warning a user and preventing unauthorized use.

FIGS. 15A and 15B are images related to the visualization of sub-information by using a background pattern. FIG. 15A shows image data in which main and sub-information are embedded, and FIG. 15B shows an image in which sub-information is revealed as a background pattern. The tint block in the printed material appears when the printed material is copied and reprinted. If an equal density pattern is reproduced with halftone dots and dither output with coarse and fine dots, the fine dots cannot be read with a scanner and are difficult to reproduce with a printer. This is mainly due to a decrease in MTF (Modulation Transfer Function) in a high frequency range of an image input / output device such as a printer, and a phenomenon in which an image is not reproduced in a high frequency dot portion occurs.

Such a phenomenon is due to the non-linear effect of the image input / output device, and depends on the type and characteristics of the scanner or printer, and in some cases, the tint block is difficult to appear or is unstable. Therefore, some edge emphasis is given to the image data at the time of printer output so that a background pattern is always generated at the time of print output. That is, a Laplacian filter consisting of the following kernel values is added to the image.
｜ −0.25 0 −0.25 ｜
| 0 1 0 |
｜ −0.25 0 −0.25 ｜
Such a filter is not emphasized by the checkered pattern of the sub information, but the green noise pattern is emphasized. Therefore, a density difference is generated between the checkered pattern and the green noise pattern, and the information is visualized with the sub-information portion being white (or the density is lowered).

By doing as described above, the sub information that has not been visualized in the display becomes visible at the time of printing. For this reason, sub-information such as “copy prohibited” appears, which is a warning against illegal copying.

As described above, the present invention provides an invisible digital watermarking apparatus and method that enable embedding of multiplexed main and sub information in an image.
As usage, normal embedded information is embedded in the main information, and logos and patterns are embedded as sub-information, making it possible to apply to copyright protection, tampering detection, and illegal leak tracking. In addition, the sub information can be used as a main information synchronization signal or as a print ground pattern. Since it has two different characteristics of watermark information, it can be used for various purposes in combination.

The watermark embedding algorithm may be widely disclosed. In order to remove the watermark, it is necessary to obtain a green noise pattern, which cannot be easily generated. This exists in the secret key, and even if it is lost or stolen, another image cannot be opened if a different key is used for each image. It's like a fingerprint. Security is centralized in the key, and the algorithm may be disclosed for dissemination.

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 a monitor, 9 is a keyboard, 10 is a communication function, 11 is CPU, 12 is Internet, 13 distribution of image data, 14 contact for purchase request, 15 send private key,
16 is a copyright owner's computer, 17 is a purchaser's computer, 21 is image data, 22 is image correction processing, 23 is FFT (Fast Fourier Transform), 24 is histogram equalization processing, 26 is character conversion processing, and 27 is watermarking , 32 represents histogram equalization of the entire image, 33 represents mask convolution operation processing, and 34 represents extraction pattern display processing.

Claims (6)

In the digital watermark method in which confidential information or copyright information is embedded in digital image data i (x, y), the image data is expanded into a plurality of blocks, and embedded bit information i (i = 0, 1) for each block Is a digital watermarking method that multiplexes and embeds a main information pattern corresponding to and a sub-information pattern of a binary figure,
The spectral distribution of the main information pattern and the sub information pattern is not overlapped with each other, and the embedding operation is performed in the real space (pixel space) between the pattern combining the main information and the sub information and the image data.
An electronic watermarking apparatus and method characterized in that watermark extraction is performed in frequency space for extraction of main information and in real space for extraction of sub information. The embedding pattern of the main information corresponds to the bit information i (i = 0, 1), and the dot pattern pi (x, y) (i = 0,1)
The sub-information embedding pattern is a pattern m (x, y) whose spectrum is distributed in the high frequency range.
Create a pattern pi '(x, y) in which sub information is embedded in the main information pattern,
The embedded image w (x, y)
w (x, y) = i (x, y) + gain ・ pi '(x, y)
(Where gain is the composition ratio, less than 1),
The watermark information is extracted by dividing the embedded image data into blocks,
Extraction of main information is performed by discriminating and estimating using the mask pattern from the shape of the spectrum distribution.
Sub-information is extracted by convolution processing of a mask pattern whose spectrum is distributed in the high frequency range.
The digital watermarking apparatus and method according to claim 1. The image i (x, y) before embedding is restored by using a secret key including at least the dot pattern p0 (x, y) at the time of embedding with respect to the image w (x, y) in which the watermark is embedded. The digital watermarking apparatus and method according to claim 2, wherein the digital watermarking apparatus is capable of embedding new information as a digital watermark using the key. The embedded dot pattern of the main information is two types of dot profiles p0 (x, y) and p1 (x, y) having different spectral distributions,
3. The digital watermarking apparatus and method according to claim 2, wherein the sub-information embedding pattern is a checkered binary pattern. 5. The digital watermark according to claim 4, wherein the sub information includes a tamper detection pattern, and in the sub information watermark extraction, the presence or absence of tampering or the location of tampering can be specified depending on whether or not the pattern is detected. Apparatus and method. 5. The sub-information is characterized in that a size of an original image can be obtained by embedding a synchronization pattern with respect to a block at a constant interval and measuring the interval at the time of extraction. The digital watermarking apparatus and method according to claim 1.