JP2023068677A - Digital watermark embedding and extracting method - Google Patents

Abstract

To obtain a digital watermark method which has print and scan resistance and correctly extracts watermark information even when it is printed in a curved surface.SOLUTION: Image data is divided into a plurality of blocks, and bit information is multiplexed by using two different kinds of digital watermark embedding techniques for each block, and the bit information is embedded. Extraction of a watermark from the embedded image obtains a high percentage of a correct answer by calculating the embedded bit information and its reliability from each digital watermark method and selecting the highly reliable watermark information.SELECTED DRAWING: Figure 1

Description

The present invention relates to an electronic watermark for embedding information in a document image, and more particularly to an electronic watermark embedding and extraction method capable of extracting accurate watermark information by multiplexing and embedding watermark information using different embedding methods.

Invisible digital watermarking technology, which embeds information in an image, has various applications for using the embedded information. Among them, digital watermarking technology for copyright protection is considered to be the most useful application. In such applications, when copyright information is embedded as invisible information in image data and illegally used, the copyright is protected by prosecution for copyright infringement. In this operation, it is necessary to constantly track and monitor image data on the Internet and image data in print in order to sue illegal use, so it is necessary to establish a monitoring system. In addition, the embedded watermark information tends to dissipate when the image is partially cropped, enlarged or reduced, or subjected to various editing and processing processes, as well as image compression using JPEG. This makes it difficult to monitor and track.

On the other hand, progress is being made in the application of digital watermarking, in which images embedded with information are photographed with a smartphone camera or the like, and extracted information can be used to launch a specific application or direct users to websites on the Internet.

For example, by using a smartphone to read an image in which a purchaser has embedded information in a product package at a store, etc., it is possible to obtain a variety of information about the product, such as the price, expiration date, and place of production of the product. Stores can also use it for sales and inventory management, just like barcodes attached to products. In addition, applications such as customers at fast-food restaurants using smartphones to scan product photos and place orders using the restaurant's app are conceivable.

Such applications are called annotation applications because they embed meta information and the like related to image data. This annotation-based digital watermark has slightly different requirements from the digital watermark for copyright protection in operation.

One is that in the case of copyright protection, the quality of the embedded image must be high because the image itself is a product, but in the case of annotations, the embedded information is the main factor, and the image is simply embedded. Since it is a medium, it does not necessarily have to be of high quality. Instead, the embedded information must be read 100% correctly, requiring high reliability. In particular, there are trapezoidal distortions when reading with a camera, uneven illumination during photography, and stains and scratches on paper.

Second, in the case of copyright protection, it is necessary for the purchaser to be able to remove the watermark as necessary using a key, restore the original image, or add the second copyright holder. must be. On the other hand, in the case of annotation application, there is no such need, and digital watermarking technology does not necessarily need to be reversible.

In this way, digital watermarking technology for annotation applications has different technical requirements compared to copyright protection. Various measures are required to improve reliability.

In the case of this annotation application, the readout is the printed material and the image displayed on the monitor. Therefore, digital watermarking technology must be print- and scan-resistant. Moreover, in the case of printing on a package, the printed surface may be cylindrical or spherical. Furthermore, when printed on plastic wrapping paper, the embedded watermark information is distorted due to the flexibility of the shape and the curved surface of the image. It is desirable that the watermark information can be read accurately even in such cases.

As an electronic watermarking method that satisfies such printing/scanning resistance, we have proposed a green noise diffusion type electronic watermarking method. For example, in Patent Documents 1 and 2, an image is divided into blocks, and a green noise pattern corresponding to the bit string (0, 1) of the embedded information is embedded in each block. There is one that obtains a bit string. In this method, the dominant frequency of the spatial frequency of the green noise pattern is set so that the responsiveness of the printer is high and the responsiveness of the human eye is low. It can be strengthened, and print/scan resistance is obtained.

However, when there is distortion such as printing on plastic wrapping paper, there is a problem that the embedded information cannot be obtained correctly. This is because the two-dimensional image captured by the camera is geometrically distorted, desynchronized, and the apparent spatial frequency is also changed.

Japanese Patent Application Laid-Open No. 2018-182471 Japanese Patent Application No. 2017-117460

It is an object to obtain a digital watermarking method that solves this problem, has print/scan resistance, and can extract watermark information correctly even when printed on a curved surface.

In order to solve the above-mentioned problems, the present invention provides an electronic watermarking method for embedding watermark information expanded into bit strings into digital image data. The embedded bit information is extracted by dividing the embedded image into blocks. Extraction by the second digital watermarking method calculates the embedded bit information and its reliability from the distribution of the luminance pattern in the real space (pixel space), and determines the watermark information from the obtained reliability of both. It is characterized by

According to this method, when embedding by two different embedding methods, the extraction of watermark information by the first digital watermark method is performed by extracting the watermark information by the second digital watermark method from the distribution of the spectrum pattern converted into the frequency space. is extracted from the distribution of the luminance pattern in the real space (pixel space) by arithmetic processing. Since the second digital watermark embeds a low-frequency pattern, the first digital watermark method has almost no effect on the spectral distribution. In addition, the first digital watermark embedding does not affect the second digital watermark embedding because the spatially averaged embedding data value is zero. That is, each method can be prevented from affecting each other. Also, by selecting the one with the higher reliability from the two, it is possible to increase the correct answer rate of the embedded information.

In addition, among the two different electronic watermarking methods, the first electronic watermarking method uses a plurality of dot patterns that exhibit green noise characteristics whose spectrum is reduced in the low and high frequency regions, and one dot pattern corresponding to the embedded bit information. One is selected and embedded in each block, and the extraction obtains the spectral distribution pattern of the image for each block, and calculates the bit information and reliability from the pattern. Pixel data is embedded by adding or subtracting a constant value according to the bit information in a partial area consisting of pixels, and extraction is performed by extracting bit information and reliability from the average data of pixels in the embedded partial area and the average data outside the area. is calculated, and the reliability is compared, and the bit information with the higher reliability is selected.

According to this method, in the first digital watermarking method, the embedded dot pattern can maintain high image quality and high durability, the embedding strength can be increased, and an electronic watermark with printing and scanning resistance is possible. The electronic watermarking method 2 enables electronic watermarking that is resistant to displacement and distortion of the detected pattern. Therefore, it is possible to extract from a curved printed image by combining these and complementing each other. In addition, since the reliability of both is extracted by different methods and arithmetic processing, it cannot be compared unconditionally. A digital watermarking method can be realized.

Next, in the electronic watermarking method, before embedding watermark information, a smoothing operation is performed on the embedded image. The digital watermark embedding of is characterized by embedding in the blue data of the smoothed image.

According to this method, it is possible to increase the rate of correct answers for extracting watermark information regardless of the image. In the first digital watermarking method, the spectrum is obtained by Fourier transform of the dot pattern, but since it overlaps with the spectrum pattern of the image, judgment errors are likely to occur in the case of images with large high-frequency components. Therefore, by performing a smoothing operation on the entire image in advance, the high-frequency components of the image are shifted to the low-frequency side, and this problem can be avoided regardless of the image. Further, in the first electronic watermarking method, even if embedded in the luminance data of the image, the green noise pattern is difficult to be visually recognized, so that the image quality does not deteriorate. On the other hand, in the second digital watermarking method, since the embedded pattern is visible when embedded in luminance information, it is difficult to be seen when embedded in a blue (B) signal because it is embedded in a relatively large area. can be maintained.

Furthermore, in the two different types of digital watermarking methods, embedding is performed in different blocks, and when embedding watermark bit information in each block, watermark bits and reliability are calculated from the embedding information immediately after embedding. If it is lower than a predetermined threshold, the embedding strength is repeatedly increased until the threshold is reached.

In the first digital watermarking method, the reliability at the time of extraction is lowered where the high-frequency component of the image is large, and judgment errors are likely to occur. In addition, in the second digital watermarking method, determination errors are likely to occur in locations where there is a large change in luminance within a block, edge portions of an image, and the like. Usually, both occur at the same place in the image, so if the blocks embedded with the respective digital watermarks are different, the correct answer rate and reliability can be improved. In addition, the reliability is calculated immediately after embedding in the block, and if the reliability is low, the embedding strength is increased so that the reliability is set to a predetermined level, making it possible to further increase the correct answer rate in watermark extraction. .

As described above, by using these two different digital watermarking methods and taking various measures, it is possible to maintain high image quality, have printing and scanning resistance, improve the correct answer rate of extracted data, and obtain highly reliable watermark information. Obtainable. It is also possible to extract watermark information from an image that has been distorted into a curved surface.

FIG. 2 is a diagram of the operation form of digital watermarking according to the present invention; Diagram of the digital watermark embedding system of the present invention A diagram showing the embedding pattern of the first digital watermarking method and its spectral characteristics. A diagram showing a sub-region for embedding in the second digital watermarking method A diagram showing the flow of watermark embedding in the digital watermarking method of the present invention. A diagram showing the flow of watermark extraction in the digital watermarking method of the present invention. A diagram showing a partial region at the time of extraction in the second digital watermarking method A diagram showing an image embedded with watermark information according to the digital watermarking method of the present invention.

FIG. 2 is a configuration diagram of an electronic watermark system for creating an electronic watermark embedded image according to the present invention. In a personal computer (PC) 1 of a creator, image data to be embedded is stored in a data memory 2 such as a hard disk. The image data is image-processed by the image processing program in the program memory using the CPU, ROM, RAM, etc., and the embedded image is displayed on the monitor 3 . A printer 4 is connected to the PC 1, and the embedded image is output as a printed matter 5 from the printer. Also, the embedded image can be transmitted via the network and displayed on the external display 7 .

In order to embed information in the image, character information is entered from the keyboard. The number of characters that can be embedded varies depending on the image size and block size. When the block size is 32.times.32 with 512 pixels.times.512 pixels, a maximum of 32 bytes can be embedded in each block, and when the block size is 16.times.16, a maximum of 128 bytes is possible. As far as the ASCII code is concerned, these are 32 characters and 128 characters, respectively.

In order to improve the accuracy and reliability of information, error correction codes are introduced. This is to eliminate extraction errors due to flaws and stains on the paper surface, uneven lighting, reflection of ambient light and external light in the display image, and the like. Here, RS (16, 6) by Reed-Solomon code is used. It inserts 6 words of check code out of 16 words (thus reducing the information code to 10 words) and is able to correct 3 words of error. The encoded bit information is embedded within the image.

FIG. 1 shows an operational mode for annotation. An image 9 embedded with watermark information 8 including an error correction code is output/displayed on a print 5 or a display 7 . The user takes a picture of it with a camera 10 such as a smart phone, and extracts watermark information using watermark extraction software incorporated in the smart phone. As a result, a link to a home page 12 on the net or a specific application 13 is activated. This watermark extraction software can be freely downloaded from, for example, the distributor's home page.

Examples of the present invention will be described below.
The green noise pattern used in the first electronic watermarking method is generated in the following steps.
Step1: First, as an initial value, R^2/2 random dots are placed in an R×R area and p(x,y). The initial value indicates white noise characteristics.
Step2: Next, we perform the two-dimensional Fourier transform of the dot pattern to obtain P(fx,fy).
Step3: For P(fx,fy), the frequencies fx,fy are
fx,min≤fx≤fx,max
fy,min≤fy≤fy,max
is multiplied by a filter D(fx,fy) that limits the band to obtain a new spectral characteristic P'(fx,fy).
Step4: Perform inverse Fourier transform on P'(fx,fy) to obtain multivalued dot pattern p'(x,y).
Step5: Error function: e(x,y)=p'(x,y)-p(x,y)
are obtained, and white and black are reversed in order of large error.
Steps 2 to 5 are repeated, and when the error function becomes below a certain value, the process ends and the final pattern is obtained.

By adjusting the maximum frequency fmax and minimum frequency fmin, the dot pattern obtained by such a method is difficult to be visually recognized by the human eye, and has sufficient response characteristics for printers, scanners, cameras, etc., that is, sufficient resolution. It is possible to obtain the watermark information and extract the watermark information. Two types of anisotropic patterns are created because it is necessary to correspond to embedded bits "0" and "1" in order to create an embedding pattern for digital watermarking. FIG. 3 shows a dot pattern of 32×32 pixels and its spectrum. The two types of patterns pi(x,y) (i=0,1) have elliptical spectra, each of which is rotated by 90°.

Here, in Step 1, by using different random numbers for the initial values, different dot patterns are obtained. In other words, the dot pattern can be changed by changing the SEED value for random number generation in the pseudo-random number generator. However, the spectral shape does not change. Therefore, the extraction software can be used in common without modification.

This protects third parties from modifying or falsifying the data. Visible patterns such as two-dimensional patterns can be easily modified or forged, but green noise patterns are complicated and can be changed by changing the SEED value for each embedded image. Modification is difficult and security is improved.

Next, each of the two types of digital watermarking methods will be described. First, embedding of watermark data in the first electronic watermarking method will be described.
In embedding watermark information, color image data is converted into Y, Cb, and Cr, which are luminance and color difference signals, and watermark information is embedded in luminance Y. This is to facilitate extraction from print data during printing. When embedded in color information, image data consisting of red, green, and blue (RGB) color channels is converted into subtractive cyan, magenta, yellow, and black (CMYK) colorants in printing. This is because the characteristics overlap and crosstalk occurs with each other, resulting in poor color separation accuracy.

For embedding watermark information, digital image data i(x,y) is divided into R pixel × R pixel blocks, and a green noise pattern is embedded in each block corresponding to watermark bit information (0 or 1). shall be embedded. Let pi(x,y) (i=0,1,2) be the dot pattern showing the green noise characteristics, then the watermark embedded image w(x,y) is
w(x,y)=i(x,y) + g・pi(x,y)
becomes. where g is the embedding strength. pi(x,y) shall take values ±1/2 to keep the local mean density of the embedded image unchanged.

Next, extraction of watermark information will be described. Images read by a camera or scanner are corrected for size, tilt, trapezoidal distortion, etc., and then spectrally transformed by FFT (Fast Fourier Transform) in block units. The converted spectral pattern W(x,y) becomes an elliptical ring-shaped spectral pattern shown in FIG. 3 according to the patterns p0 and p1 used at the time of embedding. The embedded bits are determined by pattern recognition of the shape of this elliptical ring.

Using these two elliptical ring-shaped masks D0 and D1, the cumulative luminance values Q0 and Q1 in the mask are obtained as follows.
Q0=(1/Z0)ΣD0(x,y)・W(x,y)
Q1=(1/Z1)ΣD1(x,y)・W(x,y) (1)
However, Zi = ΣDi(x, y) (i = 0, 1).
From the above equation, the embedded bit is determined by the experimentally determined threshold value Th.
Bit = 0 when Q0-Q1>Th
When Q1-Q0>Th Bit = 1 (2)
Otherwise the bit gets undefined. Also, the reliability Rel1 is
Rel1 = |Q0-Q1| (3)
and

Next, the embedding of the second electronic watermarking method will be described. FIG. 4 shows the partial area S(16) in the block (14), which is embedded in the pixels included in the S area. A constant value .sigma. is added to or subtracted from the image data corresponding to watermark bit information 0 or 1. FIG. That is, embed by the following method.
When the embedded bit is 0, σ is added to the image data of area S. w(x,y)=i(x,y)+σ
When the embedded bit is 1, σ is subtracted from the image data of area S. w(x,y)=i(x,y)-σ (4)
If the area other than S in the block is T, then nothing is done for area T(15).

The image data used for embedding is blue (B) channel image data. This is because when it is displayed on a display, it is difficult to distinguish from the human visual characteristics, and when it is printed, it is printed with a yellow color material, so it is similarly difficult to distinguish. For example, when the block size is 32 pixels × 32 pixels, the partial region S (16) is a large region such as 16 pixels × 16 pixels to 20 pixels × 20 pixels, so its spatial frequency is direct current (DC) It has a low frequency close to that of the component, is significantly different from the spectrum distribution in the first electronic watermarking method, and hardly interferes with each other, so it does not affect the extraction accuracy of the first electronic watermarking method.

For watermark extraction, the embedded image is divided into blocks of R pixels × R pixels, and the added value A of the image data in the partial area S in the block and the added value B of the image data in the area T outside S are calculated by the following formula. Ask for
A=(1/Ns)Σsw(x,y)
B=(1/Nt)Σtw(x,y) (5)
where Ns and Nt are the number of pixels in the partial areas S and T respectively, and Σs and Σt are the additions in the partial areas S and T.
Bit judgment is
Bit = 0 when AB>0
When AB<0 Bit = 1 (6)
becomes. Reliability Rel2 is
Rel2=|AB| (7)
and

Each of the digital watermarking methods has been described above. Subsequently, a method combining the two types of digital watermarking methods of the present invention will be described.
FIG. 5 shows the flow of watermark embedding in the digital watermarking method of the present invention. First, as preprocessing (20), the embedded character string is developed into bit information, and a correction code is added using Reed-Solomon code. Since the amount of embedded information is reduced by using the error correction code, it is necessary to consider the bit information to be embedded in advance.
As a pre-processing, the embedded image itself is slightly smoothed. This is to reduce the spectral interference between the green noise pattern and the image in the first digital watermarking method. Applying a small smoothing filter, say 2 pixels wide, reduces the Nyquist frequency of the image by a factor of 2 and almost eliminates the overlap with the spectrum of the green noise pattern. By doing so, image dependence is eliminated, and any image can be stably embedded and extracted. It has the same effect as a low-pass filter used in optical systems such as digital cameras.

After these preprocessing, the image data is divided into blocks (21). For example, when an image of 512×512 pixels is divided into blocks of 32×32 pixels, the total number of blocks is 256. Block numbers are assigned to each block in raster order, and embedding is performed sequentially for each block.

First, the first embedding (23) is performed for one block with an embedding strength g (g<<1). Next, fast Fourier transform (FFT) is performed on the image data of one block just embedded (24), and bit information and reliability are obtained (25). If the image is uniform and flat, the embedded bit values and high reliability are returned, but if the image has strong high-frequency components, incorrect values and low reliability may be returned. In this case, the embedding strength is increased by Δg (27) so that the embedding value is equal to or greater than the predetermined constant confidence value T1, and the embedding is attempted again. This operation is repeated, and when it reaches T1 or more, it ends (26).

Next, embedding by the second digital watermarking method is performed in the (n+k)th block shifted by the kth. This is to improve the correct answer rate of the extracted data by changing the embedding position, which will be described later.
Embedding in the second digital watermarking method is performed according to equation (4) with the embedding strength σ for the partial area S in the block.
When the embedded bit is 0: σ is added to the image data of the area S. When the embedded bit is 1: σ is subtracted from the image data of the area S (29). Next, the embedded bit and reliability are extracted from the image data immediately after the embedding (30). If the image is uniform and flat, the embedded bit value and high reliability are returned, but if there are edges in the image, etc., an incorrect value and low reliability are returned. The reliability is compared with a predetermined threshold value T2 (31), and if it is equal to or less than T2, the embedding strength is increased by Δσ (32), and the embedding is tried again. This operation is repeated until the reliability reaches T2 or higher.

When one block ends, the next block (n+1th block) is moved to (34) and the same embedding operation is performed. In the second digital watermark, it is the n+k mod N-th block (where N is the total number of blocks, mod is the remainder modulo N, and k is an arbitrary integer that is predetermined). Either of the two types of digital watermarking can be embedded first. This is because the embedding into pixels is all linear, so even if the order of embedding changes, the result is the same.
A judgment is made as to whether or not all blocks have been completed (33), and the embedding is completed when completed.

FIG. 6 shows the flow of watermark extraction. First, since the embedded image is photographed by a camera or the like, trapezoidal distortion occurs, so image correction (39) is performed. This correction detects the four corners of the image, corrects the trapezoidal distortion by projective transformation, and makes it a predetermined image size.
Subsequently, the image data is divided into blocks (40), and extraction processing is performed from the first block (41).

First, extraction is performed by the first digital watermarking method. The image data in the block is transformed (42) into a spectral pattern by FFT. After that, spectral pattern identification (44) is performed using equations (1) and (2), and embedded bit b1 and reliability Rel1 are calculated (44).

Next, in a block shifted by k (45), extraction is performed by the second digital watermarking method. Divide into partial areas S and T (46), find the average pixel data of each area by equation (5), and calculate embedded bit b2 and reliability Rel2 by equations (6) and (7) (47 ). At this time, as shown in Fig. 7, the partial areas S and T are set to areas S (18) and T (17) smaller than the boundary line (dashed line) at the time of embedding in order to avoid synchronization deviation. Extraction is possible even when there is distortion. As the partial area S is made smaller, the resistance to synchronization errors improves, but the accuracy of the statistics of the average pixel data decreases and the variance increases, so it is determined experimentally.

Next, we compare the reliability of both. Since the derivation formulas for the reliability of the two are different, they cannot be compared unconditionally, so they must be standardized. In the first digital watermarking method, |Q0-Q1| obtained by extracting from the image embedded in the uniform image is used as the normalization coefficient, and in the second digital watermarking method, the embedding strength σ is used as the normalization coefficient, and non-linear A coefficient is also introduced to transform confidence. Nonlinear coefficients are determined experimentally. The following bit determination is performed using the converted reliability as Rel1 and Rel2.
First, if the extracted bits of both are b1=b2 (48), it is determined that bit=b1 (or b2) (49). Otherwise, it is checked whether the reliability is Rel1>Rel2 (50), and if so, bit=b1 (52), otherwise bit=b2 (51). Subsequently, it is determined whether or not all blocks are finished (53), and if not finished, the next block (n+1th block) is moved to (54). As with embedding, the second digital watermark is the n+1+k mod N-th block (where N is the total number of blocks and mod is the remainder). After all blocks are completed, the extracted bit string is subjected to error correction reverse encoding (55) to obtain an embedded character string (56).

FIG. 8 shows an embedded image obtained by the above processing. An image of 256 pixels×256 pixels is smoothed by 2 pixels, and 8 characters (64 bits) of ASCII code are embedded in a block size of 32 pixels×32 pixels. The embedding intensity is intentionally increased (g=0.25, σ=48) to make the embedding pattern easier to see. In the figure, (A) is an embedded image when only the first electronic watermarking method is used, and (B) is an embedded image when only the second electronic watermarking method is used. Although they are actually embedded in a color image, (A) displays the luminance component and (B) displays the blue channel component in black and white. The lower part shows an enlarged view of a part of the image (approximately 64×64 portion surrounded by a white frame in the upper left part). A green noise pattern is visually recognized from the enlarged view of (A), but it is difficult to be visually recognized by the human eye in observation at a distance of clear vision or longer unless the image is enlarged. In embedding by the second digital watermarking method in (B), the patch pattern in the embedded area is visible. hard to get rid of.

(C) shows the embedding using both digital watermarking methods. Although these patterns overlap each other, they can be extracted independently and do not adversely affect the extraction accuracy. In other words, the embedding pattern in the first digital watermarking method has a spatial frequency limited to the band from fmax to fmin, and the second digital watermarking method has the spatial frequency characteristics of the embedded subregion (patch). has a much lower frequency than the first digital watermarking method, and is close to a direct current (DC) component. Therefore, no interference or crosstalk occurs in the spatial frequencies of the two, and the extraction accuracy in the first electronic watermarking method does not deteriorate.
As for the extraction accuracy in the second digital watermarking method, the average pixel value of the partial area (patch) is not affected by the first watermarking method. This is because the embedding pattern pi(x,y) has a value of ±1/2 in the first digital watermarking method, so the average density (or average brightness) does not change with the patch size. Therefore, extraction accuracy in the second digital watermarking method does not deteriorate.

Watermark information is extracted from the image (C). With two types of digital watermarking methods, various measures are taken to improve the reliability of correct answers, so correct answers can be obtained with high accuracy. The reason is that two types of digital watermarking methods are used to complement each other's weak points. Moreover, since they act independently of each other, one method does not adversely affect the other. Therefore, it has printing and scanning resistance, and it is possible to extract watermark information correctly even when it is printed on a curved surface.

In the present invention, the same embedded information is embedded by two types of electronic watermarking methods by each watermarking method. If it is possible to embed different watermark information for each. In this case, the amount of information that can be embedded is doubled.

As described above, the present invention develops image data into a plurality of blocks, multiplexes and embeds bit information using two different digital watermark embedding techniques for each block, and thus has print/scan resistance and curved surface. It is possible to obtain a digital watermarking method that correctly extracts watermark information even when printed in a shape.
As a form of use, product introduction at a store or the like is conceivable. Images embedded in packages, etc., can be photographed with a smartphone camera, etc., and various information about the product, such as the price, expiration date, and place of production, can be obtained. It is also possible to read watermarks from non-flat printed matter such as plastic-wrapped products. It is also possible to launch a specific application from the extracted information, or to navigate to a home page on the Internet. It can also be used for sales and inventory management like POS in shops.
It can also be used like a two-dimensional barcode. Compared to two-dimensional barcodes, the embedded capacity is smaller, but it is more secure because it is difficult to tamper with or modify the embedded information.

1 is a personal computer (PC); 2 is a data memory; 3 is a monitor; 4 is a printer; 11 is camera photography, 11 is watermark extraction, 12 is a home page, 13 is application software, 14 is an embedded block, 15 and 16 are partial areas at the time of embedding, and 17 and 18 are partial areas at the time of extraction.

Claims (4)

In an electronic watermarking method for embedding watermark information expanded into bit strings in digital image data, image data is expanded into a plurality of blocks, and bit information is multiplexed and embedded using two different electronic watermark embedding methods for each block,
The embedded bit information is extracted by dividing the embedded image into blocks, and by the first digital watermarking method, the image data in the block is converted into frequency space. Calculate the embedded bit information and its reliability from the distribution of the luminance pattern in the real space (pixel space),
A method of embedding and extracting a digital watermark, characterized in that watermark information is determined from the obtained reliability of both. Of the two different electronic watermarking methods, the first electronic watermarking method selects one from a plurality of dot patterns whose spectrum exhibits reduced green noise characteristics in the low and high frequency regions according to the embedded bit information. Select and embed in each block, extract the spectral distribution pattern of the image for each block, and calculate bit information and reliability from that pattern.
The second digital watermarking method embeds pixel data by adding or subtracting a constant value according to bit information in a partial area consisting of a plurality of pixels in a block, and extracts the average data of the pixels in the embedded partial area. and the average data outside the area to calculate bit information and reliability,
2. The method of embedding and extracting a digital watermark according to claim 1, wherein the respective reliability levels are compared and bit information with higher reliability is selected. In the above-described electronic watermarking method, the embedded image is smoothed before watermark information is embedded, and the embedding in the first electronic watermarking method embeds luminance information in the smoothed image, and embeds the second electronic watermark in the luminance information of the smoothed image. 3. The method of embedding and extracting a digital watermark according to claim 2, wherein the embedding of is embedded in the blue data of the smoothed image. In the two different types of digital watermarking methods, embedding is performed in different blocks, and when watermark bit information is embedded in each block, the watermark bits and reliability are calculated from the embedded information immediately after embedding, and the reliability is set to a predetermined value. 3. A digital watermark embedding and extracting method according to claim 2, wherein if it is lower than the threshold, increasing the embedding strength is repeated until the threshold is reached.

Images