JP2003143388A - Image processing method and apparatus, and printed matter - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To improve the reliability of reading a digital watermark. SOLUTION: In step 202, original image data 100
Color space conversion. In step 204, the patch information is embedded in the original image data. Step 2
At 10, the Fourier transform is performed on the patch-embedded image data 101 to extract amplitude information 111 and phase information 112. On the other hand, in step 218, a hash value is added to the watermark information 120 and error correction encoding is performed. In step 220, spectrum spreading is performed to obtain spread watermark information 121. Further, in the processing of steps 242, 244, and 246, the patch-embedded image data 101
Create a weight mask from. The weight mask, the amplitude information 111, and the spread watermark information 121 are synthesized. At step 262, the color space is restored.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method and apparatus, and printed matter. In particular, the present invention relates to an image processing method for embedding and embedding digital watermark information in digitized image data, and an image processing method for detecting digital watermark information from an image in which digital watermark information is embedded.

[0002]

2. Description of the Related Art Conventionally, digital watermarks have been used for the purpose of finding illegal copies of digital information. This digital watermark embeds information into the digital data in a form that cannot be perceived on the surface, and detects the embedded information to detect illegal copies of the digital data. Is. Here, typical examples of the target digital data are still image data, moving image data, audio data, etc., and information unique to the right-related person of these digital data (for example, copyright information) is embedded as a digital watermark. , The protection of these rights-related persons is being carried out.

However, in the above-mentioned conventional technique, the embedded digital watermark information can be detected only from the digital data copied as digital data. For example, even if digital watermark information is embedded in still image data, if the still image data is printed using a printer, the embedded watermark information can be restored from the printed matter on which the still image data is printed. Was impossible. This is because some of the still image data may be lost or the contrast of the image may change when the printer is used for printing and the printed matter is digitized again by using the scanner. That is the reason.

Therefore, the group of the inventors of the present application aims to provide a digital watermarking method and apparatus that are robust to contrast conversion and printing (more generally, D / A (digital / analog) conversion). As a result, we developed a technology that can detect a digital watermark from a printed matter by printing the image on a digital image by synthesizing the digital watermark with a predetermined method. The invention relating to a digital watermark synthesizing method, a digital watermark detecting method, devices for these, and printed matter according to this technology is described in 2000 (AD 2000).
The patent application has been completed on August 18th.

Here, the outline of the above-mentioned prior art will be described.
FIG. 13 is a schematic diagram showing a digital watermark synthesizing method according to the conventional technique, and FIG. 14 is a schematic diagram showing a digital watermark detecting method according to the conventional technique.

<Digital Watermark Synthesis Procedure (Prior Art)> In FIG. 13, reference numeral 100 represents original image data. The amplitude information 111 and the phase information 112 are obtained by Fourier transforming (210) the original image data 100. Reference numeral 120 is watermark information embedded in the original image data 100. The watermark information 12 spread by spectrally spreading (220) this watermark information 120
1 is obtained. Next, the spread watermark information 121 is combined with the amplitude information 111 (225). Further, an inverse Fourier transform (230) is executed using the phase information 112, and image data in which watermark information is combined is obtained.

On the other hand, the original image data 100 is wavelet transformed (24) in order to reduce the noise by matching the noise due to the watermark composition with the local features of the image.
2) Then, the original image data 100 is decomposed into sub-band images and represented by partial images orthogonal to each other. Then, based on this partial image, energy map data having the same size as the original image is obtained (calculation of high-frequency energy data 24
2).

Next, the difference between the image obtained by combining the watermark information obtained by the inverse Fourier transform (230) and the original image data 100 is obtained (235). Further, the above energy map data weight mask is created (246).
Processing is performed, and the watermark information is added again to the original image data 100.
(250, 255). In this way, the watermark embedded image data 150 is obtained.

In summary, the digital watermark synthesizing procedure according to the prior art is characterized in that watermark-embedded image data is synthesized by embedding watermark information only in a predetermined intermediate frequency band in original image data. To do. Since the watermark information of the watermark-embedded image data 150 created by such a procedure is excellent in robustness, it can be detected even if an image is printed once and then digitized again by a scanner or the like.

<Digital Watermark Detection Procedure (Prior Art)> In FIG. 14, reference numeral 150s is watermark-embedded image data to be a digital watermark detection target. The watermark-embedded image data 150s is, for example, digital data obtained by scanning a printed matter with a scanner or the like. Watermark embedded image data 150
By performing the discrete Fourier transform (270) on s, the amplitude information 160 of the image data 150s can be obtained. on the other hand,
A symbol value detecting sequence 170 is obtained based on the pseudo random number sequence.

Next, the symbol response value is calculated based on the amplitude information 160 and the symbol value detecting sequence 170 (2
80), and symbol value detection (290) is performed. For the symbol value 290 thus obtained,
Radix conversion is performed to obtain the detected watermark information 120.

[0012]

As described above, the technique for enabling the detection of digital watermark information from the printed matter even if the image data combined with the digital watermark is once printed has been completed in principle. It was However, for practical use, it is desired to further improve the reliability of digital watermark detection by using various methods. This is because unlike digital data, which can be copied and stored as it is, the analog information in the form of a printed matter is subject to various changes during each process or over the years. It is necessary to do so.

Specifically, there are the following problems. First
In addition, when the digital data in which the watermark information is embedded is once printed and then digitized again by a scanner or the like, an error may occur in the reading direction or scale of the scanner or the like. In addition, the scale and the like may be completely different because the resolution (the number of dots per unit length) at the time of printing is different from the resolution at the time of scanning.
If there is such an error, the watermark information may not be detected correctly.

Secondly, if the strength of the watermark is increased in order to improve the reliability of the detection of the watermark information, the watermark information becomes noise and the image quality of image data which is digital data is deteriorated. . In order to prevent the deterioration of image quality, it is desirable to embed the watermark information in a data space that is inconspicuous to humans, but it is not always necessary that the original image data is represented as digital data that is convenient for inserting the watermark. Not necessarily.

Thirdly, if the peak value of the symbol value is not good even if the watermark information detection process is performed, the detected watermark information may be erroneous. Therefore, it is required to prevent such an error, or at least make it possible to recognize that the watermark information is erroneously detected.

Fourthly, in the process of printing (D / A conversion) and scanning (A / D (analog / digital) conversion), a low-pass filter works on the image data, whereby the watermark information can be detected correctly. There is a possibility that it will not exist.

The present invention has been made in consideration of the above circumstances, and an image processing method capable of solving the above problems and improving the reliability of reading watermark information, and an image processing method thereof. It is intended to provide a device and a printed matter.

[0018]

In order to solve the above-mentioned problems, the present invention provides a spectrum spreading section for generating spread watermark information by spectrum spreading watermark information, and a geometric spread on the original image data. A patch embedding unit for repeatedly embedding patch information for giving a geometrical reference in the original image data in a predetermined pattern, and Fourier transforming the original image data in which the patch information is embedded by the patch embedding unit. A Fourier transform unit for extracting amplitude information and phase information, a synthesizing unit for synthesizing the diffused watermark information and the amplitude information, an inverse Fourier transform based on the output of the synthesizing unit, and the phase information. Then, a difference image for generating difference image data which is a difference between the result of the inverse Fourier transform and the original image data in which the patch information is embedded. And an energy map data calculation unit that performs wavelet transformation on the original image data in which the patch information is embedded and that calculates energy map data of the original image data in which the patch information is embedded, and the patch information is embedded. An image processing apparatus is characterized by comprising: a source for generating watermark-embedded image data based on the original image data, the difference image data, and the energy map data.

Further, the image processing apparatus of the present invention includes a color space conversion unit for performing a color space conversion process for embedding watermark information in the original image data in which patch information is embedded, and the watermark information embedded state. A color space restoration unit for converting the image data into a color space before conversion by the color space conversion unit, wherein the Fourier transform unit is an original in which the patch information converted by the color space conversion unit is embedded. By subjecting the image data to Fourier transform, the amplitude information and the phase information are extracted, and the difference image generation unit,
Inverse Fourier transform based on the output of the synthesizing unit and the phase information, and the difference that is the difference between the result of this inverse Fourier transform and the original image data in which the patch information after conversion by the color space converting unit is embedded. Image energy is generated, and the energy map data calculation unit embeds the converted patch information by performing a wavelet transform on the original image data in which the patch information converted by the color space conversion unit is embedded. Is to calculate energy map data of the original image data, the generation unit is based on the original image data embedded patch information after conversion, the difference image data, and the energy map data. In this way, the watermark embedded image data is generated.

Further, the present invention is an image processing apparatus for reading detection target image data obtained based on an image in which patch information for giving a geometrical reference to original image data is embedded and detecting watermark information. Therefore, the pattern of the patch information is calculated based on the position of the peak of the value of the autocorrelation and the autocorrelation calculation unit that detects the change in the pixel value in the detection target image data and calculates the autocorrelation. A patch detection unit, a geometric conversion unit that corrects the detection target image data by using a magnification and a rotation angle calculated based on the calculated patch information pattern, and a geometric conversion unit that corrects the geometric conversion unit. A Fourier transform unit for performing Fourier transform on the detection target image data to extract amplitude information, and a despreading unit for detecting watermark information based on the amplitude information. It is intended to.

The image processing apparatus of the present invention further comprises a color space conversion unit for converting the detection target image data into a color space for detecting watermark information, and the Fourier transform unit includes the color space conversion unit. It is characterized in that the detection target image data in the color space after conversion by the unit is Fourier-transformed to extract amplitude information.

Further, according to the present invention, in an image processing method for generating watermark-embedded image data by embedding watermark information only in a predetermined intermediate frequency band in original image data, the watermark information is spectrally spread. According to the above, a spread spectrum process of generating diffused watermark information, and a patch information embedding process of repeatedly embedding patch information for giving a geometrical reference to the original image data in the original image data in a predetermined pattern,
In the patch information embedding step, the original image data in which the patch information is embedded is Fourier-transformed to extract amplitude information and phase information, a Fourier transform step, the diffused watermark information, and the amplitude information. And a differential image data that is a difference between the result of the inverse Fourier transform and the original image data in which the patch information is embedded, based on the output of the synthetic process and the phase information. A difference image generating step of generating, an energy map data calculating step of wavelet transforming the original image data in which the patch information is embedded, and calculating energy map data of the original image data in which the patch information is embedded, and the patch The original image data in which the information is embedded, the difference image data, and the energy map Based on the chromatography data, and having a generating process of generating the watermark information embedded image data.

Further, the present invention is an image processing method for generating watermark-embedded image data by embedding watermark information only in a predetermined intermediate frequency band in the original image data, in the original image data geometrically. A color space conversion process of converting the original image data in which patch information for giving a reference is embedded into a color space for embedding watermark information, and the watermark-embedded image data,
A color space restoration process for converting the color space into a color space before conversion in the color space conversion process, and the Fourier transform process Fourier transforms the original image data in which the patch information converted by the color space conversion process is embedded. By transforming, the amplitude information and the phase information are extracted, and the difference image generating process performs an inverse Fourier transform based on the output of the combining process and the phase information, and the result of the inverse Fourier transform. The difference image data that is the difference from the original image data in which the patch information after the conversion by the color space conversion process is embedded is generated, and the energy map calculation process is performed after the conversion by the color space conversion process. The original image data in which the patch information is embedded is subjected to a wavelet transform, so that the original image data in which the converted patch information is embedded is processed. It is for calculating the Energy map data, the generating process, the the original image data embedded patch information after conversion, and the difference image data,
Image data with watermark information embedded therein is generated based on the energy map data.

Further, the image processing method of the present invention is characterized by further comprising a printing step of printing using the generated watermark-embedded image data.

The present invention is also an image processing method for detecting watermark information by reading detection target image data obtained based on an image in which patch information for giving a geometrical reference to original image data is embedded. Therefore, the pattern of the patch information is calculated based on the autocorrelation calculation process of detecting the change in the pixel value in the detection target image data and calculating the autocorrelation and the position of the peak of the value of the autocorrelation. A patch detection process, a geometric conversion process for correcting the detection target image data by using a magnification and a rotation angle calculated based on the calculated patch information pattern, and a geometric conversion process corrected in the geometric conversion process. It has a Fourier transform process of performing Fourier transform on the detection target image data to extract amplitude information, and a despreading process of detecting watermark information based on the amplitude information. It is characterized in.

The present invention is also an image processing apparatus for processing original image data, wherein patch information for giving a geometrical reference to the original image data is repeated in a predetermined pattern in the original image data. It is characterized by comprising a patch information embedding unit for embedding.

The present invention is also an image processing apparatus which receives as input the detection target image data in which patch information for giving a geometrical reference to the original image data is embedded, and the pixels in the detection target image data are input. An autocorrelation calculation unit that detects a change in value and calculates an autocorrelation, a patch detection unit that calculates a pattern of patch information based on the position of the peak of the value of the autocorrelation, and a calculated pattern. And a geometric conversion unit that corrects the detection target image data using the calculated magnification and rotation angle.

Further, according to the present invention, in a printed matter on which an image is printed, patch information and watermark information for giving a reference to original image data are embedded in the image of the printed matter. Is.

[0029]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below with reference to the drawings.

First, an outline of a procedure for embedding watermark information and a procedure for detecting watermark information will be described with reference to FIGS. 1 and 2. FIG. 1 is a flowchart showing an outline of a procedure for embedding watermark information. Figure 2
It is a flowchart which shows the outline of the procedure which detects a watermark.

<Procedure for Embedding Watermark Information (Outline)> As shown in FIG.
The process of embedding patch information (step 204) is performed so that the scale (also referred to as magnification) and the rotation angle (image orientation) of the original image data can be detected later. That is, the patch information is for giving a geometrical reference to the original image data. Next, color space conversion (step 20)
Perform the processing of 2). As a result, the original image data in which the patch information is embedded is converted from the image data in the original color space into the image data in the color space that is convenient for combining watermarks.

On the other hand, an encoding process (step 218) is performed based on the watermark information 120. Then, the watermark information embedded in the error correction code in step 218 is embedded in the image data (output from step 202) in which the patch information is embedded and the color space is converted by the process of embedding watermark information (step 1200). . In the process of step 1200, the watermark information coded in step 218 is spread in a wider band to spread the spectrum, and is divided into a predetermined block image and spread into the original image data in which the patch information is embedded. A process of embedding the watermark information is performed, the blocks obtained by this process are integrated, and the process of making the original image data in which the watermark information and the patch information are embedded is performed.

Then, the inverse conversion of the color space conversion performed in step 202 is performed on the original image data in which the watermark information and patch information obtained in step 1200 are embedded to restore the color space (step 262). And then
As a result, the watermark-embedded image data 150
Is obtained. In this embodiment, an example in which the watermark information is embedded after the patch information is embedded is shown, but the order is not particularly limited.

<Watermark Information Detection Procedure (Outline)> As shown in FIG. 2, the watermark-embedded image 150a (detection target image) is A / D-converted (step 264). The watermark embedded image 150a is analog information represented as, for example, a printed matter. Also, step 26
The A / D conversion process of No. 4 is performed using, for example, a scanner. Next, the digitized image data obtained from the watermark-embedded image 150a is subjected to an emphasis filter (step 265). As a result, the watermark-embedded image data 150s (detection target image data) is obtained. When the detection target image data 150s is originally digitalized image data, step 264
The A / D conversion process of is omitted. In that case,
Further, the processing of the emphasis filter in step 265 may be omitted.

Next, a process of detecting a patch (step 266) from the detection target image data 150s is performed. This patch information has been embedded in the watermark-embedded image 150a in step 204 shown in FIG. By detecting this patch information, the detection target image data 1 currently targeted for watermark information detection
The difference in scale and image orientation between 50s and the original image data 100 shown in FIG. 1 can be seen. The difference between these is
It is obtained as numerical information of scale and rotation angle, respectively.

Next, the geometric conversion (step 267) is performed using the above-mentioned numerical information of the scale and the rotation angle. Thereby, the detection target image data 150s
Are corrected to have the same scale and image orientation as the original image data 100.

Next, the encoded watermark information is detected (step 1270), and the detected watermark information is decoded (step 292) to be combined with this image. Watermark information 120
Can be extracted correctly.

Although not described in the above general procedure, if the color space represented by the input watermark-embedded image data and the color space in which the watermark is embedded are different, watermark information is detected. Therefore, conversion processing is performed so that the color spaces are the same.

Next, the details of each processing used in the above-mentioned outline procedure will be described. <Color Space Conversion and Restoration> Here, the color space conversion shown in step 202 of FIG. 1 and the same step 2
The restoration of the color space indicated by 62 will be described. Generally, digitized image data can be represented by various color spaces. R is a commonly used color space.
There are GB, CMYK, YIQ, gray scale, and the like.
RGB is R (red, red) and G (green, green) and B (blue, bl)
ue) is a color space that uses each component. CMYK is C
It is a color space that uses (cyan, cyan), M (magenta, magenta), Y (yellow, yellow), and K (black, black) components. YIQ is a color space that uses Y (luminance) and I and Q (each color tone) components. Grayscale is a color space that uses only luminance and is used to represent monotone images. Note that there is YUV as a color space similar to YIQ, and this YUV also uses Y (luminance) and U and V (each color tone) components.

FIG. 3 is a schematic diagram showing an outline of color space conversion and color space restoration. (A) to (c) shown in FIG.
Indicates the case where the color space on which the original image data is based (which is referred to as “actual image color space” for convenience) is different from the color space for embedding the watermark information.
In the example shown in FIG. 3A, the actual image color space is grayscale, and the color space for embedding watermark information is a color space having a luminance (G) component, a color space having a frequency (F) component, or the like. It is a color space. In the example shown in FIG. 3B, the actual image color space is RGB, and the color space for embedding watermark information has an RGB color space, a color space having a luminance (G) component, and a frequency (F) component. A color space such as a color space. In the example shown in FIG. 3C, the actual image color space is CMYK.
And the color space for embedding watermark information is CMYK
Color space having a density (G ′) component, a color space having a frequency (F ′) component, and the like.

In the digital watermark composition according to this embodiment, the actual image color space is converted into a color space for embedding watermark information, and the watermark information is embedded in a predetermined component in the color space for embedding the watermark information. And then restore to the actual image color space. The "predetermined component" is preferably a component that cannot be perceived by humans or is difficult to perceive even if watermark information is embedded, but it is not limited to such a component and any component may be used. .

That is, in principle, the watermark information may be embedded in any component of any color space. However, in a verification experiment conducted by the inventors, when watermark information is embedded in the Y (luminance) component (channel) in the color space of the YIQ model, there is little deterioration in image quality when a human sees an image, It has been found that the watermark information is particularly robust.

As an example, the actual image color space is RGB,
A method of converting and restoring the color space when the color space for embedding the watermark is YIQ will be described. This conversion and restoration is calculated by the following equation (1).

[0044]

[Equation 1]

Here, A is a predetermined transformation matrix of 3 rows and 3 columns. A ⁻¹ is an inverse matrix of the transformation matrix A. In the mathematical expression (1), each component of Y, I, and Q is labeled as Y ′, I ′, and Q ′ for convenience.

As another example, the actual image color space is CM.
YK, and the color space for embedding watermark information is YI
A method of color space conversion and restoration in the case of Q will be described. The relationship between RGB and CMYK is C = (G + B) /
Since 2, M = (B + R) / 2 and Y = (R + G) / 2, the conversion between CMYK and YIQ is the above conversion matrix A
Is calculated by the following equation (2). When converting the actual image color space CMYK to the color space YIQ, the K component is omitted, and the pseudo relationship between RGB and CMY is C =
Using (G + B) / 2, etc., the conversion relationship between CMY and Y'I'Q 'is shown.

[0047]

[Equation 2]

In the equation (2), for convenience, Y, I,
Each component of Q is labeled as Y ', I', Q '.

<Patch Information Embedding> Here, the patch information embedding shown in step 204 of FIG. 1 will be described. The patch is information that is used as a reference for the magnification and rotation angle of the original image data. Then, by embedding this patch at a predetermined position in the original image data, it serves as this reference. In this embedding method, the patch is repeatedly embedded in the original image data in a predetermined pattern (a predetermined cycle).
Hereinafter, the patch is referred to as patch information.

The image data for which the patch information is embedded is represented by the function f (x, y, c). here,
(X, y) is the coordinate value of the target image data, and c
Is a color channel (color component) in a predetermined color space. Let s be the color channel selected for embedding the patch information. In principle, patch information may be embedded in any color channel. However, as a result of the proof experiment conducted by the inventors, there is little deterioration in image quality when a human sees the image data and the printed matter on which the image data is printed, and it is easy to detect the patch information. , The following color channels are known to be effective.

That is, the color channel Y in the YIQ color space and the color channels B and R in the RGB color space.
A color channel G when the GB color space is converted into a color space including luminance (G), a color channel Y in the CMYK color space, a color channel D converted into the density in the CMYK color space, and the like for embedding patch information. It is effective as a color channel.

Next, the procedure for embedding patch information will be described. The image data represented by the function f (x, y, s) is converted into a block image f of P size Mp pixels × Np pixels.
It is decomposed into i (xp, yp, s). However, 0 ≦ xp <
Mp, 0 ≦ yp <Np, and 0 ≦ i <P.

FIG. 4 is a schematic diagram showing division into block images for embedding patch information. The entire image data (original image data) shown in FIG. 4 has horizontal (8 × Mp) pixels,
It is composed of a total of (64 × Mp × Np) pixels of vertical (8 × Np) pixels. In FIG. 4, the hatched portion is one block image, and this block image is composed of horizontal Mp pixels and vertical Np pixels. The other block images have the same size. And the whole image data is 64
Is divided into block images (that is, P = 6)
4).

Then, using a pseudo random number sequence generated based on the secret key Key1, Pn integer coordinate values (xps, y)
ps) without duplication. However, 0 ≦ p <Pn,
This Pn is the number of patch information to be embedded in one block image. Since the block size is Mp × Np, it is necessary to satisfy Pn ≦ Mp × Np. In addition, Mp
Although Np and Np may be different from each other, the following description will be given using Np with Mp = Np for convenience.

Next, the Pn pixels (xp
s, yps), and patch information p of weight w (xp, yp)
Embed (xp, yp). However, 0 ≦ xp <Np,
0 ≦ yp <Np. That is, the action of changing the value of each of the Pn pixels selected using the pseudo-random number sequence as shown by the following equation (3) or equation (4) is added.

[0056]

[Equation 3]

[0057]

[Equation 4]

However, the function p (xp, yp) is the brightness G (xp, yp) of the block image fi (xp, yp, s).
/ This is a function that takes into consideration the density D (xp, yp) and the edge E (xp, yp). An example of the function p (xp, yp) is
It is like the following formula (5).

[0059]

[Equation 5]

The coordinate values (xps, yps) of the pixels in which the patch information is embedded are common to all P block images.

FIG. 5 is a graph showing an example of a weighting pattern of embedded patch information. In the example shown in FIG. 5A, the change of the weight α is added only to the selected pixel (xps, yps). That is, fi (xps, yps, s) ← fi (xps, yps,
s) + α. In the example shown in FIG. 5B, the selected pixel (x
ps, yps) is added with the weight α, and pixels (xps−1, yps) and (xps + 1,
yps) is added with a weight (-α / 2). That is fi (xps, yps, s) ← fi (xps, yps, s) ＋ α fi (xps-1, yps, s) ← fi (xps-1, yps, s) −α / 2 fi (xps + 1, yps, s) ← fi (xps + 1, yps, s) −α / 2. The patch information weighting pattern shown in FIG. 5B is more emphasized than the patch information weighting pattern shown in FIG. 5A and is embedded in the selected pixel. The feature is that it is easy to detect patch information. The weighting pattern of the patch information is not limited to the one illustrated in FIG. 5, and other weighting patterns may be used.

FIG. 6 is a flowchart showing the procedure of the patch information embedding process described above. Explaining with reference to FIG. 6, the given original image data is divided into block images (step 2041). Note that in FIG. 6, division into block images for embedding patch information is represented as “block division-1”.

On the other hand, the patch parameter value is determined (step 2042). Here, the patch parameter is the size of the block image (the number of vertical and horizontal pixels), 1
The number of patch information to be embedded in each block image, embedding strength (eg, α in FIG. 5), weighting pattern of embedding patch information (eg, weighting pattern of (a) or (b) in FIG. 5), patch information Is information including a pixel position and the like in which is embedded.

The following values are used as examples of patch parameter values. Original image data is vertical 1024
In the case of about × 1024 horizontal pixels, an example of the size of the block image is 64 vertical pixels × 64 horizontal pixels. As the size of the block image becomes relatively smaller, that is, as the number of divisions increases, the number of peaks of the autocorrelation function when detecting patch information increases, but the peak property deteriorates. On the contrary, as the size of the block image becomes relatively large, that is, the number of divisions becomes smaller, the peak property of the autocorrelation function at the time of detecting the patch information becomes better, but the number of peaks decreases. That is, since the peak property and the number of peaks are in a trade-off relationship, it is necessary to divide the original image data into block images of appropriate size. The number of pieces of patch information to be embedded in one block image is, for example, 1024 for a block image of vertical 64 pixels × horizontal 64 pixels. The embedding strength of the patch information is, for example, about 10 gradations for a 256 gradation color channel.

The pixel position where the patch information is embedded is
The pseudo random number sequence based on the secret key Key1 is used to determine the coordinate value in the block image.

Then, using the patch parameter values determined in step 2042, a process of embedding patch information is performed for one block image (step 2043). Then, it is determined whether or not the patch information embedding process has been completed for all the divided block images, and if not completed, the process returns to step 2043 to process the remaining block images (step 2044). By this iterative process, the process of embedding patch information is performed for all block images. After the processing for all the block images is completed, the divided block images are combined (step 2045). In FIG. 6, the block synthesis corresponding to the “block division-1” is referred to as “block synthesis-1”.
Is represented.

<Encoding of Watermark Information> Here, the encoding of the watermark information shown in step 218 of FIG. 1 will be described.

FIG. 7 is a flowchart showing the procedure of the watermark information encoding process. Underlying watermark information M0
Is a bit string having a length of m bits, and is expressed as M0 = (b01, b02, ..., b0n). First, a hash value having a length of H bits is generated by a predetermined hash function based on the watermark information M0. By adding this hash value to the watermark information M0, M0 '= (b01, b02, ..., B0n, h01, h02, ...
,, h0H) is obtained (step 2181).
Here, of M0 ', (b01, b02, ..., b0n)
Is called an information bit part, and (h01, h02, ..., H0H) is called a hash value bit part. Note that step 21
The hash function used in 81 outputs fixed-length data from an indefinite-length input, cannot predict input data from the output, cannot easily create input data having the same output, and cannot output the same output data. Try to use one that satisfies the requirement that it is very unlikely to occur. Specifically, a hash function such as MD5 (Message Digest 5) devised by Ronald L. Rivest or SHA-1 (Secure Hash Algorithm 1) devised by NSA is used.

Next, the watermark information M0 'is coded into a bit string M1 = (b11, b12, ..., B1m) having a length of m bits by using a coding method having an error correction capability. (Step 2182). This encoded bit string is used as watermark information M1. Note that step 21
The error correction coding method used in 82 is, for example, a Hamming code or a BCH code (Bose-Chaudhuri-Hocquengh).
em Code), Reed-Solomon code, turbo code, etc. are used.

<Diffusion of Watermark Information> Here, the spread spectrum of the watermark information, which is a part of the processing for embedding the watermark information shown in step 1200 of FIG. 1, will be described. First, the watermark information M1 which is binary data is encoded using the key information Key2. A spreading sequence xn of length N is calculated as follows using a pseudo random number sequence mn having the key Key2 as an initial value and a predetermined real number parameter α.

[0071]

[Equation 6]

The pseudo random number sequence mn generated and used here is preferably a cryptographically secure one. To generate this pseudo-random number sequence, for example, Go
ldCode, Kasami Code, M-sequ
ence, Perfectmap, etc. are used.

Next, the watermark information M1 is converted into an L-adic number (L ≧ 2).
The radix is converted into a value, and the converted value of each digit (hereinafter referred to as “symbol value”) is set to Sk (0 ≦ k <K). Then, for k = 0, 1, ..., K-1, the process represented by the following mathematical formula is executed.

[0074]

[Equation 7]

That is, the spreading sequence x expressed in the equation (6)
Based on n, the symbol value Sk
Spreading watermark information M (b1, b2, ..., By adding a pseudo-random number sequence according to (0 ≦ k <K)
bL) is calculated.

As described above, the watermark information M1 coded by the error correction coding method is diffused into a pseudo-random number sequence of length N, and the watermark information M (b1, b2, ..., B).
L) can be obtained.

A method of extracting the watermark information M1 from the spreading sequence xn generated by the above procedure will be described. First, the pseudo random number sequence mn is obtained based on the key information Key2. Further, the obtained diffused watermark information is tentatively expressed as yn. Then, the pseudo random number sequence mn
Using, the symbol value detection sequence qn (k, l) of length N is calculated by the following equation.

[0078]

[Equation 8]

After the processing of equation (8) is executed for all n, shift is performed so that the average value of qn (k, l) becomes zero. Then, from this qn (k, l) and the diffused watermark information yn, for l = 0, 1, ..., L-1,
The symbol response value Zl is calculated by the following formula.

[0080]

[Equation 9]

When Zl thus obtained has the maximum value, l is the detected symbol value, and the original watermark information M1 is obtained by performing the radix conversion of this symbol value.
Can be taken out.

<Embedding of Watermark Information> Here, FIG.
A procedure for embedding the diffused watermark information in the image in the watermark embedding process shown in step 1200 of FIG. First, as shown in FIG. 8, target image data is a block image fb of horizontal M0 pixels × vertical N0 pixels.
Divide into (x, y). The watermark information is embedded for each block image. The size of the block image for embedding the watermark information may be the same as or different from the size of the block image for embedding the patch information described above. However, since FFT (Fast Fourier Transform) is used in the process of embedding watermark information, it is desirable that M0 = N0 and M0 be a power of two. Below, it demonstrates on the assumption that M0 = N0.

The Fourier transform Fb (u, v) of the divided block image fb (x, y) is calculated by the following equation (10).

[0084]

[Equation 10]

Then, the amplitude component of Fb (u, v) is Am
p (u, v), and the phase component is Phase (u,
v). Then, this amplitude component Amp (u, v)
The watermark information is embedded in the formula (11).

[0086]

[Equation 11]

However, in equation (11), pow
r is an arbitrary real number and represents the watermark strength. Secret key K
By using the pseudo-random number sequence generated from ey3, the pair of integers (un, vn) (0 ≦ n <N) is set to be in the intermediate frequency band for embedding the predetermined watermark information. Select without duplication. Finally, the inverse Fourier transform is performed to embed the watermark information in the block image fb ′.
Get (x, y). This process is performed on all block images and they are combined to obtain an image f '(x, y) in which watermark information is embedded.

<Reduction of Image Noise by Watermark Information> Next, a procedure for reducing the noise by matching the noise due to the embedding of the watermark information with the local characteristics of the image data will be described. FIG. 9 is a flowchart showing a processing procedure for generating image data in which patch information and watermark information are embedded, including processing for reducing this noise.

First, the image data 101 having the patch information embedded therein is subjected to the wavelet transform (step 242) to extract the local feature of the image data 101 having the patch information embedded therein. By this wavelet transform, the image data 101 in which the patch information is embedded is represented as a patch image f (x, y), and the patch image f (x, y) is decomposed into sub-band images and orthogonal partial images ( Dj (n) f) (x, y). Then, using this partial image (Dj (n) f) (x, y), the energy Pjf of the component of Dj at the resolution j
(X, y) is calculated by the following mathematical expression (12). Note that in FIG. 9, the image data 10 in which patch information has been embedded
In No. 1, it is assumed that the color space conversion 202 suitable for embedding watermark information has been performed.

[0090]

[Equation 12]

The Pjf (x, y) thus obtained is magnified 2 ^ (-j) times by the nearest neighbor method to obtain the patch image f.
Energy map data En having the same size as (x, y)
(X, y) is obtained (calculation of high-frequency energy data, step 244). This energy map data En
The following mathematical expression (13) is used so that (x, y) becomes a predetermined value or less.
Thresholding is performed by the method shown in.

[0092]

[Equation 13]

Next, as expressed by the following equation (14),
Difference image data Diff (x, y) between the image f '(x, y) in which the watermark information is embedded and the patch image f (x, y) in which the patch information is embedded is obtained (step 23).
5).

[0094]

[Equation 14]

Finally, the energy map data En
Using (x, y), the weight mask En (x, y) · p
power2 (power2 is an arbitrary positive real number) is created (step 246), and the watermark information is again applied to the patch image f.
(X, y) are combined (step 250, step 25)
5). That is, it is as in the following mathematical expression (15).

[0096]

[Equation 15]

F '(x, y) obtained by the equation (15)
Is watermark-embedded image data 151.
As described above, by using the wavelet transform, it is possible to obtain the image data 151 in which the watermark information that is hard to be perceived by humans and has high robustness is embedded. By converting (restoring) the image data 151 in which the watermark information is embedded into the original color space, the watermark information-embedded image data 150 is obtained.

That is, the method of generating the watermark information embedded image data according to the present embodiment is a spectrum spreading process of generating spread watermark information by spectrally spreading the watermark information, and a patch image (patch information is already embedded. Image data) is subjected to a Fourier transform to extract amplitude information and phase information, a synthesizing step of synthesizing the diffused watermark information and the amplitude information, an output of the synthesizing step, and the phase. Inverse Fourier transform based on the information, a difference image generation process for generating difference image data which is the difference between the result of this inverse Fourier transform and the patch image, and energy map data of the image by wavelet transforming the patch image Energy map data calculation process for calculating the patch image and the difference image data. It is assumed to have a generating process of generating the watermark information embedded image data based on the energy map data with. Based on the above premise, the first feature of the method for generating watermark information embedded image data of the present embodiment is that patch information is embedded in the original image data. The second feature is that
That is, a hash value calculated by a predetermined hash function based on the watermark information is added to the watermark information, and the watermark information with the added hash value is embedded in the patch image. The third feature is that the watermark information is encoded by an error correction code, and then the watermark information encoded in this way is embedded in the patch image. Also, the fourth
The feature is that color space conversion is performed based on the patch image, and watermark information is embedded in a predetermined component in the converted color space. In addition to this, after embedding the watermark information, the original color space is restored.

<Emphasis Filter> Here, the emphasis processing shown in step 265 of FIG. 2 will be described. D / A conversion process (printing process from digitized image data) or A / D conversion process (scanner, digital camera, etc. are used to form an image as digitized image data from an analog medium such as a printed matter. It is conceivable that the watermark information is difficult to be extracted in the process of taking in). In particular, when applying the weight mask created in step 246 of FIG. 9, it is conceivable that an action equivalent to a low-pass filter works and watermark information becomes difficult to be extracted. Therefore, in order to make it easier to detect the watermark information, it is conceivable to apply this enhancement filter process. As the processing of this emphasis filter, for example, the processing represented by the following formula (16) or formula (17) is performed.

[0100]

[Equation 16]

[0101]

[Equation 17]

<Detection of Patch Information> Here, the processing of detecting the patch information shown in step 266 of FIG. 2 will be described. Scanners and digital cameras are often used for A / D conversion of images on analog media such as printed matter, but the size and direction of the images must be strictly digitized and faithfully taken into the original image data. Is very difficult and usually causes some distortion, rotation and scale changes. Also, when the resolution when printing (D / A conversion) and the resolution when digitizing and capturing (A / D conversion) are different, the scale changes. Therefore, by detecting the patch information embedded in the process of generating the watermark information embedded image data, the magnification and the rotation angle between the digitized image data and the original digitized image data are calculated.

Since the patch information is inserted on a pixel-by-pixel basis, it is necessary to detect a change in the value between adjacent pixels even when detecting the patch information. To do so, use the following formula (1
It is effective to perform the process using the filter h (x, y) as shown in 8). The formula for h (x, y) changes depending on the patch information, but the formula shown here is an example in the case of being entered in pixel units.

[0104]

[Equation 18]

Then, the filter h (x, y) is added to the image ff (x, y) (image data digitized and taken in) whose watermark information is to be detected, as in the following equation (19).
y) Calculate the convolution with.

[0106]

[Formula 19]

As described above, since the same patch information is embedded in the block image obtained by dividing the original image data, that is, the patch information is embedded in the original image data in tile form, the image f (x, y ), The size of the block image containing the patch information in the image ff (x, y), that is, the repeating period and the repeating direction of the patch information can be obtained. That is, the image f (x,
By detecting the peak of the autocorrelation of y), the image f
The magnification and rotation angle between f (x, y) and the original image data can be calculated.

The autocorrelation of the image f (x, y) is calculated by the following equation (20).

[0109]

[Equation 20]

From the obtained peak position (xm, ym) of the value of F1 (x, y), the magnification (scale) and the rotation angle (angle) can be calculated as in the following formula (21).

[0111]

[Equation 21]

However, in the equation (21), xm = y
m ≠ 0, and Np is the size of the block image when the patch information is embedded.

Since the magnification and the rotation angle are calculated as described above, the scale and orientation of the image ff (x, y) can be corrected.

<Geometric Transformation Processing> Here, the geometric transformation processing shown in step 267 of FIG. 2 will be described. Using the scale factor and the rotation angle obtained by the patch information detection process,
A geometric conversion process as shown in FIG. 10 is performed on the image ff (x, y).

The image before geometric transformation is converted into the image ff (x, y).
And the converted image is the converted image fo (x, y). The pixel (xi, yi) of the image ff (x, y) corresponding to the pixel (xo, yo) of the converted image fo (x, y) is
It can be calculated by the following formula (22). That is, FIG.
Is the image ff (x, y) in FIG. 10A, and the transformed image fo (x, y) in FIG.
y).

[0116]

[Equation 22]

Further, if the coordinate value of the pixel calculated by the equation (22) is not an integer, interpolation processing is performed. Although various methods are conceivable as the interpolation processing, as an example, the nearest adjacent pixel is used to calculate by the formula (2
The interpolation process represented by 3) is performed.

[0118]

[Equation 23]

Other interpolation processing methods include, for example, bilinear interpolation and three-dimensional curve interpolation.

In summary, the geometric transformation processing is performed on the basis of the size of the block image in which the patch information calculated by detecting the patch information is embedded, and the image data of the watermark information to be detected and the original image data. The image data to be detected is corrected by obtaining the scale factor and the rotation angle (direction, angle) between the two.

<Detection and Decoding of Encoded Watermark Information> Here, the process of detecting the encoded watermark information shown in step 1270 of FIG. 2 and the same step 29.
The decoding process shown in 2 will be described. FIG. 11 is a flowchart showing a procedure of a process of detecting watermark information. Image data with watermark information embedded therein is input to the processing shown in FIG. This image data has already been subjected to the detection of patch information and the processing of geometrical conversion as necessary.

First, in step 12701 of FIG. 11, the input image data in which the watermark information is embedded is divided into a plurality of block images. The size of each block image divided here is the same as the size of the block image when the watermark information embedding process is performed. Note that in FIG. 11, division into block images for detecting watermark information is represented as “block division-2”.

Next, in step 12702, Fourier transform processing is performed on one block image,
Thereby, amplitude information of the block image is obtained. Next, in step 12703, it is determined whether the processing has been completed for the necessary block image. When finished, the process proceeds to the next step 12704, and when there are still block images to be processed, the process returns to step 12702 to process them. Then, step 12704.
In, a multi-channel encoding process is performed based on the amplitude information of each block image obtained above.

As the contents of the multi-channel encoding process in step 12704, any one of the following first to fourth patterns is performed.

In the first pattern, a process of averaging the amplitude information obtained by Fourier transform from each block image is performed. On the other hand, a symbol value detection sequence is obtained based on a predetermined pseudo random number sequence. Then, the symbol response value is calculated based on the averaged amplitude information and the symbol value detection sequence, and the symbol value is detected. The watermark information obtained by converting the symbol value to the radix is the detected watermark information (hereinafter, the processing until the watermark information is detected based on the amplitude information is referred to as “despreading processing”). However, since the watermark information detected here is the information coded by the error correction code, the decoding process is further performed. Further, since the hash value is added to the information obtained by this decoding, the hash value is used again to obtain the hash value from the original watermark information and the matching is performed. This first pattern has an advantage that the amount of calculation processing until finally obtaining the watermark information is small.

In the second pattern, amplitude information obtained by Fourier transform from each block image is used to
The despreading process is performed for each block image. Then, by performing majority processing (simple majority vote or weighted majority vote) on the basis of the plurality of watermark information (information of the state encoded by the error correction code) obtained as a result, a single watermark information (by the error correction code (Encoded state information) is obtained. Then, the obtained watermark information is decrypted, and the hash value is collated. The second pattern has a merit that it is resistant to errors because despreading processing is performed for each block image and the majority processing is performed on the result.

In the third pattern, amplitude information obtained by Fourier transform from each block image is used to
The despreading process is performed for each block image. Further, each watermark information obtained as a result is decoded. Then, a single piece of watermark information (decoded information) is obtained by performing a majority decision process (simple majority decision or weighted majority decision) based on the plurality of pieces of watermark information (decoded information). Then, hash value matching processing is performed on the decrypted watermark information. Since the third pattern is performed up to the decoding process for each block image, it is possible to perform the majority decision process with the information corrected by the capability of the error correction code, and there is an advantage that it is more resistant to errors. .

In the fourth pattern, despreading processing, decoding processing, and hash value matching processing are performed for each amplitude information obtained from each block image. In the fourth pattern, since the hash values are collated individually for each block image, even if many errors are included in the image data in which the watermark information is embedded, the hash values for one or more block images are If the collation is successful, there is a high possibility that the information is watermark information that has been correctly detected, and there is an advantage that it is more resistant to errors.

The flow from the watermark despreading process to the hash value matching process will be further described. FIG. 12 is a flowchart showing a detailed procedure of watermark information detection processing. In step 12702 of FIG. 12, Fourier transform is performed on the block image. Step 1
In 2705, the despreading process of the watermark information is performed based on the amplitude information obtained as a result of the Fourier transform and the pseudo random number sequence obtained based on the secret key information (Key4 in this case). In step 2921, decoding processing of the error correction code is performed based on the watermark information (information coded by the error correction code) obtained as a result of the despreading processing. In step 2922, M0 ′ = obtained by the above processing
, B0n, h01, ..., h0H of (b01, ..., b0n) (information bit part), a hash value is calculated by a predetermined hash function and calculated. Hash value and M0 '=
, (B0n, h01, ..., b0H) are compared with (h01, ..., h0H) (hash value bit part). In step 2923, it is determined whether the collation results match. If they match, the information bit part of the above M0 ′ is M0 = (b01, ...,
b0n), it is determined that this M0 is the correctly detected watermark information. If they do not match, the detected watermark information is judged to be an error.

The image processing method for embedding watermark information and the image processing method for detecting embedded watermark information have been described above. Next, an image processing apparatus that executes the processes of these image processing methods will be described.

The image processing apparatus for embedding watermark information is provided with the following units. The hash value adding unit calculates a hash value based on the watermark information using a predetermined hash function, and adds the calculated hash value to the watermark information. The error correction coding unit performs a process of coding the watermark information (this watermark information may be a hash value added. This is the same in the following) with an error correction code. The spread spectrum unit spreads the watermark information (this watermark information may be coded by an error correction code. This also applies below) to generate spread watermark information. To do. The patch information embedding unit performs a process of embedding patch information for the purpose of calculating a magnification and a rotation angle for each block image of the divided original image data. That is, the patch information embedding unit performs a process of embedding patch information in the original image data in a predetermined direction and cycle. The color space conversion unit performs processing of converting original image data in which patch information is embedded or original image data into a color space for embedding watermark information. The Fourier transform unit may be original image data (patch information may be embedded in the original image data. In some cases, the original image data may be post-color space conversion. The same applies below. ) Is Fourier transformed to extract amplitude information and phase information. The synthesis section
Processing for combining the diffused watermark information and the amplitude information is performed. The difference image generation unit performs an inverse Fourier transform based on the output of the synthesis unit and the phase information, and performs a process of generating difference image data that is a difference between the result of this inverse Fourier transform and the original image data. The energy map data calculation unit calculates the energy map data of the image by wavelet transforming the original image data. The generating unit generates watermark information-embedded image data based on the original image data, the difference image data, and the energy map data. The color space restoration unit performs a process of re-converting the watermark information embedded image data into the color space before conversion by the color space conversion unit.

By connecting a printing device to the above image processing device and performing printing using the watermark-embedded image data generated by the above image processing device,
Printed matter can be manufactured. The printed matter can detect the watermark information by being digitized, and further has the resistance by the magnification and the rotation angle by the patch information.

The image processing device for detecting the embedded watermark information includes the following parts. The color space conversion unit performs a process of converting image data for which watermark information is detected into a color space for detecting watermark information. The enhancement filter unit is for the image data for which the watermark information is to be detected (this image data may have been converted to a color space for detecting the watermark information. This is the same in the following). And an enhancement filter device for performing an enhancement filter process. The autocorrelation calculation unit has a pixel value of image data of which the input watermark information is to be detected (this image data may have been subjected to an emphasis filter process. This also applies to the following). Is detected by using a filter h (x, y) or the like, and a process of calculating an autocorrelation is performed. The patch detection unit, based on the position of the peak of the value of the autocorrelation, the size of the block image in which the patch information is embedded, that is,
The repetition cycle and the repetition direction of the patch information are calculated. The geometric conversion unit performs a process of correcting the magnification (scale) and the rotation angle (angle) of the image data for which the watermark information is detected, based on the repetition period and the direction calculated by the patch detection unit. . The Fourier transform unit performs a Fourier transform on the image data for which the watermark information is to be detected (this image data may have been corrected by the geometric transform unit. This is the same in the following), and the Fourier transform is performed in advance. A process of extracting amplitude information of the obtained intermediate frequency band is performed. The despreading unit obtains a symbol value detection sequence based on a predetermined pseudo random number sequence. Then, the despreading unit calculates a symbol response value based on the amplitude information and the symbol value detection sequence obtained from the Fourier transform unit. Further, the despreading unit detects a symbol value and performs radix conversion on the symbol value to calculate watermark information. The error correction decoding unit performs a decoding process of the error correction code based on the watermark information obtained by this despreading, and obtains the decoded watermark information. The hash value collating unit uses a predetermined hash function based on the information bit part included in the detected watermark information (this watermark information may be after decoding of the error correction code) to use the hash value. Is calculated, and the calculated hash value is compared with the hash value bit part included in the watermark information.

The information processing apparatus described above is realized by using, for example, a computer. In this case, the process of the processing by each unit described above is stored in a computer readable recording medium in the form of a program, and the above process is performed by the computer reading and executing the program. Here, the computer-readable recording medium refers to a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, or the like. Further, the computer program may be distributed to the computer via a communication line, and the computer that receives the distribution may execute the program.

Although the embodiments of the present invention have been described in detail with reference to the drawings, the specific configuration is not limited to these embodiments, and a design etc. within a range not departing from the gist of the present invention are also possible. included.

For example, in the above embodiment, the color space in which the patch information is embedded and the color space in which the watermark information is embedded are different, but they may be the same color space. If an appropriate color space is required for embedding patch information or watermark information, color space conversion processing is added or deleted as appropriate.

In the above embodiment, the original image data is divided into block images to embed the patch information and then the color space is converted for each block image. However, the color space conversion is performed for the entire original image data. May be performed, and then the block image may be divided for embedding the patch information. Further, in the above-described embodiment, the color space conversion is performed on the entire original image data and then divided into block images for embedding watermark information. Conversely, on the contrary, the original image data is blocked for embedding watermark information. The color space may be converted for each block image after dividing the image.

In the above embodiment, the watermark information is embedded after the patch information is embedded in the original image data. Conversely, the watermark information is embedded in the original image data and then the patch information is embedded. You may do it.

Further, when embedding the patch information, a process of removing the patch information may be performed by applying a change opposite to the change of the pixel added to the image data such as the original image to the pixel. The purpose of embedding patch information and its detection is to obtain the magnification and rotation angle of an image that changes by undergoing D / A conversion and A / D conversion. Therefore, after the magnification and the rotation angle are obtained, by removing the patch information, noise on the image due to the patch information can be removed and a high quality image can be reproduced. It should be noted that this patch information is preferably removed from image data such as an original image in which watermark information is embedded and then patch information is embedded.

In the above embodiment, the patch information is
Although it has been decided to repeatedly insert the image data such as the original image in the vertical and horizontal directions at a predetermined cycle (the same cycle may be used), the direction in which the patch information is repeated is not limited to this, and may be another direction. Is also good. As long as the direction in which the patch information is embedded is known, it is possible to correct the digitized image data by incorporating it based on the detected patch information. The two directions in which patch information is repeatedly inserted may or may not be orthogonal to each other.
Furthermore, the patch information may be repeatedly inserted only in one direction instead of two directions.

In the above embodiment, the patch information is inserted and detected in order to improve the reliability of the watermark information detection. However, the patch information is not limited to the purpose of the watermark information detection, and the D / A conversion is simply performed. Alternatively, the patch information may be inserted and detected for the purpose of obtaining the amount of change in the magnification and / or the rotation angle of the image and the image data that have undergone A / D conversion.

[0142]

As described above, when the image processing apparatus of the present invention is used, patch information for giving a geometrical reference to the original image data is repeatedly embedded in the original image data in a predetermined pattern. Since the patch embedding unit is provided, the patch information is detected even when the image in which the patch information is embedded undergoes D / A conversion and A / D conversion and its magnification and rotation angle change from the original image. By doing so, a geometric reference can be obtained, and changes in the magnification and the rotation angle can be obtained.

If the image processing apparatus of the present invention is used,
Since the original image data in which the patch information is embedded includes the color space conversion unit that performs the process of converting the color space for embedding the watermark information, the watermark information can be embedded in the color space different from the original image data. . Further, since the image data having the watermark information embedded therein is provided with a color space restoration unit for converting the image data into the color space before conversion by the color space conversion unit, image data represented in the same color space as the original image data is output. You can Thereby, the watermark information can be embedded in the original image data of various data formats. Further, by properly selecting the color space and the channel, it is possible to make the noise of the image due to the insertion of the watermark inconspicuous.

Further, the image processing apparatus of the present invention detects the change of the pixel value in the image data to be detected and calculates the autocorrelation based on the position of the peak of the autocorrelation value. A patch detection unit that calculates the pattern of the patch information, and a geometric conversion unit that corrects the detection target image data using the magnification and the rotation angle calculated based on the calculated pattern of the patch information. Since it is provided, it becomes possible to restore the detection target image data to the magnification and the rotation angle of the original image data based on the detected patch information, which improves the accuracy of watermark information detection.

Further, since the image processing apparatus of the present invention comprises the color space conversion unit for converting the detection target image data into the color space for detecting the watermark information, it is different from the color space represented by the detection target image data. Even when the watermark information is embedded in the color space, it is possible to detect the watermark information by first converting to the color space in which the watermark information is embedded.

Also, according to the image processing method of the present invention, since there is a printing process in which the generated watermark-embedded image data is used for printing, patch information is embedded and watermark information is embedded as described above. Printed matter can be manufactured. When such a printed matter is A / D converted by reading with a scanner and converted into digital image data again, the watermark information can be detected with good accuracy.

Further, the image processing apparatus of the present invention detects the change of the pixel value in the image data to be detected and calculates the autocorrelation, and based on the position of the peak of the autocorrelation value, A patch detection unit that calculates a pattern of patch information, and a geometric conversion unit that corrects the detection target image data by using a magnification and a rotation angle calculated based on the calculated pattern are included. ,In general,
Only the image data in which the patch information is embedded can be used to restore the image to the original magnification and rotation angle.

Further, in the printed matter of the present invention, since the patch information and the watermark information for giving a reference to the original image data are embedded in the image of the printed matter, it is possible to detect the watermark information with higher accuracy. You can

[Brief description of drawings]

FIG. 1 is a flowchart showing an outline of a procedure of a watermark synthesizing process according to an embodiment of the present invention.

FIG. 2 is a flowchart showing an outline of a procedure of watermark detection processing according to the same embodiment.

FIG. 3 is a schematic diagram showing an outline of color space conversion and color space restoration according to the embodiment.

FIG. 4 is a schematic diagram showing division into block images for patch embedding according to the same embodiment.

FIG. 5 is a graph showing an example of a patch embedding pattern according to the same embodiment.

FIG. 6 is a flowchart showing a procedure of patch embedding processing according to the same embodiment.

FIG. 7 is a flowchart showing a procedure of a process of encoding watermark information according to the same embodiment.

FIG. 8 is a schematic diagram showing division into block images for watermark embedding according to the same embodiment.

FIG. 9 is a flowchart showing a procedure of the entire watermark combining process according to the same embodiment.

FIG. 10 is a schematic diagram showing an outline of a geometric conversion process for image data to be detected according to the same embodiment.

FIG. 11 is a flowchart showing a procedure of a process of detecting watermark information according to the same embodiment.

FIG. 12 is a flowchart showing a more detailed procedure of a process of detecting watermark information according to the same embodiment.

FIG. 13 is a schematic diagram showing a digital watermark combining method according to a conventional technique.

FIG. 14 is a schematic view showing a digital watermark detection method according to a conventional technique.

[Explanation of symbols]

100 original image data 101 patch embedded image data 111 amplitude information 112 phase information 120 watermark information 121 diffused watermark information 150, 150s, 151 watermark embedded image data 150a watermark embedded image (analog information)

─────────────────────────────────────────────────── ─── Continuation of front page (51) Int.Cl. ⁷ Identification code FI theme code (reference) H04N 1/41 H04N 1/41 Z 5J104 7/08 7/08 Z 7/081 (72) Inventor Masatoshi Ohtake 1-5-1 Taito, Taito-ku, Tokyo Within Toppan Printing Co., Ltd. (72) Inventor Kenji Matsudo 1-5-1 Taito, Taito-ku, Tokyo Within Toppan Printing (72) Inventor Takuya Oshige Tokyo 1-5-1 Taitō, Taito-ku, Tokyo, Topic Printing Co., Ltd. F-term (reference) BC07 BC08 BD07 5J104 AA14 HA15 NA12 NA13 NA14

Claims (11)

[Claims] 1. A spread spectrum unit for generating spread watermark information by spectrum spreading of the watermark information, and patch information for giving a geometrical reference to the original image data in the original image data. A patch embedding unit that repeatedly embeds a predetermined pattern; a Fourier transform unit that extracts amplitude information and phase information by performing a Fourier transform on the original image data in which patch information is embedded by the patch embedding unit; The watermark information, the combining unit for combining the amplitude information, the output of the combining unit, and the inverse Fourier transform based on the phase information, the result of this inverse Fourier transform, and the patch information is embedded. A difference image generation unit that generates difference image data that is a difference from the original image data; and original image data in which the patch information is embedded An energy map data calculator that performs wavelet transform and calculates energy map data of the original image data in which the patch information is embedded, original image data in which the patch information is embedded, the difference image data, and the energy map An image processing device comprising: a generation unit that generates watermark information-embedded image data based on the data. 2. A color space conversion unit that performs a process of converting a color space for embedding watermark information in the original image data in which patch information is embedded, and the color space conversion for the image data in which the watermark information is embedded. A color space restoration unit for converting into a color space before conversion by the unit, wherein the Fourier transform unit Fourier transforms the original image data in which the patch information converted by the color space conversion unit is embedded. By the above, the amplitude information and the phase information are extracted, the difference image generation unit performs an inverse Fourier transform based on the output of the synthesis unit and the phase information, and the result of the inverse Fourier transform and the color space conversion. For generating difference image data which is a difference from the original image data in which the patch information after conversion by the unit is embedded, and the energy map data calculation unit By performing wavelet transformation on the original image data in which the patch information after conversion by the color space conversion unit is embedded, energy map data of the original image data in which the patch information after conversion is embedded is calculated. The generation unit is configured to generate watermark-embedded image data based on the original image data in which the converted patch information is embedded, the difference image data, and the energy map data. Claim 1
The image processing device according to item 1. 3. An image processing device for detecting watermark information by reading detection target image data obtained based on an image in which patch information for giving a geometrical reference to original image data is embedded, An autocorrelation calculation unit that detects a change in the pixel value in the detection target image data and calculates an autocorrelation, and a patch detection unit that calculates the pattern of the patch information based on the position of the peak of the value of the autocorrelation. A geometric conversion unit that corrects the detection target image data by using a magnification and a rotation angle calculated based on the calculated patch information pattern; and detection target image data corrected by the geometric conversion unit. An image characterized by comprising: a Fourier transform unit for performing a Fourier transform on the information to extract amplitude information; and a despreading unit for detecting watermark information based on the amplitude information. Processing equipment. 4. A color space conversion unit for converting the detection target image data into a color space for detecting watermark information, wherein the Fourier transform unit is in the color space after conversion by the color space conversion unit. The image processing apparatus according to claim 3, wherein the detection target image data is Fourier-transformed to extract amplitude information. 5. An image processing method for generating watermark-embedded image data by embedding watermark information only in a predetermined intermediate frequency band in original image data, wherein the watermark information is spread by spectrum spreading. A spread spectrum step of generating watermark information, a patch information embedding step of repeatedly embedding patch information for giving a geometrical reference to the original image data in a predetermined pattern in the original image data, and embedding the patch information By the process, a Fourier transform process of extracting the amplitude information and the phase information by Fourier transforming the original image data in which the patch information is embedded, a synthesis for combining the diffused watermark information and the amplitude information. Process, inverse Fourier transform based on the output of the synthesis process and the phase information, A difference image generation process of generating difference image data that is a difference between the result of Fourier transform and the original image data in which the patch information is embedded, and the original image data in which the patch information is embedded is wavelet-transformed, and the patch Based on the energy map data calculation process of calculating energy map data of the original image data in which the information is embedded, the original image data in which the patch information is embedded, the difference image data, and the energy map data, An image processing method comprising: a generation step of generating image data in which watermark information is embedded. 6. An image processing method for generating watermark-embedded image data by embedding watermark information only in a predetermined intermediate frequency band in the original image data, for giving a geometrical reference to the original image data. A color space conversion process for converting the original image data in which the patch information is embedded into a color space for embedding watermark information, and the watermark-embedded image data in a color space before conversion in the color space conversion process. A color space restoration process for converting; and the Fourier transform process, by performing Fourier transform on the original image data in which the patch information converted by the color space conversion process is embedded, to obtain amplitude information and phase information. For extracting the difference image, the difference image generation process, an inverse Fourier transform based on the output of the combining process and the phase information, For generating difference image data that is a difference between the result of the inverse Fourier transform of the image and the original image data in which the patch information after conversion by the color space conversion process is embedded. Energy map data of the original image data in which the patch information after the conversion is embedded is calculated by performing a wavelet transform on the original image data in which the patch information after the conversion in the spatial conversion process is embedded. The process is characterized by generating watermark information embedded image data based on the original image data in which the converted patch information is embedded, the difference image data, and the energy map data. The image processing method according to claim 5. 7. The image processing method according to claim 5 or 6, further comprising a printing step of printing using the generated watermark-embedded image data. Method. 8. An image processing method for detecting watermark information by reading detection target image data obtained based on an image in which patch information for giving a geometrical reference to original image data is embedded. An autocorrelation calculation process of detecting a change in pixel value in the detection target image data and calculating an autocorrelation, and a patch detection process of calculating a pattern of the patch information based on a peak position of the autocorrelation value. A geometric conversion process for correcting the detection target image data by using a magnification and a rotation angle calculated based on the calculated patch information pattern; and detection target image data corrected in the geometric conversion process. By performing a Fourier transform on the amplitude information to extract the watermark information, and a despreading process for detecting the watermark information based on the amplitude information. Image processing method to be considered. 9. An image processing apparatus for processing original image data, wherein patch information for embedding a geometrical reference to the original image data is repeatedly embedded in the original image data in a predetermined pattern. An image processing apparatus comprising: a unit. 10. An image processing apparatus, which receives as an input image data to be detected in which patch information for giving a geometrical reference to original image data is embedded, a change in a pixel value in the image data to be detected is detected. An autocorrelation calculation unit that detects and calculates an autocorrelation, a patch detection unit that calculates a pattern of patch information based on the position of the peak of the value of the autocorrelation, and a calculation based on this calculated pattern. An image processing apparatus, comprising: a geometric conversion unit that corrects the detection target image data using a magnification and a rotation angle. 11. A printed matter on which an image is printed, characterized in that patch information and watermark information for giving a reference to original image data are embedded in the image of the printed matter.