JP2011147017A - Digital watermark embedding method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a watermark embedding method which allows a user to grasp how a pixel value is varied or is limited during embedding on a color space. <P>SOLUTION: For pixels g(m, n), g^(m, n) in each position (m, n) of a source image and an index image, points P<SB>mn</SB>(L<SB>mn</SB>, a<SB>mn</SB>, b<SB>mn</SB>), P^<SB>mn</SB>(L<SB>mn</SB>, a<SB>mn</SB>, b<SB>mn</SB>) on a Lab space, wherein pixel values of the pixels are to be indicated, are adopted. An average W<SP>-</SP>of a pixel norm total sum is calculated, using pixel norms D(m, n), D^(m, n) which is a distance to an origin of the color space, a further quantized pixel norm W<SP>-</SP><SB>Q</SB>is calculated, bit information is embedded therein, and arithmetic operation is performed so as to move a change of the image at that time on a line connecting P<SB>mn</SB>and P^<SB>mn</SB>. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an electronic watermark embedding method for embedding an electronic watermark in image data, and more particularly, to an electronic watermark embedding method capable of determining the form and limit of image quality degradation when embedding an electronic watermark.

Digital watermarking is a technique for embedding some information in a digital content such as an electronic image so that it cannot be perceived by humans. The embedded information is called “watermark”. Digital watermarks are attracting attention as one of the technologies that can be used for copyright protection, and various researches and developments have been made.
By the way, when an electronic watermark is embedded in an electronic image, the embedded watermarked image deteriorates in image quality from the original image.

Therefore, conventionally, in order to suppress the influence of image quality degradation, a method of avoiding a luminance component with high visual sensitivity and minutely changing a color component such as a color difference has been used. For example, in image data using Y (yellow), M (magenta), C (cyan), and K (black), there is a technique for adding digital watermark information using yellow that is difficult to detect for human vision. (See Patent Document 1). In addition, in R (red), G (green), and B (blue) color system image data, there is a technology that uses digital watermark information using a blue component that is difficult to detect by human vision (Patent Document 2). reference).
Another method is to analyze the frequency components of the image data, and to increase the embedding strength in the high-frequency region where the visual characteristics are insensitive, considering the human visual characteristics, and decrease the embedding strength in the low-frequency region. (See Patent Document 3).

Japanese Patent No. 3251348 JP 2001-148776 A JP 2004-023184 A

By the way, in these conventional digital watermark techniques, image quality degradation is unique to each embedding method, and the degree of image quality degradation has not been determined until actual embedding. Thus, the fact that the form of image quality deterioration due to watermark embedding and the limit of deterioration are not determined is one of the factors hindering practical use. In order for the electronic watermarking technique to be widely used, it is desirable to be able to predict to some extent how the watermarked image will change when the watermark is embedded in the original image.
Therefore, the present invention changes the color space so that the pixel value change (image quality degradation) in the color space when the watermark information is embedded in the original image approaches the pixel value of the index image set in advance. It is an object of the present invention to provide a watermark embedding method that can grasp the method of change and the limit of change.

  From a practical viewpoint, the original image can be processed into various images such as a JPEG compressed image, a blurred image, an edge enhancement image, and a filter image by image processing. Using the post-image as the index image, the change format (image quality degradation format) can be determined so that the change in the watermarked image approaches the index image. It is desirable that it can be defined.

  In view of this, the present invention can preliminarily define a format of image quality degradation and a limit that is an allowable range of image quality degradation (a lower limit at which image quality degradation is most severe). An object of the present invention is to provide a digital watermark embedding technique capable of guaranteeing a certain level of image quality.

  Further, in the conventional digital watermark technology, when extracting (detecting) watermark information from a watermarked image, some reference image information used at the time of embedding is often required. An object of the present invention is to provide a digital watermark embedding technique capable of extracting watermark information without preparation.

  First, an outline of the digital watermark embedding method of the present invention will be described. In the present invention, an index image (an arbitrary image different from the original image) that defines the type of image quality deterioration due to watermark embedding and the allowable range of the deterioration degree is specified in advance. The index image is an image that is composed of pixels that are paired with each pixel of the original image and is defined as an image of the limit of image quality deterioration. When watermark information is embedded by moving the pixel value of the original image in the color space, In this image, the pixel value of the original image is changed so as to approach the pixel value of the index image.

In the following description, “^” is added to the symbol and parameter of the index image paired with the symbol and parameter used for the original image.
In the present invention, for the pixels g (m, n) and g ^ (m, n) at the respective positions (m, n) of the original image and the index image, the pixel values of the respective pixels are indicated on the Lab color space. Take points P _mn (L _mn , a _mn , b _mn ), P _mn (L _mn , a _mn , b _mn ). Here, the “Lab color space” indicates the entire color space using color information (pixel values) L, a, and b as parameters. Specifically, there are L ^* a ^* b ^* color system color space, Luv color system color space, XYZ color system color space, CIE-RGB color system color space, sRGB color system color space, etc. (A typical color space in which the present invention is used is an L ^* a ^* b ^* color system color space, and is expressed as a Lab color space similar to this).

In the present invention, pixel norms D (m, n) and D ^ (m, n), which are the distances between the points P _mn and P ^ _mn of the color space and the origin, are calculated, and an operation using this is performed.
The original image is divided into block units B (i, j) that are units for embedding watermark bit information, and further divided into sub-block units Bs (k, l) so that the influence of the embedding process can be adjusted. The total pixel norm W (k, l) is calculated in units. Further, an average W ⁻ of the sum of pixel norms for the sub-blocks included in the block is calculated. Subsequently, the average W ⁻ of the sum of the pixel norms is quantized with a quantization value Q that gives embedding strength, and this is set as a “watermark evaluation value W ⁻ _Q ”. An operation for embedding bit information (watermark) is performed on the watermark evaluation value W ^- _Q to obtain an adjustment value. The adjustment value is one of W ⁻ _Q , W ⁻ _Q + 1, and W ⁻ _Q− 1. Which value is obtained depends on whether the bit information to be embedded is even or odd, and W ⁻ _Q + 1, W in relation to the pixel norms D (m, n) and D ^ (m, n) of all pixels in the block. ^- _Q allowable change amount of it is calculated to determine whether beneficial to approximate the index image to be adjusted to any of -1 Σ _+, Σ _- (details will be described later) is determined in accordance with classification case of.
Then, the “total change amount K of pixel norm” after embedding bit information and before embedding is calculated, and the total change amount K of pixel norm is distributed to each pixel as “change amount K (m, n) for each pixel”. To do.
The pixel norm D (m, n) is changed by pixel on a straight line from P _mn (L _mn , a _mn , b _mn ) to P ^ _mn (L _mn , a _mn , b _mn ) in the color space. A point P ′ _mn (L _mn , a _mn , b _mn ) on the color space that becomes the pixel norm D ′ (m, n) changed by K (m, n) is calculated.
A pixel value represented as a point P ′ _mn (L _mn , a _mn , b _mn ) on the color space is set as the pixel value of the watermarked image.
Performed in units of blocks of the above procedure, on the Lab color space, _P a point of the original image _{mn (L mn, a mn,} b mn) from the index image point _{P ^ mn (L mn, a} mn, b mn) A watermark-embedded image having the color information (pixel value) as a point P ′ _mn (L _mn , a _mn , b _mn ) on the straight line directed to is obtained.

That is, the digital watermark embedding method of the present invention made to solve the above problems has the following configuration.
An electronic watermark embedding method for embedding watermark information in an original image, an index image setting step for setting an index image different from the original image, and a unit for embedding the original image and the index image in 1-bit watermark information A block / subblock dividing step of dividing each block B (i, j), B ^ (i, j) and dividing each block into subblocks Bs (k, l), Bs ^ (k, l); In the Lab color space in which color information is specified by pixel values L, a, and b for pixels g (m, n) and g ^ (m, n) at each position (m, n) of the original image and the index image Pixel norms D (m, n), D ^ (D (), which are the distances between the points P _mn (L _mn , a _mn , b _mn ), P _mn (L _mn , a _mn , b _mn ) and the origin of the color space a pixel norm calculation step for calculating m, n); A difference ΔD (m, n) between pixel norms D (m, n) and D ^ (m, n) of corresponding pixels g (m, n) and g ^ (m, n) in the original image and the index image is obtained. A pixel norm difference calculating step to calculate, a sum Σ ₊ of pixel norm differences of pixels having a positive pixel norm difference ΔD (m, n) in the block, and a negative pixel norm difference ΔD (m, n) in the block Huh a, it is possible to change the original image respectively allowance the sum of pixel norm in blocks when close to the index image can be changed to the positive side, the negative side _- the sum Σ of pixels norm difference of pixels having A change allowance calculating step for calculating as a capacity, a pixel norm sum calculating step for calculating a sum W (k, l) of pixel norms D (m, n) in units of sub-blocks, and for each sub-block included in the block Pixel norm sum W (k , The average embedding previous pixel norm total average W a l) ^- Quantum using the quantization value Q give embedding strength against ^- and pixel norm sum average calculation step of calculating a pixel norm total average W before embedding used as it is as the adjustment value _Q ^- a quantization step of calculating a _Q, watermark evaluation value W ^{- -} phased watermark evaluation value W watermark evaluation value W when the parity of _Q of even-odd and the watermark bit information matches In addition, when the even-oddness of the watermark evaluation value W ^- _Q does not match the even-oddity of the bit information of the digital watermark, the watermark evaluation value W ^- _Q is set according to the magnitude relationship between the positive and negative pixel norm difference sums Σ +, Σ−. +1 increase or -1 Gensa value was used as the adjustment value, the adjustment value and a quantized value Q and based on watermarking after the pixel norm total average W ^- 'average calculated after implantation pixel norm sum to calculate the A step, pixel norm total average W after implantation ^- 'and Embedding previous pixel norm total average W ^- the total change amount calculation step of calculating a total amount of change in pixel norm requiring a difference in the embedded bit information of the K, the A pixel-specific change amount K (m, n) that changes the pixel norm D (m, n) in accordance with the positive / negative of the total change amount K and the positive / negative of the pixel norm difference ΔD (m, n) for each pixel g (m, n). ) And a point P _mn (L _mn , a _mn , b _mn ) of the pixel g (m, n) of the original image in the color space is _converted into a pixel g ^ (m, n) a pixel that is moved on a straight line toward a point P ^ _mn (L _mn , a _mn , b _mn ), and a pixel norm D (m, n) is changed by a pixel-specific change amount K (m, n). norm D '(m, n) point in the color space which is a _{P' mn (L mn, a} mn, b mn) Pixel g '(m, n) after implantation by calculating so that and a color information calculating step after implantation of calculating color information.

According to the present invention, a change in pixel value (change in the position of a point in the color space) when watermark information is embedded in an original image can be changed so as to approach a pixel value of a preset index image. Because it can, you can understand how to change in color space and the limit of change.
Therefore, using various images such as a JPEG compressed image, a blurred image, an edge enhancement image, and a filter image as an index image, a change format (image quality degradation format) can be determined so that a watermarked image approaches the index image. The limit of image quality deterioration can be specified so that the index image does not exceed the index image even when the maximum change occurs.

In addition, when extracting watermark information from a watermarked image, the average W ⁻ of the pixel norm sum W (k, l) is calculated for the watermarked image, and even / odd when quantized with the quantization value Q is extracted. By doing so, the watermark information can be obtained, so that the watermark information can be extracted without preparing image information for reference such as an index image.

It is a figure which shows the procedure of the electronic watermark embedding method which is one Embodiment of this invention. It is a figure which shows an example of an original image and a parameter | index image. It is a figure for showing block division and subblock division about an original picture and an index picture. It is the figure which showed the movement (change of a pixel value) of the point in L ^* a ^* b ^* color space. It is a figure which shows the relationship between the watermark evaluation value and adjustment value when embedding bit information (watermark).

  Hereinafter, embodiments of the present invention will be described. First, a color system used in the present invention will be described, and then a digital watermark embedding method, extraction method, and error reduction processing will be described.

(Color system)
The color system includes an XYZ color system related to the color stimulus value, a CIE-RGB color system, an sRGB color system that is an RGB value displayed on the monitor screen, a YCbCr color system used for JPEG compression, There are various color systems such as the L ^* u ^* v ^* color system and the L ^* a ^* b ^* color system which are uniform color spaces, and they are used in applications suitable for their respective characteristics.

  In the present invention, regardless of which color system is used as the color system used when embedding watermark information, the original image is changed so as to approach the pixel value of the index image in the color space of the color system. It is possible to grasp how to change in the color space of the color system and the limit of the change.

In particular, when the L ^* a ^* b ^* color system, which is one of uniform color spaces adapted to human vision, is applied, color information can be changed in line with vision. In the following description, L ^* a ^* b ^* color system is used.
Since the color systems can perform data conversion with each other via, for example, the XYZ color system, if only a little complicated coordinate conversion is performed, the watermark embedding process (color information changing process) that can be executed in one color system is It can be executed with other color systems.

In general, the sRGB color system is used as a pixel value when an image is displayed on a monitor screen. Therefore, in order to associate the sRGB color system with the L ^* a ^* b ^* color system, data conversion is performed via the XYZ color system and the CIE-RGB color system. The conversion formula at this time is shown below.

A conversion formula between the CIE-RGB color system and the XYZ color system is shown in (1).
R _γ , G _γ , and B _γ are CIE-RGB values, which are different from RGB (sRGB values) stored as pixel values in a digital image. A conversion formula between sRGB and CIE-RGB is shown in (2).
Here, R ′, G ′, and B ′ are values obtained by dividing each value of sRGB by 255 and correcting it to a range of [0.0, 1.0].

Next, (3) shows a conversion formula from the XYZ coordinate system to the L ^* a ^* b ^* color space.

Under the standard light D65 light source, Xn = 95.045, Yn = 100, Zn = 108,892. The L ^* component is lightness, the a ^* component is chromaticity coordinates from green to red, and the b ^* component is chromaticity coordinates from yellow to blue.

Characterized in the L ^* a ^* b ^* color space, when a distance on the color space and between the color _{C 1} and the color _C distance and color _{C 3} in the color space between the ₂ and the color _{C 4} is equal, the color _{C 1} The color difference between the color C ₂ and the color C ₂ and the color difference between the color C ₃ and the color C ₄ can be perceived by humans.

(How to embed digital watermark)
Next, a digital watermark embedding method according to the present invention will be described with reference to the drawings. FIG. 1 is a flowchart showing the procedure of the digital watermark embedding method of the present invention.

As shown in FIG. 2, an original image 1 and an index image 1 ^ having a width W and a height H used for watermark embedding are prepared (St1).
The index image 1 ^ may be an image that is different from the original image 1 and that can associate a pair of pixels with each pixel of the original image 1. For example, a JPEG compressed image obtained by processing the original image 1, a blurred image, an edge enhancement image, a filter image, or the like may be used. In an extreme case, a subject image different from the original image may be used as the index image.

As shown in FIG. 3, the original image 1 and the index image 1 ^ are divided into L × L pixel blocks, and each block is further divided into Ls × Ls pixel sub-blocks (St2).
The i-th block in the horizontal direction and the j-th block in the vertical direction in the original image 1 and the index image 1 ^ are B (i, j) and B ^ (i, j), and B (i, j) and B ^ (i , J) are denoted by Bs (k, l) and Bs ^ (k, l), respectively, in the horizontal direction and the l-th sub-block in the vertical direction.
here,
0 ≦ i <W / L, 0 ≦ j <H / L,
0 ≦ k <L / Ls, 0 ≦ l <L / Ls
(In FIG. 3, 0 ≦ i <4, 0 ≦ j <5, 0 ≦ k <2, 0 ≦ l <2).

Bit information (0 or 1) embedded in a certain block B (i, j) is defined as b _ij . The following procedure is used to embed 1-bit information in one block.

A pixel at a position (m, n) (0 ≦ m <L, 0 ≦ n <L) in B (i, j) is g (m, n), and L represents a pixel value of g (m, n). A point on the ^* a ^* b ^* color space is a point P _mn (L ^* _mn , a ^* _mn , b ^* _mn ) (see FIG. 4). Similarly, a point on the L ^* a ^* b ^* color space of a pixel g ^ (m, n) in B ^ (i, j) is defined as P ^ _mn (L ^* _mn , a ^* _mn , b ^* _mn ). .
The distance between the origin O and the point P _{mn in the} color space is D (m, n), the distance between the origin O and the point P ^ _mn is D ^ (m, n), and these are the pixel norms D (m, n). n) and D ^ (m, n).
The pixel norms D (m, n) and D ^ (m, n) of each pixel g (m, n) are calculated using the following equation (4) (St3).

For each pixel, a difference ΔD (m, n) between the pixel norm D (m, n) and the pixel norm D ^ (m, n) is calculated using Expression (5) (St4).
ΔD (m, n) is referred to as “pixel norm difference”.

Then, on a block basis, each pixel in the block is divided into a pixel group having a positive pixel norm difference ΔD (m, n) and a pixel group having a negative pixel norm difference ΔD (m, n). Then, using equation (6), the sum Σ _{+ of} the positive pixel norm difference ΔD (m, n) corresponding to the former pixel group and the negative pixel norm difference ΔD (m, n) corresponding to the latter pixel group the sum of) Σ _- to calculate the (St5).
The sigma ₊ "positive pixel norm difference sum", sigma _- referred to as "negative pixel norm difference sum".
Σ ₊ and Σ ₋ can be changed to an allowable amount that can change the sum of pixel norms to the positive side and to the negative side in units of blocks when the pixel value of the original image approaches the pixel value of the index image. Allowable amount. For this reason, Σ ₊ and Σ ₋ are collectively referred to as allowable change amounts. That is, the sum of the pixel norms can be moved within the range of positive and negative change allowances Σ _{− to} Σ ₊ . As will be described in detail later, Σ ₊ and Σ ₋ are used as criteria for determining whether to operate the “watermark evaluation value” on the positive side or the negative side when performing an operation of embedding bit information. .

Separately from the calculation of the positive and negative pixel norm difference sums Σ ₊ and Σ ₋ , the sum W of D (m, n) in the sub-block Bs (k, l) is calculated by the equation (7) (St6). .
The sum W (k, l) of D (m, n) in the sub-block Bs (k, l) is referred to as “pixel norm sum”.

Then, the average value W of the pixel norm summation W contained in the block (k, l) ^- a, is calculated using the following equation (8) (St7).
W ^{− is referred} to as “pixel norm sum average before embedding”.
The pixel norm sum average W ⁻ before embedding is used for calculation between the pixel norm sum average W ⁻ ′ after embedding.

Next, the pixel norm sum average W ⁻ before embedding is quantized by dividing by the quantized value Q and rounding off (St8). Specifically, the calculation according to the equation (9) is performed. The magnitude of the quantized value Q gives the watermark embedding strength. Increasing Q can increase the embedding strength. However, (in relation to the block division and the sub block division) Too much larger Q, points on the pixel value after embedding ^L * ^a * ^{b *} color space _{^{P mn (L * mn, a}} * mn, b ^* _mn ) and the point P ^ _mn (L ^* _mn , a ^* _mn , b ^* _mn ) cannot be present, so an appropriate Q is selected.
The quantized pixel norm sum average W ^- _Q is called a “watermark evaluation value” because bit information is embedded in the next process.

Subsequently, in the watermark evaluation value W ^- _Q , the watermark evaluation value after the bit information b (b is bit information b _ij of the block (i, j) and becomes 0 or 1) is embedded in the watermark evaluation value W ^- _Q. The corresponding “adjustment value” is calculated. Furthermore, using this adjustment value and the quantized value Q, a value at the same level as the pixel norm sum average before embedding is calculated (St9).
Specifically, as shown in FIG. 5, the watermark evaluation value W ^- _Q is an even number (2Q, 4Q,...) For the block with b = 0, and the watermark evaluation value W for the block with b = 1. ^The adjustment value is calculated by performing an operation of adjusting the watermark evaluation value W ^- _{Q according} to the case classification described below so that _Q is an odd number (1Q, 3Q,...). Further, this adjustment value is multiplied by the quantized value Q as shown in the equations (10) and (11), and returned to the value W ⁻ ′ at the same level as the pixel norm sum average W ⁻ . W ⁻ ′ is referred to as “pixel norm sum average after embedding”.

First, W ^- _Q ≡b when mod 2 ((9) at the calculated watermark evaluation value W ^- when parity of _Q matches the bit information b to be embedded) is directly W ^- adjustment value the value of _Q And

On the other hand, W - Q ≠ b when mod 2 ((9) in computed watermark evaluation value W - when parity of Q does not coincide with the bit information b to be embedded) is, W - Q +1 or,, W - Let Q -1 be the adjustment value. W - Q + 1, W - is one or take the Q -1 sign of the allowable change amount sigma + absolute value, sigma - determined by the magnitude of the absolute value of.
That is,
| Σ + | ≧ | Σ - | When the, - the adjustment value (W Q +1).
| Σ + | <| Σ - | When the, - the adjustment value (W Q -1).
The positive and negative permissible change amounts Σ + and Σ − can be changed to the negative side, which is a permissible amount that can change the sum of the pixel norms to the positive side when the pixel value of the original image approaches the pixel value of the index image. Since this is an allowable amount, by setting the adjustment value as described above, it is possible to select the one with the larger room for change. The large room for change means that the embedding strength can be set more flexibly.

Subsequently, a difference K between the post-embedding pixel norm sum average W ⁻ ′ and the pre-embedding pixel norm sum average W ⁻ is calculated (St10).
Since the difference K is a value corresponding to the total change amount of the pixel norm necessary for embedding bit information, it is called “the total change amount of pixel norm”.

Subsequently, the pixel norm total change amount K calculated by Expression (12) is distributed and changed to the pixel norm of each pixel g (m, n) in the block (St11). The pixel norm total change amount K is not distributed evenly to all the pixels in the block, but is classified according to the sign of K. When K is positive, the pixel norm difference ΔD (m, n) is positive. The total amount of change K is proportionally distributed (ΔD / Σ ₊ ) to the group of g (m, n), and the amount of change K (m, n) obtained by doubling the number of sub-blocks is distributed. When K is negative, the total variation K is proportionally distributed (ΔD / Σ ₋ ) to the group of pixels g (m, n) in which the pixel norm difference ΔD (m, n) is negative, and the sub-block The amount of change K (m, n) that is doubled by the number is distributed. This change amount K (m, n) is referred to as “a change amount for each pixel”.
Specifically, the pixel-by-pixel change amount K (m, n) for changing D (m, n) of each pixel in the block is calculated by equations (13) and (14).
Thus, when K is positive, only D (m, n) of the pixel group in which ΔD is positive is changed, and when K is negative, only D (m, n) of the pixel group in which ΔD is negative is changed. By changing, only the pixel whose pixel value when the bit information is embedded approaches the pixel value of the index image in the color space is changed and the bit information is embedded.

Next, as shown in FIG. 4, in the color space, from the point P _mn (L _mn ^* , a ^* _mn , b ^* _mn ) of the pixel g (m, n) of the original image, the pixel g ^ (m , N), the point is moved on a straight line toward the point P ^ _mn (L ^* _mn , a ^* _mn , b ^* _mn ), and the pixel norm D (m, n) is changed per pixel K (m, n). ), A point P ′ _mn (L _mn ^* , a ^* _mn , b ^* _mn ) on the color space that becomes the pixel norm D ′ (m, n) changed by () is calculated (St12).

The point P ′ _mn (L _mn ^* , a ^* _mn , b ^* _mn ) after the movement moves to a point that satisfies Expression (15).

Here, t is the ratio closer to _{P mn} to _{P ^ mn,} it satisfies the equation (16), and is a value satisfying 0 ≦ t ≦ 1.

Therefore, the point P ′ _mn on the L ^* a ^* b ^* color space of the watermarked image in which bit information (watermark) is embedded is changed from the point P _{mn of the} original image to the point P of the index image as shown in FIG. ^ On the straight line toward _mn , the pixel norm D (m, n) comes to the position where it is changed to D '(m, n), and the method of change in color space and the limit of change are grasped. it can.
Further, since the color information is changed in the L ^* a ^* b ^* color space, which is a uniform color space, the color information changes in accordance with human perception, and the embedded image can be predicted. Further, since the image quality degradation at the maximum change is P ′ _mn , it is possible to guarantee the image quality degradation.

(Watermark information extraction method)
Next, a process of extracting bit information from a watermarked image in which bit information (watermark) is embedded will be described.

As with embedding, the watermarked image is divided into blocks and sub-blocks. Below, the process which extracts bit information _bij from block B (i, j) is _demonstrated .

First, for the pixel g (m, n) in the block B (i, j), a point P _mn (L _mn ^* , a ^* _mn , b ^* _mn ) on the L ^* a ^* b ^* color space representing the pixel value. Get.

  Subsequently, the pixel norm sum W (k, l) in the sub-block is calculated using the equations (4) and (7).

Subsequently, an average value W ⁻ of the pixel norm sum W (k, l) included in the block is calculated using Expression (8).
Subsequently, the quantized pixel norm sum average W ^- _Q is calculated using Equation (9).
At this time, if the value of W ⁻ _Q is an even number, b _ij = 0 is extracted as bit information, and if the value of W ⁻ _Q is an odd number, b _ij = 1 is extracted as bit information.
All the watermark information embedded in the image is extracted by executing the above processing for all the blocks.

  Therefore, extraction of bit information does not require a reference image such as an index image, and extraction is possible if only information on the quantization value Q used at the time of embedding is given.

(Error reduction processing)
As described above, the present invention can be used by using various color spaces (explained as Lab spaces) including the L ^* a ^* b ^* color space. Therefore, when a color other than the sRGB color space is used, it is necessary to convert the component value such as the L ^* a ^* b ^* color space into the component value of the sRGB color space and further convert them into integer values. For this reason, an error occurs between the L ^* a ^* b ^* component value calculated from the RGB value of the watermarked image at the time of extraction and the L ^* a ^* b ^* component value immediately after embedding the watermark.

Therefore, it is desirable to add an error reduction process in order to increase the extraction rate of watermark information. Conventionally, when converting component values between different color spaces and converting them to integer values, all pixels are rounded to integer values.
However, if there is no linear relationship between the color spaces before and after conversion, rounding off cannot remove the error, but it may promote the error. Therefore, in the present invention, the accuracy of watermark information extraction is improved by performing error reduction processing according to the following procedure.

L ^* a ^* b ^* component values L ^* _rs ′, a ^* _rs ′, b ^* of the pixel at the position (r, s) (0 ≦ r <W, 0 ≦ s <H) where the watermark bit information is embedded _The component values obtained by converting _rs ′ to the sRGB color system are R ^to _rs , G ^to _rs , and B ^to _rs .

Round up each component value to the nearest decimal point. That is, X _rs ^u , X _rs ^d (Xε {R, G, B}) is calculated using the equation (17).

Subsequently, the distance d _i (pixel) from the origin calculated through the conversion from R _rs ^p , G _rs ^p , B _rs ^p (pε {u, d}) to the L ^* a ^* b ^* system color space Norm) is calculated for all p combinations.
That is, the distance d _i values from the eight origins are calculated using equation (18).

Here, f (x, y, z) represents a distance from the origin when x, y, z is converted into the L ^* a ^* b ^* color space.

Then, using Equation (19), calculates the absolute value Delta] E _i of the difference between the distance and d _i values from the origin pixel position (r, s).

  Here, 0 ≦ i ≦ 7.

From the eight ΔE _i obtained by the equation (19), the minimum value ΔE _min is obtained by the equation (20).

X _rs ^p (X∈ {R, G, B}, p∈ {u, d}) used to calculate ΔE _min is expressed as R, G, and R of the pixel at the position (r, s) of the watermarked image. B value is determined.

  In this way, the error can be minimized by selecting one closest to the distance from the origin from eight combinations obtained by rounding up and down instead of rounding off.

(Modification)
The embodiment of the present invention has been described above. However, the present invention is not limited to this, and modifications can be made without departing from the spirit of the invention. Therefore, modifications will be described.

  In the digital watermark embedding method described above, the block division is divided into 4 × 5 blocks, and one block is divided into 2 × 2 sub-blocks. Even if the sub-block is a minimum of 1 × 1 sub-blocks (that is, the whole block is one sub-block), the calculation can be performed theoretically, and this case is also included in the present invention.

In the above-described embodiment, the L ^* a ^* b ^* color space is used. However, when the present invention is applied to another color space, the point of the original image is directed to the point of the index image on the color space. Since a point of a watermarked image can be represented on a line, a change in the image on the color space can be predicted (a monitor image can be predicted if converted to the sRGB color system). .

  The present invention can be used as a technique for embedding a digital watermark in image data.

1 Original image 1 ^ Index image

Claims (3)

An electronic watermark embedding method for embedding watermark information in an original image,
An index image setting step for setting an index image that is different from the original image, and
The original image and the index image are divided into blocks B (i, j) and B ^ (i, j), which are units for embedding 1-bit watermark information, and subblocks Bs (k, l), Block / sub-block dividing step of dividing into Bs ^ (k, l);
On the Lab color space in which color information is designated by the pixel values L, a, and b for the pixels g (m, n) and g ^ (m, n) at each position (m, n) of the original image and the index image. Pixel norms D (m, n), D ^ (D (), which are distances between the points P _mn (L _mn , a _mn , b _mn ), P _mn (L _mn , a _mn , b _mn ) and the origin of the color space. a pixel norm calculation step for calculating m, n);
A difference ΔD (m, n) between pixel norms D (m, n) and D ^ (m, n) of corresponding pixels g (m, n) and g ^ (m, n) in the original image and the index image is obtained. A pixel norm difference calculation step to calculate;
Sum of pixel norm differences of pixels having positive pixel norm difference ΔD (m, n) in the block Σ ₊ , Sum of pixel norm differences of pixels having negative pixel norm difference ΔD (m, n) in the block Σ _- a tolerance, the allowable change amount calculation step of calculating a permissible amount which can be changed to the negative side, which can alter the summation of the pixels norms positive side in blocks when the respective approximate the original image as an index image When,
A pixel norm sum calculating step of calculating a sum W (k, l) of pixel norms D (m, n) in sub-block units;
A pixel norm sum average calculating step of calculating an average of pixel norm sums W (k, l) of each sub-block included in the block as a pixel norm sum average W ⁻ before embedding;
A quantization step of calculating a _Q, ^- embedding the previous pixel norm total average W ^- quantized using the quantization value Q that gives the embedding strength to watermark evaluation value W
Watermark evaluation value W ^- _Q of even-odd and watermark evaluation value when the the even-odd bit information of the digital watermark matches W ^- _Q used directly as adjustment values, also watermark evaluation value W ^- _Q bits of even-odd and watermark When the even / odd information does not match, the value obtained by increasing the watermark evaluation value W ⁻ _Q by +1 or −1 according to the magnitude relationship between the positive and negative pixel norm difference sums Σ ₊ and Σ ₋ is used as the adjustment value. A post-embedding pixel norm sum average calculating step of calculating a pixel norm sum average W ⁻ ′ after watermark embedding based on the adjustment value and the quantized value Q;
The total change amount calculation step of calculating a total amount of change in pixel norm requiring a difference in the embedded bit information with K, ^- pixel norm total average W after implantation ^- 'and Embedding previous pixel norm total average W
A pixel-specific change amount K (m, n) that changes the pixel norm D (m, n) in accordance with the positive / negative of the total change amount K and the positive / negative of the pixel norm difference ΔD (m, n) for each pixel g (m, n). n) a pixel-by-pixel change amount calculating step for calculating n);
Point _P mn of pixels g of the original image on the color space _{(m, n) (L mn} , a mn, b mn) pixels of the index image g ^ (m, n) of a point _P ^ _{mn (L mn,} a _mn , b _mn ), and a color space that becomes a pixel norm D ′ (m, n) obtained by moving the pixel norm D (m, n) by a pixel-specific change amount K (m, n). And a post-embedding color information calculation step of calculating color information of the pixel g ′ (m, n) after embedding by calculating the upper point P ′ _mn (L _mn , a _mn , b _mn ). A digital watermark embedding method as a feature. The L ^* a ^* b ^* color system color space is used as the Lab color space, and the calculated point P ′ _mn (L ^* _mn , a ^* _mn , b ^* _mn ) on the L ^* a ^* b ^* color space The digital watermark embedding method according to claim 1, wherein the image is output by converting the color information into color information of an sRGB color system color space. Integer value conversion when converting from L ^* a ^* b ^* color system color space to color information of the sRGB color system color space is performed as P ′ _mn (L ^* _mn , a ^* _mn , b ^* on the color space) ^. _mn ) is converted into the sRGB color system color space, and eight R, G, B integer value groups are obtained by rounding up and rounding off the decimals of the three component values R, G, B. ,
For each integer value group, the distance d from the origin when converted back to the L ^* a ^* b ^* color system color space is obtained, and the point P ′ _mn (L ^* _mn , a ^* _mn , b ^* _mn ) The digital watermark embedding method according to claim 2, wherein an absolute difference ΔE with respect to a distance from the origin is calculated, and R, G, B of an integer value group that minimizes ΔE is used as an RGB value of the watermarked image.

Images