JP3910987B2 - Digital watermark embedding method and apparatus - Google Patents

Description

  The present invention relates to a digital watermark embedding device that embeds watermark information in digital data converted into contents such as voice, music, moving images, still images, and the like, and a digital watermark detection device that detects watermark information from embedded content.

  Digital watermarking is necessary for identification of copyright holders and users such as voice, music, video and still images converted into digital data, copyright holder's rights information, usage conditions for the data, and use Information such as confidential information and copy control information (these are called watermark information) is embedded so that it is not easy to perceive, and usage control is performed by detecting the watermark information from within the data as necessary. This technology protects copyright including copy control and promotes secondary use.

[Digital watermark classification]
Conventionally, digital watermarking schemes for images can be broadly divided into a pixel domain utilization type and a frequency domain utilization type. The digital watermarking method using a pixel area directly embeds watermark information by changing pixel values. On the other hand, the digital watermarking method using the frequency domain is one that moves from the pixel domain to the frequency domain by orthogonal transformation, and after embedding in the frequency domain, returns to the pixel domain from the frequency domain again by inverse orthogonal transformation. is there. The watermark information is embedded as a wave.

[Digital watermarking using frequency domain]
As a digital watermarking method using frequency domain, for example, reference [1] Koch, E. and Zhao, J., “Towards Robust and Hidden Image Copyright Labeling,” Proceedings of 1995 IEEE Workshop on Nonlinear Signal and Image Processing, 452 -455, June 20-22, 1995.) [2] Cox, IJ, Kilian, J., Leighton, T. and Shamoon, T., “Secure Spread Spectrum Watermarking for Multimedia,” NEC Research Institute, Technical There is a method described in Report 95-10, 1995. In these methods, the frequency component to be embedded is set from a low frequency to an intermediate frequency that is less affected by irreversible compression.

[Digital watermarking using pixel area]
As a digital watermarking method using a pixel area, there is a method of embedding by changing the LSB of a pixel value. The change is performed by determining whether to add or reduce a certain amount of variation according to a pseudo-random number sequence (PN sequence) or according to a mask pattern prepared in advance.

[Digital watermarking by spread spectrum]
Spread spectrum refers to a communication method in which information is widely distributed and transmitted in a sufficiently large band as compared with a band necessary for a signal to be communicated. Excellent resistance to noise on the transmission line. By considering the original content as a carrier wave, watermark information as a desired wave, and the influence of irreversible compression as interference waves (noise), the spread spectrum concept is applied to digital watermark technology to improve robustness to irreversible compression, etc. There is a method.

As a digital watermarking method using spread spectrum, pixel region diffusion (Ref. [3] Smith, JR and Comiskey, BO, “Modulation and Information Hiding in Images,” Information Hiding, 207-226, 1996. and Reference [4] Onishi, Kazuhiro Oka and Kokoo Matsui, “Watermark signatures for images using PN sequences,” SCIS 97, 26B, 1997.) and spreading in the frequency domain (see the previous document [2]) have been proposed. The Cox method and Onishi method are sometimes referred to as a perturbation method and a direct sequence spread spectrum, respectively.
[Diffusion in the frequency domain (perturbation method)]
In the method of document [2], watermark information is embedded by orthogonally transforming pixel values, and watermark information is diffused and embedded in the frequency domain. Spreading is performed by changing the values of a plurality of frequency components according to a random number sequence in the frequency domain. After spreading, inverse orthogonal transform is performed. The watermark information is detected by performing orthogonal transform on the pixel value, and determining the correlation value between the embedded frequency component value and the random number sequence used for embedding. Since the embedded watermark information is distributed over the entire image (block) in the pixel area, it is robust to various operations. Further, if the frequency component in which the watermark information is embedded is in the low intermediate frequency region, the watermark information is not easily lost even by the low frequency pass filter.

[Diffusion in the pixel area (direct diffusion method)]
On the other hand, in the method of document [4], embedding of watermark information is directly diffused by multiplying a pixel value by a PN (pseudo-random noise) sequence. Orthogonal transformation is performed on the obtained image, watermark information is embedded in the frequency domain, and inverse orthogonal transformation is performed again. Then, despreading is performed by multiplying the pixel value by the same PN sequence. In the detection of watermark information, pixel values are directly diffused by a PN sequence. An orthogonal transform is performed on the obtained image, and determination is made from the value of the frequency component into which watermark information is embedded.

[Digital watermark detection]
In the digital watermark method using spread spectrum, the digital watermark is extracted by evaluating a correlation value between the pseudo random number sequence used in embedding and the value of the pixel or frequency component of the image to be detected.

For example, in the method of document [2], the frequency component value represented by the following equation is obtained from the pixel value f (x, y).

  From this frequency component value, a predetermined number (n) of frequency components having a large absolute value are selected from the frequency components excluding the DC component, and according to a pseudo random number sequence Wi according to N (0, 1), Embedding is performed by changing to F ′ (ui, vi) = F (ui, vi) × (1 + αwi) and then returning to the pixel value by inverse orthogonal transformation.

On the other hand, when detecting a digital watermark, a frequency component F ^* (u _i , v _i ) is obtained from an image to be detected, and whether an amount called similarity expressed by the following formula exceeds a certain threshold value: It is determined whether or not a digital watermark is embedded depending on whether or not.

Also, in the method of document [4], in embedding, after multiplying a pixel value f (x, y) by a pseudo random number sequence p (x, y) having a value of {−1, 1}, Find the frequency component of

Here, m is a parameter representing the strength of embedding, and its code represents 1 and 0 of bits of watermark information.
Then, the direct current component is embedded by changing the following equation.

On the other hand, when detecting a digital watermark, the frequency component F ^* (u _i , v _i ) is obtained from the image to be detected, and the watermark information is determined to be 1, 0 by determining the sign of the following equation.

  In the two conventional examples described above, 1-bit information indicating whether a digital watermark is embedded or whether 0 or 1 is embedded is determined.

In the literature [3], a plurality of mutually orthogonal basis functions φ _i (x, y) are prepared. These basis functions are pseudo-random numbers. The embedding is performed according to the following formula.

On the other hand, detection is performed by obtaining a value of the following equation and comparing it with a certain threshold value.

Here, f ^* (x, y) is a pixel value of the detection target image. In this method, as many as the number of bases to be prepared can be embedded. However, it was necessary to prepare as many pseudorandom functions assured orthogonality.
In the previous proposal by the inventors (Reference [5] Hirofumi Muratani, Taku Kato, Naoki Endo, “Digital watermark embedding device and digital watermark detection device”, Japanese Patent Application No. 10-340019, 1998.) Based on the method, the embedding is performed for the frequency component F (u _i , v _i ) for each bit b _{i of the} watermark information with respect to the frequency component of the following equation obtained by orthogonal transform of the image after multiplication of the pseudo random number sequence. The watermark information is expressed by changing the value.

  Compared with the method of Reference [3], this method automatically satisfies the orthogonality of the basis from the property of the basis of the orthogonal transformation, regardless of the property of the pseudorandom number sequence, and the method of Reference [4]. Compared with the method, it has a feature that there is a method in which the number of bits is increased and the cost of orthogonal transformation is omitted.

[Multi-bit digital watermark]
In actual application of digital watermarking, it is necessary to embed multi-bit information such as right holder information, user information, and usage control information. In response to this requirement, there has been considered a method of diverting a digital watermark method for embedding 1-bit watermark information to a method for embedding multi-bit watermark information.

  The first method is a method in which an embedding target image is divided into a plurality of blocks, and watermark information is embedded in each block bit by bit. However, this method has a drawback that, when a part of an embedding target image is cut out, information on bits embedded in a block in that part is accidentally lost.

  The second method is to use the space of the pseudo random number sequence. In this method, n pseudo-random number sequences having small correlations are prepared, each bit of watermark information is assigned to each of them, and they are embedded so as to be superimposed.

  In the case of the space domain utilization type, the fully spread version described in the document [3] and the embedding for a plurality of frequency components in the document [5] are examples of the second method. In the case of using the frequency domain, for example, reference [6] Herrigel, Alexander, Nasir Memon and Minerva M. Yeung, “Secure Copyright Protection Techniques for Digital Images,” Second Information Hiding Workshop, 169-190, 1998. Is disclosed. According to this second method, even if a part of the image is lost, there is a possibility that any bit of the watermark information can be restored from the other part. However, in the method of document [6], instead of preparing a plurality of pseudo-random number sequences, one pseudo-random number sequence is shifted and used, and encoding is performed using the offset value. In this method, the pseudo-random number sequence needs to have no autocorrelation.

[Discrete value superposition problem]
When the second method described above is applied to the frequency domain as in the document [6], the frequency component value is expressed by a continuous value (decimal number), so that each piece of watermark information as shown in FIG. Even if the basis function corresponding to the bit is superimposed and then proper normalization is performed in consideration of the influence on the image quality as shown in FIG. 10B, the shape of the superimposed function is maintained. There is no fear that the watermark information will be lost greatly.

  On the other hand, when the same method is applied to the spatial region (pixel region) as in [3], after the basis function waveform corresponding to each bit of the watermark information is superimposed as shown in FIG. In consideration of the influence on the image quality, normalization is performed as shown in FIG. 11 (b), and when re-discretization is performed as shown in FIG. 11 (c), the shape of the superimposed function waveform is obtained. Is greatly impaired, and watermark information may be largely lost due to rounding errors. If normalization is not performed in order to leave the superimposed function waveform shape, the effect on the image becomes large.

[Determination and extraction of digital watermark]
In the digital watermark technology, there are processes of detection and extraction in a digital watermark detection apparatus. Determination is processing for determining whether or not a digital watermark is embedded in the content. On the other hand, extraction is a process of extracting and reading embedded watermark information.

  For this reason, there is a method of embedding watermark information for determination (information specifying the presence or absence of a digital watermark) in the content in addition to the original watermark information to be extracted. The latter watermark information identifies whether the extracted watermark information is correct watermark information or whether it is accidentally read as watermark information even though the watermark information is not actually embedded. Because. However, embedding two types of different watermark information has a problem of increasing the influence on the quality of content.

  Further, in a digital watermark detection apparatus that extracts watermark information after determining whether or not a digital watermark is embedded and determining that the digital watermark is embedded, extraction processing is performed after waiting for a determination process. There is a delay in the processing time for performing. This problem can be solved by performing the extraction process in parallel without waiting for the determination process, but there is a problem that the hardware cost for parallelization increases.

  As described above, in the conventional digital watermark technique, in the method of normalizing and re-discretizing by superimposing the basis function waveform corresponding to each bit of the watermark information, the watermark information may be lost due to a rounding error.

  Accordingly, an object of the present invention is to provide a digital watermark embedding method and apparatus in which watermark information is not easily lost when embedding watermark information consisting of a plurality of bits.

  In order to solve the above-described problem, the digital watermark embedding apparatus according to the present invention is provided with a plurality of basis function waveforms prepared corresponding to each bit of a plurality of bits of watermark information to be embedded in the embedding target content. Basis function waveform synthesizing means for synthesizing after normalizing according to the value of the corresponding bit and normalizing and outputting the synthesized waveform; and discretizing means for discretizing the synthesized waveform and outputting a discretized waveform; Superimposing means for superimposing the discretized waveform on the embedding target content.

  Here, the discretization means separates the amplitude value of the input composite waveform into an integer part and a decimal part, and a value that takes 1 or 0 according to the appearance frequency that is predetermined according to the value of the decimal part and the integer part. Are added to output a discretized waveform.

  According to this digital watermark embedding device, since the fractional part of the amplitude value of the composite waveform is reflected in the amplitude value of the discretized waveform, the influence of the rounding error, which has been a problem in the conventional re-discretization, can be reduced. This solves the problem of loss of watermark information.

  Furthermore, according to the present invention, there is provided a storage medium that stores content embedded with watermark information by the above-described digital watermark embedding apparatus.

  According to the present invention, a plurality of basis function waveforms prepared corresponding to each bit of watermark information of a plurality of bits to be embedded with respect to the embedding target content are synthesized after imparting polarity according to the value of the corresponding bit. In addition, for the synthesized waveform obtained by normalization, re-discretization is performed by performing re-discretization so that the fractional part of the amplitude value of the synthesized waveform is reflected in the amplitude value of the discretized waveform and superimposing it on the embedding target content. Since the fractional part of the amplitude value of the combined waveform is reflected in the amplitude value of the digitized waveform, it is possible to prevent the disappearance of the watermark information due to the rounding error that has been a problem in the conventional re-discretization.

Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 is a conceptual diagram of a content distribution system to which a digital watermark embedding device 1 and a digital watermark detection device 2 of the present invention are applied.
Content to be embedded, such as images and sounds, and watermark information (multi-bit watermark information) consisting of a plurality of bits are input to the digital watermark embedding device 1, and the embedded content obtained here includes a storage medium for storing the content It distributes via the route 3. The digital watermark detection apparatus 2 detects multi-bit watermark information.

Embodiments of a digital watermark embedding device and a digital watermark detection device according to the present invention will be described below.
(First embodiment)
In the present embodiment, a digital watermark embedding device and a digital watermark detection device are described in which, when embedding watermark information consisting of a plurality of bits, even if a part of an embedding target image is cut off, the watermark information is not easily lost.

First, the digital watermark embedding apparatus according to the first embodiment of the present invention will be described with reference to FIG.
An embedding target image 10 is input to the digital watermark embedding apparatus. The embedding target image 10 includes a plurality of portions each including four small regions written as (1,1) component, (2,1) component, (1,2) component, and (2,2) component, respectively. Are divided into regions 11A, 11B,..., 11M, 11N,.

  For example, the small area has a size of, for example, 4 pixels × 4 pixels, but is not limited thereto, and may be 1 pixel × 1 pixel or may be larger. The partial areas 11A, 11B,..., 11M, 11N,... Include four small areas in total, two in the horizontal direction and in the vertical direction, but may include a larger number of small areas. For example, the shape may be an elongated shape such that there are two small areas in the horizontal direction and one in the vertical direction.

  The embedding target image 10 includes a (1,1) component extraction unit 12-1, a (2,1) component extraction unit 12-2, a (1,2) component extraction unit 12-3, and a (2,2) component extraction unit. 12-4, a (1,1) component set 13-1, a (2,1) component set 13-2 selected from the partial areas 11A, 11B,..., 11M, 11N,. 1,2) component set 13-4 and (2,2) component set 13-4 are extracted. That is, in each of the component extraction units 12-1, 12-2, 12-3, and 12-4, as in the block division of the first method described in the section of [Multi-digitization of digital watermark] in the conventional technique, The components of the image data are extracted from the small areas distributed over the entire embedding target image 10, not from the area close to the area.

  Further, in each of the component extraction units 12-1, 12-2, 12-3, and 12-4, a small relative position in each partial region is set from the partial regions 11A, 11B,..., 11M, 11N,. Although the component of the area is extracted, the component of the small area having a different relative position in the partial area may be extracted. For example, the (1,1) component of the partial region 11A, the (2,1) component of the partial region 11B, the (1,2) component of the partial region 11M, and the (2,2) component of the partial region 11N are combined into one. Extracted by the component extraction unit, the (2,1) component of the partial region 11A, the (1,1) component of the partial region 11B, the (2,2) component of the partial region 11M, and the (1,2) component of the partial region 11N Are extracted by another component extraction unit as a set.

  The division of the embedding target image 10 and the extraction by the component extraction units 12-1, 12-2, 12-3, and 12-4 are actually buffers that temporarily store the data of the embedding target image 10. This can be realized by an operation for controlling reading from the memory (frame memory).

  The respective sets 13-1, 13-2, 13-3 of the (1,1) component, (2,1) component, (1,2) component, and (2,2) component extracted in this way 13-4 is input to the first basis function superimposing unit 14-1, the second basis function superimposing unit 14-2, the third basis function superimposing unit 14-3, and the fourth basis function superimposing unit 14-4, respectively. Bit watermark information, in this example, a basis function waveform whose polarity changes in accordance with each value of the first bit, the second bit, the third bit, and the fourth bit of the watermark information 15 consisting of 4 bits is superimposed. The

  However, in some cases, the embedded image 20 is subjected to operations such as irreversible compression and D-A-D (Digital-Analog-Digital) conversion after embedding, and is processed into data slightly different from that immediately after embedding. There may be.

  After the basis function waveforms are superimposed in this manner, the (1,1) component set 16-1, the (2,1) component set 16-2, the (1,2) component set 16-3, (2,2 ) The component set 16-4 is input to the reconstructing unit 17 and synthesized so as to be located in a region corresponding to each of the images, thereby having the same content as the embedding target image 10 and the watermark information 15 Embedded in the embedded image 18 is reconstructed.

Next, a digital watermark detection apparatus corresponding to the digital watermark embedding apparatus of FIG. 2 according to the present embodiment will be described with reference to FIG.
The embedded image 20 is input to the digital watermark detection apparatus. The embedded image 20 is the same image as the embedded image 18 output from the digital watermark embedding apparatus of FIG. 1, and each includes a (1,1) component, a (2,1) component, and a (1,2) It is divided into a plurality of partial areas each including four small areas written as a component and (2,2) component.

  The embedded image 20 includes a (1,1) component extraction unit 22-1, a (2,1) component extraction unit 22-2, a (1,2) component extraction unit 22-3, and a (2,2) component extraction unit. 22-4, a (1,1) component set 23-1, a (2,1) component set 23-2, and a (1,2) component set 23-3 selected from each partial region, The (2,2) component set 23-4 is extracted.

  The division of the embedded image 20 as described above and the extraction by the component extraction units 22-1, 22-2, 22-3, and 22-4 are actually buffers that temporarily store the data of the embedded image 20. This can be realized by an operation for controlling reading from the memory (frame memory).

  The (1,1) component set 23-1, the (2,1) component set 23-2, the (1,2) component set 23-3, and the (2,2) component extracted in this way The set 23-4 is assigned to the first basis function correlation determining unit 24-1, the second basis function correlation determining unit 24-2, the third basis function correlation determining unit 24-3, and the fourth basis function correlation determining unit 24-4. The information of the first bit, the second bit, the third bit, and the fourth bit of the 4-bit watermark information 25, which is multi-bit watermark information, is respectively input.

  That is, in the basis function correlation determination units 24-1, 24-2, 24-3, and 24-4, the input sets 23-1, 23-2, 23-3, and 23-4 and watermark information are input. Embedded by determining the correlation with the first, second, third, and fourth basis function waveforms prepared corresponding to the first bit, second bit, third bit, and fourth bit, respectively. The watermark information 25 is output by determining “1”, “0” of the first, second, third, and fourth bits of the watermark information.

According to the configuration of the present embodiment described above, the problems of the conventional digital watermark multi-bit technology can be solved.
First, in the first method described in the section of [Multi-bit digital watermarking], the embedding target image is divided into a plurality of blocks, and watermark information is embedded in each block one bit. When a part of the image is lost, for example, by cutting the completed image, the bits of the watermark information embedded in the block included in the lost part are lost, and the watermark information cannot be detected correctly.

  On the other hand, in the present embodiment, the basis function waveform corresponding to each bit of the watermark information is embedded in a small area dispersed throughout the embedding target image 10, so that the embedded image 18 is lost due to partial clipping or the like. Even in this case, it is possible to easily reproduce the watermark information from the information of the basis function waveform remaining in other portions.

  Further, in the conventional second method described in [Multi-biting of digital watermark], watermark information embedding by the fully spread version described in the document [3] applied to the spatial domain is shown in FIG. As explained, after superimposing the basis function waveform corresponding to each bit of the watermark information on the same pixel, normalization and re-discretization, the shape of the basis function waveform superimposed by the rounding error is greatly impaired, There was a risk that the watermark information would be lost.

  On the other hand, in this embodiment, the basis function waveform corresponding to each bit of the watermark information is distributed in different small areas, so that a plurality of basis function waveforms are formed on the same pixel as in the case of embedding watermark information by a fully spread version. The normalization and re-discretization processes that are necessary in accordance with the superposition are not necessary. Therefore, there is no problem that the watermark information is lost due to a rounding error caused by re-discretization.

(Second Embodiment)
Next, as a second embodiment of the present invention, when embedding a plurality of bits of watermark information as in the first embodiment, the watermark information is lost even when a part of the embedding target image is cut off. A digital watermark embedding apparatus and a digital watermark detection apparatus which are difficult to be described will be described.
First, the digital watermark embedding apparatus according to the present embodiment will be described with reference to FIG.
The digital watermark embedding apparatus receives an embedding target image 30 and watermark information 31 including a plurality of bits. The watermark information 31 is input to the basis function waveform synthesizer 32, and a plurality of basis function waveforms prepared corresponding to each bit of the watermark information 31 are superimposed after the polarity corresponding to the value of the corresponding bit is given ( Synthesized) and normalization is performed, so that a synthesized waveform 33 is generated.

  That is, in the basis function waveform synthesizer 32, the basis function waveform prepared corresponding to each bit of the watermark information 31 corresponds to the value of the bit corresponding to the watermark information 31, for example, the value of the bit is “1”. If it is “0”, the polarity (sign) is inverted and then superimposed on the basis function waveform corresponding to the value of other bits, and further normalized in consideration of the effect on image quality. As a result, a composite waveform 33 is output.

  The synthesized waveform 33 output from the basis function waveform synthesizing unit 32 is input to the probabilistic discretizing unit 34, and is re-discretized so that the decimal point is reflected in the output as described later. A normalized waveform 35 is output. The discretized waveform 35 output from the stochastic discretizer 34 is superimposed on the embedding target image 30. Unlike the conventional discretization unit that performs re-discretization, the stochastic discretization unit 34 is configured to reduce the influence of rounding errors, and its specific configuration is shown in FIG.

  In the stochastic discretization unit 34 shown in FIG. 5, the composite waveform 33 from the basis function waveform synthesis unit 32 is input to the fraction part separation unit 41, and each amplitude value of the composite waveform 33 is separated into an integer part and a fraction part. Is done. The integer part is directly input to the adder 44, and the decimal part is input to the adder 44 via the comparator 43. The comparison unit 43 compares the decimal part with a given decimal number sequence, in this example, a uniform random number that takes a value within the range [0, 1] output from the random number generation unit 42, and A value that takes 0 or 1 according to the magnitude relationship is output.

  Then, the output of the comparison unit 43 and the integer part are added by the addition unit 44 and become the output of the stochastic discretization unit 34. By performing such processing over the entire synthesized waveform 33, that is, for each pixel, a discretized waveform 35 is obtained.

  FIG. 6 shows an example of the synthesized waveform 33 and the discretized waveform 35. As a result of the above-described processing of the stochastic discretization unit 34, the fractional part of the amplitude value of the composite waveform 33 is reflected in the amplitude value of the discretization waveform 35. Therefore, this occurs in the conventional re-discretization as described in FIG. There is no rounding error due to truncation of the fractional part, and the problem of loss of watermark information due to this is solved.

  In the above example, the output is determined by the comparison unit 43 using the random number generated by the random number generation unit 42. Instead, for example, [0, 1] is equally divided into m sections, When the decimal part takes a value within the interval of [k / m, (k + 1) / m], the output of the comparison unit 43 becomes “1” at a frequency of (k + 1) times (m + 1) times, otherwise It may be set to “0”. In that case, a round robin circuit may be used instead of the random number generator 42.

  As described above, according to the present embodiment, the basis function waveform synthesis unit 32 superimposes a plurality of basis function waveforms prepared corresponding to each bit of the watermark information 31 after the polarity corresponding to the value of the corresponding bit is given. Then, by re-discretizing the synthesized waveform 33 obtained by normalization by the stochastic discretization unit 34 having the configuration shown in FIG. 5, no rounding error occurs in the re-discretization. Therefore, it is possible to avoid the problem that the watermark information embedded in the embedded image 37 obtained by the waveform superimposing unit 36 is largely lost due to a rounding error.

(Third embodiment)
Next, as a third embodiment of the present invention, a digital watermark embedding device and a digital watermark detection device for embedding watermark information having both functions of digital watermark determination and extraction with little influence on contents will be described.

First, the digital watermark embedding apparatus according to the present embodiment will be described with reference to FIG.
The digital watermark embedding device receives an embedding target image 50, 1-bit detection watermark information a, and extraction watermark information including a plurality of (n) bits b1, b2,. Each of the bits b1, b2,..., Bn of the detection watermark information a and the extraction watermark information is information having a value of 1 or 0, and the watermark information conversion unit 51 uses these for determination and extraction. After being converted into each bit (ternary value) B1, B2,..., Bn of n-bit watermark information reflecting both of the watermark information, the first basis function superimposing unit 52-1, the second basis function superimposing unit 52 are converted. -2, the third basis function superimposing unit 52-3, ..., and the nth basis function superimposing unit 52-n.

  On the other hand, the embedding target image 50 is input to the first component extraction unit 53-1, the second component extraction unit 53-2,..., The nth component extraction unit 53-n, and these component extraction units 53-1, 53-. 2,..., 53-n, information on components (regions) on which the first, second,. The extracted components are input to the basis function superimposing units 52-1, 52-2,..., 52-n, and the first, second,.

  The components after the basis function waveform is superimposed by the basis function superimposing units 52-1, 52-2,..., 52-n are synthesized by the synthesis unit 54, reconstructed as one embedded image 55, and output. Is done.

Next, the processing of the watermark information conversion unit 51 will be described using the flowchart shown in FIG.
The watermark information conversion unit 51 first checks whether the value of the detection watermark information a is 1 or 0 (step S1). When a = 1, that is, when a digital watermark is to be embedded, each bit bi ( It is checked whether the value of i = 1, 2,..., n) is 1 or 0 (step S2). As a result, if bi = 1, Bi = 1 is set (step S4), and if bi = 0, Bi = -1 is set (step S5). On the other hand, if a = 0 in step S1, that is, if no digital watermark is embedded, Bi = 0 is set (step S3).

  In the basis function superimposing units 52-1, 52-2,..., 52-n, the embedding target image 50 is converted according to the values (Bi) of the corresponding bits B 1, B 2,. The superposition of the basis function waveform to the corresponding component is controlled. That is, in the basis function superimposing units 52-1, 52-2,..., 52-n, when Bi = 0, the basis function waveform is not superimposed on the corresponding component of the embedding target image 50, and Bi = 1. In this case, the basis function waveform is superimposed with the same polarity, and when Bi = -1, the basis function waveform is superimposed after the polarity is inverted.

  Therefore, the embedded image 55 output from the synthesizing unit 54 is a watermark composed of a basis function superimposed waveform reflecting both information of the detection watermark information a and the extraction watermark information (b1, b2,..., Bn). Information will be embedded.

  As described above, in the digital watermark embedding device of this embodiment, the watermark information conversion unit 51 generates watermark information reflecting both the detection watermark information and the extraction watermark information of a plurality of bits, and the watermark information after the conversion Is used as a basis function waveform to be superimposed on each component extracted from the embedding target image 50, thereby creating an embedded image 55, which is compared with the conventional method of embedding individual watermark information for determination and for extraction. Thus, the influence on the quality of the embedded image 55 can be minimized.

  In this embodiment, when the basis function waveform corresponding to each bit of the watermark information is embedded in the embedding target image 50, it may be embedded in the same region as in the above-described fully spread version, or different blocks may be embedded. Alternatively, it may be embedded in tiles, or may be embedded in a set of spatially dispersed small regions as in the first embodiment.

  Furthermore, in the digital watermarking method using the spread spectrum using the space domain or the frequency domain, the pseudo random number sequence to be used may be divided into n partial pseudo random number sequences, and the bits of the watermark information may be associated with each. I do not care.

  Next, the digital watermark detection apparatus according to the present embodiment will be described with reference to FIG. This digital watermark detection apparatus is suitable for the embedded image obtained by the digital watermark embedding apparatus of FIG. 8, but is not limited to this, and the embedded image in which multi-bit watermark information is embedded. If so, it can be basically applied.

  The embedded image 60 is input to the digital watermark detection apparatus of FIG. This embedded image 60 is, for example, an image similar to the embedded image 18 output from the digital watermark embedding device of FIG. 1, and includes detection watermark information a and extraction watermark information (b1, b2,..., Bn). Watermark information reflecting both pieces of information is embedded as a basis function superimposed waveform.

  The embedded image 60 is input to the first component extraction unit 61-1, the second component extraction unit 61-2, ..., the nth component extraction unit 61-n, and these component extraction units 61-1 and 61- 2,... 61-n, information on the first, second,. The components extracted by the component extraction units 61-1, 61-2,..., 61-n are the first correlation value calculation unit 63-1, the second correlation value calculation unit 63-2,. The first, second,..., Nth correlation values ci (i), which are respectively input to the calculation unit 63-n and are correlation values with the first basis function waveform, the second basis function waveform,. = 1, 2,..., N) are obtained by calculation.

The first, second,..., N-th correlation values c i thus obtained are the total correlation calculation unit 64, the first bit extraction unit 66-1, the second bit extraction unit 66-2,. 66-n. The overall correlation calculation unit 64 calculates a correlation value (overall correlation value) with the basis function waveform of the entire embedded image from the first, second,..., Nth correlation values ci. The overall correlation value C is calculated as the sum of absolute values of the first, second,..., Nth correlation values ci as shown in the following equation, for example.

  The calculated overall correlation value C is input to the digital watermark determination unit 65, where it is determined whether or not there is a digital watermark, that is, whether or not watermark information is embedded. This determination can be made based on whether or not the overall correlation value C exceeds a certain threshold value T. For example, if C> T, it is determined that the watermark information is embedded, otherwise it is not embedded.

  On the other hand, in the first bit extraction unit 66-1, the second bit extraction unit 66-2,..., The nth bit extraction unit 66-n, for example, by comparing each correlation value ci with a predetermined threshold value ti, Each bit of the watermark information for extraction is determined. For example, if ci> ti, the corresponding bit of the extraction watermark information is determined to be “1”, and if ci <ti, it is determined to be “0”. The first bit extraction unit 66-1, the second bit extraction unit 66-2,..., The nth bit extraction unit 66-n output these determination results as each bit bi of the watermark information (extraction watermark information). Is done.

  Here, the extraction watermark information extraction processing described above in the first bit extraction unit 66-1, the second bit extraction unit 66-2,..., The nth bit extraction unit 66-n is performed by the digital watermark detection unit 65. It is desirable to carry out according to the detection result of the presence or absence of the digital watermark. That is, the extraction watermark information is extracted only when the digital watermark detection unit 65 determines that the watermark information is embedded. Thereby, useless processing can be omitted.

  As described above, in the digital watermark detection apparatus of the present embodiment, the presence / absence of digital watermarks by the overall correlation determination unit 64 and the digital watermark detection unit 65 and the bit extraction units 66-1, 66-2,. The processing of extracting (bit determination) a plurality of bits of watermark information according to is performed based on the correlation values obtained by the correlation value calculation units 63-1, 63-2,..., 63-n. Compared with the method to be performed, the processing cost is low, and the processing delay is also reduced. In addition, the processing speed can be improved while avoiding the problem of an increase in hardware due to parallel processing.

  Furthermore, it is generally required that the determination of whether or not the watermark information is embedded has fewer errors than the extraction of the embedded watermark information (bit determination). The correlation value of many basis function waveforms is accumulated by the above determination to obtain the overall correlation value, and it is determined whether or not the watermark information is embedded from the overall correlation value. Can do.

  In each of the above-described embodiments, the case where the content is an image (particularly a still image) has been described. However, the present invention can be similarly applied to content such as a moving image and audio, and similar effects are expected. be able to. For example, when the first embodiment is applied to speech that is a one-dimensional signal, the speech waveform is divided into, for example, frames in the time axis direction, and each frame is divided into small regions such as subframes. The same processing may be performed.

  The first and second embodiments are basically digital watermarks using a spatial area (pixel area) of content, but the third embodiment also applies to digital watermarks using the frequency domain of content. Applicable.

The figure which shows schematic structure of the content distribution system containing the digital watermark embedding apparatus and detection apparatus which concern on this invention 1 is a block diagram showing a configuration of a digital watermark embedding device according to a first embodiment of the present invention. 1 is a block diagram showing a configuration of a digital watermark detection apparatus according to a first embodiment of the present invention. The block diagram which shows the structure of the digital watermark embedding apparatus based on the 2nd Embodiment of this invention. The block diagram which shows the structure of the stochastic discretization part in FIG. Waveform diagram explaining the operation of the stochastic discretization unit The block diagram which shows the structure of the digital watermark embedding apparatus based on the 3rd Embodiment of this invention. Flowchart for explaining processing of the watermark information conversion unit in FIG. The block diagram which shows the structure of the digital watermark detection apparatus which concerns on the 3rd Embodiment of this invention. The figure which shows the mode of the superimposition of the basis function waveform with a decimal value for demonstrating a prior art The figure which shows the mode of the superimposition of the basis function waveform with a discrete value for demonstrating another prior art

Explanation of symbols

10 ... Embedding target images 11A, 11B, ..., 11M, 11N ... Partial regions 12-1, 12-2, 12-3, 12-4 ... Small region component extraction units 14-1, 14-2, 14-3, 14-4: Basis function superimposing unit 15 ... Watermark information 17 ... Reconstructing unit 18 ... Embedded image 20 ... Embedded image 22-1, 22-2, 22-3, 22-4 ... Small region component extracting unit 24- 1, 24-2, 24-3, 24-4 ... basis function correlation determination unit 25 ... watermark information 30 ... embedding target image 31 ... watermark information 32 ... basis function superimposed waveform synthesis unit 34 ... probabilistic discretization unit 36 ... waveform Superimposition unit 37 ... Embedded image 41 ... Decimal part separation unit 42 ... Random number generation unit 43 ... Comparison unit 44 ... Addition unit 50 ... Embedding target image 51 ... Watermark information conversion unit 52-1, 52-2, ..., 52-n ... Basis function weight 53-3, 53-2,..., 53-n... Component extraction unit 54... Synthesis unit 55. Embedded image 60. Embedded image 61-1, 61-2,. -1, 63-2, ..., 63-n ... correlation value calculation unit 64 ... overall correlation calculation unit 65 ... digital watermark detection unit 66-1, 66-2, ..., 66-n ... watermark information extraction unit

Claims (4)

In a digital watermark embedding apparatus that embeds a plurality of bits of watermark information into an embedding target content and obtains embedded content,
A plurality of basis function waveforms prepared corresponding to each bit of the watermark information are superimposed after being given a polarity according to the value of the corresponding bit, and normalized to output a composite waveform;
An amplitude value of the composite waveform is separated into an integer part and a decimal part, and a value that takes 1 or 0 according to an appearance frequency determined in advance according to the value of the decimal part and an integer part are added to the composite waveform. Discretizing and outputting a discretized waveform;
And a step of superimposing the discretized waveform on the content to be embedded. In a digital watermark embedding apparatus that embeds a plurality of bits of watermark information into an embedding target content and obtains embedded content,
Basis function waveform synthesizing means that superimposes a plurality of basis function waveforms prepared corresponding to each bit of the watermark information after applying a polarity according to the value of the corresponding bit and normalizes and outputs a synthesized waveform When,
An amplitude value of the composite waveform is separated into an integer part and a decimal part, and a value that takes 1 or 0 according to an appearance frequency determined in advance according to the value of the decimal part and an integer part are added to the composite waveform. A discretization means for discretizing and outputting a discretized waveform;
An electronic watermark embedding apparatus comprising: a superimposing unit that superimposes the discretized waveform on the embedding target content.   A storage medium storing content embedded with watermark information by the digital watermark embedding method according to claim 1.   A storage medium storing content embedded with watermark information by the digital watermark embedding apparatus according to claim 2.

Images