JPWO2006027823A1 - Digital watermarking device, authentication processing device, decoding processing device, and requantization device - Google Patents

Abstract

An image signal is quantized, an authenticator is calculated from the quantization index of the image signal, and a signature circuit 1 for embedding the authenticator in the quantized index is provided, while a difference signal between the image signal and the reference signal is provided. An encoder 2 is provided that quantizes the difference signal according to the quantization index in which the authenticator is embedded by the signature circuit 1 and outputs the quantization index of the difference signal.

Description

  The present invention provides, for example, a digital watermark device that embeds a digital watermark in multimedia data such as an image signal and an audio signal, an authentication processing device that authenticates multimedia data in which the digital watermark is embedded, and a multi-media in which the digital watermark is embedded. The present invention relates to a decoding processing device that decodes media data and a requantization device that requantizes multimedia data in which a digital watermark is embedded.

As a general method for authenticating that digital data has been properly issued and has not been tampered with thereafter, there is a method such as an electronic signature.
The principle of a method for implementing an electronic signature is that an issuer creates an authentication code by encrypting a hash value of digital data and transmits the authentication code together with the digital data.
On the receiving side, the authentication code sent from the issuer is decrypted, the hash value of the digital data is recalculated, and the authenticity of the digital data is confirmed by comparing the hash value with the authentication code. To do.

When digital data is image data or audio data, many digital watermarking apparatuses have been proposed in which the authentication code is securely combined with the digital data by embedding the authentication code as digital watermark in the digital data ( For example, refer nonpatent literature 1.).
In Non-Patent Document 1, a hash value is calculated from a DCT (Discrete Cosine Transform) coefficient for moving image data encoded by MPEG (Motion Pictures Experts Group), and the hash value is embedded as a digital watermark. A method is disclosed. According to this method, if the value of the DCT coefficient in the encoded data changes even a little, it is possible to detect the change on the receiving side.
Note that in the digital watermark device disclosed in Non-Patent Document 1, in order to reduce the amount of code, in the moving image, only the difference is transmitted using the property that temporally adjacent images are similar to each other. Predictive encoding is performed.

R. Du and J.J. Fridrich, “Lossless authentication of MPEG-2 video”, Proceedings of IEEE International Conference on Image Processing, September 2002.

  Since the conventional digital watermarking device is configured as described above, when the amount of code is reduced by transmitting a difference signal between adjacent moving image data, moving image data is obtained from the difference signal on the receiving side. Once decoded, the difference signal cannot be reproduced. For this reason, the hash value embedded in the difference signal also disappears simultaneously with the decoding of the moving image data, and there is a problem that the authentication function is lost.

The present invention has been made in order to solve the above-described problems. An electronic watermarking apparatus capable of providing multimedia data that does not lose the authentication function even when multimedia data such as images and sounds is decoded. The purpose is to obtain.
The present invention also provides an authentication processing device for authenticating multimedia data provided from the digital watermark device, a decoding processing device for decoding the multimedia data, and requantization for requantizing the multimedia data. The object is to obtain a device.

  The digital watermarking apparatus according to the present invention includes a signature unit that quantizes an input signal, calculates an authenticator from the quantized index of the input signal, and embeds the authenticator in the quantized index. There is provided encoding means for obtaining a difference signal between the signal and the reference signal, quantizing the difference signal according to the quantization index in which the authenticator is embedded by the signature means, and outputting the quantization index of the difference signal. It is a thing.

  Thus, there is an effect that it is possible to provide multimedia data that does not lose the authentication function even if multimedia data such as images and sounds are decoded.

It is a block diagram which shows the digital watermark apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows an example which quantizes a kth conversion coefficient. It is a flowchart which shows the processing content of the correction circuit in a signature circuit. It is a block diagram which shows the electronic watermarking apparatus by Embodiment 2 of this invention. It is a block diagram which shows the authentication processing apparatus by Embodiment 3 of this invention. It is a block diagram which shows the inside of a coefficient acquisition part. It is a block diagram which shows the inside of a coefficient acquisition part. It is a block diagram which shows the decoding processing apparatus by Embodiment 4 of this invention. It is a block diagram which shows the inside of a decoding circuit. It is a flowchart which shows the processing content of a correction circuit. It is a block diagram which shows the requantization apparatus by Embodiment 5 of this invention. It is a block diagram which shows the inside of a coefficient acquisition circuit.

Hereinafter, in order to describe the present invention in more detail, the best mode for carrying out the present invention will be described with reference to the accompanying drawings.
Embodiment 1 FIG.
FIG. 1 is a block diagram showing a digital watermarking apparatus according to Embodiment 1 of the present invention. In the figure, a signature circuit 1 quantizes an image signal as an input signal, and an authentication code is obtained from the quantization index of the image signal. The calculation is performed, and the authentication code is embedded in the quantization index. The signature circuit 1 constitutes a signature means.
The encoder 2 obtains a difference signal between the image signal and the reference signal, quantizes the difference signal in accordance with the quantization index in which the authenticator is embedded by the signature circuit 1, and outputs the quantization index of the difference signal. The encoder 2 constitutes an encoding means.

The converter 11 of the encoder 2 converts the input image signal X (where X is a vector of the image signal) into a frequency domain signal T (X) in units of blocks.
The subtracter 12 of the encoder 2 is a reference image signal Xp (Xp is a vector of the reference image signal) encoded in the past from the signal T (X) in the frequency domain that is the conversion result of the converter 11. The frequency domain signal T (Xp) is subtracted to obtain the frequency domain prediction error T (X) -T (Xp), which is the difference signal.

The quantizer 13 of the signature circuit 1 quantizes the frequency domain signal T (X) output from the converter 11 and outputs a quantization index Qw (T (X)) of the signal T (X).
The correction circuit 14 of the signature circuit 1 corrects the quantization index Qw (T (X)) output from the quantizer 13, and uses the corrected quantization index Qw (T (X)) as an authenticator operation circuit 15 and Output to the modulator 16.
The authenticator calculation circuit 15 of the signature circuit 1 calculates an authenticator W (where W is a vector of authenticators) from the quantized index Qw (T (X)) corrected by the correction circuit 14.
The modulator 16 of the signature circuit 1 modulates a part of the corrected quantization index Qw (T (X)) according to the authenticator W calculated by the authenticator arithmetic circuit 15 and outputs the modulation result Yw. .

The quantizer 17 of the encoder 2 is a frequency domain prediction error T () calculated by the subtractor 12 according to the modulation result Yw of the modulator 16 (quantization index in which the authenticator W is embedded by the modulator 16). X) -T (Xp) is quantized and a quantization index Y of the prediction error T (X) -T (Xp) (where Y is a vector of quantization indexes) is output.
The inverse quantizer 18 of the encoder 2 inversely quantizes the prediction error quantization index Y output from the quantizer 17 to decode the prediction errors T (X) -T (Xp).
The adder 19 of the encoder 2 receives the prediction error T (X) −T (Xp) decoded by the inverse quantizer 18 and the frequency domain signal T (Xp) of the reference image signal Xp output from the switch 23. Addition is performed and the addition signal T (X) is output.

The inverse converter 20 of the encoder 2 converts the addition signal T (X) output from the adder 19 into the time domain, and outputs the signal in the time domain as the reference image signal Xp.
The storage circuit 21 of the encoder 2 stores and outputs the reference image signal Xp output from the inverse converter 20 for the moving time.
The converter 22 of the encoder 2 converts the reference image signal Xp output from the storage circuit 21 into the frequency domain, and outputs a signal T (Xp) in the frequency domain.
The switch 23 of the encoder 2 supplies the frequency domain signal T (Xp) output from the converter 22 to the subtractor 12 and the adder 19. However, when encoding a moving image, it is necessary to appropriately mix images that can be decoded without referring to other images. In this case, “0” is sent to the subtracter 12 and the adder 19. give. At this time, the quantizer 17 operates so as to encode the current input image as it is.

Next, the operation will be described.
First, when the converter 11 of the encoder 2 receives the image signal X, the converter 11 converts the image signal X into a frequency domain in units of blocks and converts the signal T (X) in the frequency domain into the subtractor 12 and the signature circuit 1. To the quantizer 13.
Here, T () is an operator of the converter 11.

The subtracter 12 of the encoder 2 receives the frequency domain signal T (X) from the converter 11 and the frequency domain of the reference image signal Xp encoded in the past from the converter 22 via the switch 23. When the signal T (Xp) is received, the signal T (Xp) is subtracted from the signal T (X) in the frequency domain to obtain a frequency domain prediction error T (X) -T (Xp).
On the other hand, when the quantizer 13 of the signature circuit 1 receives the frequency domain signal T (X) from the converter 11, the quantizer 13 quantizes the frequency domain signal T (X). The index Qw (T (X)) is output.
Here, Qw () is an operator of the quantizer 13.

The correction circuit 14 of the signature circuit 1 corrects the quantization index Qw (T (X)) output from the quantizer 13 and uses the corrected quantization index Qw (T (X)) as an authenticator operation circuit 15. And output to the modulator 16.
Specific processing contents of the correction circuit 14 will be described later.

The authenticator calculation circuit 15 of the signature circuit 1 calculates the authenticator W from the quantization index Qw (T (X)) corrected by the correction circuit 14.
For the calculation of the authenticator, for example, the method described in Ito et al., “Digital Watermarking Method for Proving Authenticity of JPEG Images” (National Conference of the Institute of Electronics, Information and Communication Engineers, March 2003) may be used.
In this method, N transform blocks are grouped together, the hash values of all the quantization indexes are calculated, and the hash value encrypted is used as a signature. If the number of bits of the hash value is equal to N, a 1-bit authenticator is calculated for one conversion block.

When the authenticator operation circuit 15 calculates the authenticator W, the modulator 16 of the signature circuit 1 modulates a part of the corrected quantization index Qw (T (X)) according to the authenticator W, The modulation result Yw is output.
Yw = Qw (T (X)) + W (1)
Specifically, the modulator 16 embeds the authenticator W in the quantization index Qw (T (X)) using the method described in the above document.
In the method of the above document, in each transform block, the value of the last quantization index of the zigzag scan is set to “−1” or “+1” depending on whether the bit of the authenticator W is “0” or “1”. To change.

As a result, the quantization index in which the authenticator W of the input image is embedded is output from the modulator 16.
Here, the method of embedding a 1-bit signature in each conversion block has been described. However, if the value of N is set larger than the number of bits of the hash value, the digital watermark is preferentially given to the block that changes rapidly. Can be embedded.

The quantizer 17 of the encoder 2 receives the modulation result Yw from the modulator 16 and receives the frequency domain prediction error T (X) −T (Xp) from the subtractor 12. The frequency domain prediction error T (X) -T (Xp) is quantized according to the result Yw, and the quantization index Y of the prediction error T (X) -T (Xp) is output.
That is, the quantizer 17 outputs a quantization index Y that minimizes d in the equation (3) under the condition that the following equation (2) is satisfied.
Qw (Qc- ¹ (Y) + T (Xp)) = Yw (2)
d = | Qc ⁻¹ (Y) − (T (X) −T (Xp)) | (3)

However, Qc ⁻¹ () represents an inverse quantization operator for the quantizer 17, and || represents a vector norm.
In Expression (2), Qc ⁻¹ (Y) + T (Xp) is obtained by adding the inverse quantized prediction error signal to the reference frame conversion coefficient, and the conversion coefficient of the image decoded on the receiving side. Represents.
Equation (2) gives a condition that the transform coefficient to be decoded becomes Yw when quantized with Qw, and Equation (3) gives the difference T (X between the restored value of the prediction error and the true value. ) -T (Xp).
Accordingly, the quantizer 17 predicts the prediction error T (X) −T that minimizes the distortion caused by encoding under the condition that the modulation result Yw of the signature circuit 1 is stored in the decoded image signal. The quantization index Y of (Xp) is output.
The quantizer 17 may perform the above operation for each of the transform coefficients.

FIG. 2 is an explanatory diagram showing an example of quantizing the kth transform coefficient.
The upper side shows the Qw quantization characteristic, and the lower side shows the Qc quantization characteristic, which is represented by a quantization index and its quantization range.
A point on the central number line is a quantized representative value of Qc. Also, the subscript k on the right shoulder of the vector indicates the k-th component of the vector. q _c ^k and q _w ^k are the quantization widths of Qc and Qw with respect to this transform coefficient, and are set to satisfy Equation (4).
q _c ^k ≦ q _w ^k (4)

For example, if Yw ^k = i and T (X) ^k and T (Xp) ^k are at the positions in FIG. 2, the prediction error T (X) ^k −T (Xp) ^k is quantized by Qc. The quantization index is Y = 1.
However, the quantization representative value is not in the i-th quantization range of Qw, and Y = 1 does not satisfy the condition of Expression (2). Since the quantization index closest to T (X) ^k belonging to the i-th quantization range with the representative value Qw is Y = 2, this is the output of the quantizer 17.

Here, in order for Y to always exist, at least one quantization representative value of Qc must exist in all quantization ranges of Qw. As shown in FIG. 2, when Qc is uniform quantization, equation (4) may be established for all the coefficients k. However, in encoding such as MPEG, non-uniform quantization characteristics that are coarse near 0 are used. In this case, if the interval of the coarsest part of the quantized representative value of Qc is Tmax ^k, it is a sufficient condition that Y exists to satisfy the following expression (5).
q _w ^k > = Tmax ^k (5)

For example, H.M. In the case of a quantizer according to the H.263 coding standard, Qw is set as follows. H. If the quantization parameter for controlling the code amount is Q (= 1,..., 31), the inverse quantization operation of H.263 is defined by the following equation (6) if Q is an odd number, and Q is an even number. Then, it is prescribed | regulated by Formula (7).
y = (2Y + sign (Y)) Q (6)
y = (2Y + sign (Y)) Q-sign (Y) (7)
Here, y is a quantized representative value to be restored, and sign (Y) is a function that is −1 when Y <0, 0 when Y = 0, and 1 when Y> 0.

In this quantizer, since Tmax = 3Q (Q is an odd number) or Tmax = 3Q-1 (Q is an even number), Qw may be set as follows. That is, when Q is an odd number, the following equation (8) is set, and when Q is an even number, the following equation (9) may be set.
q _w ^k = 3Q (8)
q _w ^k = 3Q−1 (9)

The inverse quantizer 18 of the encoder 2 generates a reference image to be used for the next digital watermark embedding process when the quantizer 17 outputs the quantization index Y of the prediction error as described above. Then, the prediction error quantization index Y is inversely quantized to decode the prediction error T (X) -T (Xp).
When the inverse quantizer 18 decodes the prediction error T (X) -T (Xp), the adder 19 of the encoder 2 converts the prediction error T (X) -T (Xp) from the converter 22 to the switch 23. The signal T (Xp) in the frequency domain of the reference image signal Xp output via is added, and the added signal T (X) is output.

When receiving the addition signal T (X) from the adder 19, the inverse converter 20 of the encoder 2 converts the addition signal T (X) into the time domain, and uses the time domain signal as a reference image signal Xp. Store in the memory circuit 21.
The converter 22 of the encoder 2 reads the reference image signal Xp shifted by the moving time from the storage circuit 21, converts the reference image signal Xp into the frequency domain, and passes through the switch 23 in the frequency domain. The signal T (Xp) is supplied to the subtracter 12 and the adder 19.
Note that when the moving image is encoded, the switch 23 needs to appropriately mix images that can be decoded without referring to other images. In this case, “0” is subtracted from the subtractor 12. This is given to the adder 19. At this time, the quantizer 17 operates so as to encode the current input image as it is.

In the description of the signature circuit 1 described above, N transform blocks are grouped together and the hash values of all the quantization indexes are calculated. However, for the number N of blocks in which the hash values are calculated, When the number of bits M of the hash value is small (when M <N is satisfied), the modulator 16 may select M blocks from N and apply modulation only to those blocks.
Since there is a degree of freedom in this block selection method, it is possible to reduce image quality degradation due to embedding of the authenticator W by selecting a block that is as obscure as possible (a portion that is less affected by modulation) and applying modulation. it can. However, it must be possible to detect which block is modulated at the time of decoding.

  There are several such methods. When the above method for modulating the value of the last quantization index of the zigzag scan is used, for example, the number of non-zero quantization indexes (significant coefficients) is obtained, and the significance is obtained. A block having more than a predetermined number of coefficients may be selected. At this time, the number of significant coefficients increases by 1 due to the embedding of the digital watermark, but since the rank of the significant coefficients of each block does not change, the embedded block can be uniquely identified at the time of decoding. In addition, what is necessary is just to select arbitrary blocks for the block with the same significant coefficient. Since the number of significant coefficients is generally small in a flat portion of an image where deterioration is conspicuous, the probability of embedding a digital watermark is low. Therefore, it is possible to generate an encoded image in which interference due to embedding of a digital watermark is difficult to be perceived.

Finally, the processing content of the correction circuit 14 will be specifically described.
As described above, the authenticator operation circuit 15 calculates the authenticator W based on the currently input image signal X regardless of the reference image, and the quantizer 17 has the authenticator W embedded therein. Since the prediction error T (X) -T (Xp) is quantized so as to store the quantization index Yw, the signal output from the encoder 2 is a complete signature for authenticating the image signal in units of frames. Holds information. As will be described later, this signature information is extracted when a part of a frame is extracted from a moving picture and its conversion coefficient is re-encoded (for example, a part of a moving picture encoded by MPEG is used). In the case of re-encoding with JPEG), the re-encoded signal can be inherited without omission.

However, when decoding an image into a time domain signal, signature information may disappear due to errors introduced in the process of decoding. Such an error generally occurs when a transform coefficient is converted from a real number to an integer and when the integer value is further clipped between a certain dynamic range. Therefore, if sufficient countermeasures are not taken, signature information cannot be inherited by the time domain image.
The correction circuit 14 of the signature circuit 1 corrects the quantization index Qw (T (X)) so that the decoded signal does not greatly exceed the dynamic range.
This correction can be performed, for example, by a procedure as shown in FIG.

When receiving the quantization index Qw (T (X)) from the quantizer 13, the correction circuit 14 of the signature circuit 1 performs a process of embedding 0 in the quantization index Qw (T (X)) (step ST1). ).
Next, the correction circuit 14 determines that the embedded result of 0 does not greatly exceed the dynamic range (step ST2).
That is, the correction circuit 14 performs inverse quantization on the quantization index Yw in which the digital watermark is embedded corresponding to Qw, and inversely transforms the quantization index Yw to obtain a time domain vector x.
x = T ⁻¹ (Qw ⁻¹ (Qw (T (X)) + W)) (10)

The correction circuit 14 calculates a shortest distance d (x, A) between the vector x and the set A, where A is a set of vectors in which all components fall within the dynamic range, and the shortest distance d (x, A) is calculated. It is determined whether or not it is within a certain value D.
d (x, A) ≦ D (11)
If the shortest distance d (x, A) is within a certain value D, the correction circuit 14 recognizes that the embedded result of 0 does not greatly exceed the dynamic range.
When D = 0, the determination condition is satisfied only when x is included in A.

Next, when the correction circuit 14 determines that the embedded result of 0 does not greatly exceed the dynamic range, the correcting circuit 14 performs a process of embedding 1 in the quantization index Qw (T (X)) (step ST3).
Next, the correction circuit 14 determines that the result of 1 embedding does not greatly exceed the dynamic range (step ST4). Since this determination method is the same as the determination method in step ST2, description thereof is omitted.
If the correction circuit 14 recognizes that the result of 1 embedding does not greatly exceed the dynamic range, the quantization index Qw (T (T)) after embedding the authenticator W, which is a digital watermark, in the quantization index Qw (T (X)). Since (X)) is guaranteed to be close to the set A, the quantization index Qw (T (X)) is output to the authenticator operation circuit 15 and the modulator 16.

  However, when it is determined that the embedded result of 0 greatly exceeds the dynamic range, or when it is determined that the embedded result of 1 greatly exceeds the dynamic range, the quantization index Qw (T (X)) is an authentication that is a digital watermark. After the child W is embedded, it is not guaranteed that the quantization index Qw (T (X)) is close to the set A, so that the quantization index Qw (T (X)) is corrected (step ST5).

That is, the correction circuit 14 calculates a target vector t given by the following equation and stores the target vector t.
t = T (P (T- ¹ (Qw- ¹ (Qw (T (X)))))) (12)
However, P () is an operator representing an orthogonal projection with respect to the set A, and t is a vector obtained by inversely quantizing Qw (T (X)) and inversely transforming, and s is all vectors belonging to the set A. Among them, the vector closest to s is converted by T.

Next, the correction circuit 14 inversely quantizes the current quantization index Qw (T (X)) to calculate Qw ⁻¹ Qw (T (X)), compares this with the target vector t, The quantization index of Qw (T (X)) corresponding to the component farthest from the vector t is corrected by 1 in the direction of the target vector t.
With this modification, the quantization index Qw (T (X)) approaches the set A. If the quantization index Qw (T (X)) is already close enough to the target vector t, the target vector t is changed to a vector at the center of the set A (a vector in which all components are intermediate gray levels). By switching to the converted one, the modification of the quantization index Qw (T (X)) is repeated.

  As described above, the correction circuit 14 corrects the quantization index Qw (T (X)) so that the modulation result Yw of the modulator 16 does not greatly exceed the dynamic range. In particular, if D = 0, the quantization index Qw (T (X)) is corrected so that the modulation result Yw of the modulator 16 is always within the dynamic range.

The above-mentioned Ito et al. Document shows that the problem of rounding error due to integerization can be avoided by setting Qw sufficiently coarse.
That is, the following equation (13) is a sufficient condition that the digital watermark is not lost due to a rounding error.
q _w ^k ≧ R (13)
In the case of 8 × 8 DCT, it can be proved that the influence of the rounding error can be eliminated by setting R = 8 and setting the quantization width of all the coefficients to 8 or more.

As described above, in order to save the signature information in the image signal decoded in the time domain, the effect of rounding error is eliminated by setting the limit of Qw to the expression (13), and the correction circuit 14 is mounted on the signature circuit 1 and clipping is performed. It is sufficient to take countermeasures. For example, H.M. In the case of using the H.263 quantization, Qw may be set as follows from the equations (8), (9), and (13).
That is, when Q is an odd number, it is set according to equation (14), and when Q is an even number, it is set according to equation (15).
Qw = max (8, 3Q) (14)
Qw = max (8,3Q-1) (15)

  As is apparent from the above, according to the first embodiment, the image signal is quantized, an authenticator is calculated from the quantized index of the image signal, and the authenticator is embedded in the quantized index. While the circuit 1 is provided, the difference signal between the image signal and the reference signal is obtained, the difference signal is quantized according to the quantization index in which the authenticator is embedded by the signature circuit 1, and the quantization index of the difference signal is obtained. Since the encoder 2 for output is provided, it is possible to provide multimedia data that does not lose the authentication function even if multimedia data such as images and sounds are decoded.

  Further, according to the first embodiment, since the quantization index Qw (T (X)) output from the quantizer 13 is corrected, an image embedded with a digital watermark has a large dynamic range. Can not exceed. Therefore, even when an image is decoded in the time domain, there is an effect that the authenticity of the image can be confirmed.

Embodiment 2. FIG.
4 is a block diagram showing a digital watermark apparatus according to Embodiment 2 of the present invention. In the figure, the same reference numerals as those in FIG.
The quantizer 31 of the encoder 2 quantizes the frequency domain prediction error T (X) -T (Xp) calculated by the subtracter 12 and quantizes the prediction error T (X) -T (Xp). The index Y is output.
The inverse quantizer 32 of the signature circuit 1 decodes the quantization index Y output from the quantizer 31 and outputs a frequency domain prediction error T (X) -T (Xp).

The adder 33 of the signature circuit 1 adds the prediction error T (X) −T (Xp) output from the inverse quantizer 32 and the frequency domain signal T (Xp) of the reference image signal Xp output from the switch 23. Then, the addition signal T (X) is output.
The correction circuit 34 of the encoder 2 corrects the quantization index Y output from the quantizer 31 according to the modulation result Yw of the modulator 16 (the quantization index in which the authenticator W is embedded by the modulator 16). .

In Embodiment 1 described above, the authenticator W is directly calculated from the input image. However, the authenticator W may be calculated based on the decoded image once decoded from the prediction error signal.
Specifically, it is as follows.

When the subtracter 12 outputs the prediction error T (X) −T (Xp) in the frequency domain, the quantizer 31 of the encoder 2 outputs the prediction error T (X) as in the first embodiment. -T (Xp) is quantized, and the quantization index Y of the prediction error T (X) -T (Xp) is output to the inverse quantizer 32 and the correction circuit 34.
When receiving the quantization index Y from the quantizer 31, the inverse quantizer 32 of the signature circuit 1 decodes the quantization index Y and adds the frequency domain prediction error T (X) -T (Xp) to the adder. To 33.
The adder 33 of the signature circuit 1 uses the prediction error T (X) −T (Xp) output from the inverse quantizer 32 and the frequency domain signal T (Xp) of the reference image signal Xp output from the switch 23. Addition is performed, and the addition signal T (X) is output to the quantizer 13.
Since the processing contents from the quantizer 13 to the modulator 16 are the same as those in the first embodiment, description thereof is omitted.

When the correction circuit 34 of the encoder 2 receives the quantization index Y from the quantizer 31 and also receives the modulation result Yw from the modulator 16, the correction circuit 34, like the quantizer 17 of the first embodiment, The quantization index Y output from the quantizer 31 is corrected according to the modulation result Yw.
That is, the correction circuit 34 outputs a quantization index Y that minimizes d in the equation (17) under the condition that the following equation (16) is satisfied.
Qw (Qc- ¹ (Y) + T (Xp)) = Yw (16)
d = | Qc ⁻¹ (Y) − (T (X) −T (Xp)) | (17)

  In encoding a moving image, a prediction error signal having a small value is roughly quantized in order to suppress noise. In the second embodiment, an authenticator W is generated from the decoded image signal. There is an effect that an encoded image with apparently little noise can be obtained without impeding the noise suppression function.

Embodiment 3 FIG.
FIG. 5 is a block diagram showing an authentication processing apparatus according to Embodiment 3 of the present invention. In the figure, the coefficient acquisition unit 41 uses, for example, an encoded signal in which an authenticator is embedded by the digital watermarking apparatus of FIGS. When received, the conversion coefficient of the encoded signal is acquired.
The quantizer 42 quantizes the transform coefficient acquired by the coefficient acquisition unit 41 and outputs a quantization index to the authenticator operation circuit 43 and the authenticator detector 44. The coefficient acquisition unit 41 and the quantizer 42 constitute a quantizing unit.

The authenticator calculation circuit 43 calculates an authenticator from the quantization index output from the quantizer 42. The authenticator operation circuit 43 constitutes an authenticator operation means.
The authenticator detector 44 detects an authenticator embedded in the quantization index output from the quantizer 42. The authenticator detector 44 constitutes an authenticator detector.
The comparator 45 compares the authenticator detected by the authenticator detector 44 with the authenticator calculated by the authenticator calculation circuit 43, and determines that the authenticity of the image is maintained when the two match. If they do not match, it is determined that the authenticity of the image is lost. The comparator 45 constitutes a comparison means.

FIG. 6 is a block diagram showing the inside of the coefficient acquisition unit 41. In the figure, an inverse quantizer 51 decodes a quantization index that is an encoded signal, and restores a prediction error transform coefficient.
The adder 52 adds the conversion coefficient of the prediction error restored by the coefficient acquisition unit 41 and the conversion coefficient of the reference image converted by the converter 55, and outputs the addition result as a conversion coefficient of the encoded signal.
The inverse converter 53 decodes the time domain image signal from the conversion coefficient output from the adder 52 and stores the image signal in the storage circuit 54. The converter 55 refers to the image signal stored in the storage circuit 54. The image is converted into an image conversion coefficient, and the conversion coefficient is output to the adder 52.

Next, the operation will be described.
For example, when the coefficient acquisition unit 41 receives an encoded signal in which an authenticator is embedded by the digital watermarking apparatus of FIGS. 1 and 4, the coefficient acquisition unit 41 acquires a conversion coefficient of the encoded signal.
If the input signal is a signal sequence in which transform coefficients are encoded, the corresponding part may be extracted from the signal sequence, but the input signal is decoded in the time domain. In the case of an image, the coefficient acquisition unit 41 may acquire the conversion coefficient using a converter 56 (see FIG. 7) that converts the input image into the frequency domain.

When the coefficient acquisition unit 41 acquires the transform coefficient, the quantizer 42 quantizes the transform coefficient and outputs the quantization index to the authenticator calculation circuit 43 and the authenticator detector 44.
When receiving the quantization index from the quantizer 42, the authenticator operation circuit 43 calculates an authenticator from the quantization index.
However, since the digital watermark may be embedded in the quantization index, the portion in which the digital watermark is embedded is not used for calculating the authenticator. For example, when a digital watermark is embedded by the method described in Embodiment 1, the authenticator is calculated with the last quantization index of the block embedded with the digital watermark set to 0.

When the authenticator detector 44 receives the quantization index from the quantizer 42, the authenticator detector 44 detects the authenticator embedded in the quantization index.
For example, if a digital watermark is embedded by the method described in Embodiment 1, the authenticator is restored from the last quantization index value of the block in which the digital watermark is embedded.

  When the authenticator detector 44 detects the authenticator and the authenticator computing circuit 43 computes the authenticator, the comparator 45 compares both authenticators, and if both match, the authenticity of the image is maintained. Authenticate that you are leaning. On the other hand, if they do not match, it is determined that the authenticity of the image is lost.

  As apparent from the above, according to the third embodiment, since the authentication code is calculated depending on the input image, not only from the encoded signal sequence, but also from the decoded image. Also has the effect of being able to prove its authenticity.

Embodiment 4 FIG.
FIG. 8 is a block diagram showing a decryption processing apparatus according to Embodiment 4 of the present invention. In the figure, the decryption circuit 61 receives a signal sequence in which an authenticator is embedded by the digital watermarking apparatus of FIGS. The signal sequence is decoded into a time domain signal. Note that the decoding circuit 61 constitutes a decoding means.
When the coefficient acquisition circuit 62 receives the signal sequence in which the authenticator is embedded, the coefficient acquisition circuit 62 restores the conversion coefficient used for calculating the authenticator from the signal sequence.
The quantizer 63 quantizes the transform coefficient restored by the coefficient acquisition circuit 62 with the same characteristics as those used for calculating the authenticator. The coefficient acquisition circuit 62 and the quantizer 63 constitute a quantization means.
The correction circuit 64 corrects the decoding result of the decoding circuit 61 according to the transform coefficient quantized by the quantizer 63. The correction circuit 64 constitutes correction means.

FIG. 9 is a block diagram showing the inside of the decoding circuit 61. In the figure, when the inverse quantizer 71 receives a quantization index of a prediction error that is a signal sequence in which an authenticator is embedded, prediction is made from the quantization index. Restore the error signal value.
An adder 72 adds the signal value of the prediction error restored by the inverse quantizer 71 and the signal value of the reference image, and outputs a transform coefficient of the decoded image.

The inverse converter 73 converts the transform coefficient of the decoded image output from the adder 72 into an image signal in the time domain. Therefore, since the output of the inverse transformer 73 is an image signal expressed in the time domain, and the output of the inverse transformer 73 is generally a real number, it is converted into an integer by rounding off or the like. When the result exceeds the specified numerical range, clipping is performed to correct it to the specified range. The output of the inverse converter 73 is corrected as necessary by the correction circuit 64 and becomes an image signal having an authenticator.
The storage circuit 74 stores the time domain image signal converted by the inverse converter 73.
The converter 75 converts the time domain image signal stored in the storage circuit 74 into a reference image signal value.

Next, the operation will be described.
When receiving the signal sequence in which the authenticator is embedded by the digital watermarking device of FIGS. 1 and 4, for example, the decoding circuit 61 decodes the signal sequence into a time domain signal according to a certain rule.
On the other hand, when the coefficient acquisition circuit 62 receives the signal sequence in which the authenticator is embedded, the coefficient acquisition circuit 62 restores the transform coefficient used for calculating the authenticator from the signal sequence. However, as shown in FIG. 9, when the adder 72 of the decoding circuit 61 restores the transform coefficient and outputs it to the quantizer 63, the coefficient acquisition circuit 62 is unnecessary.

The quantizer 63 quantizes the transform coefficient restored by the coefficient acquisition circuit 62 with the same characteristics as those used for calculating the authenticator in the digital watermark apparatus.
The transform coefficient Yw quantized by the quantizer 63 is given to the correction circuit 64 as a target for correcting the decoding result of the decoding circuit 61.

When receiving the quantized transform coefficient Yw from the quantizer 63, the correction circuit 64 corrects the decoding result of the decoding circuit 61 according to the transform coefficient Yw.
That is, when the correction circuit 64 receives the integer vector X (decoding result) of the image signal output from the decoding circuit 61, when the image signal is converted and quantized, the image signal matches the conversion coefficient Yw. Then, the integer vector X of the image signal is corrected. The reason why this correction is necessary is that if integer processing or clipping processing is included in the decoding process, the information of the authenticator embedded as a digital watermark may be lost due to an error caused by the processing.

FIG. 10 is a flowchart showing the processing contents of the correction circuit 64.
Hereinafter, the processing content of the correction circuit 64 will be described in detail with reference to FIG.
The correction circuit 64 determines whether or not the integer vector X of the image signal satisfies the condition of the following equation (18) (step ST11).
Qw (T (X)) = Yw (18)
Here, T () is an operator representing transformation, and Qw () is an operator that performs quantization with the same characteristics as those used in the calculation of the authenticator.

When the equation (18) is satisfied (step ST11), the correction circuit 64 ends the correction process because the vector X holds information on the authenticator.
However, when Expression (18) is not satisfied (step ST11), the vector X is corrected (step ST12).
The correction of the vector X is performed by bringing the component farthest from the real vector Xw by 1 closer to the real vector Xw among the components of the vector X, where Xw is the real vector from which the transform coefficient Yw is restored. Here, the real vector Xw is given by the following equation.
Xw = T ⁻¹ (Qw ⁻¹ (Yw)) (19)
For the quantized Qw, when equation (12) holds, it can be proved that the integer vector closest to the real vector Xw always holds the authenticator information. Therefore, by repeating the correction of the vector X, equation (18) can be obtained. An integer vector X that satisfies can always be found.

  As apparent from the above, according to the fourth embodiment, since the authentication code is calculated depending on the input image, there is an effect that the authentication code can be inherited in the decoded image signal. . In addition, since the decryption vector is corrected so as to have an authenticator, an effect that the authenticator can be inherited in an image decoded into a time domain signal is obtained.

Embodiment 5 FIG.
FIG. 11 is a block diagram showing a requantization apparatus according to Embodiment 5 of the present invention. In the figure, a coefficient acquisition circuit 81 receives an image signal in which an authenticator is embedded by, for example, the digital watermark apparatus of FIGS. When received, the conversion coefficient used for calculating the authenticator is restored from the image signal. The coefficient acquisition circuit 81 constitutes conversion coefficient acquisition means.
The quantizer 82 quantizes the transform coefficient acquired by the coefficient acquisition circuit 81 with the same characteristics as those used for calculating the authenticator. Note that the quantizer 82 constitutes a first quantizing means.
The quantizer 83 quantizes the transform coefficient acquired by the coefficient acquisition circuit 81 according to the quantization result of the quantizer 82. Note that the quantizer 83 constitutes a second quantizing means.

FIG. 12 is a block diagram showing the inside of the coefficient acquisition circuit 81. In FIG. 12, when the inverse quantizer 91 receives a quantization index of a prediction error that is a signal sequence in which an authenticator is embedded, Restore the signal value of the prediction error.
The adder 92 adds the signal value of the prediction error restored by the inverse quantizer 91 and the signal value of the reference image, and outputs a decoded image transform coefficient.

The inverse converter 93 converts the conversion coefficient of the decoded image output from the adder 92 into a time domain image signal.
The storage circuit 94 stores the time-domain image signal converted by the inverse converter 93.
The converter 95 converts the time domain image signal stored in the storage circuit 94 into a reference image signal value.

Next, the operation will be described.
For example, when the coefficient acquisition circuit 81 receives an image signal in which an authentication code is embedded by the digital watermarking apparatus of FIG. 1 or FIG.
However, the input image signal may be an encoded signal sequence or a decoded time domain signal.

When the quantizer 82 receives the conversion coefficient from the coefficient acquisition circuit 81, the quantizer 82 quantizes the conversion coefficient acquired by the coefficient acquisition circuit 81 with the same characteristics as those used for calculating the authenticator by the digital watermarking device.
When the quantizer 83 receives the transform coefficient Yw quantized by the quantizer 82, the quantizer 83 quantizes the transform coefficient so that the transform coefficient Yw is stored in the transform coefficient acquired by the coefficient acquisition circuit 81. To do.
That is, for example, the quantizer 83 outputs the transformation coefficient acquired by the coefficient acquisition circuit 81 so as to output the vector Y that minimizes d in the expression (21) under the condition of the following expression (20). Quantize
Qw (Qc ⁻¹ (Y)) = Yw (20)
d = | Qc ⁻¹ (Y) −T (X) | (21)

  As apparent from the above, according to the fifth embodiment, since the authentication code is calculated depending on the input image, it is possible to inherit the authentication code to the re-encoded signal sequence. Play. This is useful when, for example, a part of a frame of a moving image encoded by MPEG or the like is re-encoded by JPEG or the like as a still image.

  In the first to fifth embodiments, the input signal is an image signal. However, the present invention can be widely applied to multimedia data such as an audio signal.

  As described above, the digital watermark device according to the present invention is suitable for a device that needs to embed a digital watermark in the multimedia data so that the authentication function is not lost even if the multimedia data is decoded. Yes.

Claims (14)

  The input signal is quantized, an authenticator is calculated from the quantization index of the input signal, a signing means for embedding the authenticator in the quantization index, a difference signal between the input signal and the reference signal is obtained, and An electronic watermarking device comprising: encoding means for quantizing the difference signal according to a quantization index in which an authenticator is embedded by a signature means, and outputting the quantization index of the difference signal.   The encoding means includes a converter that converts an input signal into a frequency domain signal, a subtractor that obtains a difference signal between the frequency domain signal output from the converter and a reference signal, and an authenticator embedded by a signature means. 2. The electronic apparatus according to claim 1, further comprising: a quantizer that quantizes the difference signal obtained by the subtractor according to the quantization index and outputs a quantization index of the difference signal. Watermarking device.   The signing means quantizes the frequency domain signal output from the converter of the encoding means, outputs a quantization index of the frequency domain signal, and the quantization index output from the quantizer An authenticator calculating circuit for calculating an authenticator from the modulator, and a modulator for modulating a part of the quantization index output from the quantizer according to the authenticator calculated by the authenticator calculating circuit. The digital watermarking device according to claim 2, wherein:   4. The signature means comprises a correction circuit that corrects the quantization index output from the quantizer and outputs the corrected quantization index to the authenticator operation circuit and the modulator. Digital watermarking device.   4. The digital watermarking apparatus according to claim 3, wherein the quantizer of the signing unit has a quantization characteristic set depending on the quantization characteristic of the quantizer of the encoding unit.   4. The digital watermarking apparatus according to claim 3, wherein the modulator modulates a portion of the quantization index output from the quantizer that is less affected by the modulation.   The encoding means includes a converter for converting an input signal into a frequency domain signal, a subtractor for obtaining a difference signal between the frequency domain signal output from the converter and a reference signal, and a difference obtained by the subtractor. A quantizer that quantizes the signal and outputs the quantization index of the difference signal, and a modification that corrects the quantization index output from the quantizer according to the quantization index in which the authenticator is embedded by the signature means 2. The digital watermarking apparatus according to claim 1, wherein the electronic watermarking apparatus comprises a circuit.   The signing means decodes the quantization index output from the quantizer of the encoding means and outputs a difference signal, and adds the difference signal output from the inverse quantizer and the reference signal. An adder, a quantizer that quantizes the addition signal output from the adder and outputs a quantization index of the addition signal, and an authentication that calculates an authenticator from the quantization index output from the quantizer And a modulator that modulates a part of a quantization index output from the quantizer according to an authenticator calculated by the authenticator arithmetic circuit. Item 8. The electronic watermarking device according to Item 7.   9. The signature means includes a correction circuit that corrects the quantization index output from the quantizer and outputs the corrected quantization index to the authenticator operation circuit and the modulator. Digital watermarking device.   9. The digital watermarking apparatus according to claim 8, wherein a quantization characteristic of the quantizer of the signature unit is set depending on a quantization characteristic of the quantizer of the encoding unit.   9. The digital watermarking apparatus according to claim 8, wherein the modulator modulates a portion of the quantization index output from the quantizer that is less affected by the modulation.   When a signal with an authenticator embedded is received by the digital watermarking device, the transform coefficient of the signal is obtained, the quantizer that quantizes the transform coefficient and outputs a quantization index, and is output from the quantizer Authenticator computing means for computing an authenticator from the quantized index, authenticator detecting means for detecting an authenticator embedded in the quantized index output from the quantizing means, and detected by the authenticator detecting means. An authentication processing apparatus comprising: a authenticator that compares the authenticator calculated by the authenticator calculating means;   Upon receiving the signal with the authenticator embedded by the digital watermarking device, the decoding means for decoding the signal, the conversion coefficient of the signal, the quantization means for quantizing the conversion coefficient, and the quantization means A decoding processing apparatus comprising correction means for correcting a decoding result of the decoding means in accordance with a quantized transform coefficient.   When a signal with an authenticator embedded therein is received by the digital watermarking apparatus, a transform coefficient acquisition unit that acquires a transform coefficient of the signal, and a first quantization unit that quantizes the transform coefficient acquired by the transform coefficient acquisition unit And a second quantization means for quantizing the transform coefficient acquired by the transform coefficient acquisition means in accordance with the quantization result of the first quantization means.

Images