JP2002304184A - Device for embedding electronic watermark information - Google Patents

Abstract

PROBLEM TO BE SOLVED: To resolve the problem of a conventional device which performs approximate substitution of a vector value or a code book degrades quality and brings about degradation of only a specific pattern and enables positions over the syntax to be specified easily and is delicate for attack with respect to electronic watermark information. SOLUTION: A part 11 calculating MNR for decoding re-quantizes a signal quantized by a quantizer 5 and calculates energy of each subband, to obtain a signal power and calculates a masking level on the basis of the signal power and uses an average quantization error derived from the quantization width of auxiliary information, to calculate the noise level of each subband and calculates an MNR for decoding from these values. An assigned information volume adjustment part 10 discriminates whether a signal value of watermark information to be embedded and a flag (0/1) obtained according to whether the MNR quality level at the time of decoding of a prescribed sub band of a pertinent frame is an odd-numbered level or an even-numbered level coincide with each other, and the assigned information volume of each sub band is corrected, until they coincide with each other.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a digital watermark information embedding apparatus, and more particularly to a digital watermark information embedding apparatus that embeds digital watermark information in a voice compression system using audibility characteristics.

[0002]

2. Description of the Related Art In recent years, with the development of digital technology, it has become possible to easily distribute contents via a network. However, in spite of the convenience of content distribution, the development of digital technology has caused a problem that copyrights of authors are infringed because a large number of copies of the content can be easily generated. In particular, music data can be stored in a smaller capacity than conventional compact discs (CDs) by information compression using a high-efficiency encoding technique, and illegal copying is a serious situation.

In order to prevent such illegal copying, a digital watermarking technique has been conventionally used. This conventional digital watermarking technique can be classified into a method of embedding digital watermark information in music data itself and a method of digital signal processing, that is, a method of embedding digital watermark information in compressed music data.

In the latter system, for example, in a system having a codebook for converting vector values and signal values, a certain rule (key) is used to replace a vector value or a codebook which should be originally selected. A digital watermark information embedding device that embeds digital watermark information by replacing data (such as another data with a very close value) with a document (“ITU G.
729 ", The 55th Annual Convention of Information Processing Society of Japan, vol. 4, pp. 428
-429).

A digital watermark information embedding device suitable for encoding digital audio data with frequency conversion has also been known (Japanese Patent Laid-Open No. 11-3165).
No. 99). In this conventional device, the perceptual importance of each frequency component of audio data that has been converted from a time-domain signal into a frequency-domain signal and subjected to quantization processing is determined,
Digital watermark information is inserted into data of frequency components that are important for human auditory characteristics, so that when the digital watermark information is replaced, decoding is likely to occur as a change or deterioration in sound quality.

[0006]

However, in the conventional digital watermark information embedding apparatus which performs approximate replacement of a vector value or a codebook, the replacement data causes deterioration in quality due to an increase in an error amount and deterioration in only a specific pattern. Occurs. In addition, the position on the syntax is easily specified, and it is vulnerable to an attack on digital watermark information.

Further, the conventional digital watermark information embedding apparatus using the frequency conversion coding method described in the above publication involves replacement position information and replacement value information, which are MPEG2-AA.
C, which has a problem of lack of versatility. Furthermore,
In this conventional apparatus, the position on the syntax is easily specified, and there is a problem in attacking the digital watermark information.

[0008] The present invention has been made in view of the above points,
The ratio (MN) between the masking level determined by the psychoacoustic model and the noise level determined from the quantization error amount determined by the bit allocation or the number of quantization steps
A digital watermark information embedding device that can suppress sound quality degradation by embedding digital watermark information using signal values obtained by grouping (numerizing) the MNR quality ranks in specific subbands using R (masking level to noise level ratio). The purpose is to provide.

[0009]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention provides a conversion unit for converting an input audio signal from a time domain to a frequency domain, and a method for quantizing a frequency domain signal from the conversion unit. A quantizer which receives an input audio signal as an input, and a masking level calculated for each of a plurality of subbands using human auditory characteristics, and a quantization noise level when a spectrum in the subband is quantized. Is calculated (MNR), and the MN
Control means for performing bit allocation control of the number of quantization bits or control of the number of quantization steps for the quantizer based on R, and the power value of the signal obtained by requantizing the quantized signal output from the quantizer Calculating means for calculating the MNR at the time of decoding for each subband from the bit allocation value or the auxiliary information indicating the number of quantization steps output from the control means; The calculated MNR quality at the time of decoding of a predetermined subband allocated to transmit watermark information in advance is relative to the calculated MNR quality at the time of decoding of all other subbands of the same frame. And the signal value obtained by grouping the MNR quality ranks and the signal sequence of the desired watermark information signal sequence.
A comparison unit that compares bits or a signal value for each predetermined number of bits, and groups one bit or a signal value for each predetermined number of bits of the signal sequence of the desired watermark information and the MNR quality rank. And adjusting the bit allocation control amount or the quantization step number control amount for the quantizer by the control means based on the comparison result of the comparison means so that the relation with the signal value obtained thereby always becomes the set relation. Means for embedding a signal sequence of desired watermark information for each frame or a quantized signal of an intermittent frame for each bit or a predetermined number of bits.

According to the present invention, a signal value obtained by grouping the MNR quality ranks at the time of decoding calculated for a predetermined subband, and one bit or a predetermined number of bits of a signal sequence of desired watermark information By controlling the bit allocation control amount or the quantization step number control amount for the quantizer based on the comparison result, a predetermined sub-signal of the quantized signal of each frame or intermittent frame is compared. A signal value obtained by grouping the MNR quality ranks at the time of band decoding can be adjusted so as to match a signal value of desired watermark information.

Here, the calculating means calculates a masking level based on a power value of a signal obtained by requantizing the quantized signal output from the quantizer, and outputs a bit allocation value output from the control means. Alternatively, a quantization noise level for each sub-band is calculated using an average quantization error derived from auxiliary information indicating the number of quantization steps, and a decoding noise level for each sub-band is calculated from the masking level and the quantization noise level. It is characterized in that the MNR is calculated for each frame.

Further, the above-mentioned comparison means in the present invention comprises:
Out of all the sub-bands of the same frame, the calculated MNR quality rank at the time of decoding of a predetermined sub-band allocated to transmit watermark information in advance is changed to two ranks of an odd-numbered rank and an even-numbered rank. M when grouping and decoding
Means for comparing a signal value of desired watermark information as a first signal value when the NR quality rank is an odd-numbered group and a second signal value when the NR quality rank is an even-numbered group; When the signal value of the signal sequence of the desired watermark information is the first signal value, the adjusting unit becomes an odd-numbered group in the MNR quality order at the time of decoding, and the signal value of the signal sequence of the desired watermark information is the second signal value. When the signal value is M
The control unit adjusts the bit allocation control amount or the quantization step number control amount for the quantizer based on the comparison result of the comparison unit until the NR quality rank becomes an even-numbered group.

[0013]

Next, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram of an embodiment of a digital watermark information embedding device according to the present invention, and FIG. 2 is a block diagram of an example of a general music signal information compressing device using audibility characteristics. In both figures, the same components are denoted by the same reference numerals.

First, a general music signal information compression apparatus will be described with reference to FIG. 2. Typical music signal information compression systems include MPEG-Audio and ATRAC.
(Adaptive Transform Acoustic Coding), dolby
-Digital, etc., and weights the amount of information in a unit band on a frequency axis called a sub-band in a model with a frequency band narrower as a critical band, which is a critical bandwidth which is a resolution in human hearing, in any system. Information is reduced.

The weighting is based on the fact that when a signal larger than the surroundings exists at a certain frequency, a small sound existing around the frequency is masked and cannot be heard, and a relative masking amount (relative audible threshold) and a frequency Determined according to the
The sound below a certain sound pressure is given for each sub-band from the corrected masking amount obtained from the absolute masking amount (absolute audible threshold value) which is not perceived.

In FIG. 2, the digital audio signal is supplied to the time-frequency converter 2 and the psychoacoustic model 3 after a predetermined number of frames are temporarily stored in the input signal buffer 1. The time-frequency conversion unit 2 includes a polyphase filter and an MDCT (Modified Discrete
ne transform), the input digital audio signal in the time domain is converted into a signal in the frequency domain and supplied to the quantizer 5.

On the other hand, the psychoacoustic model 3 uses the human auditory characteristics of the input digital audio signal and makes it inaudible for each frequency band called a masking level, which is a basis for cutting the inaudible sound. The quantization error is calculated from the signal value calculated for each subband, and is output to the allocation information amount control unit 4.

The above-mentioned masking level is determined by the parameters of the psychoacoustic model 3, and the same signal does not mean that the masking levels match. This psychoacoustic model 3 is a part that determines sound quality, and occupies the most important position for the encoder.

The allocation information amount control unit 4 receives the masking level output from the psychoacoustic model 3 as an input,
The quantizer 5 determines how many bits the signal from the time-frequency conversion unit 2 should be quantized to prevent noise from being perceived, and generates an assigned bit amount (or quantization step width) to generate a quantized step size. 5 The quantizer 5 quantizes the signal from the time-frequency conversion unit 2 with the allocated bit amount (quantization step width) from the allocation information amount control unit 4, and converts the obtained quantized signal into a bit stream forming unit. 7 is output.

The auxiliary information encoding unit 6 generates a normalized value, a quantization width and the like as auxiliary information for each sub-band of the frequency signal based on the allocated bit amount from the allocation information amount control unit 4, It is supplied to the stream forming unit 7. The auxiliary information includes a codebook number when Huffman encoding is performed, and further includes a flag and a filter coefficient value when auxiliary encoding such as stereo correlation or noise shaping is employed. The bit stream forming unit 7 multiplexes the header information and the side information including the content of the synchronization signal and the auxiliary information with the quantized signal from the quantizer 5 to generate and output a bit stream of a predetermined standard. .

As described above, in this general music signal information compression apparatus, bit allocation control for judging whether or not limited information is allocated to which band and how to improve the overall sound quality in audibility, or The amount of information is weighted and information is reduced by controlling the number of quantization steps to determine how much quantization accuracy is required for the corresponding band. The criterion at this time is a value obtained by comparing a masking level obtained from a signal value calculated for each subband with a quantization noise level generated by quantizing a spectrum in the subband, that is, an MNR.
Is used. Theoretically, if the noise level obtained from the error amount between the original signal and the quantized signal is equal to or lower than the masking level, it can be said that the quantization noise is not perceived.

Note that the parameters of the psychoacoustic model are not required to be disclosed, and even if the parameters are unknown, there is no problem to the decoder. Therefore, the masking level amount is a value that only the party owning the encoding device can know. That is, the MNR using the masking level is also known only to the party who owns the encoding device.

Next, the configuration and operation of the digital watermark information embedding device according to the present invention using the above-mentioned general music signal information compression device will be described with reference to the block diagram of FIG. 1, the signal schematic diagram of FIG. 3, and FIG. It will be described together with the flowchart of FIG. The embodiment shown in FIG. 1 is different from the general music signal information compression device shown in FIG.
0 and a decoding-time MNR calculation unit 11 are further added.

The watermark information generator 9 generates, as watermark information, copyright information for asserting the copyright of the digital audio signal, a copy permission signal indicating whether or not copying is permitted, and the like. This watermark information is, for example, a signal of about 2 to 60 bits. The watermark information generator 9 also generates and outputs an n-bit synchronization signal of a predetermined fixed pattern in time series with the m-bit watermark information, as shown in FIG.

The allocation information amount adjustment unit 10 assigns, for example, one frame of digital watermark information to one frame (determined by a coding method based on a block length divided in time. For example, ATRAC is 512 audio samples). When bits are embedded, whether it is 0 or 1 is determined based on the MNR rank (odd / even). The signal value (for one bit) to be embedded matches the flag (0/1) obtained from the MNR rank of the corresponding frame. It is determined whether or not this is the case, and the allocation information amount for each frequency band (sub-band) is corrected until they match.
Note that, as described above, MNR refers to the ratio between the masking level and the quantization noise level.

The decoding-time MNR calculator 11 requantizes the signal quantized by the quantizer 5 and calculates energy for each frequency band (sub-band) to obtain signal power. And a noise level for each frequency band (sub-band) using an average quantization error derived from the quantization width (quantization step width) of the auxiliary information. And the decryption MN
The R calculation unit 11 calculates the MNR at the time of decoding from these values.

Next, the operation of this embodiment will be described with reference to FIG.
This will be described with reference to FIG. Now, it is assumed that the state is the same as before the conventional speech coded bit stream is formed (step S1 in FIG. 4). In this state, the quantized signal extracted from the quantizer 5 is supplied to the decoding-time MNR calculation unit 11. Is done. The decoding-time MNR calculator 11 requantizes the above-mentioned quantized signal, obtains signal power from the energy, calculates a masking level based on the signal power,
Based on the bit allocation information or the quantization step value information from the auxiliary information encoding unit 6, a quantization noise level for each frequency band (sub-band) is calculated using an average quantization error based on the minimum quantization accuracy. , And the decoding MNR, which is the ratio between the masking level and the quantization noise level,
The calculation is performed on a frame-by-frame basis for all subbands of the same frame for each subband (step S2 in FIG. 4).

Subsequently, the decoding-time MNR calculating unit 11 determines whether the MNR quality of a predetermined subband assigned to transmit watermark information in advance among all the subbands of the same frame is different from that of the same frame. Are ranked relative to the MNR quality in all the sub-bands, and the signal value A of the MNR quality rank in the above-mentioned predetermined sub-band is obtained (step S in FIG. 4).
3).

Here, the above-mentioned MNR quality rank becomes higher as the noise level becomes lower than the masking level. The signal value A of the MNR quality rank is “1” if the quality rank of the MNR is an odd rank, and is “0” if the MNR quality rank is an even rank. In other words, the present embodiment is characterized in that the MNR quality rank is divided into two groups: an odd rank group and an even rank group.

Next, the allocation information amount adjuster 10 serially inputs the n-bit synchronization signal and the m-bit watermark information from the watermark information generator 9 as shown in FIG. The 1-bit value of the sync signal or the watermark information (hereinafter, also referred to as “watermark signal value”) is compared with the above-described signal value A input from the decoding-time MNR calculating unit 11 to determine whether or not they match (see FIG. Step S4 in FIG. 4).
The synchronization signal is a signal for improving the detection accuracy of the watermark information on the decoding device side, and is a predetermined fixed pattern.

When the watermark signal value and the signal value A match, the allocation information amount adjustment unit 10 notifies the allocation information amount control unit 4 of the comparison result of the matching. Accordingly, the allocation information amount control unit 4 supplies the allocation bit amount (quantization step width) to the quantizer 5 as described with reference to FIG.
The quantizer 5 quantizes the signal from the time-frequency converter 2 and outputs the obtained quantized signal to the bit stream forming unit 7 (step S6 in FIG. 4).

On the other hand, if the result of the comparison in step S4 shows that the watermark signal value and the signal value A do not match, the comparison result of the mismatch is notified to the allocation information amount control unit 4. This allows
When the coding scheme is a fixed rate, the allocation information amount control unit 4 allocates information one immediately before (one one) in order from the subband to which information allocation was lastly performed (the quantization step width was last changed). The decoding-time MNR calculator 11 is controlled through the auxiliary information encoder 6 so as to substitute for the sub-band whose quantization step width has been changed before (step S in FIG. 4).
5).

That is, although the embedding apparatus of this embodiment performs encoding, the encoder allocates information in order from the subband having the lower MNR, so that the information is allocated. The subband assigned last (the quantization step width was last changed) has the best quality, and the assignment of the information amount of this subband to other subbands hardly affects the quality of the reproduced signal. On the other hand, MN
The quality order of R changes according to the excitation signal to be encoded and the amount of allocation information, and the quality order of each subband can be changed by changing the amount of allocation information.

Therefore, in this embodiment, when the watermark signal value and the signal value A do not match, and when the encoding method is the fixed rate, the information amount to be lastly allocated is allocated immediately before. The bit allocation control of the quantizer 5 is changed so as to substitute for the sub-band (the quantization step width was changed before). Further, when the encoding method is a variable rate, the encoding is further continued from the state where the encoding was once interrupted at a certain reference rate (or reference quality) (the number of quantization bits is further increased from the reference value). Do).

Accordingly, the quality of the MNR of a predetermined subband allocated to transmit watermark information in advance among all the subbands of the same frame is changed to the MNR of all the other subbands of the same frame. , And the quality of the MNR of a given subband changes from odd to even or from even to odd with a probability of 1/2.

Therefore, the method of modifying the amount of allocated information varies depending on whether the encoding method is a fixed rate or a variable rate.
After calculating the corrected MNR for each sub-band of the same frame, the NR calculation unit 11 (step S in FIG. 4)
2), the allocation information amount adjustment unit 10 compares whether the signal value A for each MNR quality order of the subband after the correction matches the watermark signal value (steps S3 and S4 in FIG. 4). Less than,
The same operation as above is repeated until the corrected subband MNR quality order signal value A matches the watermark signal value (steps S5, S2, S3, S4 in FIG. 4). In addition,
In the case of a variable rate, the number of quantization bits may be reduced.

As described above, in the present embodiment, the allocation information amount control unit 10 controls the allocation information amount control unit 10 so that the MNR of the predetermined subband has a quality order as shown in Table 1, for example. By instructing No. 4, watermark information can be easily embedded.

[0038]

[Table 1] Here, the MNR quality ranks in Table 1 are grouped into two simple groups, odd ranks or even ranks.
The assigned information amount adjustment unit 10 and the assigned information amount control unit 4 perform predetermined sub-processing so that the desired watermark information having a bit value of “1” is embedded when the MNR quality rank is an odd group and “0” when the MNR quality order is an even group. By finely adjusting the MNR of the band, watermark information can be inserted.

For example, to embed an 8-bit synchronization signal and 30-bit watermark information in a predetermined subband,
If one bit is embedded in one frame, the shortest is 3
With eight frames, watermark information can be embedded. Further, it may be embedded continuously in each frame or intermittently in each frame.

In the present embodiment, in the fixed rate method, bits are not used for subbands for which the amount of information is originally required, and are given instead to subbands required next. Although the quality is slightly deteriorated, the sub-bands where the deterioration occurs are not concentrated on the fixed sub-bands, and the overall quality is maintained. Also, at the variable rate, the quality degradation caused by embedding the watermark information is almost equal. In both methods, the processing is not performed on a predetermined subband in which watermark information is embedded, so that deterioration is not concentrated on a specific band.

The MNR can be obtained by the decoder as follows. The masking level can be obtained by requantizing the quantized signal value and employing the same psychoacoustic model as the encoder. As the noise level, the minimum quantization precision △ is naturally given from the bit allocation amount or the quantization step width obtained as the auxiliary information, and the average error △ / 2 can be obtained as the quantization noise. From these two, the MNR can be calculated in the encoder.

When the MNR is obtained for each sub-band, the quality rank of a predetermined sub-band is detected, and the quality rank is grouped into a numerical value and a watermark is formed based on the synchronization signal from the signal sequence quantified. Information can be easily extracted.

The present invention is not limited to the above embodiment. For example, the grouping of MNR quality ranks is not limited to the grouping of odd ranks or even ranks, and a multiple of 1 is set to "1". A multiple of 2 is "2", a multiple of 3 is "3",
It is also possible to group four values by setting a multiple of 4 to “0”. The quality ranking is determined by the allocation information amount adjusting unit 1.
It may be performed at 0. Further, a plurality of sub-bands may be used to transmit the watermark information. In this case, the same watermark information is embedded while shifting the plurality of sub-bands in time, thereby improving resistance to transmission errors, attacks, signal conversion, and the like. it can.

[0044]

As described above, according to the present invention,
Adjusting the signal value obtained by grouping the MNR quality rank at the time of decoding a predetermined subband of the quantized signal of each frame or intermittent frame so as to match the signal value of the desired watermark information Thus, the desired watermark information can be embedded in the quantized signal.

In embedding the desired watermark information, since only the quality of a predetermined subband is not changed, only a specific frequency band is not deteriorated, and only a small number of bits are averaged. The effect on the quality of the reproduced signal is very small just because of the loss, and thus the sound quality deterioration is solved by solving the quality deterioration by the substitute information and the attack resistance by embedding in a specific place, which were inevitable in the past. And a digital watermark information embedding device that is robust against attacks can be realized.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of a digital watermark information embedding device according to the present invention.

FIG. 2 is a block diagram of a general music signal information compression device using the audibility characteristic.

FIG. 3 is a diagram illustrating an example of a signal configuration of information to be embedded.

FIG. 4 is a flowchart for explaining the operation of FIG. 1;

[Description of Code] 1 Input signal buffer 2 Time-frequency conversion unit 3 Psychological psychology model 4 Assignment information amount control unit 5 Quantizer 6 Auxiliary information encoding unit 7 Bit stream forming unit 9 Watermark information generator 10 Assignment information amount adjustment Unit 11 Decoding MNR calculation unit

Claims (4)

[Claims] A conversion unit configured to perform a time-domain to frequency-domain signal conversion on an input audio signal; a quantizer configured to quantize the frequency-domain signal from the conversion unit; And calculates a ratio (MNR) between a masking level calculated for each of a plurality of sub-bands utilizing human hearing characteristics and a quantization noise level when a spectrum in the sub-band is quantized, and based on the MNR, Control means for performing bit allocation control of the number of quantization bits for the quantizer, or control of the number of quantization steps, and a power value of a signal obtained by requantizing the post-quantization signal output from the quantizer. And the auxiliary information indicating the number of quantization steps output from the control means, the MNR at the time of decoding for each subband, Calculating means for calculating in units of, the calculated MNR quality at the time of decoding of a predetermined subband allocated to transmit watermark information in advance among all subbands of the same frame, A signal value obtained by ranking the order of the MNR quality at the time of decoding calculated in all the subbands and grouping the MNR quality order, and a desired watermark Comparing means for comparing one bit of the signal sequence of information or a signal value for each predetermined number of bits, and one bit of the signal sequence of the desired watermark information or the signal value for each predetermined number of bits, On the basis of the comparison result of the comparing means, the control unit performs the previous processing so that the relation with the signal value obtained by grouping the quality ranking of the MNR always becomes the set relation. Adjusting means for adjusting the bit allocation control amount or the quantization step number control amount with respect to the quantizer, wherein the signal sequence of the desired watermark information is set to 1 for the post-quantization signal of each frame or intermittent frame. A digital watermark information embedding device, wherein the digital watermark information is embedded for each bit or a predetermined number of bits. 2. The calculating means calculates the masking level based on a power value of a signal obtained by requantizing the quantized signal output from the quantizer, and assigns a bit to the bit value output from the control means. A quantization noise level for each sub-band is calculated using an average quantization error derived from auxiliary information indicating a value or the number of quantization steps, and decoding is performed for each sub-band from the masking level and the quantization noise level. 2. The digital watermark information embedding device according to claim 1, wherein the MNR at the time is calculated for each frame. 3. The comparing means determines the calculated MNR quality rank at the time of decoding of a predetermined subband allocated to transmit watermark information in advance among all the subbands of the same frame in odd-numbered order. And an even-numbered rank, and when the MNR quality rank at the time of decoding is an odd-numbered group, the first signal value is used.
When the group has an even-numbered rank, the signal value of the desired watermark information is compared with the signal value of the desired watermark information as the second signal value. When the signal value is a signal value, the quality order of the MNR at the time of decoding is the odd-numbered group, and when the signal value of the signal sequence of the desired watermark information is the second signal value, the quality order of the MNR at the time of decoding is the second signal value. Until an even-numbered group, based on the comparison result of the comparing means,
3. The digital watermark information embedding device according to claim 1, wherein the control unit adjusts the bit allocation control amount or the quantization step number control amount for the quantizer. 4. The signal sequence of the watermark information is a signal sequence in which a synchronization signal of a fixed pattern and a watermark information signal are synthesized in a time-series manner. Electronic watermark information embedding device according to the item.