JP3365330B2 - Vector quantization apparatus and vector quantization method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vector quantizing device and a vector quantizing method, and in particular, it is possible to embed watermark information in an encoded bit string when encoding a digital signal such as voice or image. The present invention relates to a vector quantization device and a vector quantization method.

2. Description of the Related Art As a conventional vector quantizer, watermark information is embedded in a designated bit of a 5-bit index by embedding position designating information in a digital code string composed of vector-quantized indexes, and a fixed index is provided. There is proposed a method of selecting a quantized vector having the minimum distortion with respect to an input vector and outputting the index value thereof (Japanese Patent Laid-Open No. 10-224).
342 publication).

[0002]

In this case, since the embedding position is fixed, the data and watermark information obtained by encoding with the watermark information embedded in the same input signal sequence of finite length. There is a problem that the place where the watermark information is embedded is specified from the difference data by comparing the data obtained by encoding without embedding. Therefore, an object of the present invention is to make it difficult to easily specify the embedding position of watermark data, and to accurately grasp the unauthorized use of information after vector quantization, and a vector quantization device and vector quantization device. To provide a method.

[0003]

[Means for Solving the Problems]
In a vector quantizer for vector quantizing input signal data, an index for waveform cutout data that is not embedded with watermark information, of waveform cutout data obtained by performing waveform cutout of the input signal data. When selecting data, any one of a plurality of index data whose error with the original waveform cutout data is within a predetermined error range when decoded is used as index data corresponding to the waveform cutout data. It is characterized in that a random selection means for randomly selecting is provided.

According to a second aspect of the present invention, in the vector quantizer for quantizing the conjugate vector of the input signal data,
Of the waveform cutout data obtained by performing the waveform cutout of the input signal data, when selecting the combination of the index data for the waveform cutout data that is the non-embedding target of the watermark information, the original when decoding From the combination of a plurality of index data whose error with the waveform cutout data is within a predetermined error range, any one combination of index data is randomly selected as a combination of index data corresponding to the waveform cutout data. It is characterized by having a combination random selection means.

According to a third aspect of the present invention, in the configuration of the first or second aspect, when the waveform cutout data is generated, dummy data is added to a part of the waveform cutout data. It is characterized by having means.

According to a fourth aspect of the present invention, in the first aspect, the first vector quantizing means to the n-th vector quantizing means (n: integer of 2 or more) for performing multi-stage vector quantization are provided. It is characterized by that.

According to a fifth aspect of the present invention, in the vector quantizer according to the fourth aspect, at least one of the first vector quantizing means to the n-th vector quantizing means has the vector quantizing means. When generating the waveform cut-out data, a dummy data adding means for adding dummy data to a part of the waveform cut-out data is provided.

According to a sixth aspect of the present invention, in the third or fifth aspect of the invention, the waveform cutout data is composed of a limited number of predetermined sample data, and the dummy data adding means is the The input signal data is divided into divided data composed of a predetermined number of sample data smaller than the finite number, and the dummy data is added to each of the divided data to form the finite number of sample data. It is characterized by generating waveform cutout data.

According to a seventh aspect of the present invention, in the structure according to the sixth aspect, the number of the sample data in the dummy data is the division data for embedding watermark information and the non-embedding target for watermark information. The feature is that it is different for divided data. According to the structure of claim 8, in the structure of claim 6, as the dummy data,
It is characterized in that a predetermined fixed value of sample data is repeatedly added to the divided data until the number of sample data forming the divided data reaches the finite number.

According to a ninth aspect of the present invention, in the configuration according to the sixth aspect, the sample data constituting the waveform cutout data, which is obtained by performing a predetermined operation on the divided data as the dummy data. The calculation data is composed of a number of sample data corresponding to the difference between the number of the sample data and the number of the sample data forming the divided data.

According to a tenth aspect of the invention, in the configuration of the ninth aspect, the operation data is calculated by extrapolation operation.

According to an eleventh aspect of the present invention, in the configuration of the ninth aspect, the operation data is calculated by interpolation operation.

According to a twelfth aspect of the present invention, in the vector quantization method for vector quantizing the input signal data, the watermark information non-existence is included in the waveform cutout data obtained by performing the waveform cutout of the input signal data. When selecting the index data for the waveform cutout data to be embedded, when decoding, select one of the multiple index data whose error from the original waveform cutout data is within the specified error range. It is characterized by comprising a random selection step of randomly selecting as index data corresponding to the waveform cutout data.

According to a thirteenth aspect of the present invention, in the vector quantization method for quantizing the conjugate vector of the input signal data, the watermark information of the waveform cutout data obtained by performing the waveform cutout of the input signal data. When selecting a combination of index data for waveform cut-out data that is not embedded, select from among a combination of multiple index data that the error with the original waveform cut-out data is within the specified error range when decoded. A combination random selection step of randomly selecting one of them as a combination of index data corresponding to the waveform cutout data is provided.

According to a fourteenth aspect of the present invention, in the configuration of the twelfth or thirteenth aspect, when the waveform cutout data is generated, dummy data is added to a part of the waveform cutout data. It is characterized by having a process.

[0016]

BEST MODE FOR CARRYING OUT THE INVENTION Next, preferred embodiments of the present invention will be described with reference to the drawings. [1] First Embodiment [1.1] Configuration of Vector Quantization Device FIG. 1 shows a schematic configuration block diagram of the vector quantization device. The vector quantizer 10 cuts out the waveform of the input signal data DIN and outputs the cut-out waveform data DP as a waveform cut-out device 11, and generates an index based on the cut-out waveform data DP to generate bit stream data DBS. And an index generating unit 12 for outputting.

FIG. 2 is a detailed block diagram of the index generator 12. The index generation unit 12 includes an embedded position information storage unit 21 that stores embedded position data DFP of watermark information.
And a random number generator 22 for generating and outputting random number data DRN
The watermark information storage unit 23 that stores the watermark information data DWM in advance, the codebook 24 that stores the index data, and the cut-out waveform based on the cut-out waveform data DP, the random number data DRD, and the index data DCB that are input. The first encoder 25 that selects the nth candidate index data (n is a natural number) corresponding to the random number data from the plurality of candidate index data corresponding to the data DP and outputs it as the first index data DIDX1 is input. The index data corresponding to the cut-out waveform data and the watermark information data DWM is selected to select the second index data D.
Based on the second encoder 26 that outputs as IDX2 and the embedding position data DFP, the cutout waveform data DP that does not correspond to the embedding position is output to the first encoder 25, and the cutout waveform that corresponds to the embedding position. A selector 27 that outputs the data DP to the second encoder 26, a header information generation unit 28 that generates and outputs the header information data DHD based on the embedding position data DFP, a header information data DHD, and the first encoder 25. From the first index data DIDX1 outputted from the second encoder 26 and the second index data DIDX2 outputted from the second encoder 26 are arranged in a serial manner and output as bitstream data DBS. It is configured.

[1.2] Operation Waveform Cutout Device 11 of Vector Quantizer Cuts the waveform of the input signal data DIN, and the index generator 12 outputs the cutout waveform data DP.
To the selector 27. On the other hand, the embedding position information storage unit 21 of the index generating unit 12 outputs the embedding position data DFP of the stored watermark information to the selector 27 and the header information generating unit 28. This allows the selector 27
Outputs the cut-out waveform data DP not corresponding to the embedding position to the first encoder 25 based on the embedding position data DFP, and outputs the cut-out waveform data DP corresponding to the embedding position to the second encoder 26.
Output to. In addition, the random number generation unit 22 uses the random number data DRN.
Is generated and output to the first encoder 25, and the watermark information storage unit 23 sequentially outputs the stored watermark information data DWM to the second encoder 26. As a result, the first encoder 25 inputs the cut-out waveform data DP, the random number data DRD, and the index data DCB read from the codebook 24.
The index data of the nth candidate (n is a natural number) corresponding to the random number data is selected from the plurality of candidate index data corresponding to the cut-out waveform data DP based on
The index data DIDX1 is output to the bitstream generation unit 29.

More specifically, of the index data corresponding to the cut-out waveform data DP, the index data with the smallest error is set as the first candidate index data,
Assuming that the index data in which the error increases sequentially (including the error with the same error) is the second candidate index data → the third candidate index data → ... For example, a predetermined error range predetermined based on the random number data DRD Among the first to third candidate index data belonging to the above, any one of the candidate index data is selected and output to the bitstream generation unit 29 as the first index data DIDX1. Therefore, since the candidate index data with the smallest error is not necessarily selected, it becomes difficult to find the watermark information actually embedded as the decoy index data. Further, the second encoder 26 reads out index data corresponding to the input cut-out waveform data DP and watermark information data DWM from the codebook 24, selects it, and outputs it as second index data DIDX2 to the bitstream generation unit 29. To do. More specifically,
For example, if the watermark information data DWM has the 4th bit of the index data set to “1”,
The index data having the smallest error among the index data having the 1st bit as “1” is output to the bitstream generation unit 29 as the second index data DIDX2.

The operation of the header information generator 28 will be described with reference to FIG. FIG. 3 is an explanatory diagram of the configuration of the bit stream data DBS. The bit stream data DBS is roughly composed of header information data DHD and vector index information data DVI. The header information data DHD corresponds to, for example, ID data DID having a value that does not overlap in the world with which the vector quantizer 10 can be specified, encoding method data DEM indicating an encoding method, and the bit stream data DBS. Title data DTI representing a title and m pieces of watermark position information data DWMP1 to DWMPm representing positions including watermark information. In this case, the ID data DID is "GUID (Glov
al Unique ID) ”or the like is used. Further, the header information data DHD other than the ID data DID may be encrypted by a predetermined encryption method. The data included in the header information data DHD is not limited to these, and various data can be included as necessary.

[1.3] Vector Decoding Device Next, the vector decoding device will be described. [1.3.1] Configuration of Vector Decoding Device FIG. 4 shows a schematic configuration block diagram of the vector decoding device. The vector decoding device 15 converts the input bitstream data DBS (see FIG. 3) into the encoding method data DEM.
The output signal data D is decoded by a decoding method corresponding to the encoding method corresponding to
Decoder 16 that outputs as OUT, watermark position information data DWMP (= any one of watermark position information data DWPM1 to DWPMn in bit stream data DBS, or watermark position information input from an external watermark position information database (not shown) Data) and watermark information decoder 17 which outputs output watermark information data DWMOUT. The watermark information decoder 17 does not necessarily need to be included in a device such as a decoding device used by a general user that does not necessarily require watermark information. Here, prior to the description of the vector decoding device, the bitstream data DBS ′ when the watermark position information data is input from the external watermark position information database.
The structure of will be described. Bitstream data DB
S'is roughly composed of header information data DHD and vector index information data DVI. The header information data DHD is, for example, ID data DID having a value that does not overlap in the world with which the vector quantizer 10 can be specified, and watermark position storage information data DWMST indicating the storage position of the watermark information on the watermark position information database. And are provided.

Also in this case, "GUID (Gloval Unique ID)" or the like is used as the ID data DID.
Further, the watermark position storage information data DWMST may be encrypted by a predetermined encryption method. The data included in the header information data DHD is not limited to these,
It is possible to include various data as needed. When the watermark position storage information data DWMST is passed to the watermark position information database side, the watermark position information database outputs the watermark position information data corresponding to the watermark position storage information data DWMST to the watermark information decoder 17 as the watermark position information data DWMP. Will be done. Therefore, the bit stream data DBS does not directly include the information regarding the position where the watermark information is embedded, so that the data security can be further ensured.

[1.3.2] Motion of Vector Decoding Device Decoder 16 of vector decoding device 15 decodes input bit stream data DBS to a coding system corresponding to coding system data DEM. Output signal data DOUT by decoding using a decoding method or a predetermined decoding method.
Output as. As a result, the user can reproduce the image, sound, etc. corresponding to the output signal data DOUT. On the other hand, the watermark information decoder 17 receives the watermark position information data DWMP (= any one of the watermark position information data DWPM1 to DWPMn in the bit stream data DBS, or the watermark position information input from an external watermark position information database (not shown). Output watermark information data DWMOUT based on the data). Thus, if the copyright information is included as the output watermark information data DWMOUT, the source of the output signal data DOUT can be easily specified.

[1.4] Effects of the First Embodiment As described above, according to the first embodiment, the index data corresponding to the non-embedding position of the watermark information is selected in the vector quantization. , Not only the first candidate index data having the smallest error but also other index data belonging to a predetermined error range are randomly selected by random numbers, so that the bitstream data is generated without including the watermark information. It is difficult to extract the watermark information because the difference data including the information other than the watermark information will be obtained even if the difference data is obtained by comparing the generated data with the bitstream data generated by including the watermark information. Bitstream containing stronger watermark information against malicious attacks such as falsification or deletion of watermark information. Ream data can be obtained.

[1.5] Modification of First Embodiment [1.5.1] First Modification In the above description, only the case of embedding watermark information has been described, but the case of not embedding watermark information at all. In the same manner, not only the first candidate index data having the smallest error when selecting the index information on the first encoder side, but also other index data belonging to a predetermined predetermined error range are randomly generated by random numbers. If the configuration is selected, a malicious attacker compares the bitstream data generated without including the watermark information with the bitstream data including the watermark information, and compares the difference data. Even if it is obtained, the difference data including the information other than the watermark information is obtained in the same manner, which makes it difficult to extract the watermark information. , It is possible to obtain a strong bit stream data in such alteration or deletion of the watermark information. [1.5.2] Second Modification In the above description, the watermark information is embedded in a part of the index data, but the watermark information may be embedded in all the index data. It is possible.

[2] Second Embodiment In the above first embodiment, no processing was performed on the input signal data DIN input to the waveform slicing device 11, but the second embodiment is , Input signal data D
IN is one or more finite length sample strings (for example, in the case of audio data, one sample string is 16 bits), and watermark information is embedded at the beginning of each finite length sample string and no watermark information is embedded. It is an embodiment in the case of performing waveform cutout after adding dummy data having different numbers of sample columns. In the following description, a case where audio data is used as the input signal data will be described for easy understanding, but the present invention can be applied to other data such as image data. [2.1] Description of Principle of Second Embodiment Here, prior to specific description of the second embodiment, a description of the principle of the second embodiment will be given. FIG. 5 shows a principle explanatory diagram of the second embodiment. Assume that the input signal waveform corresponding to the input signal data DIN is as shown by the solid line in FIG.
Input signal data DIN input from time t0 to time t1
Is one finite number of sample sequences, the index data obtained when vector quantization is performed from the beginning of the finite length sample sequence becomes index data ~ according to the corresponding input signal waveform. Become.
On the other hand, when dummy data of a predetermined number of samples (in FIG. 5A, the signal waveform corresponding to the dummy data is shown by a chain line) is added to the head of the finite length sample sequence, Time t2 to time t3 (t3-t2 =
The index data obtained when vector quantization is performed from the beginning of a finite number of sample sequences corresponding to the input signal data DIN input up to t1-t0) is the index data '... ', Which is completely different from the index data ~. Therefore, in the second embodiment, by utilizing this, the number of samples of dummy data added to the head of the finite length bit string when watermark information is embedded and the dummy data added to the head of the finite length sample string when watermark information is not embedded By differentiating the number of samples, the malicious attacker compares the bitstream data generated without the watermark information with the bitstream data generated with the watermark information. Even when the difference data is obtained, the difference data including the information other than the watermark information is obtained, which makes it difficult to extract the watermark information.

[2.2] Operation of Vector Quantization Device Next, the operation of the vector quantization device of the second embodiment will be described. The operation of the vector quantization device of the second exemplary embodiment is the same as that of the first exemplary embodiment with respect to the operation after the waveform cutout, and therefore detailed description thereof will be omitted, and the process up to the waveform cutout will be described. explain. FIG. 6 shows a processing flowchart until the waveform cutting according to the second embodiment. When the input signal data DIN is input (step S1), it is determined whether or not the input data is the head of a finite number of sample strings (step S2). Step S2; No), the process is step S
Go to 6. In this case, input signal data DIN
All may be set to be treated as one finite number of sample strings, or the input signal data DIN may be divided into a plurality of pieces and each obtained divided data may be set to be treated as a finite number of sample strings. good. For example, if the sample data number L of a finite number of sample rows is set as L = sample data number of cut-out waveform data DP−sample data number of dummy data, etc., sample data of cut-out waveform data after adding dummy data The number can be made the same as the number of sample data of the cut-out waveform data when the dummy data is not added. In the determination of step S2, when the input data is the head of a finite number of sample strings (step S2; Yes), it is determined whether or not the watermark information is to be encoded (step S3). Step S3
If the watermark information is to be embedded (watermark) in the determination (step S3; Yes), X (X)
Is a natural number), for example, "00h" (h represents a hexadecimal number. The same applies hereinafter) is added to the beginning of the data as dummy data. On the other hand, in the determination of step S3, when normal encoding without embedding watermark information is performed (step S3; No), Y (Y is a natural number and Y ≠ X) constant value, for example, “00h "(= 8 bits) is added to the beginning of the data as dummy data.

Next, the waveform of the data with or without the dummy data is cut out (step S6),
The process moves to the encoding process. A more specific description will be given here. In the following description, the data length of one sample data is assumed to be 8 bits, but the data length is not limited to this. For example, an arbitrary data length of 16 bits, 32 bits, 64 bits or the like is possible. (1) When embedding dummy data at the beginning of input data When watermark information is embedded (watermark) encoding, a finite number of input sample strings are “08h” as shown in FIG. , "F7h""04h""06
h ”,“ FAh ”,“ 01h ”,“ 02h ”“ FCh ”
"04h", "06h", "FAh", ..., X
= 2, the number of samples of the cut-out waveform data is 6 samples (corresponding to a data length of 48 bits), and the dummy data is “## h” (actually, arbitrary sample data such as “00h”). , “## h” for the sake of description. The same shall apply hereinafter.), The first extracted waveform data obtained is “## h” as shown in FIG. 11B. ","## h "," 08h "," F7h "," 0
4h ”and“ 06h ”. Accordingly, the second cut-out waveform data is shown in FIG.
As shown in (c), "FAh", "01h", "02h", "FCh", "0"
4h ”and“ 06h ”.

On the other hand, when dummy data is not added (conventional example), the obtained first cut-out waveform data is "08h""F7h" as shown in FIG. 11 (f). , "04h", "06h", "F
As shown in FIG. 11 (g), the second cut-out waveform data becomes “02h”, “FCh”, “04h”, “06h”, “FA”.
h ″, ..., which is completely different from the cut-out waveform data to which dummy data is added and the cut-out waveform data (conventional example) obtained when dummy data is not added. Ordinarily, watermark information is not embedded. In the case of performing the coding of, as shown in FIG.
"08h", "F7h", "04h", "06h", "FA"
h ”,“ 01h ”,“ 02h ”“ FCh ”“ 04h ”,
“06h”, “FAh”, ..., Y = 3,
If the number of samples of the cut-out waveform data is 6 and the dummy data is “## h”, the first cut-out waveform data obtained is similarly “## h”, “ ## h ”,“ ## h ”,“ 08h ”“ F
7h ”and“ 04h. ”Along with this, the second cut-out waveform data is“ 06h ”“ FAh ”,“ 01h ”,“ 02h ”“ FC
h ”and“ 04h ”.

Accordingly, the obtained cut-out waveform data is the cut-out waveform data obtained when dummy data is not added, that is, "08h", "F7h", "04h", "06h", "F".
Ah ”,“ 01h ”and“ 02h ”“ FCh ”,“ 04h ”,“ 06h ”“ FA
h ″, ... And cut-out waveform data obtained when watermark information is embedded (watermark) encoding, “## h”, “## h”, “08h”, “F7h”,
"04h", "06h" and "FAh", "01h", "02h", "FCh",
It is completely different from "04h" and "06h". (2) When embedding dummy data at the beginning of each cut-out waveform data When watermark information is embedded (watermark) encoding, a finite number of input sample sequences are as shown in FIG. "08h", "F7h""04h""06
h ”,“ FAh ”,“ 01h ”,“ 02h ”“ FCh ”
"04h", "06h", "FAh", ..., X
= 2, the number of samples of the cut-out waveform data is 6, and the dummy data is “## h”, the obtained first cut-out waveform data is as shown in FIG. As shown, “## h”, “## h”, “08h”, “F7h”, “0
4h ”and“ 06h ”. Accordingly, the second cut-out waveform data is shown in FIG.
As shown in (e), “## h”, “## h”, “FAh”, “01h”,
“02h” and “FCh”. Therefore, by embedding dummy data at the beginning of each cut-out waveform data, it is possible to obtain completely different data from the cut-out waveform data obtained when the dummy data is not added.

[2.3] According to the effect of the second embodiment, the bitstream data generated without the watermark information is compared with the bitstream data generated with the watermark information. Even if the difference data is obtained, difference data including information other than the watermark information will be obtained, making it difficult to extract the watermark information, and against malicious attacks such as falsification or deletion of the watermark information, Bitstream data including stronger watermark information can be obtained. [2.4] Modification of Second Embodiment In the above description, the number of sample data having a predetermined constant value (the number of sample data) is set as the dummy data, but a dummy number is used to set the dummy data. It is also possible to determine the number of samples of data.

[3] Third Embodiment In the second embodiment described above, the dummy data is set as a fixed number of predetermined values, but in the third embodiment, the actual data of a finite number of sample strings is used. This is an embodiment in which the obtained interpolation data is added as dummy data. [3.1] Operation of Third Embodiment FIG. 7 shows a processing flowchart of the waveform cutting according to the third embodiment. When the input signal data DIN is input (step S11), it is determined whether or not the input data is the head of a finite number of sample strings (step S12). Step S1
2; No), the process proceeds to step S16. In this case, as the finite number of sample strings, all the input signal data DIN may be set to be treated as one finite number of sample strings as in the second embodiment, or a plurality of input signal data DIN may be used. It may be set so that each of the divided blocks is treated as a finite number of sample strings. In the determination in step S12, if the input data is the head of a finite number of sample strings (step S1
2; Yes), and it is determined whether or not the encoding for embedding the watermark information is performed (step S13). When the watermark information is encoded in the determination of step S3 (step S13; Yes), N samples (N is a natural number of 2 or more) of data from the beginning of the finite number of sample strings are used to immediately X pieces of data (X is a natural number) are estimated by an extrapolation method, and the obtained X pieces of sample data are added to the beginning of the data as dummy data.

On the other hand, in the determination of step S13, when normal encoding without embedding watermark information is performed (step S13; No), N samples from the head of the finite number of sample strings (N is a natural number of 2 or more). The last Y data (Y is a natural number greater than or equal to 2 and Y ≠ X)
Data is estimated by the extrapolation method, and the obtained Y sample data are added to the beginning of the data as dummy data. Next, the waveform of the data with or without the dummy data is cut out (step S16), and the process proceeds to the encoding process. [3.2] Effects of Third Embodiment Therefore, also in the third embodiment, it is possible to generate bitstream data without including watermark information and to generate bitstream data including watermark information. Even if the difference data is obtained by comparing the above, the difference data including information other than the watermark information will be obtained, making it difficult to extract the watermark information and tampering with or erasing the watermark information. Bitstream data including stronger watermark information can be obtained against an attack.

[3.3] Modification of Third Embodiment [3.3.1] First Modification In the above description, a predetermined fixed number of sample data are determined by extrapolation as dummy data. However, it is also possible to use pseudo-random numbers to determine the number of samples of dummy data. A more specific description will be given with reference to the processing flowchart of FIG. First, a random number is generated (step S22), and the number of sample data of dummy data (dummy data number DC) is set based on the obtained random number (step S21). Next, a data counter for counting the generated dummy data. Initialize C.
That is, C = 0. Next, it is determined whether the count value of the data counter C is equal to the number of generated dummy data (step S).
24). When the count value of the data counter C and the number of generated dummy data are not equal in the determination of step S24 (step S24; No), it is still necessary to generate dummy data, so the interpolation data that is one sample data Calculate (step S25). Then, the obtained interpolation data is set in the clipping buffer for waveform clipping (step S26).

Then, the count value of the counter C is incremented by 1 (step S29), the process is returned to step S24, and the same process is repeated.
The process is repeated until the interpolation data of the number equal to the dummy data number DC is obtained. When the count value of the data counter C is equal to the number of generated dummy data in the determination in step S24 (step S24; Yes), the storage of the dummy data in the cutout buffer is completed, and thus the input signal data DIN The next one sample data of is set in the cutout buffer (step S27). Subsequently, it is determined whether or not the count value of the counter C has become equal to a finite word length equal to the predetermined number of sampled pieces of cut-out waveform data (step S28). In the determination of step S28,
If the count value of the counter C is still not equal to the finite word length corresponding to the predetermined number of sampled waveform data (step S28; No), the count value of the counter C is incremented by 1 (step S28). S2
9) The process is returned to step S27, and the same process is repeated until the number of data samples stored in the cutout buffer becomes equal to the finite word length corresponding to the number of cutout waveform data samples. Store the input signal data in the clipping buffer. When the count value of the counter C becomes equal to the predetermined number of samples of the cut-out waveform data in the determination in step S28 (step S28; Ye
s), the cut-out waveform data stored in the cut-out buffer is output (step S30), and thereafter, the encoding process is similarly performed. According to the first modification configured as described above, it becomes more difficult to extract watermark information, and bitstream data including stronger watermark information against malicious attacks such as falsification or deletion of watermark information. Can be obtained.

[3.3.2] Second Modification In the above description, the dummy data is calculated by the extrapolation method (extrapolation method), but the last data of the last finite sample sequence and this time are calculated. It is also possible to calculate the dummy data by the interpolation method (interpolation method) based on the data at the head of the finite sample sequence of. As the interpolation method, general interpolation methods such as linear interpolation, Lagrange interpolation, and spline interpolation can be used.

[4] Fourth Embodiment In each of the above embodiments, the basic vector quantizer was described as an example. However, in the fourth embodiment, a multistage vector quantizer is used as the vector quantizer. It is an embodiment in the case of using a chemical conversion device. [4.1] Configuration of Multistage Vector Quantization Device FIG. 9 shows a configuration block diagram of the multistage vector quantization device. The multi-stage vector quantizer 40 uses the input signal data D
A waveform cutout device 41 that cuts out a waveform of IN and outputs as cutout waveform data DP, and an index is generated based on the cutout waveform data DP, and bitstream data DBS1
And an index generation unit 42 for outputting. The index generation unit 42 outputs the first index data DIDX11 based on the cut-out waveform data DP, and the cut-out waveform data D.
A first adder 46 that adds P and the first index data DIDX11 to output the added first index data DIDX11 ', and a second output the second index data DIDX12 based on the added first index data DIDX11'.
The index selection unit 47 and the first index data D
A second adder 48 that adds the IDX11 and the second index data DIDX12 and outputs the added second index data DIDX12 ', and a third index that outputs the third index data DIDX13 based on the added second index data DIDX12'. The selector 49 and the multiplexer 50 for generating and outputting the bit stream data DBS1 from the first index data DIDX11, the second index data DIDX12, and the third index data DIDX13 by a time division method or the like are configured.

In this case, the first index selecting section 45, the second index selecting section 47, and the third index selecting section 49 have the embedded position information storage section 21, the random number generation section 22, and the watermark information storage in the first embodiment. Part 2
3, codebook 24, first encoder 25, second encoder 2
6 and selector 27. [4.2] Operation of Vector Quantization Device The waveform cutout device 41 of the multistage vector quantization device 40 performs waveform cutout of the input signal data DIN and cutout waveform data DP.
Is output to the first index selection unit 45 and the first addition unit 46 of the index generation unit 42. Accordingly, the first index selection unit 45 of the index generation unit 42
Generates an index based on the cut-out waveform data DP and outputs the first index data DIDX11 to the first adder 4
6, output to the second adder 48 and the multiplexer 50. In this case, the first index selection unit 45
When selecting the first index data DIDX11 corresponding to the watermark information non-embedded position, not only the first candidate index data with the smallest error but also other index data belonging to a predetermined error range are randomly generated by random numbers. Therefore, the same effect as that of the first embodiment can be obtained. The first adding unit 46 adds the cut-out waveform data DP and the first index data DIDX11 and outputs the added first index data DIDX11 ′ to the second index selecting unit 47.

The second index selector 47 adds the first index
The second index data DIDX12 is generated based on the index data DIDX11 ', and is output to the second adding unit 48 and the multiplexer unit 50. Also in this case, the second index selecting unit 47, like the first index selecting unit 45, selects the second index data DIDX12 corresponding to the watermark information non-embedded position, and has the smallest error in the first candidate index data. Not only
Other index data belonging to a predetermined predetermined error range is randomly selected by random numbers. Second adder 48
Adds the first index data DIDX11 and the second index data DIDX12 and outputs the added second index data DIDX12 ′ to the third index selection unit 49 and the multiplexer unit 50. The third index selection unit 49 outputs the third index data DIDX13 to the multiplexer unit 50 based on the added second index data DIDX12 ′. Also in this case, the third index selection unit 49, like the first index selection unit 45 or the second index selection unit 47, the third index data DIDX1 corresponding to the watermark information non-embedding position.
At the time of selecting 3, not only the first candidate index data having the smallest error, but also other index data belonging to a predetermined predetermined error range are randomly selected by random numbers.

[4.3] Effects of the Fourth Embodiment As a result, the multiplexer unit 50 causes the multiplexer 50 to process the first index data DIDX11, the second index data DIDX12, and the third index data DIDX13 in the bit stream data DBS1 by a time division method or the like. Is generated and output, the first index selecting section 45, the second index selecting section 47, or the third index selecting section 49.
In selecting the index data corresponding to the non-embedded position of the watermark information in (1), not only the first candidate index data having the smallest error but also other index data belonging to a predetermined error range are randomly selected by random numbers. Therefore, even if the difference data is obtained by comparing the bitstream data generated without watermark information and the bitstream data generated with watermark information, information other than the watermark information is compared. The difference data that is included in a complicated state can be obtained, making it difficult to extract the watermark information, and bits that include stronger watermark information against malicious attacks such as falsification or deletion of the watermark information. Stream data can be obtained. [4.4] Modification of Fourth Embodiment In the above description, as in the case of the first embodiment,
The case where index data is randomly selected by random numbers has been described, but the same mode as in the second and third embodiments can be applied to multistage vector quantization.

[5] Fifth Embodiment In the above-described first to third embodiments, the basic vector quantizer was described as an example, but the fifth embodiment is a vector quantizer. This is an embodiment in which a conjugate vector quantization device is used. [5.1] Configuration of Conjugate Vector Quantization Device FIG. 10 shows a configuration block diagram of the conjugate vector quantization device. The conjugate vector quantizer 60 uses the input signal data D
A waveform cutout device 61 that cuts out a waveform of IN and outputs it as cutout waveform data DP, and an index is generated based on the cutout waveform data DP, and bitstream data DBS2
And an index generation unit 62 for outputting. The index generation unit 62 stores a first codebook 65 that stores the first index data DCB1 having the first system and a second code data 65 that stores the second index data DCB2 having a second system different from the first system. Among the combinations of the codebook 66 and the first index data DCB1 and the second index data DCB2 corresponding to the same cut-out waveform data DP, among a plurality of combinations in which the amount of error is included in a predetermined error range determined in advance. A combination selection in which any one of the combinations is selected by a random number and the first index data DCB1 and the second index data DCB21 corresponding to the selected combination are output as the first selection index data DIDX21 and the second selection index data DIDX22. Part 66
And a multiplexer unit 68 for generating and outputting bit stream data DBS2 by a time division method or the like based on the first selection index data DIDX21 and the second selection index data DIDX22.

[5.2] Operation of Conjugate Vector Quantization Device The waveform cutting device 61 of the conjugate vector quantization device 60 performs waveform cutting of the input signal data DIN and cut-out waveform data DP.
To the index generation unit 62. The combination selection unit 66 of the index generation unit 62 uses the first index data DC corresponding to the input cut-out waveform data DP.
Among the combinations of B1 and the second index data DCB2, any one of a plurality of combinations in which the amount of error is included in a predetermined error range is selected by a random number and corresponds to the selected combination. The first index data DCB1 and the second index data DCB21 are set to the first selected index data DIDX21.
And output to the multiplexer unit 68 as the second selection index data DIDX22. As a result, the multiplexer unit 68 generates and outputs the bit stream data DBS2 based on the first selection index data DIDX21 and the second selection index data DIDX22 by a time division method or the like.

[5.3] Effects of the Fifth Embodiment As described above, the amount of error in the combination of the first index data DCB1 and the second index data DCB2 corresponding to the cut-out waveform data DP is preset. Since it is configured to select any one combination from a plurality of combinations included in the predetermined error range by random numbers, it is possible to generate the bitstream data without including the watermark information and the watermark information. Even if the difference data is obtained by comparing with the bitstream data generated by including, the difference data including information other than the watermark information will be obtained, making it difficult to extract the actual watermark information. Bitstream data containing stronger watermark information can be used against malicious attacks such as falsification or deletion of watermark information. Rukoto can.

[5.4] Effect of Fifth Embodiment In the above description, when selecting the combination of the first index data DCB1 and the second index data DCB2 corresponding to the cut-out waveform data DP, the combination is randomly selected. However, as in the case of the second and third embodiments, when obtaining the cut-out waveform data DP, by including the dummy data, it becomes more difficult to extract the watermark information. can do.

[0045]

According to the present invention, the bitstream data generated without the watermark information and the bitstream data generated with the watermark information are compared to obtain the difference data. Even in this case, it is possible to obtain difference data that includes information other than the watermark information in a complicated manner, making it difficult to extract the watermark information and against malicious attacks such as falsification or deletion of the watermark information. It is possible to obtain bitstream data including stronger watermark information, and it is possible to accurately grasp illegal use of information after vector quantization.

[Brief description of drawings]

FIG. 1 is a schematic configuration block diagram of a vector quantization device according to a first embodiment.

FIG. 2 is a detailed configuration block diagram of an index generation unit 12 of the first embodiment.

FIG. 3 is a diagram illustrating the configuration of bitstream data.

FIG. 4 is a schematic configuration block diagram of a vector decoding device.

FIG. 5 is a diagram illustrating dummy data.

FIG. 6 is a processing flowchart of the second embodiment.

FIG. 7 is a processing flowchart of a third embodiment.

FIG. 8 is a processing flowchart of a first modified example of the third embodiment.

FIG. 9 is a schematic configuration block diagram of a multistage vector quantization device according to a fourth embodiment.

FIG. 10 is a schematic configuration block diagram of a conjugate vector quantization device of a fifth embodiment.

FIG. 11 is a specific explanatory diagram of cutout waveform data.

[Explanation of symbols]

10 ... Vector quantizer, 11 ... Waveform cutting device, 12
... index generation unit, 15 ... vector decoding device, 1
6 ... Decoder, 17 ... Watermark information decoder, 21 ... Embedding position information storage unit, 22 ... Random number generation unit, 23 ... Watermark information storage unit, 24 ... Codebook, 25 ... First encoder, 26 ... Second Encoder, 27 ... Selector, 28-point header information generation unit,
29 ... Bit stream generation unit, 40 ... Multi-stage vector quantization device, 41 ... Waveform cutting device, 42 ... Index generation unit, 45 ... First index selection unit, 46 ... First addition unit, 47 ... Second index selection unit , 48 ... Second adding unit, 49 ... Third index selecting unit, 50 ... Multiplexer unit, 60 ... Conjugate vector quantizer, 61 ... Waveform cutting device, 62 ... Index generating unit, 65 ... First codebook, 66 ... Combination selection section, 67 ... Second codebook, 68 ... Multiplexer section, DIN ... Input signal data, DP ... Waveform cutout data, DBS ... Bitstream data, DFP ... Watermark embedding position information data, DWM ... Watermark information data , DHD ... Header information data, DIDX1 ... First index data, DIDX2 ... Second index data,
DRD ... random number data.

─────────────────────────────────────────────────── ─── Continuation of the front page (56) Reference JP-A-11-316599 (JP, A) JP-A-11-272299 (JP, A) JP-A-11-205153 (JP, A) JP-A-10- 224342 (JP, A) JP 10-65912 (JP, A) JP 10-112657 (JP, A) JP 11-259068 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) H03M 7/30 G09C 5/00 G10L 19/00

Claims (14)

(57) [Claims] 1. A vector quantizer for vector quantizing input signal data, wherein waveform cutout data obtained by performing waveform cutout of the input signal data is a waveform cutout that is not a target for embedding watermark information. When selecting the index data for the output data, one of the plurality of index data whose error with the original waveform cutout data is within a predetermined error range when decoded is used as the waveform cutout data. A vector quantization device comprising random selection means for randomly selecting corresponding index data. 2. A vector quantizer for quantizing a conjugate vector of input signal data, wherein the waveform of the input signal data, which is the waveform cut-out data obtained by cutting the waveform of the input signal data, is a non-embedding target of watermark information. When selecting a combination of index data for cutout data, any one of the index data combinations that has an error with the original waveform cutout data within a specified error range when decoded 5. The vector quantizer according to claim 1, further comprising a combination random selection unit that randomly selects the combination of the above as a combination of index data corresponding to the waveform cutout data. 3. The vector quantization device according to claim 1, further comprising a dummy data adding unit that adds dummy data to a part of the waveform cutout data when the waveform cutout data is generated. A vector quantization device characterized by the above. 4. The vector quantizing device according to claim 1, further comprising: first vector quantizing means to n-th vector quantizing means (n: integer of 2 or more) for performing multistage vector quantization. Characteristic vector quantizer. 5. The vector quantization apparatus according to claim 4, wherein at least one vector quantization means of the first vector quantization means to the n-th vector quantization means,
A vector quantization apparatus, comprising: dummy data adding means for adding dummy data to a part of the waveform cutout data when the waveform cutout data is generated. 6. The vector quantizing device according to claim 3 or 5, wherein the waveform cut-out data is composed of a predetermined finite number of sample data, and the dummy data adding means is provided for the input signal data. Is divided into divided data composed of a predetermined number of sample data smaller than the finite number, the dummy data is added to each of the divided data, and the waveform cutout composed of the finite number of sample data is divided. A vector quantization device characterized by generating data. 7. The vector quantization device according to claim 6, wherein the number of the sample data in the dummy data is the division data of the watermark information embedding target and the division data of the watermark information non-embedding target. Vector quantizer characterized in that it is different. 8. The vector quantization device according to claim 6, wherein as the dummy data, sample data having a predetermined constant value is converted into the divided data until the number of sample data forming the divided data becomes the finite number. A vector quantizer characterized by being repeatedly added. 9. The vector quantizing device according to claim 6, wherein the number of sample data forming the waveform cutout data, which is obtained by performing a predetermined calculation based on the divided data, as the dummy data, A vector quantization device characterized by using operation data composed of a number of sample data corresponding to a difference from the number of sample data forming the divided data. 10. The vector quantization device according to claim 9, wherein the calculation data is calculated by extrapolation calculation. 11. The vector quantization device according to claim 9, wherein the calculation data is calculated by interpolation calculation. 12. A vector quantization method for vector-quantizing input signal data, wherein the waveform cut-out data obtained by performing waveform cut-out of the input signal data is a waveform cut-off target of non-embedding of watermark information. When selecting the index data for the output data, one of the plurality of index data whose error with the original waveform cutout data is within a predetermined error range when decoded is used as the waveform cutout data. A vector quantization method comprising a random selection step of randomly selecting corresponding index data. 13. A vector quantization method for quantizing input signal data by conjugate vector quantization, wherein waveform cut-out data obtained by performing waveform cut-out of the input signal data is a waveform to which watermark information is not embedded. When selecting a combination of index data for cutout data, one of the combination of a plurality of index data in which the error with the original waveform cutout data is within a predetermined error range when decoded A vector quantization method comprising a combination random selection step of randomly selecting a combination of index data corresponding to waveform cutout data. 14. The vector quantization method according to claim 12, further comprising a dummy data adding step of adding dummy data to a part of the waveform cutout data when the waveform cutout data is generated. A vector quantization method characterized by the above.