JP3672143B2 - How to create an electronic watermark - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention simultaneously transmits and accumulates represent audio and images digital code to prevent unauthorized use by embedding the secret information (digital watermark information) on the bit sequence relates to the so-called electronic watermark creator how.
[0002]
[Prior art]
Specifically, the electronic watermark provides a mechanism that simultaneously satisfies the following two conditions.
(1) Normal decoders can reproduce audio and images, but cannot find where the secret information (electronic watermark information) is written, and if you try to rewrite or delete the secret information, The reproduced sound and image are greatly degraded.
[0003]
(2) On the other hand, the inspection decoder can read out the secret information by searching where the secret information (electronic watermark information) is embedded using another encrypted information.
The conventional digital watermark creation method selects a position to write electronic watermark information (secret information) from bit positions whose quality hardly changes when reproduced from an information compression bit sequence such as an image. A method for writing watermark information is known. However, if the information compression rate is high, or if the bit sequence of a bit sequence that has already been encoded is forcibly rewritten in a compressed bit sequence of speech or music, a bit position that often causes large distortion and does not change the quality. Is very limited.
[0004]
[Problems to be solved by the invention]
The purpose is to provide an electronic watermarking how it is possible to minimize the increase of quantization distortion of the original sound and images when writing a digital watermark information in the present invention.
[0005]
[Means for Solving the Problems]
According to the first aspect of the present invention, an input vector, electronic watermark information, and embedded position designation information for designating a bit position in which the electronic watermark information is to be embedded are input.
For a portion where embedding is not designated by the embedding position designation information, an index of a quantization vector that minimizes distortion on a predetermined scale with the input vector is obtained from the codebook, and the index is converted to the digital Output as a sign,
For the portion where embedding is designated by the embedding position designation information, the bit of the index designated for embedding in the codebook is fixed to the electronic watermark information, and the quantization vector in the fixed index The input vector with the smallest distortion on a predetermined scale is selected, and the index is output as the digital code.
[0006]
According to the invention of claim 2, a digital code string composed of a vector quantized index, electronic watermark information, and embedded position designation information for designating a bit position in which the electronic watermark information is to be embedded, are input.
For the portion where embedding is not designated by the embedding position designation information, the input digital code is output as it is,
For the portion where embedding is designated by the embedding position designation information, the bit of the index designated for embedding of the codebook used for the vector quantization is fixed to the electronic watermark information, and the fixed index From among the quantized vectors, the one that minimizes the distortion on a predetermined scale between the input index and the corresponding quantized vector is selected, and the index is output.
[0007]
According to a third aspect of the present invention, in the second aspect of the present invention, with respect to a portion for which embedding is designated by the embedding position designation information, the input digital code is decoded by the codebook used for the vector quantization and quantized. A vector is obtained, and a quantization vector that minimizes the distortion is selected from the quantization vector and the quantization vector in the fixed index.
[0008]
According to a fourth aspect of the present invention, in the second aspect of the present invention, with respect to a portion for which embedding is designated by the embedding position designation information, distortion with other quantization vectors is previously determined for each quantization vector of the codebook. An index of the codebook of the input digital code is obtained from the obtained table, and a distortion between this and the index in the codebook specified by embedding and fixed by the watermark information is obtained from the table, Find its smallest index.
[0009]
Than the mentioned manner rather than once embedding independent electronic watermark information and the input vector to the encoded bit sequence in the present invention, in the codebook search for vector quantization, watermarked information It is characterized by searching for an optimal index under conditions, and an increase in distortion can be minimized.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1A is a block diagram showing an embodiment corresponding to claim 1 of the present invention. In this example, the quantization vector is searched in consideration of watermark information at the stage of processing of the encoder 11. That is, as shown in FIG. 1B, an input vector such as an image signal and an acoustic signal from the input terminal 12 is input, and the difference between the decoded vector selected from the codebook 13 and decoded by the decoder 14 is added. A quantization vector that is taken by the means 15 and has a minimum distortion with respect to the decoded vector and the input vector at a scale determined in advance by the distortion calculation control unit 16 is selected from the codebook 13, and indicates the selected quantization vector. The index In of the codebook 13 is output from the encoding output terminal 17.
[0011]
In this embodiment, watermark information is input from the terminal 18 to the encoder 11 and embedded position designation information is input from the terminal 19. When embedding is not specified by the embedding position designation information, the input vector is encoded in the same way as before as described above, but when embedding of watermark information is designated by the embedding position designation information, The embedded designation position in the index of the codebook 13 is fixed with the watermark information at that time, and the quantization vector of the codebook 13 is searched. That is, as shown in FIG. 1B, in the selection control unit 21, the quantization vector to be searched from the codebook 13 is limited by the embedded position designation information and the watermark information.
[0012]
For example, as shown in FIG. 2A, the bit at the position designated by the embedding position designation information shown in FIG. 2B in the encoded bit series of the indexes In1, In2,. The In this example, the index In1 is originally searched for a quantized vector having the smallest distortion from among the 5-bit 32 vectors, but the third bit is fixed to 0 by the watermark information by the embedded position designation information. Accordingly, as shown in FIG. 2D, the one with the smallest distortion is selected from the quantization vectors (16 in total) for the index whose third bit in the codebook 13 is 0.
[0013]
The one selected from the 16 may be the same as the one with the least distortion of the 32, but the one selected from the 16 is the optimal quantization. On average, the difference from the vector is very small. In other words, when the encoded bit sequence in which the watermark information is embedded is reproduced by a normal decoder, the difference from the original output is small.
[0014]
In addition, it is difficult for a normal decoder to detect which bits are fixed and searched. This is because even if the encoded output bits are inverted and reproduced one bit at a time, the reproduction output deteriorates depending on the configuration of the codebook 13, but the bit of the watermark information deteriorates in the same way. This is because there is no characteristic only with bits.
In the conventional method, since the watermark information is written by inverting a part of the coded bit without considering the input, the quality may be improved by inverting the bit. Therefore, it is possible to detect the watermark by detecting the bit whose quality is improved, but in the embodiment of the present invention, there is very little possibility that the quality is improved by inversion of only one bit. Certainly, in the example of this figure, changing the index for all 32 combinations may improve the quality, but the quality improvement is negligible, so the improved index and the encoded index Even if the comparison is made, the position of the watermark information and the information cannot be specified. Also, if there is no input original sound or original picture information, there is no means to inspect whether the quality has been improved objectively, and it is difficult to falsify the watermark information.
[0015]
The above-mentioned alteration of watermark information is based on the assumption that the normal decoder operation, that is, the decoding procedure, the decoding codebook, and all the bit positions are made public, that is, all quantizations. It is assumed that the relationship between the vector and its index is known. However, if there is an undisclosed bit even in a part of the codebook, it can be considered that the watermark information cannot be falsified. For example, it is conceivable to add an error detection code (for example, a parity bit) for a specific secret bit and combine the process of stopping decoding when an error is detected there with a normal decoder. It is possible to accidentally replace the index of the watermark information part with another index with little deterioration, that is, even if it can be tampered with, it can be detected with an error detection code, the output can be stopped, and the tampered Can not play.
[0016]
As described above, in this embodiment, electronic watermark information can be embedded in the encoded bit sequence, and the encoded bit sequence in which the electronic watermark information is embedded is output to the output terminal 17. After the coded bit sequence with watermark information is transmitted or accumulated, it is decoded by a normal decoder 24, that is, by a decoder 24 for a coded bit string in which watermark information is not embedded, and is transmitted from its output terminal 25. However, regardless of the addition of information, an output vector with little distortion degradation can be obtained.
[0017]
The watermark information is input to the watermark information decoder 26 to extract the watermark information from the coded bit sequence, and the watermark information is coded from the coded bit sequence with watermark information by the embedded position designation information obtained from the encoder 11 side. However, the information is obtained at the output terminal 27 by taking out only the information bits. The embedded position designation information is usually encrypted so that only special information can be obtained, and the embedded position designation information is decrypted by the watermark information decoder 26 side to detect the embedded bit position.
[0018]
FIG. 3A shows an embodiment corresponding to the invention of claim 2.
In this example, an encoded bit sequence encoded regardless of watermark information is input to the input terminal 31. In other words, the encoded bit sequence encoded by the normal encoder 11 is transmitted, or the accumulated bit sequence is read out and input from the input terminal 31 to the watermark information embedder 32, so that there is no information on the original input vector. Information is written. At this time, in order to suppress degradation of the original voice and image, the quantization vector corresponding to the index in which the watermark information bits in the codebook are fixed is the index closest to the original reproduction vector. replace. In other words, the watermark information embedder 32 includes the same codebook as the codebook used when the encoded bit sequence input to the input terminal 31 is encoded, and reproduces the input encoded bit sequence. As described with reference to FIG. 1, the reproduced vector is a quantized vector selected from the index in which one bit of the codebook index is fixed with the watermark information of the terminal 18 based on the embedded position designation information of the terminal 19 Output the index of the medium distortion.
[0019]
Alternatively, when the determined distortion scale is constant, for example, as shown in FIG. 3B, for each quantization vector of each index In1, In2,..., Distortion between this and the quantization vector of each other index is calculated. Each calculated distortion table is prepared, and a set of indexes that can be selected for each combination of embedding position designation information and watermark information is created as a selectable table as shown in FIG. From the watermark information and the embedding position designation information, a set of selectable indexes is detected with reference to the selectable index table of FIG. 3C, and the index in the detected selectable index set is input to the terminal 31. The index with the smallest distortion with respect to the index may be searched with reference to the distortion table of FIG. 3B, and the index may be output to the output terminal 17. Alternatively, for each input index, an index with minimum distortion may be obtained in advance for each case of watermark information and its embedding position, and this may be output immediately.
[0020]
In this case, distortion caused by embedding watermark information is larger than in the case of the embodiment shown in FIG. 1 using input information, but it does not deteriorate greatly and can be decoded by a normal decoder. Further, the possibility of improving the quality by replacing the index with a normal decoder is slightly greater than in the case of FIG. However, since it is not necessary to encode each time in order to insert watermark information, the processing becomes simple.
[0021]
Note that when the distortion measure depends on other parameters in the coded bits, for example, when auditory weighted vector quantization is performed, the distortion measure is calculated by decoding the parameter, and the closest quantum in the measure is calculated. It is necessary to replace with the vector vector index.
[0022]
【The invention's effect】
According to the present invention, the following electronic watermark function can be realized, and deterioration from the original sound or image can be suppressed to a slight level.
(1) Normal decoders can reproduce audio and images, but it is difficult to find where secret information (watermark information) is written. If you try to rewrite or delete the secret information, And the image is greatly degraded.
[0023]
(2) On the other hand, the inspection decoder can read out the secret information by searching where the information is embedded by using another encrypted embedded position designation information.
[Brief description of the drawings]
FIG. 1A is a block diagram showing an embodiment of the invention of claim 1 and an embodiment of the invention of claim 5, and B is a functional block diagram showing a specific example of the encoder 11. FIG.
FIG. 2 is a diagram showing an encoded bit sequence, embedded position designation information, and watermark information, respectively, and D is a diagram showing an example of a codebook.
FIG. 3A is a block diagram showing an embodiment of the invention of claim 2, B is a diagram showing a distortion table, and C is a diagram showing a table of selectable index sets.

Claims (4)

In the method of creating an electronic watermark that digitally encodes an input vector of sound or an image and embeds electronic watermark information in the digital code.
Input the input vector, the electronic watermark information, and embedding position designation information for designating a bit position in which the electronic watermark information is to be embedded,
For a portion where embedding is not designated by the embedding position designation information, an index of a quantization vector that minimizes distortion on a predetermined scale with the input vector is obtained from the codebook, and the index is converted to the digital Output as a sign,
For the portion where embedding is designated by the embedding position designation information, the bit of the index designated for embedding in the codebook is fixed to the electronic watermark information, and the quantization vector in the fixed index A method for creating an electronic watermark comprising: selecting an input vector having a minimum distortion on a predetermined scale and outputting the index as the digital code. In an electronic watermark creation method for embedding electronic watermark information in a digital code string composed of vector quantized indexes,
Input the digital code string, the electronic watermark information, and embedding position designation information for designating a bit position in which the electronic watermark information is to be embedded,
For the portion where embedding is not designated by the embedding position designation information, the input digital code is output as it is,
For the portion where embedding is designated by the embedding position designation information, the bit of the index designated for embedding of the codebook used for the vector quantization is fixed to the electronic watermark information, and the fixed index A method for creating an electronic watermark characterized by selecting from the quantized vectors among them the one that minimizes distortion on a predetermined scale between the input index and the corresponding quantized vector, and outputting the index .   For the portion where embedding is designated by the embedding position designation information, the input digital code is decoded with the codebook used for the vector quantization to obtain a quantized vector, and this quantized vector and the fixed 3. The method of creating an electronic watermark according to claim 2, wherein a quantization vector that minimizes the distortion is selected from the quantization vectors in the index.   For the portion where embedding is designated by the embedding position designation information, for each quantization vector of the codebook, the code of the input digital code is obtained from a table in which distortion with other quantization vectors is obtained in advance. The index of the book is obtained, and the distortion between this and the index in the codebook specified by the embedding and fixed by the watermark information is obtained from the table, and the minimum index is obtained. 2. A method for creating an electronic watermark according to item 2.

Images