JP4718415B2 - Digital watermark embedding apparatus and digital watermark embedding method - Google Patents

Description

  This invention reduces leakage of the base of frequency spreading by embedding digital watermarking using frequency spreading and quantization by changing the quantization characteristics according to the quantization width and the probability density distribution of the input signal. The present invention relates to a digital watermark embedding apparatus and method.

  Conventionally, as a method of embedding information in an image or music signal with a digital watermark, a method in which a part of the input signal is replaced with another value depending on the information by quantization or the like, or a signal in which information is frequency-spread into the input signal The method of adding is representative. On the other hand, Non-Patent Document 1 proposes a method combining these two. In the method described in Non-Patent Document 1, when the signal sequence of the input signal is x (i) (i = 1,..., N), first, the input signal x (i) is frequency-converted by the following equation (1). The spread signal y is obtained.

  Here, u (i) (= {− 1, + 1}) is key information shared by information senders and receivers.

Next, the frequency-spread signal y is quantized and changed to _yw . Here, y _w depends on the information to be embedded. For example, when the information is 1 bit and the value is 0, y _w is expressed by the following equation (2).

When the information is 1 bit and the value is 1, y _w is expressed by the following equation (3).

Here, m is an integer, Δ is a quantization width, and sgn (y) is a function that is −1 when y <0, and 1 otherwise. This change is actually performed by appropriately changing the signal sequence x (i). Although there is a degree of freedom in changing the signal sequence x (i), in the following, the simplest case is assumed to be performed by adding / subtracting a constant value δ to all signal sequences x (i). Even with this assumption, the following argument holds. Also, m is selected to minimize | y−y _w |, that is, the error when changing y to y _w . Here, Non-Patent Document 1, in order to achieve the information-theoretic channel capacity, without changing the y to necessarily y _w, it has been set to a value close signal after the change is y _The distribution is centered around _w and does not affect the following discussion. The receiver of information recalculates the transmitted information by calculating again Equation (1) from the received signal and determining whether it is closer to Equation (2) or Equation (3).

Brian Chen and Gregory W. Wornell, “Achievable Performance of Digital Watermarking Systems,” Proceedings of IEEE International Conference on Multimedia Computer Systems, vol. 1, Florence, Italy, June 1999, pp. 13-18.

  A conventional digital watermark embedding method as described above is a signal in which a digital watermark is embedded by a third party (hereinafter referred to as an attacker) who does not know the key information u (i) other than the sender and receiver of information. There is a problem in that a part of the key information u (i) can be known by analyzing (hereinafter referred to as a stego signal).

  Here, the fact that a part of the key information u (i) is known means that the attacker adds a signal obtained by inverting the estimated key information u (i) to the stego signal and embeds it with a digital watermark from the attacked signal. This means that information cannot be detected, and the purpose of digital watermarking, that is, transmission of information between senders and receivers, cannot be achieved.

  The attacker's strategy for this is as follows. First, it is assumed that the attacker knows the set X that is a candidate for the variable x (i). That is, it is assumed that the digital watermark is embedded by the above-described process using N variables selected from the public set X based on the key shared by the sender and the receiver. A similar argument holds if the set X can be estimated even if it is not disclosed. Let M = | X |

  Now, since the attacker does not know which variable has been selected, m is selected from the M variables, and random noise is added to the variable. At this time, if the value of the variable is positive, the sign of the noise is negative, and if not, the noise value is positive. That is, the attacker always adds noise in the direction of decreasing the absolute value of the variable. Such noise is hereinafter referred to as “directional noise”. As shown below, an attacker can cause a large detection error only by adding directional noise to some of the variables of X. Here, for comparison, noise added with the same magnitude regardless of the sign of the variable will be referred to as “random noise”.

  FIG. 7 is a graph showing experimental data of directional noise and random noise in the conventional digital watermark embedding method. Here, the host signal (original signal to be embedded with the digital watermark: host signal) is a Gaussian signal having an average of 0 and a variance of 100, and the digital watermark is embedded with N = 64 and quantization width Δ = 32. The attack is performed by adding average 0 Gaussian noise to all the variables, and the dispersion value (distortion) is as shown on the horizontal axis of the graph. However, in the case of directional noise, the sign is reversed from that of the variable. When this experiment is performed 10,000 times, the number of errors of detection (number of errors) is on the vertical axis. In this graph, the solid line is experimental data for random noise, and the dotted line is experimental data for directional noise. It can be seen that when the noise power is the same, the detection error rate is larger when directional noise is added.

  The effect of the directional noise is considered to be caused by the difference in probability distribution between the signal changed in the positive direction and the signal changed in the negative direction due to the digital watermark. This can be considered that information about the key, which is a key code, is partially leaked and has been deciphered by a technique called directional noise. For example, FIG. 8 is a graph showing a difference in signal distribution caused by digital watermark embedding. The solid line is the distribution of variable values that have changed in the negative direction due to the embedding of the digital watermark, and the dotted line is the distribution of variable values that have changed in the positive direction. Since these distributions are different, it can be easily imagined that some of the signal changes given by the digital watermark will be canceled if noise is added in the direction of decreasing the absolute value of the variable. Since the variables are selected at random, the two distributions should match if no digital watermark is embedded.

  The reason why such a difference in distribution occurs is as follows. Gaussian noise (or its linear combination as in equation (1)) is concentrated and distributed at 0, but when such a variable is roughly quantized, the value has a probability of changing in the direction of increasing absolute value. high. FIG. 9 is a diagram illustrating the quantization characteristics when the quantization width Δ = 32. In FIG. 9, the triangular wave schematically represents the distribution of input signals (probability distribution density: pdf (y)). Here, for example, regarding the host signal y quantized to -8, the distribution of signals larger than -8 (absolute value is smaller than -8) and signals that are not so is biased. There are more variables with increasing values. This can be said for all the quantized values in FIG. 9, and if such quantization is performed, the absolute value of the signal after quantization increases on average than before the quantization.

  However, when the absolute value of y is amplified, the difference in signal distribution after embedding a digital watermark is amplified as shown in FIG. FIG. 10 is a diagram showing a difference in signal distribution caused by embedding the digital watermark after amplification of the absolute value of y. In FIG. 10, Xp represents the distribution of variables such that u (i) · x (i)> 0, and Xm represents the distribution of variables such that u (i) · x (i) <0. Xp and Xm (solid lines) represent the distribution of variables before embedding the digital watermark, while Vp and Vm (dotted lines) represent the distribution of variables after embedding the digital watermark. From FIG. 10, in order to amplify the absolute value of y regardless of the positional relationship between Xp and Xm, the signal distributed on the right side is always changed in the positive direction, and the signal distributed on the left side is changed in the negative direction. It can be seen that the difference in distribution is further facilitated by the embedding of the digital watermark. 10A shows a case where Xp is distributed to the right (positive direction) from Xm, and FIG. 10B shows the opposite case.

  As described above in detail, according to the conventional digital watermark embedding method, there is a high probability that the absolute value of y is increased by the quantization due to the bias of the input signal distribution. There is a difference between the distribution of the signal and the signal distribution changing in the negative direction, which causes a problem that a part of the key information leaks.

  In this case, for example, the detection error of the digital watermark can be increased by a simple method such as directional noise. This kind of key leakage means that if you look at the value of a variable with an embedded digital watermark, you can probabilistically know whether it has changed in the positive or negative direction due to the watermark. It is. At this time, it should be noted that if an individual variable is viewed, a strong attack is possible if many variables are combined even if it cannot be determined in which direction the variable has changed.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to obtain a digital watermark embedding apparatus and method that is difficult to detect the direction in which a variable has changed and is resistant to directional noise.

  An electronic watermark embedding device according to the present invention comprises: spreading means for frequency-spreading an input signal; quantization means for quantizing the frequency-spread input signal; and quantization of the quantization means according to an input quantization width And a control means for changing the characteristics.

  The digital watermark embedding device according to the present invention has an effect that the direction in which the variable has changed is not easily detected and is resistant to directional noise.

Embodiment 1 FIG.
A digital watermark embedding apparatus according to Embodiment 1 of the present invention will be described with reference to FIGS. FIG. 1 is a block diagram showing a configuration of a digital watermark embedding apparatus according to Embodiment 1 of the present invention. In the following, in each figure, the same reference numerals indicate the same or corresponding parts.

  In FIG. 1, a digital watermark embedding apparatus 100 according to the first embodiment includes a spreading means 10 for frequency-spreading an input signal from an input terminal 1, and a quantum for quantizing the output of the spreading means 10 and outputting the output from an output terminal 2. And a control unit 30 that changes the quantization characteristic of the quantization unit 20 based on the quantization width input from the input terminal 3.

  Next, the operation of the digital watermark embedding apparatus according to the first embodiment will be described with reference to the drawings.

  FIG. 2 is a diagram showing the quantization characteristics when the quantization width Δ = 32 and the parameter b = Δ / 4 of the digital watermark embedding apparatus according to Embodiment 1 of the present invention. In FIG. 2, the horizontal axis represents the host signal y, and the vertical axis represents the frequency of y. The triangular wave schematically represents the distribution of input signals (probability distribution density: pdf (y)). FIG. 3 is a graph showing experimental data of directional noise and random noise in the digital watermark embedding apparatus according to Embodiment 1 of the present invention. In FIG. 3, the horizontal axis represents distortion and the vertical axis represents number of errors.

  First, the spreading means 10 performs a linear combination (frequency spread signal) on the input signal x (i) (i = 1,..., N) inputted from the input terminal 1 based on the equation (1). Calculate y.

  Next, the quantization means 20 quantizes the frequency-spread signal (host signal) y and outputs it from the output terminal 2. The point of embedding information is the same as that of the conventional example, except that the characteristics are variable. The quantization characteristic of the quantization means 20 is given as shown by the following formula (4), for example.

Here, the right side of the equation (4) is the quantization characteristic of the quantization means 20, b is a parameter for making the quantization characteristic variable by the control means 30, and y _w (output of the quantization means 20) is quantized. This is for shifting the range of the value of y (the input of the quantizing means 20).

  In the quantization characteristic shown in FIG. 2, for example, variables from −32 to +16 are quantized to “−8”. Further, the variables from +16 to +48 are quantized to “+24”. Here, it should be noted that the value after quantization (referred to as a quantization representative value) is not the center of the range (quantization range) of the input signal quantized there. This shift is given by the parameter b.

  Looking at the variables quantized to “−8”, the variables from −32 to −8 and +8 to +16 have larger absolute values than “−8”. Further, the variables from -8 to +8 have smaller absolute values than "-8". Accordingly, the absolute values of the variables from −32 to −8 and +8 to +16 are decreased by quantization, and the absolute values of the variables from −8 to +8 are increased. In FIG. 2, the range of the variable whose absolute value decreases is displayed in gray (shaded).

  Compared with the conventional quantization characteristic of FIG. 9, it is clear that in the quantization characteristic of FIG. 2, the frequency of the variable in the gray range increases and the frequency of the variable other than that decreases. This can also be said for all other quantized representative values, and it can be seen that the increase in the absolute value of y after digital watermark embedding is suppressed on average by using the quantization characteristic of equation (4). . As described above, the increase in absolute value promotes the difference in the signal distribution that is changed in the opposite direction, and causes a decrease in the resistance to directional noise. The quantization characteristic shown in FIG. 2 suppresses an increase in absolute value, so that an electronic watermark embedding resistant to directional noise can be performed.

  FIG. 3 shows experimental data demonstrating this effect. The graph of FIG. 3 shows the number of errors (vertical axis) when an electronic watermark is embedded and detected under the same conditions as in the conventional example of FIG. However, the quantization characteristic is expressed by equation (4), and the parameter b = Δ / 4. It can be seen that the dotted line experimental data (the number of errors with respect to directional noise) has decreased to the same level as the solid line of experimental data (the number of errors with respect to random noise).

  As can be seen from the above explanation, in order to obtain the above effect, it is necessary to suppress the case where the absolute value increases, and to determine the quantization characteristic so as to balance the case where the absolute value decreases. I know it ’s good. The control means 30 is provided for this purpose, and the sender gives a quantization width Δ from the input terminal 3 and changes the parameter b (= Δ / 4) according to the quantization width Δ. It is a thing.

  In general, if the quantization width Δ is small (if the quantization is fine), the difference between the increase and decrease in the absolute value is small, so the value of the parameter b may be small. The quantization width Δ is a parameter that gives the strength of the digital watermark. When the quantization width Δ is increased, the strength increases (information is not easily destroyed), but the deterioration (quantization distortion) due to the digital watermark embedding increases. By providing the control means 30, it is possible to always select an optimum quantization characteristic in terms of key security for a wide range of watermark strengths.

  As described above, according to the first embodiment, since the absolute value of the variable is prevented from increasing on average in quantization, leakage of the key for frequency spreading is prevented, and digital watermark embedding that is strong against directional noise There is an effect that can be performed.

  In the above description, the parameter b is described as being the same for all quantized representative values. However, the value may be changed according to the quantized representative value. In addition, the effect of the first embodiment can be obtained by optimizing the quantization characteristic and suppressing the increase in the absolute value of the variable after quantization. The quantization characteristic is limited to the equation (4). It is not something that can be done.

  In the above description, the input signal of the diffusing unit 10 is a signal to be embedded with a digital watermark. However, this signal may be a converted signal or a part thereof. For example, a transform coefficient of DCT (Discrete Cosine Transform) or wavelet transform and a part thereof are often used for embedding a digital watermark. Since the distribution of such variables is concentrated at 0 as in the case of the Gaussian signal described above, the experimental effects shown in FIGS. 3 and 7 can be obtained similarly.

  In addition, in the first embodiment, the correspondence relationship between the input signal and the quantized representative value is changed, and the (secret) information is transmitted by the quantized representative value. In the case of detecting information from a signal in which a digital watermark is embedded in the first embodiment, a conventional detector (not shown) on the receiver side can be used as it is.

Embodiment 2. FIG.
A digital watermark embedding apparatus according to Embodiment 2 of the present invention will be described with reference to FIG. FIG. 4 is a block diagram showing the configuration of the digital watermark embedding apparatus according to Embodiment 2 of the present invention.

  In FIG. 4, the digital watermark embedding apparatus 100A according to the second embodiment has a spreading means 10 for frequency-spreading an input signal from the input terminal 1, and a quantum for quantizing the output of the spreading means 10 and outputting it from the output terminal 2. , Means 20 for calculating the statistical distribution (probability distribution density) of the input signal from the input terminal 1, and means for quantizing based on the output of the analyzing means 40 and the quantization width inputted from the input terminal 3. 20 control means 30A for changing the characteristics of 20 is provided.

  Next, the operation of the digital watermark embedding apparatus according to the second embodiment will be described with reference to the drawings.

  When the distribution of the input signal is known in advance, the control means 30 may determine the quantization characteristic based only on the quantization width Δ as in the first embodiment. However, if the distribution of the input signal is unknown, an electronic watermark embedding method corresponding to a wide range of signals can be obtained by providing an analysis means 40 for calculating the distribution of the input signal.

  The operation of the digital watermark embedding apparatus according to the second embodiment is the same as that of the digital watermark embedding apparatus according to the first embodiment, except that the analysis unit 40 is provided. The analysis unit 40 calculates the distribution of the input signal. For example, the analysis unit 40 may measure a histogram (frequency) of the signal input in a certain period. This result is given to the control means 30A. The control unit 30A optimally adjusts the characteristics of the quantization unit 20 in accordance with the quantization width input from the input terminal 3. That is, the control unit 30A determines the parameter b based on the quantization width Δ and the statistical distribution (probability distribution density) of the input signal. This adjustment is the same as in the first embodiment, and the increase in absolute values of variables after quantization may be suppressed.

  As described above, in the second embodiment, since the analysis means 40 for calculating the distribution of the input signal is provided, it is possible to prevent the leakage of the key of the frequency spread corresponding to a wide range of signals, and to resist the directional noise. There is an effect that watermark embedding can be performed.

Embodiment 3 FIG.
A digital watermark embedding apparatus according to Embodiment 3 of the present invention will be described with reference to FIG. FIG. 5 is a block diagram showing a configuration of a digital watermark embedding apparatus according to Embodiment 3 of the present invention.

  In FIG. 5, a digital watermark embedding device 100B according to the third embodiment includes a spreading means 10 for frequency spreading an input signal from the input terminal 1, an adder 50 for adding the output of the spreading means 10 and a parameter b, The quantizing means 20A for quantizing the output of the adder 50 and outputting it from the output terminal 2, and the characteristics of the quantizing means 20A are changed based on the input signal from the input terminal 1 and the quantization width inputted from the input terminal 3. Control means 30B for outputting the parameter b to the adder 50 is provided.

  Next, the operation of the digital watermark embedding apparatus according to the third embodiment will be described with reference to the drawings.

  The quantization characteristic of equation (4) has a certain effect on reducing key leakage, as shown in FIG. If this quantization characteristic is explicitly realized, a digital watermark embedding method that is resistant to directional noise can be obtained with a simpler configuration.

  The digital watermark embedding apparatus according to the third embodiment explicitly executes the quantization of Expression (4), and the adder 50 adds the shift parameter b to the frequency-spread input signal y. The output of the adder 50 is quantized by the quantizing means 20A. Since the input has already been shifted, the quantizing means 20A in this case is the normal (explained in the conventional example) ( Any quantization may be used as long as the median of the quantization range is the representative quantization value. In order to realize Equation (4), the value of the shift parameter b must be determined depending on the input signal. Therefore, the input signal from the input terminal 1 is given to the control means 30B together with the quantization width from the input terminal 3. That is, the control unit 30B obtains the parameter b based on the quantization width Δ and the input signal.

  As described above, in the third embodiment, the adder 50 is provided in front of the quantizing means 20A so as to give a shift. Therefore, with a simple configuration, leakage of the key of the frequency spread can be prevented, and the directionality can be improved. This has the effect of embedding digital watermarks that are resistant to noise.

Embodiment 4 FIG.
A digital watermark embedding apparatus according to Embodiment 4 of the present invention will be described with reference to FIG. FIG. 6 is a block diagram showing a configuration of a digital watermark embedding apparatus according to Embodiment 4 of the present invention.

  In FIG. 6, a digital watermark embedding apparatus 100C according to the fourth embodiment includes a spreading means 10 for frequency spreading the input signal from the input terminal 1, an adder 50 for adding the output of the spreading means 10 and the parameter b, Quantizing means 20A for quantizing the output of adder 50 and outputting it from output terminal 2, analyzing means 40 for calculating a statistical distribution (probability distribution density) of an input signal from input terminal 1, There is provided a control means 30C for outputting to the adder 50 a parameter b for changing the characteristics of the quantization means 20A based on the input signal, the output of the analysis means 40 and the quantization width inputted from the input terminal 3.

  Next, the operation of the digital watermark embedding apparatus according to the fourth embodiment will be described with reference to the drawings.

  In the third embodiment, another form can be obtained by further providing an analysis means 40 for calculating the distribution of the input signal.

  The operation of the digital watermark embedding apparatus according to the fourth embodiment is the same as that of the third embodiment described above, except that the analysis means 40 for obtaining the distribution of the input signal is added. The operation of the analyzing unit 40 is the same as that of the analyzing unit 40 of the second embodiment. The control unit 30C obtains the parameter b based on the quantization width Δ, the statistical distribution (probability distribution density) of the input signal, and the input signal.

  In the fourth embodiment, the adder 50 is provided in front of the quantizing means 20A so as to give a shift, and the analyzing means 40 for calculating the distribution of the input signal is provided, so that a wide range of signals can be handled with a simple configuration. Thus, the leakage of the key of the frequency spread can be prevented, and the digital watermark embedding resistant to directional noise can be performed.

It is a block diagram which shows the structure of the digital watermark embedding apparatus based on Embodiment 1 of this invention. It is a figure which shows the quantization characteristic in case the quantization width | variety (DELTA) = 32 and parameter b = (DELTA) / 4 of the digital watermark embedding apparatus based on Embodiment 1 of this invention. It is a graph which shows the experimental data of the directional noise and the random noise in the digital watermark embedding apparatus which concerns on Embodiment 1 of this invention. It is a block diagram which shows the structure of the digital watermark embedding apparatus based on Embodiment 2 of this invention. It is a block diagram which shows the structure of the digital watermark embedding apparatus based on Embodiment 3 of this invention. It is a block diagram which shows the structure of the digital watermark embedding apparatus based on Embodiment 4 of this invention. It is a graph which shows the experimental data of the directional noise and the random noise in the conventional digital watermark embedding method. It is a graph which shows the difference of the signal distribution produced by digital watermark embedding. It is a figure which shows the quantization characteristic in the case of quantization width (DELTA) = 32 in the conventional digital watermark embedding method. It is a figure which shows the difference of the signal distribution produced by the digital watermark embedding after amplification of the absolute value of y.

Explanation of symbols

  1 input terminal, 2 output terminal, 3 input terminal, 10 diffusion means, 20, 20A quantization means, 30, 30A, 30B, 30C control means, 40 analysis means, 50 adder, 100, 100A, 100B, 100C electronic watermark Implantation device.

Claims (8)

Spreading means for frequency spreading the input signal;
A quantization means for quantizing the frequency-spread input signal;
An electronic watermark embedding apparatus comprising: a control unit that changes a quantization characteristic of the quantization unit according to an input quantization width. Spreading means for frequency spreading the input signal;
A quantization means for quantizing the frequency-spread input signal;
Analyzing means for calculating a statistical distribution of the input signal;
A digital watermark embedding apparatus comprising: a control unit that changes a quantization characteristic of the quantization unit according to the statistical distribution and an input quantization width. Spreading means for frequency spreading the input signal;
An adder that adds a constant to the frequency-spread input signal;
Quantization means for quantizing the output of the adder;
A digital watermark embedding apparatus comprising: control means for changing a constant applied to the adder in accordance with the input signal and the input quantization width. Spreading means for frequency spreading the input signal;
An adder that adds a constant to the frequency-spread input signal;
Quantization means for quantizing the output of the adder;
Analyzing means for calculating a statistical distribution of the input signal;
A digital watermark embedding apparatus comprising: control means for changing a constant applied to the adder according to the input signal, the statistical distribution, and the input quantization width. A spreading step for frequency spreading the input signal;
A quantization step for quantizing the frequency-spread input signal;
A digital watermark embedding method comprising: a control step of changing a quantization characteristic in accordance with an input quantization width. A spreading step for frequency spreading the input signal;
A quantization step for quantizing the frequency-spread input signal;
An analysis step of calculating a statistical distribution of the input signal;
And a control step of changing a quantization characteristic in accordance with the statistical distribution and the inputted quantization width. A spreading step for frequency spreading the input signal;
An addition step of adding a constant to the frequency spread input signal;
A quantization step for quantizing the signal after addition;
A digital watermark embedding method comprising: a control step of changing the constant according to the input signal and an input quantization width. A spreading step for frequency spreading the input signal;
An addition step of adding a constant to the frequency spread input signal;
A quantization step for quantizing the signal after addition;
An analysis step of calculating a statistical distribution of the input signal;
And a control step of changing the constant according to the input signal, the statistical distribution, and the input quantization width.

Images