JP3490332B2 - Video watermarking system - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital watermark technique for embedding additional information in a digital moving image in an invisible state.
In particular, the present invention relates to a detection technique by signal accumulation for achieving a high detection reliability rate even with a weak embedded signal.

[0002]

2. Description of the Related Art Generally, a moving image has a higher perceptible resolution than a still image. Therefore, in order to maintain the image quality, it is necessary to weaken an embedded signal as compared with the case of a still image. Also, even for strongly embedded signals, low pass filters and MP
After undergoing EG2 compression, etc., it is greatly attenuated.
Therefore, in the method of detecting a signal independently from each frame, or the method of detecting from a fixed number of frames, the strength of the measured signal depends on the compression process, etc., so that the reliability of detection is uniform. Is not guaranteed. At this time,
On the contrary, if a threshold value for increasing the reliability is set, on the contrary, a problem (false negative error) occurs in which the embedded signal should not be detected at all. Therefore, there is a need for a method in which the reliability of detection does not depend on the strength of the signal measured from each frame. Japanese Patent Application No. 8-348426 "Data hiding method and data extraction using statistical test" Method ", Japanese Patent Application No. 8-345568," Data hiding method and system using statistical properties ", and Japanese Patent Application No. 9-8849.
3 "Multiple frame data hiding method and detection method" does not describe any means for solving it.

[0003]

SUMMARY OF THE INVENTION Therefore, the problem to be solved by the present invention is that the reliability of detection of embedded information does not depend on the strength of the signal measured from the frame.
A digital watermarking method and system are provided. Another object is to provide a digital watermark detection method and system for detecting embedded information based on a statistical amount of observation in each moving image frame. Yet another object is to provide a digital watermark detection method and system that changes the code when accumulating the observed amount in a frame of a moving image. Another object is to provide a digital watermark detection method and system that have an effect of canceling the correlation (non-independence) of consecutive frames of a moving image. Another object is to provide a digital watermarking method and system that changes the sign of an embedded signal when embedding information. Yet another object is to provide a digital watermarking method and system that changes the code of an embedded signal with a certain period. Yet another object is to provide a digital watermark detection method and system that prevents the difference between successive frames of a moving image from increasing. Another object is to provide a method and system effective for a compression method using a difference such as MPEG2.
Yet another object is to provide a method and system that eliminates code-related synchronization between the embedding process and the detection process. Yet another object is to provide a method and system that eliminates the need for position synchronization in the embedding process and the detection process. Another issue is
An object of the present invention is to provide a digital watermark detection method and system capable of coping with a large change in the playback speed of a moving image. Another object is to provide a digital watermark detection method and system for detecting watermark information at low cost.

[0004]

In order to solve the above problems, in embedding, information to be embedded is prepared as a bit string, the sign of the bit string is changed according to the cycle of sign inversion, and the bit string is embedded in a frame. To do so. Figure 1
A block diagram for embedding information according to the present invention is shown in FIG. First, the block 110 is a bit string preparation block that prepares embedded information as a bit string. Then block 120
Changes the sign of the bit string according to the cycle of sign inversion,
It is a bit string code change block. Finally block 13
0 is a bit string embedding block for embedding a bit string in a frame of a moving image. In addition, at the time of detection, the observation amount in the frame is accumulated, the accumulated observation amount is compared with a variable threshold value corresponding to the accumulated observation amount, and the embedded information is detected based on the comparison result. To do. FIG. 2 shows a block diagram for detecting information according to the present invention. Block 2
Reference numeral 10 is an observation amount accumulation block for accumulating the observation amount in the frame. Next, block 220 is a variable threshold acquisition block that obtains a variable threshold corresponding to the accumulated observation amount. Next, block 230 is a comparison block for comparing the accumulated observation amount with a variable threshold value corresponding to the accumulated observation amount. Finally, block 240 is an information detection block for detecting embedded information according to the result of comparison.

[0005]

BEST MODE FOR CARRYING OUT THE INVENTION A digital watermarking method using the stochastic statistical properties of observed quantities (Japanese Patent Application No. 8-348426, Japanese Patent Application No. 8-345568).
Etc.), m quantities (V [1], V [2], ..., From each frame
By measuring V [m]) and interpreting each as bit information, m-bit embedding / detection is realized. Here, the observed amount from the frame without embedding,
That is, it is premised that the amount of noise observed as a signal can be approximated by a normal distribution. If the approximation to the normal distribution is correct, it is possible to predict a false positive error in which noise is erroneously interpreted as a signal, and conversely, the threshold T can be determined so that the error rate is matched with a target. .

[0006] Now, the bit information is the observation amount (| V [i]
|) Can be embedded in an image by manipulating the image data so that it exceeds this threshold T, but | V [i] |
Cannot be manipulated to exceed T because it is attenuated by the effect of compression processing or image quality is preserved. In such a case, there is a problem of false negative error in that detection is impossible even though there is an embedded signal.

In order to solve this problem, the inventors invented a method of carrying over the observed amount in each frame to the next frame and making a judgment by using the sum of the two. If two frames are insufficient, the number of observation frames is further increased and signals are accumulated to make a determination. As a premise here, the observed quantities from each frame must follow a normal distribution and be independent of each other. If f random variables that are independent of each other and follow a normal distribution with variance 1 are added, the result will be a random variable with variance f, so in the case of noise, it is expected that the accumulated observation amount will increase at a rate proportional to √f. To be done. This phenomenon is generally a random walk (ra
ndom walk). On the other hand, in the case of a signal, since it is an amount manipulated so as not to be distributed intentionally, the observed amount is expected to increase in proportion to f. Therefore, the accumulated amount U [i] is
If you divide it by f and normalize it,

[Equation 1] In the case of noise, it can be regarded as a random variable having a normal distribution with variance 1. Here, V_j [i] represents the statistic of the i-th bit observed from the j-th frame. Therefore, it can be compared with the fixed threshold T. Further, in the case of a signal, U [i] increases in proportion to √f, so that if f is sufficiently large, the fixed threshold T can be exceeded. That is, according to this method, even if the embedded signal is greatly attenuated, if the sign of the observed amount from each frame is almost uniform, the number of accumulated frames is increased to bring false negative errors close to zero. be able to.

Elimination of Correlation of Continuous Frames Generally, there is a strong correlation between continuous frames in a moving image. Rather, the frames are reproduced as a moving image because they are frames that have only a slight difference and have a strong correlation with each other. In an extreme case, the correlation coefficient is maximum in a moving image with the same content. In such a case, the assumption that the accumulated amount U [i] follows a normal distribution with variance 1 does not hold. Therefore, processing for canceling the correlation is necessary.

Therefore, when the observation quantities from the respective frames are added together, the signs are sequentially inverted to +-and accumulated. As a result, even if the same content continues, the accumulated amount U [i] does not continue to increase, oscillates small around 0, and its amplitude decreases in proportion to 1 / √f.
Originally, even if they are independent of each other, the independence is not lost even if the sign is inverted. At the time of embedding, as in the case of detection, the embedding operation is performed while inverting the sign of the embedding signal to + -in order.

FIG. 3 shows a flowchart for performing the embedding operation by changing the sign of the embedding signal in accordance with the sign inversion cycle. The cycle of the sign inversion cycle is defined as one cycle by the set of + half cycle and − half cycle, like the wave cycle. That is, if the wavelength is 10, the code changes in steps of 5 wavelengths. In FIG. 3, the code inversion cycle is C, so the code is inverted and information is embedded every C / 2 cycle. In step 310, the counter is first set to 0.
Next, in step 320, an embedded bit string is prepared, and in step 330, it is determined whether all the frames have been processed. If the result is YES, the process ends. The result is N
If it is O, the process proceeds to step 340. Step 34
At 0, the bit string is embedded in one frame. Then, in step 350, the counter n is incremented by 1. Next, in step 360, it is judged whether or not the cycle of sign inversion has come.
If the result is NO, the process returns to step 330, and if the result is YES, the process proceeds to step 370. In step 370, all the bit strings are inverted, and then in step 380, the counter n is reset.
Then, the process returns to step 330.

Avoiding bit interpretation inversion by adding a sign bit Interpretation of a bit is performed using the sign of the observed amount U [i]. For example, if it is positive, it is interpreted as bit "1", and if it is negative, it is interpreted as bit "0". In the method, synchronization regarding the code is required between the embedding and the detection processing, but it is generally difficult to synchronize the both. However, if it is left as it is, the interpreted bit is inverted (for example, "101" becomes "010") and misinterpreted.

Therefore, the inventors invented a method of adding a bit representing a code and embedding a total of m + 1 bits. The bit inversion phenomenon is saved by interpreting the bit having the same sign as the sign bit as bit "1" and the bit having the opposite sign as bit "0". For example, if "101 + 1" is inverted to "010 + 0", the sign bit is "0", so the interpretation is "101". In this way, when the sign is inverted, the property that all m + 1 bits are inverted is used.

When the sign of the embedding process is changed for two consecutive frames, the sign inversion of a long cycle and the observation of each half cycle produce a difference twice as large as that of the embedding operation. Increasing the difference between successive frames is MP
It greatly affects the compression method using the difference such as EG2, and deteriorates the compression rate. In addition, when a filter that removes noise that reduces the difference between successive frames is applied, the embedded signal is almost attenuated and disappears.

In order to avoid such a problem, it is necessary to lengthen the code inversion cycle C so that it does not change between adjacent frames. At this time, the observation is a half cycle (C /
2) Every other time, and by changing the sign as shown in FIG. 6 and storing, the signal can be reliably recovered.

FIG. 4 shows a flow chart of signal detection by long-cycle sign inversion and half-cycle observation. First, in step 410, the counter is set to 0. Next, in step 420, the sign of detection is set to "+", and step 4
In step 30, it is determined whether all the frames have been processed. If the result is YES, the process ends. If the result is NO, the process proceeds to step 440. In step 440, bit information is measured from the frame. Next, in step 450, the measured value is accumulated in the direction of the detection code ("+" / "-"). Then, in step 460, it is determined whether the strength of the accumulated signal exceeds the threshold value. If the result is YES, the process ends,
If the result is NO, control proceeds to step 470. After incrementing the counter f by 1 in step 470, step 48
When it is 0, it is determined whether it is a sign inversion cycle. If the result is NO, the process returns to step 430. If the result is YES, the process proceeds to step 490. In step 490, the sign of detection is inverted, and in step 495, the counter is reset to 0 again. Then, the process returns to step 430.

Storage method when the embedding and detection cycles cannot be matched In MPEG2 compression, only I frame of I, P and B is self-reproducing type, and the remaining two are difference type from other frames. It is compression. Therefore, the detection in which only the I frame is an observation target is a good choice from the viewpoint of cost, however, the period in which the I frame appears is not fixed according to the MPEG2 standard, and therefore the period of the embedded signal is Not necessarily in agreement. If the cycles do not match, the codes cancel each other out, and the problem of accumulation cannot be obtained. Therefore, in order to store the codes without canceling them, two storage buckets A and B are prepared, and the storage is alternately performed as shown in FIG. Allow collection to take place.

For example, the first I is + in bucket A and 2
The second is stored in the bucket B with-, but the bucket B crosses the boundary where the sign of the embedding changes, so that the second and the fourth must be stored with the same sign, and the second sign is stored with the opposite sign to cancel. However, on the other hand, since the bucket A has no code deviation from the embedding period, it can be reliably stored. In terms of the number of code deviations, the bucket A is 0, so all 3 work for recovery, while the deviation of B is 1, so 1 after cancellation.
Will be collected.

FIG. 5 shows a flow chart of signal detection in the storage method when the embedding and detection periods cannot be matched. First, assuming that the cycle of sign inversion is C, step 510
Initialize the bucket counter with. Initialized is an array representing the number of updates of buckets A and B, the number of positive updates of buckets A and B, and the number of negative updates of buckets A and B. In step 520 it is determined whether all frames have been processed. If the result is YES, the process ends. If the result is NO, the process proceeds to step 530.
Go to. In step 530, bit information is measured from the nth frame, and the bucket and direction to be stored in step 540 are determined. Next, in step 550, the upper limit of the code bias when S is smaller than 2 is confirmed. Similarly, in step 560, the upper limit of the code bias when S is 2 or more is confirmed. If the decision result in the step 550 is YES, the process returns to the step 520. The same applies to step 560.
If the restriction of the code bias is not applied, the processing is step 57.
Go to 0. In step 570, measurement values are accumulated.
If the measured value is s = 0, then in the positive direction (addition) to bucket A, s = 1
Then, the measured value is accumulated in the bucket B in the positive direction (addition), the negative value is stored in the bucket A in the negative direction (subtraction) if s = 2, and the negative value is stored in the bucket B in the negative direction (subtraction) if s = 3. Next, at step 570, it is determined whether the strength of the accumulated signal exceeds the threshold value. If the result is YES, the process ends. If the result is NO, the process proceeds to step 590. In step 590, the bucket counter is updated. Then, the process returns to step 520.

Accumulation method 2 when the embedding and detection cycles cannot be matched In the previous section, the two buckets were exclusively used so as not to overlap in time, but in FIG. 8, the cycles are shifted by 1/4. Here, we will show the method to be used redundantly. For example, bucket A
Is +/- shifted by 2 of half cycle 6, so the offset is subtracted by 2, while bucket B is shifted by 1 and the remaining 4 after offset is 4, that is, at least one is more than half recovered. I see that I have the ability. Generally, this recovery capacity is higher than the exclusive recovery shown in the previous section.

Upper limit for suppressing sign bias of accumulation In the case of the preceding term and the preceding term, accumulation with the sign reversed does not necessarily occur alternately. When the accumulation is continuously performed with one code, the purpose of canceling the correlation due to the sign inversion is not achieved, thus causing a problem that the false positive error rate is difficult to predict. In order to avoid this problem, it is necessary to skip the storage when the stored codes are biased. Bias is judged by the following judgment formula. | fp-fm | <= √f Here, fp and fm are the numbers of frames accumulated by the signs + and-, respectively, and f is the sum of both (fp + fm). Since noise is accumulated at a speed of √f as described above, if the difference (| fp-fm |) does not exceed that speed, accumulating the same sign will affect the false positive prediction. Absent.

The present invention further employs the following method for detecting watermark information at a low cost in accordance with the reproduction speed of a moving image (so that it can cope with a large change in the reproduction speed). (1). If the relative positional relationship between embedding and detection does not match (position mismatch), we introduce a periodic detection mask that is insensitive to position mismatch (shift).
In order to reduce the matching search range, the data to be calculated is reduced to the value of the inner product to reduce the amount of memory and the amount of calculation. (2). By positively utilizing the fact that the detection succeeds even if the embedding position changes due to (1), the history of the relative positional relationship between embedding and detection is used, and the overlapping positional relationship is eliminated to create a moving image. Eliminate correlations that exist in consecutive frames inside. (3). The position synchronization signal and the signal representing the bit are separated, and the bit signals are adopted in the order of the strongest position synchronization signal to improve the storage efficiency.

The above-described high-speed detection processing and moving image correlation elimination method will be described in more detail below. In data hiding [1,2], by adding the embedded pattern (P) to the original content (I),
Copyright information, copy control information, etc. are inserted as a digital watermark. When the content is a two-dimensional image array, this operation is realized by addition of the two-dimensional array.

I ': = I + P

On the other hand, in detecting the digital watermark, the correlation (X) with the content (J) to be detected is measured using the detection mask (M) having a strong correlation with the embedding pattern.

[0025]         X = J dot M = I dot M (J = I, unmarked)                        OR                       I dot M + P dot M (J = I ', marked)

Here, "dot" is an operator representing an inner product. If M is selected so that "I dot M" of the first term on the right-hand side is always small, if there is no embedding (unmarked), the correlation X takes a small value close to 0 and there is embedding (mark
In the case of ed), the term “P dot M” in the second term has a large value, and thus the presence or absence of a digital watermark can be distinguished by the size of X. As for the detection mask M, generally, if the contents are randomly selected, it is possible to make the first term "I dot M" close to 0 regardless of the contents.

Although the detection mask is a fixed pattern regardless of the content, one embedding pattern does not affect the original image quality and sound quality while maintaining the property of strong correlation with the detection mask M. In addition, the content I needs to be adjusted and created according to the content. However, in the following, in order to simplify the description, the embedded pattern and the detection mask are fixed and are treated as the same.

First, a low-cost and high-speed processing method of shift positions for realizing shift conversion (deviation) tolerance will be described. Consider a case where a position (geometric) shift occurs between the detection mask M and the target content J. In general, the shift is due to processing such as transmission between the embedding operation and the detection operation, and horizontal (1
6: 9, LetterBox) Change in floating position of image or malicious operation such as hacking. As an embedding / detection method that is not affected by the shift at all, there are methods that use frequency space such as Fourier transform and DCT transform. For example, when the content is Fourier frequency transformed and decomposed into power (strength) component and phase component, the effect of the shift appears only in the phase, that is, it does not appear in the power.・ There is a method to perform detection. However, since frequency conversion generally requires high calculation cost, it cannot be used when real-time detection equivalent to the reproduction speed such as a moving image is required. In the embedding / detection method based on the pixel in the spatial domain, the first method described above is used.
The term "I dot M" shows a small value that does not change regardless of the shift, while the second term "P dot M" shows this shift (mismatch) even though the digital watermark remains sufficiently. -alignmen
Due to t), a large response cannot be output and the right side as a whole has a small value, and as a result, detection of the digital watermark fails. The correct position (alignm
ent) candidate is tried variously, and the position (shift position in the case of mark) showing the maximum reaction needs to be searched, but a simple full search in pixel units
In case of 720x480 size image, 345,600 trials are required, which is not realistic from the viewpoint of calculation amount.

Therefore, the number of searches can be significantly reduced if alignment is possible in a search in small area units instead of in pixel units. Therefore, in the present invention, the following 8x4 size block pattern (p0
~ P3)

[0030]         ++++++++ 00000000 -------- 00000000         00000000 ++++++++ 00000000 --------         -------- 00000000 ++++++++ 00000000         00000000 -------- 00000000 ++++++++            p0 p1 p2 p3

A plurality of one of these arranged vertically and horizontally, for example, a 2x4 block, is used as a basic pattern.
pattern), an embedding pattern and a detection mask will be configured. Here, "+" represents +1 and "-" represents -1. Hereinafter, the above block pattern will be referred to as a minimum pattern (primitive pattern). The basic pattern P0 created by the minimum pattern p0 is
In case of 2x4 minimum pattern, it is as follows.

[0032]

Basic patterns P0 to P created in this way
Create a detection mask (embedded pattern) by arranging 3 at random.

[0034]

When M and P are laterally displaced from each other by one pixel column, one column at the end of each basic pattern overlaps the adjacent pattern, and the columns of the basic pattern are randomly formed. Therefore, assuming that the contribution to the inner product of that part is 0 on average, the output of the entire inner product loses 1/16. Here, it is assumed that the end portions outside the area of the pattern are folded back and are connected at the top and bottom, and at the left and right, respectively. According to the same idea, a loss of 2 rows is 2/16 and a loss of 3 rows is 3/16. When 8 columns are displaced, the loss is half, but if the shift is tried in the unit of the minimum pattern (8 pixel columns horizontally), the displacement will be 0 and the loss will be 0. Therefore, if the search is performed in the unit of the minimum pattern, the maximum loss position has a deviation remainder of 4
At that position, then there is a loss of 4/16 = 25%.

Next, considering the case where M and P are vertically deviated from each other by two lines, the two lines overlap the adjacent pattern, and thus the loss is 2/16. However, when only one line is shifted, 0 and non-zero overlap each other, so the value of the inner product falls to 0. In order to save this situation, in the present invention, in addition to M, a mask M ′ vertically shifted by one line is introduced. M'is the minimum pattern unit, p0 is p1 and p1 is p2
, P2 is replaced by p3, and p3 is replaced by p0. If one row consists of 4 rows, the detection mask M ′ can be regarded as a phase relationship of 1/4 cycle (90 degrees) of the detection mask M.

[0037]

The detection mask M ′ reacts to a shift of one row with zero loss. The loss is 2/16 when three lines are offset.
When there is a deviation of 4 lines, the loss becomes 0 if the deviation is tried in the unit of the minimum pattern, so the deviation remainder is 2 in the vertical direction.
Then the maximum loss is 2/16. Therefore, when the lateral shift remainder (mod 8) is 4 and the vertical shift remainder (mod 4) is 2, the maximum loss is 1-(1-4 / 16) x (1-2 / 16) = 34.4%.

From the above, by searching the correct shift position in the unit of the minimum pattern, the reaction of the digital watermark can be obtained with a loss of about 35% at worst. At this time, the number of searches is 1 / (8x4) x 2 = 1/16 times as large as that in the pixel unit, considering that both M and M'are applied. In addition, if you calculate the inner product only once in the unit of 8x4 minimum pattern, you can use the inner product value later, which will lead to a significant reduction in memory. This will enable the detection of the digital watermark to be built into the DVD drive. The realization of chips has become possible. In other words, for a 720x240 size field image (172,800 bytes), the tiled part that is divided into 9 (3x3) and overlapped is expressed as an inner product value instead of a pixel value. / 8)
x (240/3/4) x2 x 2 byte = 2400 bytes is enough,
Also, the search range is (720/3/8) x (240/3/4) x 2 = 1200
Street and realistic. This is because direct detection from the MPEG2 stream processes I-frames that appear about twice per second, but 0.5 seconds is long enough to try 1200 times twice.

Next, a method of eliminating moving image correlation will be described. The “frame storage” method [4] is a measure against weak detection output from each frame in digital watermark detection from moving images. When it reaches, it is judged that there is a mark. On the contrary, if the threshold is not reached in a certain defined section (time), it is determined that there is no mark. If noise that occurs randomly is accumulated, the positive and negative values of the sum cancel each other out, and a random walk occurs, and the increase can be predicted to be √n. Here, n represents the number of accumulations. On the other hand, in the case of noises (embedded signals) that are not independent of each other but have a positive correlation, if the measured noises are accumulated (summed), the sums do not cancel but increase at a speed faster than √n. to continue. If the correlation is maximum positive (1.0), n
Increase at the speed of. Therefore, the premise of frame accumulation is that when there is a mark, there is a coherent positive correlation even if the output from each frame is weak, and the result of the accumulation extends almost linearly, and conversely when there is no mark. , The frame outputs are independent of each other, and the result of the accumulation is the behavior of a random walk. In particular, the latter independent randomness is an important property in determining the threshold value for determining the presence / absence of a mark. However, in general, a moving image is a moving image in successive frames (the contents of frames are similar). 2.) There is a strong positive correlation, so the outputs from those frames are not independent of each other.

In one method of the present invention, in order to guarantee its independent randomness, the sign when accumulating the frame output is forcedly inverted in a time-regular manner to cancel the positive correlation existing between the frames. The negative correlation. In order to react with the detection, the embedding side reverses the embedding sign (direction) according to the same time rule. Furthermore, a method of detecting a digital watermark even when the reproduction speed of a moving image changes and the time rule is deviated will be described.

In the present invention, in order to guarantee the randomness of the noise detected from the unmarked moving image, the history of shift positions is introduced. This is based on the idea that the detection noises from the frame images having the maximum reaction and different shift positions are independent from each other. Where the depth of history
If n, the output from a certain frame is adopted and stored or discarded.

(1) If the shift position of this frame is included in the shift positions n of the frame that have been adopted in the past, it is rejected and discarded. (2) If not, adopt and accumulate. It is also added to the history of shift positions.

In this way, the randomness of the accumulated noise is maintained in the accumulation of the past n frames from the unmarked moving image.

If the depth of the history of shift positions is made equal to the upper limit number of accumulation, noises that are not correlated with each other can be accumulated. In the embedding process, it is necessary to change the embedding position and shake the shift position as appropriate. Otherwise, the same shift position will continue, so the accumulation will not work at all and the weak signal cannot be saved. As described above, the above method does not use the time rule and does not depend on the playback speed of the moving image. Even when the moving image changes, the accumulation is correctly performed, so that the problem of the time change is solved. Further, if the embedding position is changed at intervals of I frames and the shift position is changed even at the longest time, all the observations in every I frame are adopted for accumulation, so that the accumulation efficiency is better than that of the bucket division method. The ordering of accumulation will be described below.

As a method for preferentially accumulating a strong signal in order to further improve the accumulating efficiency, a detection mask Ms for position synchronization and a detection mask Mb for bit interpretation are prepared so as to be independent from each other. And in addition
The procedure for introducing the "ranking table" and ordering the accumulation is shown below.

(1) First, the offset position O showing the maximum reaction is selected by the mask Ms. (2) The magnitude of the reaction of Ms, the offset position O, and the bit signal strength B at that position are paired. (3) If a group matching the offset position O is not registered in the ranking table, this group is registered in the ranking table. At this time, the sets of the ranking table are arranged in descending order of Ms reaction. (4) If the set that matches the offset position O has already been registered and the Ms reaction of this new set is larger than that of the registered set, replace it and update the rank.

Then, when this ranking table is updated,

(1) The bit signals B from the 1st place to the nth place in the table are added and compared with the threshold value. Here, if each bit signal follows a normal distribution of N (0,1), the value divided by √n also follows N (0,1), so it can be compared with a fixed threshold value. (2) If the threshold value is exceeded, it is determined that there is a mark and the bit is interpreted. (3) If n still does not exceed the threshold, increase n by 1 and return to (1).

When it is judged that there is a mark, or
If the threshold values are not exceeded even after adding up to the end of the table, the process is moved to the next frame. That is, in the case of a moving image with a mark, since the signal is intentionally filled, if it strongly reacts to the synchronization Ms, it is expected to respond strongly to Mb as well. Therefore, if the results of Mb reactions are added in the order of strong reaction to Ms, the effect of adding them in the order of strong Mb reaction is obtained.
This leaves weak (uncertain) Mb behind, so
Efficient storage is realized.

On the other hand, in the case of an unmarked moving image, a strong reaction to Ms does not necessarily mean a strong reaction to Mb. On the contrary, since they take mutually independent values, they are arranged in the strong order when Ms. Mb is expected to be randomly arranged regardless of Ms. Therefore, in the order of strong Ms, Mb
Even if they are added together, the behavior of the random walk of Mb is guaranteed, and does not affect the judgment threshold of mark presence / absence. As a method of making Ms and Mb independent of each other, there are a method of dividing into two in a region, a method of making mutually overlapping patterns orthogonal to each other, and the like. In the DVD copy control system, two masks M0 and M1 which are orthogonal to each other are prepared and defined as follows.

Ms = M0 + M1 Mb = M0-M1

Here, to embed the bit "0", M0 is used as an embedding pattern, and to embed the bit "1", M1 is used as an embedding pattern. No matter which bit is embedded, the detection by Ms responds positively, and the detection by Mb responds positively or negatively depending on the embedded bit. Therefore, the interpretation of the bit can be determined by its sign. The orthogonality of Ms and Mb is realized as follows.

Ms · Mb = (M0 + M1) · (M0−M1) =
| M0 | ∧2-| M1 | ∧2 = 0

[0055]

FIG. 10 shows an embodiment of a DVD-R drive system 800 having a digital watermark control block according to the present invention. In FIG. 10, the disk 910 is rotated by a motor 914 connected to a drive circuit 912, and the data recorded in the disk 910 is read by an optical electrical head 916. The drive circuit 912 operates according to a command from the DVD control block 918. The signal read by the opto-electrical head 916 is input to the DVD control block 918, amplified, converted if necessary, and decoded by the decoding block 9
Sent to 20. Decoding block 920 performs signal modulator, demodulator and error correction. The DVD control block 918 includes a servo circuit 918B for receiving the servo data recorded in the disc or a control signal from the decoding block to control the drive circuit 912. The DVD control block 918 is a signal reading circuit 9
18A is included.

The data received by the decoding block 920 is error-corrected by the buffer 920A, MPU 920B and decoder 920C in the decoding block connected by the common bus, and is decoded in real time to the digital watermark control block 930. Sent. In the digital watermark control block 930, embedding of information and detection of embedded information disclosed in the present invention are performed. Also, as appropriate
A data control circuit having copy control or the like may be added. From the digital watermark control block 930, the data is connected to the system 100 via the interface unit 921. Although the DVD-R drive system 800 can operate alone, the interface unit 92
It may be operated while communicating with a system connectable via the network 1.

FIG. 9 shows a DVD-R drive system 8
00, an example of the hardware configuration of the system connectable to 00 is shown. The system 100 includes a central processing unit (CPU) 1 and a memory 4. The CPU 1 and the memory 4 are connected via the bus 2 to the hard disk device 13 (or MO, CD-ROM 23,
Drive device for storage medium such as DVD) and IDE controller 2
It is connected through 5. Similarly, CPU 1 and memory 4
Via the bus 2, a hard disk device 30 (or MO 28, CD-ROM 23, D
A storage medium drive such as a VD-R800) is connected via a SCSI controller 27. The floppy disk device 20 is connected to the bus 2 via the floppy disk controller 19.

A floppy disk is inserted in the floppy disk device 20, and the floppy disk or the like and the hard disk device 13 (or MO, CD-ROM,
A storage medium such as a DVD-R800) and a ROM 14 can execute a command by giving a command to a CPU or the like in cooperation with an operating system to record a code or data of a computer program and loading the memory 4 into the memory 4. To be done. The code of this computer program can be compressed or divided into a plurality of pieces and recorded on a plurality of media.

The system 100 further comprises user interface hardware and a pointing device (mouse, joystick, etc.) for inputting.
7 or keyboard 6 and a display 12 for presenting visual data to the user. It is also possible to connect a printer via the parallel port 16 and a modem via the serial port 15. The system 100 can be connected to a network via a serial port 15 and a modem or a communication adapter 18 (Ethernet or token ring card) or the like, and can communicate with another computer or the like. It is also possible to connect a remote transmission / reception device to the serial port 15 or the parallel port 16 to transmit / receive data by infrared rays or radio waves.

The speaker 23 receives the audio signal D / A (digital / analog conversion) converted by the audio controller 21 via the amplifier 22 and outputs it as audio. Also, the audio controller 21
Is A / D based on the audio information received from the microphone 24.
(Analog / digital) conversion, it is possible to incorporate the audio information outside the system into the system.

As described above, the system 100 connectable (communicable) to the DVD-R drive system 800 has a built-in normal personal computer (PC), workstation, notebook PC, palmtop PC, network computer, or computer. Various home appliances such as TVs, game machines with communication functions, telephones, FA
It can be easily understood that it can be implemented by a communication terminal having a communication function including X, a mobile phone, a PHS, an electronic organizer, or the like, or a combination thereof.

DVD-R drive system 800
11 shows an external connection with the data processing system 100, and FIG. 12 shows an internal connection.

[0063]

When a signal for copy control is embedded and hidden in DVD video data, it is necessary to strongly embed the signal in advance so that the device does not misread the copy control signal. However, since a moving image generally has a higher resolution for human eyes than a still image, there is a problem in that a strongly embedded signal is perceived as noise. According to the method and system of the present invention, instead of strong embedding, weak embedding is performed in each frame within a range where quality is maintained, and at the time of detection, it is accumulated to a sufficient strength to make a determination, which is the same as strong embedding. Extraction reliability can be achieved. Although the moving image format of DVD is MPEG2, the problem that the embedded signal is weakened by the lossy compression can be solved by the storage method and system.

[0064]

[Brief description of drawings]

FIG. 1 is a block diagram for embedding information according to the invention.

FIG. 2 is a block diagram for detecting information according to the present invention.

FIG. 3 is a flowchart for performing the embedding operation by changing the sign of the embedded signal according to the cycle of sign inversion.

FIG. 4 is a flowchart of performing signal inversion by observing a sign of a long cycle and observing every half cycle.

FIG. 5 is a flowchart for performing signal detection by a storage method when the embedding and detection periods cannot be matched.

FIG. 6 is a diagram showing a sign inversion of a long cycle and an observation method for each half cycle.

FIG. 7 is a diagram showing a storage method when the embedding and detection periods cannot be matched.

FIG. 8 is a diagram showing another storage method when the embedding and detection periods cannot be matched.

FIG. 9 is a hardware configuration example of a system connectable to the system of the present invention.

FIG. 10: D with digital watermark control block of the present invention
It is an example of a VD system.

FIG. 11 is a diagram showing the DVD system externally connected to the system 100.

FIG. 12 is a diagram showing a DVD system built in the system 100.

Front page continuation (72) Inventor Akio Koide 1623 Shimotsuruma, Yamato-shi, Kanagawa 14 IBM Japan, Ltd. Tokyo Research Laboratory (56) Reference JP-A-6-46392 (JP, A) Japanese Patent Laid-Open No. 11-98341 (JP, A) International Publication 96/25005 (WO, A1) International Publication 99/45707 (WO, A1) Koshio Matsui, Image Deep Cryptography, Japan, Morikita Publishing, June 15, 1993, p. 18-34 W. Bender, D.M. Gru hl, N.M. Morimoto, A .; Lu, Technologies for data hiding, IBM SYSTEMS JOURNAL, USA, IBM, 1996, Vol. 35, NOS 3 & 4, p. 316-320 (58) Fields investigated (Int.Cl. ⁷ , DB name) H04N ⁷ /00-7/088 H04N 7/12 H04N ⁷ /24-7/68 H04N 5/765-5/775 H04N 5 / 91-5/956

Claims (10)

(57) [Claims] 1. A moving image digital watermark detection system for detecting embedded information based on a statistical observation amount of frames, comprising: (1) means for accumulating and accumulating the observation amounts of each frame; 2) A means for comparing the accumulated observation amount with a predetermined threshold value corresponding to the accumulated observation amount, and (3) a means for detecting embedded information based on the result of the comparison. A digital watermark detection system for moving images. 2. Means for accumulating the observed amount in the frame
The system according to claim 1, wherein (1) is means for normalizing and accumulating the observed amount in the frame. 3. Means for accumulating the observed amount in the frame
The system according to claim 1, wherein (1) is means for changing the sign of the observed amount of the frame when accumulating the observed amount of the frame. 4. A means for accumulating the observed amount in the frame
(1) Inverts the sign at half-cycle intervals of the sign inversion cycle and accumulates the observed amount in the frame, with the sign inversion cycle as a set of a half cycle with a positive sign and a half cycle with a negative sign. The system of claim 1, which is a means for performing. 5. A means for accumulating the observed amount in the frame
(1) sets the sign inversion cycle as a set of a half cycle in which the sign is positive and a half cycle in which the sign is negative, and when accumulating the observation amount in the frame, at a quarter cycle interval of the code inversion cycle. The system according to claim 1, which is means for accumulating each observed value in the order of addition to A, addition to B, subtraction from A, and subtraction from B in two accumulators A and B. 6. A means for accumulating the observed amount in the frame
(1), the sign inversion period, as a set of a half cycle where the sign is positive and a half cycle where the sign is negative, when accumulating the observed amount in the frame, prepare two accumulators A and B, Each observation value is added to A at a quarter cycle interval of the sign inversion cycle, A
The means for accumulating in the order of addition, subtraction from A, subtraction from A, and accumulation in the order of addition to B, subtraction from B, subtraction from B, addition to B in parallel with accumulation of A. System. 7. The system according to claim 4, further comprising means for providing an upper limit of the deviation when the accumulated values of the two accumulators are biased to one sign. 8. A moving image digital watermark detection method for detecting embedded information based on a statistical observation amount in a frame, comprising: (1) adding and accumulating observation amounts in each frame; 2) a step of comparing the accumulated observation amount with a predetermined threshold value corresponding to the accumulated observation amount, and (3) a step of detecting embedded information based on the result of the comparison. A feature is a method for detecting a moving image digital watermark. 9. A computer-readable recording medium including a program for detecting embedded information based on a statistical observation amount in a frame in a moving image, wherein the program is stored in a computer and (1) in each frame. A function of adding up the observation amounts of and accumulating, (2) a function of comparing the accumulated observation amount and a predetermined threshold value corresponding to the accumulated observation amount, (3) by the result of the comparison, A recording medium that realizes the function of detecting embedded information. 10. A DVD system, comprising: a motor for rotating a disc; a pickup for reading and writing signals in the disc; a drive circuit for controlling the motor and the pickup; and a drive circuit for the drive circuit. The digital watermark control block comprises a DVD control block that issues a command, a decoding block that performs signal conversion and error correction, a digital watermark control block that embeds or detects information, and an interface unit that communicates with the outside. , (1) means for adding up and accumulating the observed quantities in each frame, (2) means for comparing the accumulated observed quantity with a predetermined threshold value corresponding to the accumulated observed quantity, (3) ) A DVD system comprising: means for detecting embedded information based on the result of the comparison.