CN1812544A - Target sample random marking and detecting method and video frequency digital water mark enciphering application thereof - Google Patents

Abstract

A method for detecting and random-labeling object sample includes labeling Ni numbers of samples in sample space containing N numbers of samples inrepeatedly and randomly with nl being less than N and labeling ratio being k=nl/N, calculating out critical random sampling quantity N2 by utilizing specific formula as per labeling ratio K, selecting N2 numbers of samples from sample space inrepeatedly and randomly with N2being no less than critical random sampling quantity N2 for sure to find out a labeled sample .

Description

Target sample random labelling, detection method and video digital watermark thereof are encrypted and are used

(1) technical field under

The present invention relates to a kind of target sample random labelling and detection method, and the application on video digital watermark is encrypted.

(2) background technology

In the research and application in the fields such as many physics, chemistry, geology; usually can run into such class difficult problem: detect target sample how accurately, fast and efficiently, the video digital watermark encryption technology application that is used for copyright protection in the electronic information technology research field is a good example.

Along with develop rapidly and the application popularization of network technology and digital multimedia technology, digital watermark technology is subject to people's attention just day by day as a kind of effective means of digital copyright protecting. The embed category of watermark is similar to the mark of target sample, extract the detection that then is similar to target sample, the watermarking process is appreciated that and is summarized as mark and the detection of target sample, and especially selection and the search procedure of watermark embedding frame are more typical. Although the development of more than ten years has been experienced in the research of digital watermark technology, many theories and algorithm have appearred, mostly still rest on conceptual phase, the quantity of real practical application is seldom. From the content of concrete research, the overwhelming majority still is confined to image digital watermark, and relatively less to the research of multimedia (comprising Audio and Video) digital watermarking. This is owing to the demand point of the multimedia digital watermark that derives from practical application is also very not clear and definite on the one hand, does not form better technology application model; Because multimedia file quantity is many, the file data amount is large on the other hand, and be applied to more network environment, also there are many basic problems that solve that are difficult in the technical elements of its digital watermarking, as modal in the video digital watermark, also be that problem is attacked in the most scabrous frame montage.

The digital watermarking General Requirements has hiding and two critical natures of robustness, and in video digital watermark, because video file is many, data volume is large, requirement of real-time is high, except the fundamental property that possesses above two digital watermarkings, also should satisfy following some:

1. video digital watermark (in particular for the video digital watermark of copyright supervision and control) should be a kind of blind watermark, should not need the help of original video files in watermark detection process again;

2. video digital watermark should be able to be resisted based on the modal frame montage attack of video file, comprises frame replacement, frame restructuring, frame interpolation, frame deletion etc.; Otherwise, only needing simple several frame editing operations, watermark has not just existed. Therefore, attack if a kind of better video watermarking algorithm can not be resisted the frame montage, just lost its actual application value.

3. the detection speed of video digital watermark should be enough fast, can satisfy the online detection requirements of net environment. Video digital watermark more should be brought into play its function aspect copyright supervision and control except the copyright authentication function that should have image digital watermark (off-line is finished). This just requires the detection algorithm complexity of watermark enough low, and detection time is short, and the user can be restrained oneself.

At present, also find to satisfy fully embedding grammar and the detection method of the video digital watermark of above three conditions.

(3) summary of the invention

One object of the present invention, be to provide a kind of accurately, fast and efficiently target sample random labelling, detection method.

Another object of the present invention, be to provide a kind of video digital watermark encrypt in to mark and the detection method of encrypted frame, namely embed the video digital watermark method that extracts with random frame based on random frame.

A. About target sample random labelling and detection method, specifically be discussed below:

1 one kinds of target sample random labellings, detection method may further comprise the steps:

1. in a sample space that contains N sample, not repeatedly, mark n wherein randomly₁Individual sample, wherein n₁＜N, the mark ratio $k=\frac{n_1}{N};$

2. calculate minimum number of random samples n according to mark ratio k according to following formula₂′：

3. from sample space, repeatedly do not choose at random n₂Individual sample, wherein n₂More than or equal to minimum number of random samples n₂', then will inevitably find one and be labeled sample.

In the target sample random labelling and detection method of the invention described above, the general value of described mark ratio k is 10～50%; Best value 20～30%.

2 theory analysises

In a sample space that contains N sample, mark n wherein randomly₁Individual sample (not repeating label) is equivalent to choose randomly n from N sample₁Individual sample belongs to the category of Probability Theory and Math Statistics^[4], total C_N ⁿ¹Plant different choice. Similarly, randomly draw n wherein the second time₂Individual sample (not repeating to extract) has C_N ⁿ²Plant different choice. The n that randomly draws so, for the second time₂Not containing the number of combinations that is labeled sample in the individual sample is C_N-n1 ⁿ² Duplicate sample in order to make in twice stochastical sampling, the n that namely randomly draws for the second time₂At least contain a probability that is labeled sample in the individual sample $P_{n_2}(N,n_1)\≥99.995\%$ Set up, then need to satisfy:

$P_{n_2}(N,n_1)=\frac{C_N^{n_2}-C_{N-n_1}^{n_2}}{C_N^{n_2}}\×100\%\≥99.995\%...\left(1\right)$

Wherein, $C_n^m=\frac{n\×(n-1)\×(n-2)...(n-m+1)}{m\×(m-1)...\×2\×1}.$

Formula (1) has provided in twice stochastical sampling and the accurate conditional relationship that sample repeats occurred. Because this formula is too complicated, there be the computing of multistage layer, its operand and computational complexity are all larger, are difficult to directly be used to determine n₂Span. More wish in actual applications to find a kind of intuitively, method is determined n fast₂And scope.

Formula (1) has provided in fact n when occurring the sample repetition in twice stochastical sampling₂Need satisfied inequality, wherein have a minimum number of random samples n₂′。n ₂' satisfy relational expression:

$P_{n_2}(N,n_1){|}_{n_2={n_2}^{\′}}\≥99.995\%\≥P_{n_2}(N,n_1){|}_{n_2={n_2}^{\′}-1}...\left(2\right)$

Thus, can change an angle, finding the solution of formula (1) is converted to seek minimum number of random samples n₂', the accurate conditional relationship of same up-to-date style (1) is converted to n₂≥n ₂′。

3 minimum number of random samples n₂' numerical solution

Utilize Matlab 6.5 experiment porch to carry out minimum number of random samples n₂' numerical solution, concrete solution procedure is as follows:

(1) contains N=10 at one⁴In the sample space of individual sample, wherein 10% sample (not repeating label), i.e. n of mark randomly₁＝N×10％。

(2) with n₂Be spaced apart 1 since 1 and progressively increase, calculate respectively corresponding probability $P_{n_2}(N,n_1)=\frac{C_N^{n_2}-C_{N-n_1}^{n_2}}{C_N^{n_2}}\×100\%,$ Find first n that satisfies formula (1)₂Value, and be designated as n₂′。

(3) random labelling ratio k is increased to 100% successively since 10% with 10% interval, i.e. n₁=k * 10% * N, k=1,2 ..., 10. Repeating step (1) (2) records corresponding n₂′。

(4) sample space quantity N is increased to 10 successively⁶、10 ⁸, repeating step (1)～(3).

Result of calculation is as shown in table 1.

Table 1 is fixing $P_{n_2}(N,n_1)=99.995\%,$ N＝10 ⁴，10 ⁶，10 ⁸The time minimum number of random samples n₂′

N＝10 4
											n 1(×10 3)/K(×10％)	1	2	3	4	5	6	7	8	9	10
n 2′	-	45	28	20	15	11	9	7	5	1
											N＝10 6
n 1(×10 5)/K(×10％)	1	2	3	4	5	6	7	8	9	10
											n 2′	-	45	28	20	15	11	9	7	5	1
N＝10 8
											n 1(×10 7)/K(×10％)	1	2	3	4	5	6	7	8	9	10
n 2′	-	45	28	20	15	11	9	7	5	1

Can obviously find out from table 1: when occurring the sample repetition twice stochastical sampling, minimum number of random samples n₂' along with mark ratio k (perhaps marker number n₁) increase and reduce, expression mark ratio is larger, and then to detect the feasibility of target sample large; Simultaneously, with random sample marker number n₁Compare n₂' much smaller (be n₂′＜＜n ₁), expression can be used fully much smaller than marker samples quantity n₁Stochastical sampling realize the detection of sample; More importantly, in the certain situation of mark ratio, n₂' value be identical, i.e. n₂' only relevant with the mark ratio.

Provide n below by curve₂' approximate expression, and then make above-mentioned conclusion more clear.

4 n ₂' numerical analysis and curve thereof

In order to obtain n₂' approximate expression, use the way identical with upper joint, when finding the solution fix N, work as n₁N in the change procedure progressively₂' value, concrete solution procedure is as follows:

(1) sets N=10⁶， $P_{n_2}(N,n_1)=99.995\%,$ n ₁=1 (is that the mark ratio is 1/10⁶)。

(2) fixing n₁Value is with n₂Be spaced apart 1 since 1 and progressively increase, calculate respectively corresponding probability

$P_{n_2}(N,n_1)=\frac{C_N^{n_2}-C_{N-n_1}^{n_2}}{C_N^{n_2}}\×100\%,$ Find first n that satisfies formula (1)₂Value, and be designated as n₂′。

(3) with n₁Be spaced apart 1 and progressively increase until N, repeating step (1) (2) records corresponding n₂′。

When (4) drawing fix N, n₂' with n₁Relation curve n₂′～n ₁, as shown in Figure 1.

(5) change N=10⁴, repeat (1)～(4). Curve n₂′～n ₁As shown in Figure 2.

Can be found out by Fig. 1 and Fig. 2, as N fixedly the time, n₂' with n₁Roughly become reciprocal relation, and have certain relation with N. Therefore, suppose n₂' have a following form:

Wherein, α is undetermined coefficient, f (N), and g (N), h (N) is function undetermined.

On the basis of above-mentioned experimental data, Fig. 1 is carried out repeatedly piecewise fitting, obtain n₂' with N, n₁Variation relation, and consider n₂' value is necessary for integer, then fitting function (N, n₁) as follows:

Like this, just can be according to any given N, n₁Value calculates the n that satisfies condition easily by formula (3)₂' value.

The mark ratio is designated as $k=\frac{n_1}{N},$ Formula (3) is become:

Can find out n by formula (4)₂' only relevant with mark ratio k, be the approximate segmentation reciprocal function of k. This has further proved the result of calculation in 3 joints. Fig. 3 has shown n₂'～k graph of a relation.

So far can obtain as drawing a conclusion: in a sample space that contains N sample, mark n wherein randomly₁Individual sample (not repeating label). Stochastical sampling second time n wherein₂Individual sample (not repeating to extract), n₂As long as more than or equal to its minimum number of random samples n₂', just can reach that sample repeats is the purpose that target sample effectively detects. Minimum number of random samples n₂' only to account for the ratio of total sample relevant with being labeled at random sample, namely

Target sample random labelling of the present invention and detection method have accuracy height, detection time is short, reliability is high advantage.

The application of B. encrypting at video digital watermark about the present invention specifically is discussed below:

1 one kinds of mark, detection methods that video digital watermark is encrypted may further comprise the steps:

1. determine the totalframes N of video file according to original code check of video file, utilize random number generator produce at random not between the 1～N that mutually repeats n₁Individual random number, and from video file, choose randomly successively wherein n according to this random number sequence₁Frame embeds frame as watermark, wherein n₁＜N, the mark ratio $k=\frac{n_1}{N};$

2. embed the n of algorithm to choosing by image watermark₁Two field picture is watermarked respectively, and with watermarked n₁Two field picture and do not have watermarked N-n₁Two field picture reconfigures the video file that becomes with watermark information;

3. calculate critical detection frame number n according to mark ratio k₂′，

4. determine the totalframes N of video file according to original code check of video file, utilize random number generator produce at random not between the 1～N that mutually repeats n₂Individual random number, and from video file, choose randomly successively wherein n according to this random number sequence₂Frame is as watermark detection frame, wherein n₂≥n ₂′；

5. by the image watermark detection algorithm n to choosing respectively₂Two field picture carries out watermark detection, will inevitably find the watermarked image of a frame.

To mark and the detection method of encrypted frame, described mark ratio k was generally 10～50%, is preferably 20～30% during the video digital watermark of the invention described above was encrypted.

2 theory analysises

The embedding of video digital watermark and detection operation all are based on frame of video (such as I frame, B frame, P frame) basically, but they all extremely do not clearly state and should be how embed and extract watermark in a video file that contains a lot of frames, just should be watermarked in all frame of video according to default idea, but this be difficult to possible in actual applications. Reason has two: carrying out large-scale watermark embedding in a video file is a great engineering, and the user is out of patience; Simultaneously, large-scale watermark embeds and can produce than havoc video file, has a strong impact on video display effect. Therefore, a kind of good video watermarking algorithm should lack embed watermark information as much as possible, very easily detects again simultaneously, namely between the invisibility of watermark and robustness a good balance is arranged.

According to this principle, be in video file, to select a part of frame of video as embedding frame when video watermark embeds, the hiding embedding that the minor modifications by video being embedded the frame fixed position or the minor modifications after the functional transformation realize watermark information; At first to find watermarked frame of video during watermark extracting, and then go to detect the minor modifications of doing when embedding in the fixed position according to the watermark extracting algorithm. If can detect the existence that watermark information has just proved watermark. So, how in the frame of video of enormous amount, to find exactly watermarked frame of video, how in watermarked frame of video, to find exactly watermarked particular location, just seem very important. The embedded position information that the selection information, watermark that watermark of video files is embedded frame embeds watermark in the frame is referred to as the synchronizing information of watermark, and the former robustness is called the interframe robustness of watermark, and the latter's robustness is called robustness in the frame of watermark. Synchronizing information is destroyed, just can't find watermark to embed frame, can't find watermarked particular location, also just can't normally extract watermark.

Wherein a part of frame of video is watermarked as watermark embedding frame if can select randomly from video file according to a certain percentage, then select randomly the frame of video of some to carry out watermark detection, namely utilize the method based on random frame embeds and random frame is extracted to realize video digital watermark, the so embedding of video digital watermark and leaching process, or first kind robustness problem just can be summarized as that twice sample in the stochastical sampling repeat be the effective test problems of target sample.

Rationale of the present invention is: in a sample space that contains N sample, and mark n wherein randomly₁Individual sample (not repeating label). Randomly draw for the second time n wherein₂Individual sample (not repeating to extract), n₂Need to satisfy any condition and just can make in twice stochastical sampling and duplicate sample, or the n that randomly draws for the second time₂At least contain a probability that is labeled sample in the individual sample and level off to 1, be set as

$P_{n_2}(N,n_1)\≥99.995\%.$

n ₂As long as more than or equal to its minimum number of random samples n₂', just can reach that sample repeats is the purpose that target sample effectively detects. Minimum number of random samples n₂' only to account for the mark ratio of total sample relevant with being labeled at random sample, namelyAnd with total sample number amount n₁Irrelevant. Its approximate expression is:

n ₂'～k graph of a relation as shown in Figure 3.

Embedding grammar and the detection method that embeds the video digital watermark that extracts with random frame based on random frame of the present invention, be highly resistant to common in the video digital watermark, also be that the most insoluble frame montage is attacked, have high reliability and high robust, for the practical application of video digital watermark provides guarantee.

(4) description of drawings

Fig. 1 is n₂′～n ₁Graph of a relation, wherein $P_{n_2}(N,n_1)=99.995\%,$ N＝10 ⁶

Fig. 2 is n₂′～n ₁Graph of a relation, wherein $P_{n_2}(N,n_1)=99.995\%,$ N＝10 ⁴

Fig. 3 is n₂'～k graph of a relation is (fixing $P_{n_2}(N,n_1)=99.995\%$ )。

(5) specific embodiment

The below is encrypted as example with video digital watermark, and the present invention is carried out specific description, wherein mark ratio k=30%.

1. get 1 M (0.30,28) from the relational expression of formula (1) and the relation curve of accompanying drawing 3, minimum number of random samples n when this point is illustrated in random sample mark ratio k=30%₂'=28,28 samples of namely sampling randomly necessarily can detect target sample. In order to increase reliability, guarantee to detect effect, with n₂' increase a little and bring up to 30, be i.e. n₂ Equal 30.

In a video file, choose randomly wherein 30% frame and watermarked in each frame; During watermark extracting, from video file, choose randomly 30 frames, then can find at least a frame to contain watermark, and detect watermark.

2. concrete embedding grammar and detection method can be described below:

(1) random frame embeds

1. determine the frame number of video file according to original code check of video file, suppose to contain altogether the N frame.

2. utilize random number generator to produce n₁Random number between the individual unduplicated 1～N of=N*30%.

3. according to the order of the random number sequence between the 1～N that 2. generates, from video file, choose randomly successively n wherein₁Frame embeds frame as watermark.

4. embed the n of algorithm to choosing by image watermark₁Two field picture is watermarked.

5. with watermarked n₁Frame and other N-n₁Frame is reassembled into video file, uses as the video file issue with watermark information.

(2) random frame is extracted

1. determine the frame number of video file according to original code check of video file, suppose to contain altogether the N frame.

2. utilize the random number between 30 unduplicated 1～N of random number generator generation.

3. according to the order of the random number sequence between the 1～N that 2. generates, from video file, choose randomly successively 30 two field pictures wherein.

4. respectively 30 two field pictures of choosing are at random carried out watermark detection by the image watermark detection algorithm, and the record testing result.

5. as long as find that in 30 two field pictures of choosing at random detect a frame contains watermark, has just proved in this video file watermarked.

The robustness (being robustness in the frame) that detects in order to improve image watermark, can select two class robustness effects reach complementation simultaneously embedded location again not interactive watermarking algorithm repeat to embed, thereby realize complementary detection.

3. the Realization of Simulation and attack checking

It is platform that the emulation of this video watermark algorithm is based on matlab6.5, and test procedure mainly comprises:

A) read video file, produce sequence of frames of video, altogether the N frame;

B) produce n with randomizer₁Not repeating random number between=N*30% and 30 the 1～N, first group of random number is used for determining that watermark embeds the position of frame of video, second group of random number is used for determining the position of watermark extracting frame of video;

C) it is watermarked to use image watermark to embed algorithm;

D) Attack Digital Watermarking

(1) frame of video deletion: be included in forward and backward or middle any 50 frames of deletion in the watermarked video file.

(2) frame of video exchange: watermarked frame of video and the frame of video exchange sequence before or after it, the exchange frame number is no less than 20 frames.

(3) frame of video is inserted: 50 frames are inserted in the forward and backward or middle optional position that is included in watermarked video file.

E) use the image watermark extraction algorithm to extract watermark.

Above-mentioned algorithm is carried out emulation, adopt a large amount of video files in the experiment, file relates to widely scene (comprising: personage, automobile, landscape, moving object, scene switching etc.), video file is carried out the frame operation attack (comprising: frame deletion, frame exchange, frame insertion etc.), filename is respectively Akiyo, Foreman, Tempete, Mobile.

The Simulation results statistics is as shown in table 1 below, and wherein normal extraction refers to attack directly extraction watermark information without frame of video, and the watermarking algorithm that adopts in the experiment extracts similarity in single-frame images be 1.

Table 1: experiment simulation statistics

The video file title	The video file frame number	Embed the video frame number	Extract the video frame number	The validity of watermark extracting and extraction result under the different frame operational circumstances
												Normal extraction		Frame deletion		Frame inserts		The frame exchange
				Whether effective	Extract the result	Whether effective	Extract the result	Whether effective	Extract the result	Whether effective	Extract the result
												Salesman	449	135	30	Be	★	Be	★	Be	★	Be	★
Akivo	300	90	30	Be	★	Be	★	Be	★	Be	★
												Bus	150	45	30	Be	★	Be	★	Be	★	Be	★
Carphone	382	115	30	Be	★	Be	★	Be	★	Be	★
												Mobile	300	90	30	Be	★	Be	★	Be	★	Be	★
Vectra	142	43	30	Be	★	Be	★	Be	★	Be	★
												Forman	400	120	30	Be	★	Be	★	Be	★	Be	★
Tempete	260	78	30	Be	★	Be	★	Be	★	Be	★

As can be seen from the above table, the embedding ratio is 30%, extracts quantity and is in 30 the situation, and this algorithm all can effectively extract watermark information to frame operation (frame deletion, frame insert and the frame exchange), has very strong robustness.

4. conclusion: above-mentioned demonstration and emulation experiment show, at random embedding grammar and the detection method of video digital watermark, be highly resistant to common in the video digital watermark, also be that tool is attacked in the most insoluble frame montage

High reliability and high robust are arranged, for the practical application of video digital watermark provides reliable guarantee.

Claims (6)

1 one kinds of target sample random labellings, detection method is characterized in that, may further comprise the steps:

1. in a sample space that contains N sample, not repeatedly, mark n wherein randomly₁Individual sample, wherein n₁＜N, the mark ratio $k=\frac{n_1}{N};$

2. calculate minimum number of random samples n according to mark ratio k according to following formula₂′：

3. from sample space, repeatedly do not choose at random n₂Individual sample, wherein n₂More than or equal to minimum number of random samples n₂', then will inevitably find one and be labeled sample.

2. target sample random labelling according to claim 1, detection method is characterized in that, described mark ratio k is 10-50%.

3. target sample random labelling according to claim 2, detection method is characterized in that, described mark ratio k is 20-30%.

4. mark, the detection method that video digital watermark is encrypted is characterized in that, may further comprise the steps:

1. determine the totalframes N of video file according to original code check of video file, utilize random number generator produce at random not between the 1～N that mutually repeats n₁Individual random number, and from video file, choose randomly successively wherein n according to this random number sequence₁Frame embeds frame as watermark, wherein n₁＜N, the mark ratio $k=\frac{n_1}{N};$

2. embed the n of algorithm to choosing by image watermark₁Two field picture is watermarked respectively, and with watermarked n₁Two field picture and do not have watermarked N-n₁Two field picture reconfigures the video file that becomes with watermark information;

3. calculate critical detection frame number n according to mark ratio k₂′，

4. utilize random number generator produce at random not between the 1～N that mutually repeats n₂Individual random number, and from video file, choose randomly successively wherein n according to this random number sequence₂Frame is as watermark detection frame, wherein n₂≥n ₂′；

5. by the image watermark detection algorithm n to choosing respectively₂Two field picture carries out watermark detection, will inevitably find the watermarked image of a frame.

5. mark, the detection method of video digital watermark encryption according to claim 4 is characterized in that described mark ratio is 10-50%.

6. mark, the detection method of video digital watermark encryption according to claim 5 is characterized in that described mark ratio is 20-30%.

Images