CN1241388C - Video identification method with repair function based on digital watermark - Google Patents

Abstract

The present invention relates to a video identification method with recovery function based on digital watermarks, which is used for the technical field of multimedia content safety. The present invention has the steps that firstly, video frames are obtained at video identification watermark embedding terminal and single frames are protected by means of a self-recovery and identification algorithm after the video frames are read; then, the sequence numbers of the video frames are embedded into corresponding video frames by means of a mean value quantization watermark embedding method, important video frames are protected by means of an information dispersal algorithm, and recovered information is embedded into other n frames; at a video identification receiving terminal, each frame is identified by means of the self-recovery and identification algorithm and the sequence number information of each frame is extracted by means of a mean value quantization watermark extracting method so that the inner parts of the video frames and the video frames are integrally identified and falsification areas are located; finally, the important video frames which are lost are recovered by means of the information dispersal algorithm. The present invention can detect attacks which specially point to the time sequence of video signals, can distinguish legal and illegal video processing, and can recover the important video frames which are lost.

Description

Video-frequency identifying method based on digital watermarking with repair function

Technical field

The present invention relates to be used for content of multimedia safe practice field.It specifically is a kind of video-frequency identifying method with repair function based on digital watermarking.

Background technology

A perfect content of multimedia authentication method not only will have resistance to compression, can realize accurately integrated authentication under the compression of proper proportion, and should possess automatic recovery ability.The importance of the every frame of video sequence is different, can select important frame to be protected (promptly can be resumed) by the user, also can select all frames are protected.

Find by literature search, MICHAEL O.RABIN is at article " Efficient Dispersal of Informationfor Security Load Balancing; and Fault Tolerance " (" the efficient dispersion of safe bearing load equilibrium and fault-tolerant middle information ") (Journal of the Association for Computing Machinery, Vol.36.No.2, April 1989, and pp.335-348) the Information Dispersal Algorithm (information dispersal algorithm) of (ACM's magazine) proposition is a kind of safe and reliable information stores and the transmission technology that provides for distributed system.It is to transmit respectively after certain redundancy is divided into n sub-piece with data source F by introducing.At data receiver, utilize in n the sub-piece m just can reconstruct F (m≤n, suppose the individual sub-piece of n-m in transfer of data, lose).Wherein the size of each height piece is | F|/m, so this method is the space optimum.The space load of transmission can be controlled by the size of regulating parameter n and m.This image authentication method can be used for video authentication, but video data is three-dimensional information, has the characteristics different with view data.

From the authentication and recovery algorithms by the thesis for the doctorate " Watermarking and Digital Signature Techniques for Multimedia Authentication andCopyright Protection " (" watermark in multimedia authentication and copyright protection and digital signature technology ") of the Ching-Ying Lin of department of computer science of Columbia Univ USA at him in the proposition; this is a kind of still image authenticating water-mark algorithm; this algorithm makes the still image that has added watermark can accept the JPEG lossy compression method of predefine quality factor, can detect simultaneously the malice of picture material is distorted.Authenticator in this algorithm can be oriented the image block of being distorted, and utilizes the watermark recovery that embeds to go out by the rough content of tampered region.Authenticator has utilized two invariant features in the image lossy compression method, does not need the probability judgement thereby these two invariant features are the deterministic Verification Systems that make.First characteristic be if with the DCT coefficient adjustment of image for quantizing the integral multiple of step, wherein quantize step greater than the quantification step that thereafter image or video is compressed, this DCT coefficient reconstruct fully after compression so.Second characteristic utilized DCT coefficient corresponding in two DCT pieces of image to keep the constant characteristics of relation before and after compression.Can utilize second characteristic to produce authentication signature, utilize first characteristic that authentication signature is embedded in the image information.Adopt the image after this method is handled to tolerate the image processing operations that keeps content, can detect the image processing operations that changes content simultaneously.If but the rest image authentication techniques are directly used in video data, some attacks at the time still can destroy the integrality of video data easily, and can not realize the automatic recovery of lost frames.Further in the retrieval, find to have the report of the video authentication digital watermark of self-repair function as yet.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art and defective, and a kind of video-frequency identifying method with repair function based on digital watermarking is provided.The present invention can detect special attack at the vision signal time sequencing, not only can distinguish legal Video processing (as the MPEG4 compression of certain compression ratio) and illegal Video processing (content tampering of malice such as shearing, frame losing and switching frame etc.), and utilize information dispersal algorithm and digital watermark can recover the important frame of video of losing.

The present invention is achieved by the following technical solutions, and the present invention obtains frame of video at first at video authentication watermark built-in end, it read in the back utilize from recovering and identifying algorithm protection single frames, realizes the authentication to the single frames video.Use the mean quantization watermark embedding method that the sequence number of frame of video is embedded in the corresponding frame of video then.Adopt the information dispersal algorithm protection to need the important frame of video of recovering, recovering information is embedded in other n frame.At the video authentication receiving terminal, utilize from recovery and identifying algorithm and verify each frame, and use the mean quantization watermark extracting method to extract the serial number information of frame, thereby realize in the frame of video, interframe integrated authentication and orient the tampered region, adopt information dispersal algorithm to recover the important frame of video of losing at last.

Below the present invention is further illustrated, particular content is as follows:

(1) at video authentication watermark built-in end:

(1) utilize from recovering and identifying algorithm protection single frames: adopt from recover and this still image identifying algorithm of identifying algorithm in the signature described and watermark embedded mode protect each frame in the video sequence.This method can detect some attacks that destroy integrality in the frame of video, as object deletion, object replacement etc., has resistance to compression simultaneously.In the present invention, at first produce the signature of DCT coefficient, be embedded in back 6 DCT coefficients as watermark then by tactic preceding 6 coefficients of ZIG-ZAG.

(2) use the mean quantization watermark embedding method to embed frame number information: the watermark embedding method that uses previously should be used for embedding authentication information, the back is used for embedding the recovering information of other frame again, its capacity is limited, and the resistance to compression of this method is subjected to not influencing the restriction of visual effect.In order to embed 32 bit sequence of each frame of video correspondence, the present invention has adopted the incompressible mean quantization watermark embedding method of original creation.Specific practice is: a bit that embeds 32 bit sequence in 4 DC coefficient of 48 * 8DCT pieces of frame of video.If 48 * 8DCT piece DC coefficient and its average are respectively dc _iAnd dc, then $\stackrel{\‾}{\mathrm{dc}}=\underset{i=1}{\overset{4}{\mathrm{\Σ}}}{\mathrm{dc}}_i.$ If the embedment strength of mean quantization watermark is a Δ, Q _DcBe the integer part of dc divided by Δ, R _DcBe the remainder of dc divided by Δ.If Q _DcLSB identical with the indexed bits of embedding, amended dc then _iBe dc _i-(R _Dc-0.5) * and Δ, i=1...4; If Q _DcLSB different with the indexed bits of embedding, amended dc then _iBe dc _i-(R _Dc-1.5) * Δ is (at R _DcUnder 0.5 situation) or dc _i-(R _Dc+ 0.5) * Δ is (at R _DcUnder 0.5 situation).The watermark of adopting above mean quantization watermark embedding method to embed has very high resistance to compression; utilize the recovery certainly and the identifying algorithm that adopt in recovery and identifying algorithm protection single frames to be protected the DC coefficient of 8 * 8DCT piece simultaneously, the assailant is difficult to destroy the sequence number of frame of video by revising these DC coefficient.From test effect, choose suitable Δ value and can reach the visual effect of promptly not destroying image fully very high watermark robustness is provided again.The authenticator of video receiver can extract the sequence number of each frame of video, thereby detects the attack at the video information time sequencing.

(3) adopt information dispersal algorithm to protect important frame of video: to use JPEG2000 compression algorithm frame of video, two condition restriction compression ratios are arranged: the total capacity of watermarking algorithm; The quantity of the important frame of video of choosing.Adopt the specific practice of information dispersal algorithm to be:

1. the result of frame video signal after the JPEG2000 compression is big or small for the data F of N character, establishes b _iI character among the expression F.B in actual applications _iCan be taken as the byte of 8 bits, thus 0＜b _i＜255 and all calculating all be at Z ₂₅₇Perhaps GF (2 ⁸) in finish.F is divided into the piece that length is m by following mode then: F=(b ₁..., b _m), (b _M+1..., b _2m) ..., (b _N-m+1..., b _N).In order to simplify following description, establish S _i=(b _{(i-1) m+1}..., b _Im), 1≤i≤N/m.

2. select n vectorial a _i=(a _Il..., a _Im), 1≤i≤n, make any subclass that constitutes by m vector be linearity independently.

3. utilize vector set a _i=(a _I1..., a _Im), 1≤i≤n can be divided into F n sub-piece in the following manner:

F _i=(a _i.S ₁, a _i.S ₂..., a _i.S _N/m), i=1 ..., n, wherein a _i.S _k=a _I1.b _{(k-1) m+1}+ ...+a _Im.b _KmThereby,

|F _i|＝|F|/m。C _i，i＝1，...，n

4. produce n frame number with pseudo random sequence, the n that these sequence numbers an are represented frame is used for embedding n the sub-piece of F.The method that embeds protects the embedding grammar in the single frames identical with utilizing recovery certainly and identifying algorithm.Each frame embeds the sub-piece (this frame also may be used for embedding the sub-piece of other important frame) of F.

5. the information that embeds in 4. is not shielded, in order to detect F _iMalice distort, calculate F respectively _i, i=1 ..., the cyclic redundancy check (CRC) of n, the check byte C of generation _i, i=1 ..., n is embedded into corresponding frame and (embeds F _iFrame) in, the mode of embedding is with to use the mean quantization watermark embedding method to embed in the frame number information identical.C _iAnd F _iAdopting different embedding grammars is to consider for 2 reasons: utilize from the watermark information finite capacity that recovers and identifying algorithm protects the embedding grammar in the single frames to embed, should reduce the information of embedding as much as possible; The embedding grammar Billy who uses the mean quantization watermark embedding method to embed in the frame number information has higher robustness with the embedding grammar in recovery and identifying algorithm protection single frames.Identical with the sequence number of the every frame of frame of video, check byte C _i, i=1 ..., n is subjected to utilizing the protection of embedding grammar in recovery and identifying algorithm protection single frames, because the DC component of DCT coefficient is subjected to the protection of the method.

More than be the method for the present invention at video authentication watermark built-in end.In telescopiny, repeatedly used pseudo-random sequence generator.The seed of pseudo-random sequence generator is sent to video authentication watermark extracting end as key.The method that transmits comprises with vision signal and transmitting simultaneously; Behind public key digital signature, issue or the like.

(2) at video authentication watermark extracting reset terminal:

(1) utilization is verified each frame from recovery and identifying algorithm: adopt each frame in signature of describing in recovery and this still image identifying algorithm of identifying algorithm and watermark extracting method checking video sequence.In the present invention, at first produce the signature of DCT coefficient, from back 6 DCT coefficients, extract watermark by tactic preceding 6 coefficients of ZIG-ZAG, both are relatively drawn the authentication result of this frame, promptly whether distorted,, oriented 8 * 8DCT piece of being distorted if distorted.

(2) use the mean quantization watermark extracting method to extract the serial number information of frame: specific practice is: extract 1 bit of 32 bit sequence in 4 DC coefficient of 48 * 8DCT pieces of frame of video, promptly extract Q _DcLSB.If result (1) shows that this frame was illegally distorted, point out directly that then there is query in the sequence number of this frame; If result (1) shows this frame and do not distorted, then sequence number and the actual sequence number of this frame in video sequence that extracts compared.Thereby detect attack (frame losing, switching frame etc.) at the time.

(3) adopt information dispersal algorithm to recover the important frame of video of losing: lost important frame of video if the result of (2) shows, the information that can orient lost frames by key and pseudo random sequence is embedded in that n frame of video.Part in this n frame of video also may be lost, yet according to information dispersal algorithm, as long as wherein m frame of video do not lost and the process authentication, just can recover the important frame of losing.Wherein m frame of video should meet following condition: this frame passes through authentication in step (1); The sequence number of this frame is real, promptly by step (2); The F that extracts _iPass through cyclic redundancy check (CRC).The specific practice that adopts information dispersal algorithm to recover frame of video is:

1. be located at the authentication end and obtained F ₁..., F _mSub-piece adopts m sub-piece, and then following formula is set up: $A\·\left[\begin{array}{c}{b}_1\\ \·\\ \·\\ \·\\ b_m\end{array}\right]=\left[\begin{array}{ccc}{a}_1& \·& S_1\\ & \·& \\ & \·& \\ & \·& \\ a_m& \·& S_1\end{array}\right]$ A=(a wherein _Ij) _{1≤i, j≤m}Be the matrix of m * m, i is capable to be a _i

2. because A in vector be linearity independently, so A is an invertible matrix, S ₁Can obtain from following formula:

$S_1=\·\left[\begin{array}{c}{b}_1\\ \·\\ \·\\ \·\\ b_n\end{array}\right]=A^{-1}\·\left[\begin{array}{ccc}{a}_1& \·& S_1\\ & \·& \\ & \·& \\ & \·& \\ a_m& \·& S_1\end{array}\right]$

3. repeat 1,2 processes, can obtain S ₂..., S _N/m, these data are together in series can obtain F.

4. after reconstruct finishes, the frame of video that F is decompressed and can be restored out through JPEG2000.

Above process prescription the basic step that can repair authenticating water-mark based on the video of information dispersal algorithm.It should be noted that:

The first, but the recovering information of important frame be embedded in other frame as watermark, consider the capacity of watermarking algorithm, the present invention has used two kinds of methods that reduce amount of information: 1, utilize JPEG2000 that important frame is compressed; 2, utilize information dispersal algorithm IDA dispersed information, it is the space optimum that information dispersal algorithm has been mentioned in the front, and it has been reduced to minimum degree to the redundant information of introducing.The quantity of recoverable important frame of video depends on the capacity of watermarking algorithm, if the visual effect of watermarked rear video information is required to reduce, the quantity of recoverable important frame of video can increase, even comprises all frame of video.User of the present invention can make balance between visual effect and recoverable important frame of video quantity.

The second, the present invention does not make strict restriction to the authenticating water-mark embedding algorithm of single-frame images, has certain versatility, except recovering certainly and identifying algorithm.Can also select other image and video authentication half breakable watermark algorithm.

The 3rd, aspect the real-time method, real-time of the present invention depends on the real-time of watermarking algorithm.With regard to regard to recovery and identifying algorithm, its watermark extracting method can reach the requirement of real-time fully.If the recovering information of important frame is embedded in the n frame of this frame front, the receiving terminal of video information can reach fully and recover lost frames in real time, that is to say, can reach in real time to receive, authenticate and recover.

Existing video authentication technology adopts in the cryptography digital signature technology of stricti jurise to realize that this mode has following shortcoming mostly: additional digital signature can the outer transmission bandwidth of occupying volume; The conventional digital signature does not allow the mistake of a bit, does not therefore have resistance to compression; Can only error detection, can not error correction, do not have repair function.Adding watermark in vision signal also has a lot of technology, but the purpose of adding watermark is to be used for copyright protection mostly, i.e. the function of strong watermark.The present invention adds watermark in vision signal be to be used to provide a kind of video-frequency identifying method with repair function based on digital watermarking.Can detect special attack at the vision signal time sequencing, not only can distinguish legal Video processing (as the MPEG4 compression of certain compression ratio) and illegal Video processing (content tampering of malice such as shearing, frame losing and switching frame etc.), and utilize information dispersal algorithm and digital watermark can recover the important frame of video of losing.In the prior art, also there is not a kind of technology can have above all characteristics simultaneously.

Embodiment

Below in conjunction with the application of the present invention in " video content authentication ", specifically set forth its execution mode.In this example, the user is at first read in one section frame of video as video authentication watermark built-in end, processing through program, arrive receiving terminal by Network Transmission, at the video data of receiving terminal (as video authentication watermark extracting reset terminal) to receiving, by the processing of program, recover original frame of video.Concrete steps are as follows:

(1) at video authentication watermark built-in end:

(1) reads in frame of video, adopt each frame of handling incoming video signal from the watermarking algorithm of authentication and recovery algorithms, make the watermark that embeds in each frame of vision signal have resistance to compression and also can detection and location go out at the malice of frame internal information to distort.Preceding 6 quilts of DCT coefficient are protected from authentication and recovery algorithms in the frame of video;

(2) at the video sequence characteristics different with rest image, adopt incompressible mean quantization watermark embedding method will represent that the frame number of 32 bits of video frame time order is embedded in the corresponding frame of video, the value of getting the mean quantization watermark embed strength is 20, can detect the attack that destroys video sequence time integrality by extracting this sequence number.

(3) important frame (frame that need be resumed) of employing JPEG2000 compression algorithm original video data, the compression ratio of JPEG2000 is 14: 1, the PSNR of the important frame of video after being compressed is 35dB.Win the confidence parameter n=64, the m=32 of breath decentralized algorithm adopt the data message after information dispersal algorithm will compress to be divided into 64 sub-pieces then, and the mode of 64 sub-pieces with watermark is embedded in 64 other frame of video.Determine that by pseudo random sequence the sequence number of these 64 other frame of video, unauthorized user can't determine certain important frame is embedded in which 64 frame of video of video sequence.

(2) at video authentication watermark extracting reset terminal:

(1) receives the video data that arrives by Network Transmission, adopt each frame in recovery and identifying algorithm checking video sequence.Specific implementation is to produce the signature of each frame DCT coefficient by tactic preceding 6 coefficients of ZIG-ZAG, from back 6 DCT coefficients, extract watermark, both are relatively drawn the authentication result of this frame, promptly whether distorted, if distorted, orient 8 * 8DCT piece of being distorted.

(2) use the mean quantization watermark extracting method to extract the serial number information of frame, specific implementation is 1 bit that extracts 32 bit sequence in 4 DC coefficient of 48 * 8DCT pieces of frame of video, if result (1) shows that this frame was illegally distorted, point out directly that then there is query in the sequence number of this frame; If result (1) shows this frame and do not distorted, then sequence number and the actual sequence number of this frame in video sequence that extracts compared.Thereby detect attack (frame losing, switching frame etc.) at the time.

(3) adopt information dispersal algorithm to recover the important frame of video of losing.Lost important frame of video if result (2) shows, which 64 frame of video is the information of orienting lost frames by key and pseudo random sequence be embedded in.Whether wherein have 32 frame of video do not lose and process authentication, if the information that just reconstructs the important frame of losing by their employing information dispersal algorithms, the important frame that decompresses and be restored out through JPEG2000 are again arranged if judging then.

In above embodiment, whether all frame of video can detect was altered by malice, and 1/3rd important frame of video can both be resumed, and when the compression ratio of MPEG4 is higher (as 300kbit/s), the present invention still keeps good robustness.Thereby proved validity of the present invention.

Claims (8)

1, a kind of video-frequency identifying method with repair function based on digital watermarking, it is characterized in that, at first at video authentication watermark built-in end, obtain frame of video, it is read in the back utilize, realize authentication, use the mean quantization watermark embedding method that the sequence number of frame of video is embedded in the corresponding frame of video then the single frames video from recovering and identifying algorithm protection single frames, adopt the information dispersal algorithm protection to need the important frame of video of recovering, recovering information is embedded in other n frame; At the video authentication receiving terminal, utilize from recovery and identifying algorithm and verify each frame, and use the mean quantization watermark extracting method to extract the serial number information of frame, thereby realize in the frame of video, interframe integrated authentication and orient the tampered region, adopt information dispersal algorithm to recover the important frame of video of losing at last.

2, the video-frequency identifying method with repair function based on digital watermarking according to claim 1 is characterized in that, described utilization is from recovery and identifying algorithm protection single frames, and is specific as follows:

Adopt each frame in signature of describing in recovery and this still image identifying algorithm of identifying algorithm and watermark embedded mode protection video sequence; at first produce the signature of DCT coefficient, be embedded in back 6 DCT coefficients as watermark then by tactic preceding 6 coefficients of ZIG-ZAG.

3, the video-frequency identifying method with repair function based on digital watermarking according to claim 1 is characterized in that, described use mean quantization watermark embedding method embeds frame number information, and is specific as follows:

In 4 DC coefficient of 48 * 8DCT pieces of frame of video, embed a bit of 32 bit sequence: establish 48 * 8DCT piece DC coefficient and its average and be respectively dc _iAnd dc, then $\stackrel{\‾}{\mathrm{dc}}=\underset{i=1}{\overset{4}{\∑}}d{c}_i,$ If the embedment strength of mean quantization watermark is a Δ, Q _DcBe the integer part of dc divided by Δ, R _DcBe the remainder of dc, if Q divided by Δ _DcLSB identical with the indexed bits of embedding, amended dc then _iBe dc _i-(R _Dc-0.5) * and Δ, i=1...4; Otherwise then amended dc _iBe dc _i-(R _Dc-1.5) * and Δ, at R _DcUnder 0.5 situation, or dc _i-(R _Dc+ 0.5) * and Δ, at R _DcUnder 0.5 situation.

4, the video-frequency identifying method with repair function based on digital watermarking according to claim 1 is characterized in that, described employing information dispersal algorithm is protected important frame of video, and is specific as follows:

1. the result of frame video signal after the JPEG2000 compression is big or small for the data F of N character, establishes b _iI character among the expression F, b in actual applications _iBe taken as the byte of 8 bits, thus 0＜b _i＜255 and all calculating all be at Z ₂₅₇Perhaps GF (2 ⁸) in finish, F is divided into the piece that length is m by following mode then: F=(b ₁..., b _m), (b _M+1..., b _2m) ..., (b _N-m+1..., b _N), establish S _i=(b _{(i-1) m+1}..., b _Im), 1≤i≤N/m;

2. select n vectorial a _i=(a _I1..., a _Im), 1≤i≤n, make any subclass that constitutes by m vector be linearity independently;

3. utilize vector set a _i=(a _I1..., a _Im), 1≤i≤n is divided into F n sub-piece in the following manner:

F _i=(a _i.S ₁, a _i.S ₂..., a _i.S _N/m), i=1 ..., n, wherein a _i.S _k=a _I1.b _{(k-1) m+1}+ ...+a _Im.b _KmThereby,

|F _i|＝|F|/m，C _i，i＝1，...，n；

4. produce n frame number with pseudo random sequence, the n that these sequence numbers an are represented frame is used for embedding n the sub-piece of F, and the method for embedding is identical with the embedding grammar of utilization in recovery and identifying algorithm protection single frames, and each frame embeds the sub-piece of F;

5. in order to detect to F _iMalice distort, calculate F respectively _i, i=1 ..., the cyclic redundancy check (CRC) of n, the check byte C of generation _i, i=1 ..., n is embedded into F _iFrame in, the mode of embedding is with to use the mean quantization watermark embedding method to embed in the frame number information identical.

5, the video-frequency identifying method with repair function based on digital watermarking according to claim 1 is characterized in that, described utilization is verified each frame from recovery and identifying algorithm, and is specific as follows:

Adopt each frame in signature of describing in recovery and the identifying algorithm and watermark extracting method checking video sequence, at first produce the signature of DCT coefficient by tactic preceding 6 coefficients of ZIG-ZAG, from back 6 DCT coefficients, extract watermark, both are relatively drawn the authentication result of this frame, promptly whether distorted, if distorted, orient 8 * 8DCT piece of being distorted.

6, the video-frequency identifying method with repair function based on digital watermarking according to claim 1 is characterized in that, described use mean quantization watermark extracting method extracts the serial number information of frame, and is specific as follows:

In 4 DC coefficient of 48 * 8DCT pieces of frame of video, extract 1 bit of 32 bit sequence, promptly extract Q _DcLSB, if utilize from recovery and identifying algorithm and verify that the result of each frame shows that this frame was illegally distorted, there is query in the sequence number of then directly pointing out this frame, otherwise then sequence number and the actual sequence number of this frame in video sequence that extracts is compared, thereby detects the attack at the time.

7, the video-frequency identifying method with repair function based on digital watermarking according to claim 1 is characterized in that, the important frame of video that described employing information dispersal algorithm recovery is lost is specific as follows:

If showing, the result who uses the mean quantization watermark extracting method to extract the serial number information of frame lost important frame of video, the information of orienting lost frames by key and pseudo random sequence is embedded in that n frame of video, as long as wherein m frame of video do not lost and the process authentication, just recover the important frame of losing, specific practice is:

(1) is located at the authentication end and obtained F ₁..., F _mSub-piece adopts m sub-piece, and then following formula is set up:

$A\·\left[\begin{array}{c}{b}_1\\ \·\\ \·\\ \·\\ b_m\end{array}\right]=\left[\begin{array}{c}{a}_1\·S_1\\ \·\\ \·\\ \·\\ a_m\·S_1\end{array}\right]$ A=(a wherein _Ij) _{1≤i, j≤m}Be the matrix of m * m, i is capable to be a _i

(2) because A in vector be linearity independently, so A is an invertible matrix, S ₁Can obtain from following formula:

$S_1=\left[\begin{array}{c}{b}_1\\ \·\\ \·\\ \·\\ b_n\end{array}\right]=A^{-1}\·\left[\begin{array}{c}{a}_1\·S_1\\ \·\\ \·\\ \·\\ a_m\·S_1\end{array}\right]$

(3) repeat (1), (2) process, can obtain S ₂..., S _N/m, these data are together in series can obtain F;

(4) after reconstruct is finished, the frame of video that F is decompressed and can be restored out through JPEG2000.

8, the video-frequency identifying method based on digital watermarking according to claim 7 with repair function, it is characterized in that, wherein m frame of video should meet following condition: 1) this frame is utilizing recovery certainly and identifying algorithm to verify in each frame by authentication, 2) sequence number of this frame is real, promptly by using the mean quantization watermark extracting method to extract the serial number information of frame, 3) F that extracts _iPass through cyclic redundancy check (CRC).