CN103237271B - Large-capacity video embedding algorithm and detection method and system thereof - Google Patents

Abstract

The invention provides a kind of large-capacity video embedding algorithm and detection method and system thereof, embedding grammar is: video watermark W is decomposed into 1 watermark head WH and M sub-watermark WS (I); All frame of video F of original video are divided into M+1 group frame of video, obtain 1 group of frame of video group FH and M group frame of video group FS (I) respectively; Built-in template is embedded in frame of video group FH, obtains frame of video group FHT; Watermark head WH is embedded in frame of video group FHT, obtains frame of video group FHW; M sub-watermark WS (I) is embedded in M group frame of video FS (I) respectively, obtains M group frame of video FSW (I); Composite video frame group FHW and M group frame of video FSW (I), obtains watermarked vision signal FW in order.Embed large-capacity video digital watermarking in video, and effectively can resist spatial domain geometric attack, lossy compression method, frame of video is deleted, frame of video is inserted, frame of video is average and the attack such as video cutting.

Description

Large-capacity video embedding algorithm and detection method and system thereof

Technical field

The invention belongs to Digital Video Watermarking Techniques field, be specifically related to a kind of large-capacity video embedding algorithm and detection method and system thereof.

Background technology

Digital watermarking is the technology embedding and detect figure notation in digital picture, digital video, digital audio or digital document carrier, can be widely used in the applications such as digital content searching, advertisement detecting, copyright, piracy tracking.

1954, EmilHembrooke proposes patent " Identificationofsoundandlikesignals ", declared the birth of digital watermark technology, this patent by being embedded in music with non method by a kind of identification code, thus realizes the authentication of music making people.Now digital watermarking does not become an independently subject, until early 1990s, digital watermarking progressively becomes research topic and comes into one's own.In 1993, A.Z.Tirkel etc. employed " Watermark " this term at " Electronicwatermark " in literary composition first, indicated the birth of digital watermark technology as a formal research subject.

Early stage Digital Watermark Robustness is poor, and such as VanSChyndel, R.G etc. in 1994 propose the data waterprint embedded method revising image least significant bit LSB, and its advantage is that digital watermarking is better disguised, but poor robustness.After this digital watermarking algorithm based on transform domain is suggested successively, comprises DFT, DCT, DWT and KLT Digital Watermarking System in Transform Domain algorithm, improves the robustness of digital watermarking.The people such as Cox in 1997 propose a kind of digital watermarking algorithm based on spread spectrum communication thought, and be called for short spread-spectrum watermark algorithm, this algorithm has good fail safe and very high robustness, to become in digital watermark technology a more classical scheme.In raising watermark capacity, the people such as calendar year 2001 BrianChen propose the digital watermarking algorithm based on quantization index modulation (QuantizationIndexModulation), improve the robustness of large capacity digital watermark.

Video digital watermark refers to the technology of embedding and detection figure notation in digital video carrier, and video watermark not only needs the image fault of resisting spatial domain to attack, and also needs the distortion of resistance time axle to attack.Realization about video watermark technology mainly contains two large class methods, and successive video frames is considered as entirety by the former, watermarked along time shaft, by revising brightness or the colourity of consecutive frame, reaches the object embedding and detect video watermark; Each frame of video is considered as independently picture by the latter, by revising brightness or the colourity of frame of video, by watermark embedment in single frame of video.Last class methods need to obtain simultaneously and analyze some successive video frames usually, and be easily subject to the impact that time shaft distortion is attacked, watermark embedment capacity is also smaller in addition, is not suitable in short-sighted frequency watermarked; Rear class methods can detect watermark in single-frame images, have greater flexibility, and watermark capacity is also larger, class methods after the invention belongs to.

Watermark capacity, invisibility, robustness are three important indicators in video watermark field, mainly there is the problem of following four aspects in existing Digital Video Watermarking Techniques: first, the watermark capacity that can embed in single-frame images is less, be difficult to the demand meeting application, need the multiframe utilized in video, realize the expansion of watermark capacity; Secondly, frame of video normally Continuous Play, general digital watermark is only watermarked in I frame, not P frame and B frame watermarked, cause distortion degree difference between consecutive frame huge, easily occur video playback scintillation; In addition, the video time axle distortions such as frame deletion, frame are average, frame rate conversion, camera lens cutting can cause mutually disturbing between consecutive frame, and destroy the precedence between frame of video, easily cause video watermark detection mistake; Finally, the spatial domain geometric attacks such as translation, shearing, convergent-divergent, rotation, easily destroy the spatial synchronization of watermark, cause watermark normally not detect.

Summary of the invention

For the defect that prior art exists, the invention provides a kind of large-capacity video embedding algorithm and detection method and system thereof, for embedding large-capacity video digital watermarking in video, and effectively can resist spatial domain geometric attack, lossy compression method, frame of video is deleted, frame of video is inserted, frame of video is average and the attack such as video cutting.

The technical solution used in the present invention is as follows:

The invention provides a kind of large-capacity video embedding algorithm and detection method, comprise the following steps:

S1, is decomposed into 1 watermark header WH and M sub-watermark information WS (I) by video watermark information W, I ∈ 0,1 ..., M-1}, M are integer, and M >=1; Wherein, M sub-watermark information WS (I) is designated as WS (0) ~ WS (M-1);

All frame of video F of raw video signal are divided into M+1 group frame of video group by video playback time sequencing, obtain 1 group of frame of video group FH and M group frame of video group FS (I) respectively, I ∈ 0,1 ..., M-1}, M are integer, and M >=1; Wherein, M group frame of video group FS (I) is designated as FS (0) ~ FS (M-1);

S2, is embedded into built-in template in frame of video group FH, obtains frame of video group FHT;

Watermark header WH is embedded in frame of video group FHT, obtains frame of video group FHW;

M sub-watermark information WS (I) is embedded in M group frame of video group FS (I) respectively, obtains M group frame of video group FSW (I); Wherein, M group frame of video group FSW (I) is designated as FSW (0) ~ FSW (M-1);

S3, composite video frame group FHW and M group frame of video group FSW (0) ~ FSW (M-1), obtains watermarked vision signal FW in order;

S4, utilizes described built-in template to carry out geometric correction to vision signal FW;

S5, detects watermark header WH in the vision signal through geometric correction;

And, in the vision signal through geometric correction, detect sub-watermark information WS (I);

S6, merges watermark header WH and sub-watermark information WS (I), generates complete video watermark information W.

Preferably, S1, is decomposed into 1 watermark header WH and M sub-watermark information WS (I) and specifically comprises the following steps by video watermark information W:

Watermark header WH acquisition methods is:

S1-1, is equally divided into M section by the video watermark information W of L bit, obtains M localized watermark information W (I); Wherein, M localized watermark information W (I) is designated as W (0) ~ W (M-1); The then length N1 bit=L bit/M-bit of each localized watermark information W (I);

S1-2, carries out binary coding to total number M of local watermark information, obtains binary number K; Wherein, the number of bits N2 of binary number K is for being greater than the integer of log (M-1)/log (2);

Based on L and N3, calculate the CRC check code of video watermark information W, be designated as video watermark check code C; Wherein, N3 is set point, is the number of bits of video watermark check code C;

Combine K and C sequentially, obtain watermark header WH; Wherein, the number of bits of watermark header WH is N2+N3;

M sub-watermark information WS (I) acquisition methods is:

Respectively binary coding is carried out to the segment number I of M localized watermark information W (I), obtain M binary number K (I); Wherein, the number of bits of binary number K (I) is N2; M binary number K (I) is designated as K (0) ~ K (M-1);

Calculate the CRC check code of W (I) and K (I), obtain sub-watermark check code C (I); Wherein, the number of bits of sub-watermark check code C (I) is N4;

Combine W (I), K (I) and C (I) sequentially, obtain the sub-watermark information WS (I) of (N1+N2+N4) bit.

Preferably, in S1, all frame of video of raw video signal are divided into M+1 group frame of video group by video playback time sequencing, obtain 1 group of frame of video group FH and M group frame of video group FS (I) respectively, specifically comprise the following steps:

Raw video signal is divided into the video-frequency band of fixing frame number;

Each video-frequency band is subdivided into the frame of video group that (M+1) individual frame number is D again, namely obtains 1 group of frame of video group FH and M group frame of video group FS (I).

Preferably, in S2, built-in template is embedded in frame of video group FH, obtains frame of video group FHT, specifically comprise the following steps:

Read the two field picture pixel data of frame of video group FH, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Choose NT template frequency in described DFT territory;

Revise the DFT coefficient of the described template frequency position chosen, then described built-in template is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding built-in template, obtains the frame of video group FHT embedding built-in template.

Preferably, in S2, watermark header WH is embedded in frame of video group FHT, obtains frame of video group FHW, specifically comprise the following steps:

Described watermark head embeds and comprises the steps:

Read the two field picture pixel data of frame of video group FHT, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Choose NH watermark head frequency in DFT territory;

Revise the DFT coefficient of the described watermark head frequency position chosen, then described watermark header WH is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient of watermarked header, obtains the frame of video group FHW of watermarked header.

Preferably, in S2, M sub-watermark information WS (I) is embedded in M group frame of video group FS (I) respectively, obtains M group frame of video group FSW (I), specifically comprise the following steps:

Read the two field picture pixel data of frame of video group FS (I), overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Choose NS sub-watermark frequency in described DFT territory;

Revise the DFT coefficient of the described sub-watermark frequency position chosen, then described sub-watermark information WS (I) is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding sub-watermark information WS (I), obtains the frame of video group FSW (I) embedding sub-watermark information WS (I).

Preferably, S4, utilizes described built-in template to carry out geometric correction to vision signal FW, specifically comprises the following steps:

Read the two field picture pixel data of vision signal FW, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read;

The DFT coefficient obtained after overall DFT discrete Fourier transform is analyzed, judges whether to comprise built-in template; If built-in template do not detected, then keep raw video image, exit geometric correction operation; If built-in template detected, then contrasting detecting the built-in template obtained with the original built-in template prestored, calculating Image scaling coefficients and the anglec of rotation; Then the Image scaling coefficients and the anglec of rotation that calculate is preserved;

According to described Image scaling coefficients and the anglec of rotation, reverse convergent-divergent and reverse rotation are carried out to subsequent frame image, realize carrying out geometric correction to vision signal FW.

Preferably, S5, detects watermark header WH, specifically comprises the following steps in the vision signal through geometric correction:

Read the two field picture pixel data through the vision signal of geometric correction, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Analyze the DFT coefficient of NH watermark head frequency, extract watermark header WH;

From watermark header WH, extract data K and check code C, obtain sub-watermark sum M by K.

Preferably, in the vision signal through geometric correction, detect sub-watermark information WS (I), specifically comprise the following steps:

Read the two field picture pixel data through the vision signal of geometric correction, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Analyze the DFT coefficient of NS sub-watermark frequency, extract sub-watermark WS (I), I ∈ 0,1 ..., M-1};

From sub-watermark WS (I), extract W (I), K (I) and C (I), obtain segment number I corresponding to sub-watermark according to K (I);

Checking C (I) check code, if check code is incorrect, then deletes this sub-watermark; Otherwise, if check code is correct, then sub-watermark information WS (I) successfully detected.

Preferably, S6, merges watermark header WH and sub-watermark information WS (I), generates complete video watermark information W, specifically comprise the following steps:

From sub-watermark information WS (0) ~ WS (M-1), extract localized watermark information W (0) ~ W (M-1) respectively, the localized watermark information W (0) extracted ~ W (M-1) is connected to become video watermark information W;

Checking C check code, if check code is incorrect, then deletes this video watermark information W, again detects watermark header and sub-watermark information; Otherwise, if check code is correct, then video watermark information W successfully detected.

The present invention also provides a kind of large-capacity video embedding algorithm and detection system, comprises watermark embedding module (100) and watermark detection module (110);

Described watermark embedding module (100) comprising:

Video frame packet unit (101), for all frame of video F of raw video signal are divided into M+1 group frame of video group by video playback time sequencing, obtains 1 group of frame of video group FH and M group frame of video group FS (I) respectively, I ∈ 0,1 ... M-1}, M are integer, and M >=1; Wherein, M group frame of video group FS (I) is designated as FS (0) ~ FS (M-1);

Video watermark resolving cell (103), for video watermark information W being decomposed into 1 watermark header WH and M sub-watermark information WS (I), I ∈ 0,1 ..., M-1}, M are integer, and M >=1; Wherein, M sub-watermark information WS (I) is designated as WS (0) ~ WS (M-1);

Template embedded unit (102), for built-in template being embedded in frame of video group FH, obtains frame of video group FHT;

Watermark head embedded unit (104), is embedded into watermark header WH in frame of video group FHT, obtains frame of video group FHW;

Sub-watermark embedder unit (105), is embedded into M sub-watermark information WS (I) respectively in M group frame of video group FS (I), obtains M group frame of video group FSW (I); Wherein, M group frame of video group FSW (I) is designated as FSW (0) ~ FSW (M-1);

Frame of video assembled unit (106), for composite video frame group FHW and M group frame of video group FSW (0) ~ FSW (M-1) in order, obtains watermarked vision signal FW;

Watermark detection module (110) comprising:

Geometry correction unit (111), utilizes described built-in template to carry out geometric correction to vision signal FW;

Watermark head detecting unit (112), for detecting watermark header WH in the vision signal through geometric correction;

Sub-watermark detection unit (113), for detecting sub-watermark information WS (I) in the vision signal through geometric correction;

Video watermark integrated unit (114), for merging watermark header WH and sub-watermark information WS (I), generates complete video watermark information W.

Preferably, described template embedded unit (102), specifically for the two field picture pixel data that reads frame of video group FH, is carried out overall DFT discrete Fourier transform to the described two field picture pixel data read, is obtained DFT territory;

Choose NT template frequency in described DFT territory;

Revise the DFT coefficient of the described template frequency position chosen, then described built-in template is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding built-in template, obtains the frame of video group FHT embedding built-in template;

Described watermark head embedded unit (104), specifically for the two field picture pixel data that reads frame of video group FHT, is carried out overall DFT discrete Fourier transform to the described two field picture pixel data read, is obtained DFT territory;

Choose NH watermark head frequency in DFT territory;

Revise the DFT coefficient of the described watermark head frequency position chosen, then described watermark header WH is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient of watermarked header, obtains the frame of video group FHW of watermarked header;

Described sub-watermark embedder unit (105) specifically for: read frame of video group FS (I) two field picture pixel data, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Choose NS sub-watermark frequency in described DFT territory;

Revise the DFT coefficient of the described sub-watermark frequency position chosen, then described sub-watermark information WS (I) is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding sub-watermark information WS (I), obtains the frame of video group FSW (I) embedding sub-watermark information WS (I).

Preferably, described frame of video assembled unit (106) specifically for:

Raw video signal is divided into the video-frequency band of fixing frame number;

Each video-frequency band is subdivided into the frame of video group that (M+1) individual frame number is D again, namely obtains 1 group of frame of video FH and M group frame of video FS (I).

Preferably, described geometry correction unit (111) specifically for:

Read the two field picture pixel data of vision signal FW, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read;

The DFT coefficient obtained after overall DFT discrete Fourier transform is analyzed, judges whether to comprise built-in template; If built-in template do not detected, then keep raw video image, exit geometric correction operation; If built-in template detected, then contrasting detecting the built-in template obtained with the original built-in template prestored, calculating Image scaling coefficients and the anglec of rotation; Then the Image scaling coefficients and the anglec of rotation that calculate is preserved;

According to described Image scaling coefficients and the anglec of rotation, reverse convergent-divergent and reverse rotation are carried out to subsequent frame image, realize carrying out geometric correction to vision signal FW.

Preferably, described sub-watermark detection unit (113) specifically for:

Read the two field picture pixel data through the vision signal of geometric correction, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Analyze the DFT coefficient of NS sub-watermark frequency, extract sub-watermark WS (I), I ∈ 0,1 ..., M-1};

From sub-watermark WS (I), extract W (I), K (I) and C (I), obtain segment number I corresponding to sub-watermark according to K (I);

Checking C (I) check code, if check code is incorrect, then deletes this sub-watermark; Otherwise, if check code is correct, then sub-watermark information WS (I) successfully detected;

Described video watermark integrated unit (114) specifically for:

From sub-watermark information WS (0) ~ WS (M-1), extract localized watermark information W (0) ~ W (M-1) respectively, the localized watermark information W (0) extracted ~ W (M-1) is connected to become video watermark information W;

Checking C check code, if check code is incorrect, then deletes this video watermark information W, again detects watermark header and sub-watermark information; Otherwise, if check code is correct, then video watermark information W successfully detected.

Beneficial effect of the present invention is as follows:

(1) watermark capacity is large: by video watermark being resolved into 1 watermark head and some sub-watermarks, increased substantially the capacity of video watermark;

(2) effectively resisting spatial domain geometric attack: by embedding built-in template in the video frame, during watermark detection, carry out template matches, correcting geometric deformation, realizing attacks such as spatial domain translation, shearing, convergent-divergent, rotations;

(3) video distortion degree is low: be embedded in FH frame of video group by built-in template and less first of the watermark of data volume, by sub-watermark embedment larger for data volume in FS (0) ~ FS (M-1) frame of video group, the distortion of all frame of video has good consistency, effectively prevents the flicker that video watermark causes;

(4) can resistance time synchronous distortion: video watermark just detects in single frames, and then is fused into complete watermark, and therefore watermark detection does not rely on the time sequencing between frame of video, can effectively resistance time synchronous distortion.

Accompanying drawing explanation

Fig. 1 is the frame diagram of large-capacity video embedding algorithm provided by the invention and detection system;

Fig. 2 is the schematic flow sheet of large-capacity video embedding algorithm provided by the invention and detection method;

Fig. 3 is video watermark information W decomposition texture schematic diagram provided by the invention;

Fig. 4 is video frame packet structural representation provided by the invention;

Fig. 5 is that built-in template provided by the invention embeds frequency schematic diagram;

Fig. 6 is that watermark head provided by the invention embeds frequency schematic diagram;

Fig. 7 is sub-watermark embedment frequency schematic diagram provided by the invention;

Fig. 8 is the schematic flow sheet of video digital watermark detection method provided by the invention;

Fig. 9 is video watermark fusion structure schematic diagram provided by the invention.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described in detail:

As shown in Figure 1, for the frame diagram of large-capacity video embedding algorithm provided by the invention and detection system, apply this system, as shown in Figure 2, the schematic flow sheet of large-capacity video embedding algorithm provided by the invention and detection method, mainly comprise digital watermark embed process and digital watermark detecting process, introduce this two processes below in detail:

(1) digital watermark embed process

S1, is decomposed into 1 watermark header WH and M sub-watermark information WS (I) by video watermark information W, I ∈ 0,1 ..., M-1}, M are integer, and M >=1; Wherein, M sub-watermark information WS (I) is designated as WS (0) ~ WS (M-1);

All frame of video F of raw video signal are divided into M+1 frame of video group by video playback time sequencing, obtain 1 group of frame of video group FH and M group frame of video group FS (I) respectively, I ∈ 0,1 ..., M-1}, M are integer, and M >=1; Wherein, M group frame of video group FS (I) is designated as FS (0) ~ FS (M-1).

In this step, video watermark information W is decomposed into 1 watermark header WH and M sub-watermark information WS (I), as shown in Figure 3, specifically comprises the following steps:

Watermark header WH acquisition methods is:

S1-1, is equally divided into M section by the video watermark information W of L bit, obtains M localized watermark information W (I); Wherein, M localized watermark information W (I) is designated as W (0) ~ W (M-1); The then length N1 bit=L bit/M-bit of each localized watermark information W (I);

S1-2, carries out binary coding to total number M of local watermark information, obtains binary number K; Wherein, the number of bits N2 of binary number K is for being greater than the integer of log (M-1)/log (2);

Based on L and N3, calculate the CRC check code of video watermark information W, be designated as video watermark check code C; Wherein, N3 is set point, is the number of bits of video watermark check code C;

Combine K and C sequentially, obtain watermark header WH; Wherein, the number of bits of watermark header WH is N2+N3;

M sub-watermark information WS (I) acquisition methods is:

Respectively binary coding is carried out to the segment number I of M localized watermark information W (I), obtain M binary number K (I); Wherein, the number of bits of binary number K (I) is N2; M binary number K (I) is designated as K (0) ~ K (M-1);

Calculate the CRC check code of W (I) and K (I), obtain sub-watermark check code C (I); Wherein, the number of bits of sub-watermark check code C (I) is N4;

Combine W (I), K (I) and C (I) sequentially, obtain the sub-watermark information WS (I) of (N1+N2+N4) bit.

Below for 128 bit video watermark information W, the decomposable process of video watermark is described.

M value is 4, N1 value be 32, L value be 128, N2 value be 2, N3 value be 13, N4 value is 8.Video watermark information W is equally divided into 4 sections, and segment number is respectively 0,1,2,3; 4 localized watermark information W (I) are designated as W (0), W (1), W (2) and W (3) respectively, and every section of localized watermark message length is 32 bits, and video watermark information total length is 128 bits; With 2 bit binary number K, watermark information total hop count 4 in local is encoded, represent with 11 the coding that M equals 4; Calculate the CRC check of video watermark information W, obtain the video watermark check code C of 13 bits; Combine K and C sequentially, obtain the watermark header WH of 15 bits.

With 2 bit binary number K (I), the segment number I of local watermark information W (I) is encoded, I ∈ 0,1,2,3}, 00,01,10 and 11 represent that I equals 0,1,2 and 3 respectively; Calculate the CRC check of W (I) and K (I), obtain the sub-watermark check code of 8 bits, be designated as C (I); Combine W (I), K (I) and C (I) sequentially, obtain the sub-watermark information WS (I) of 42 bits.

All frame of video of raw video signal are divided into M+1 frame of video group by video playback time sequencing, obtain 1 group of frame of video group FH and M group frame of video group FS (I) respectively, specifically comprise the following steps:

Raw video signal is divided into the video-frequency band of fixing frame number;

Each video-frequency band is subdivided into the video segment that (M+1) individual frame number is D again, to the video segment grouping obtained, namely obtains 1 frame of video group FH and M frame of video group FS (I).

As shown in the example in Fig. 4, for video frame packet structural representation, when M value is 4, when D value is 2, raw video signal is divided into the video-frequency band that length is 10 frames, the video segment of each video-frequency band is split again to be 5 length be 2 frames, the 1st video segment is frame of video group FH, and 2nd ~ 5 video segments are respectively FS (0), FS (1), FS (2), FS (3).

S2, built-in template is embedded in frame of video group FH by (1), obtains frame of video group FHT.

Concrete, read the two field picture pixel data of frame of video group FH, pixel data can be luminance component or chromatic component, carries out overall DFT discrete Fourier transform, obtain DFT territory to the described two field picture pixel data read; Digital watermarking is according to the be divided into pixel domain of embedded domain, DCT domain, DFT territory and KLT territory.Consider that the translation of DFT transfer pair spatial domain has consistency, and spatial domain convergent-divergent and rotation also just cause convergent-divergent corresponding in DFT territory and rotation, very little on the impact of DFT coefficient, the present invention is based on DFT conversion and carry out video watermark embedding and detection, spatial domain geometric attack can be resisted preferably.

Choose NT template frequency in described DFT territory;

Revise the DFT coefficient of the described template frequency position chosen, then described built-in template is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding built-in template, obtains the frame of video group FHT embedding built-in template.

As shown in the example in fig. 5, for built-in template embeds frequency schematic diagram, NT value is 26, and template frequency position can be selected flexibly, and template determines that rear needs record is preserved, and remains unchanged, and carries out geometric correction after being convenient to by template.

In this step, the amendment of DFT coefficient just increases the amplitude of DFT coefficient, and keep the arc angle of DFT coefficient constant, computing formula is as follows, Parameters in Formula λ ₁be masterplate embedment strength, can require to select flexibly according to application.

${\mathrm{DFT}}^{\′}(m,n)=DFT(m,n)+{\mathrm{\λ}}_1\×\frac{DFT(m,n)}{\left|\right|DFT(m,n)\left|\right|}$

(2) watermark header WH is embedded in frame of video group FHT, obtains frame of video group FHW;

Concrete, watermark head embeds and comprises the steps:

Read the two field picture pixel data of frame of video group FHT, pixel data can be luminance component or chromatic component, carries out overall DFT discrete Fourier transform, obtain DFT territory to the described two field picture pixel data read;

Choose NH watermark head frequency in DFT territory;

Revise the DFT coefficient of the described watermark head frequency position chosen, then described watermark header WH is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient of watermarked header, obtains the frame of video group FHT of watermarked header.

As shown by way of example in fig. 6, for watermark head embeds frequency schematic diagram, NH value is 5, can embed 15 bit watermark headers.Watermark head frequency adds up to 40, and watermark head frequency position can be selected flexibly, and these frequencies are divided into 5 groups, often organizes 8 frequencies.15 bit watermark headers are also divided into 5 groups, often organize 3 bit watermark headers, have 8 kinds of combinations: 000,001,010,011,100,101,110 and 111, respectively corresponding each group 8 frequencies.

The embedding grammar of watermark head is as follows, selecting 1 frequency respectively, altogether selecting to obtain 5 frequencies according to often organizing watermark head in corresponding group.Increase the DFT coefficient amplitude of these frequencies, keep DFT coefficient arc angle constant, computing formula is as follows, Parameters in Formula λ ₂be watermark head embedment strength, can require to select flexibly according to application.

${\mathrm{DFT}}^{\′}(m,n)=DFT(m,n)+{\mathrm{\λ}}_2\×\frac{DFT(m,n)}{\left|\right|DFT(m,n)\left|\right|}$

(3) M sub-watermark information WS (I) is embedded in M group frame of video group FS (I) respectively, obtains M group frame of video group FSW (I); Wherein, M group frame of video group FSW (I) is designated as FSW (0) ~ FSW (M-1);

Concrete, read the two field picture pixel data of frame of video group FS (I), pixel data can be luminance component or chromatic component, carries out overall DFT discrete Fourier transform, obtain DFT territory to the described two field picture pixel data read;

Choose NS sub-watermark frequency in described DFT territory;

Revise the DFT coefficient of the described sub-watermark frequency position chosen, then described sub-watermark information WS (I) is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding sub-watermark information WS (I), obtains the frame of video group FSW (I) embedding sub-watermark information WS (I).

As shown in the example in Fig. 7, be sub-watermark embedment frequency schematic diagram, NS value is 14, can embed the sub-watermark of 42 bit.Sub-watermark frequency adds up to 112, and frequency position can be selected flexibly, and these frequencies are divided into 14 groups, often organizes 8 frequencies.42 bit watermark heads are also divided into 14 groups, often organize the sub-watermark of 3 bit, have 8 kinds of combinations: 000,001,010,011,100,101,110 and 111, respectively corresponding each group 8 frequencies.

The embedding grammar of sub-watermark is as follows, selects 1 frequency respectively according to often organizing sub-watermark in corresponding group, altogether selects to obtain 14 frequencies.Increase the DFT coefficient amplitude of these frequencies, keep DFT coefficient arc angle constant, computing formula is as follows, Parameters in Formula λ ₃be sub-watermark embed strength, can require to select flexibly according to application.

${\mathrm{DFT}}^{\′}(m,n)=DFT(m,n)+{\mathrm{\λ}}_3\×\frac{DFT(m,n)}{\left|\right|DFT(m,n)\left|\right|}$

S3, composite video frame group FHW and M group frame of video group FSW (0) ~ FSW (M-1), obtains watermarked vision signal FW in order.

Each group of frame of video is reconfigured the video becoming moisture print according to original time sequencing.

(2) digital watermark detecting process

Corresponding with above-mentioned watermark embedding method, as shown in Figure 8, be the schematic flow sheet of video digital watermark detection method, as seen from the figure, mainly comprise the following steps:

S4, utilizes described built-in template to carry out geometric correction to vision signal FW;

Concrete, read the two field picture pixel data of vision signal FW, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read;

The DFT coefficient obtained after overall DFT discrete Fourier transform is analyzed, judges whether to comprise built-in template; If built-in template do not detected, then keep raw video image, exit geometric correction operation; If built-in template detected, then contrasting detecting the built-in template obtained with the original built-in template prestored, calculating Image scaling coefficients and the anglec of rotation; Then the Image scaling coefficients and the anglec of rotation that calculate is preserved;

According to described Image scaling coefficients and the anglec of rotation, reverse convergent-divergent and reverse rotation are carried out to subsequent frame image, realize carrying out geometric correction to vision signal FW.

Wherein, in the present invention, zoom factor and the anglec of rotation can be calculated by multiple method contrast detection masterplate and original stencil.Such as adopt direct exhaustive search, by different zoom factors and the anglec of rotation, convergent-divergent and rotation are carried out to detection masterplate respectively, and then contrast with original stencil, until consistent with original stencil, thus obtain zoom factor and coefficient of rotary.

S5, detects watermark header WH in the vision signal through geometric correction;

And, in the vision signal through geometric correction, detect sub-watermark information WS (I);

Wherein, watermark header WH testing process is: the two field picture pixel data reading the vision signal through geometric correction, carries out overall DFT discrete Fourier transform, obtain DFT territory to the described two field picture pixel data read; Analyze the DFT coefficient of NH watermark head frequency, extract watermark header WH; From watermark header WH, extract data K and check code C, obtain sub-watermark sum M by K.

Below in conjunction with the watermark head frequency example in Fig. 6, the DFT coefficient of descriptive analysis watermark head frequency, extracts the method for watermark head.

Take out the DFT coefficient of 5 groups of totally 40 watermark head frequencies, contrast the DFT coefficient amplitude of 8 frequencies in each group, find the frequency of amplitude maximum, correspond to corresponding 3 Bit datas, 5 group of 3 Bit data is combined, obtain 15 bit watermark header WH.

Sub-watermark information WS (I) testing process is: the two field picture pixel data reading the vision signal through geometric correction, carries out overall DFT discrete Fourier transform, obtain DFT territory to the described two field picture pixel data read; Analyze the DFT coefficient of NS sub-watermark frequency, extract sub-watermark WS (I), I ∈ 0,1 ..., M-1}; From sub-watermark WS (I), extract W (I), K (I) and C (I), obtain segment number I corresponding to sub-watermark according to K (I); Checking C (I) check code, if check code is incorrect, then deletes this sub-watermark; Otherwise, if check code is correct, then sub-watermark information WS (I) successfully detected.

Below in conjunction with the sub-watermark frequency example in Fig. 7, the DFT coefficient of the sub-watermark frequency of descriptive analysis, extracts the method for sub-watermark.

Take out the DFT coefficient of 14 groups of totally 112 watermark head frequencies, contrast the DFT coefficient amplitude of 8 frequencies in each group, find the frequency of amplitude maximum, correspond to corresponding 3 Bit datas, 14 group of 3 Bit data is combined, obtains the sub-watermark WS (I) of 42 bit.

S6, merges watermark header WH and sub-watermark information WS (I), generates complete video watermark information W.

As shown in Figure 9, for video watermark fusion structure schematic diagram, specifically comprise: from sub-watermark information WS (0) ~ WS (M-1), extract localized watermark information W (0) ~ W (M-1) respectively, the localized watermark information W (0) extracted ~ W (M-1) is connected to become video watermark information W;

Checking C check code, if check code is incorrect, then deletes this video watermark information W, again detects watermark header and sub-watermark information; Otherwise, if check code is correct, then video watermark information W successfully detected.

Composition graphs 1 again, large-capacity video embedding algorithm provided by the invention and detection system: comprise

Watermark embedding module (100) and watermark detection module (110); Described watermark embedding module (100) comprising:

Video frame packet unit (101), for all frame of video F of raw video signal are divided into M+1 group frame of video group by video playback time sequencing, obtains 1 group of frame of video group FH and M group frame of video group FS (I) respectively, I ∈ 0,1 ... M-1}, M are integer, and M >=1; Wherein, M group frame of video group FS (I) is designated as FS (0) ~ FS (M-1);

Video watermark resolving cell (103), for video watermark information W being decomposed into 1 watermark header WH and M sub-watermark information WS (I), I ∈ 0,1 ..., M-1}, M are integer, and M >=1; Wherein, M sub-watermark information WS (I) is designated as WS (0) ~ WS (M-1);

Template embedded unit (102), for built-in template being embedded in frame of video group FH, obtains frame of video group FHT; Template embedded unit (102), specifically for the two field picture pixel data that reads frame of video group FH, carries out overall DFT discrete Fourier transform to the described two field picture pixel data read, obtains DFT territory;

Choose NT template frequency in described DFT territory;

Revise the DFT coefficient of the described template frequency position chosen, then described built-in template is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding built-in template, obtains the frame of video group FHT embedding built-in template;

Watermark head embedded unit (104), is embedded into watermark header WH in frame of video group FHT, obtains frame of video group FHW; Watermark head embedded unit (104), specifically for the two field picture pixel data that reads frame of video group FHT, carries out overall DFT discrete Fourier transform to the described two field picture pixel data read, obtains DFT territory;

Choose NH watermark head frequency in DFT territory;

Revise the DFT coefficient of the described watermark head frequency position chosen, then described watermark header WH is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient of watermarked header, obtains the frame of video group FHT of watermarked header;

Sub-watermark embedder unit (105), is embedded into M sub-watermark information WS (I) respectively in M group frame of video group FS (I), obtains M group frame of video group FSW (I); Wherein, M group frame of video group FSW (I) is designated as FSW (0) ~ FSW (M-1); Sub-watermark embedder unit (105) specifically for: read frame of video group FS (I) two field picture pixel data, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Choose NS sub-watermark frequency in described DFT territory;

Revise the DFT coefficient of the described sub-watermark frequency position chosen, then described sub-watermark information WS (I) is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding sub-watermark information WS (I), obtains the frame of video group FSW (I) embedding sub-watermark information WS (I).

Frame of video assembled unit (106), for composite video frame group FHW and M group frame of video group FSW (0) ~ FSW (M-1) in order, obtains watermarked vision signal FW; Frame of video assembled unit (106) specifically for:

Raw video signal is divided into the video-frequency band of fixing frame number;

Each video-frequency band is subdivided into the frame of video group that (M+1) individual frame number is D again, namely obtains 1 group of frame of video group FH and M group frame of video group FS (I).

Watermark detection module (110) comprising:

Geometry correction unit (111), utilizes described built-in template to carry out geometric correction to vision signal FW; Geometry correction unit (111) specifically for:

Read the two field picture pixel data of vision signal FW, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read;

The DFT coefficient obtained after overall DFT discrete Fourier transform is analyzed, judges whether to comprise built-in template; If built-in template do not detected, then keep raw video image, exit geometric correction operation; If built-in template detected, then contrasting detecting the built-in template obtained with the original built-in template prestored, calculating Image scaling coefficients and the anglec of rotation; Then the Image scaling coefficients and the anglec of rotation that calculate is preserved;

According to described Image scaling coefficients and the anglec of rotation, reverse convergent-divergent and reverse rotation are carried out to subsequent frame image, realize carrying out geometric correction to vision signal FW.Geometry correction unit (111) specifically for:

Read the two field picture pixel data of vision signal FW, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read;

The DFT coefficient obtained after overall DFT discrete Fourier transform is analyzed, judges whether to comprise built-in template; If built-in template do not detected, then keep raw video image, exit geometric correction operation; If built-in template detected, then contrasting detecting the built-in template obtained with the original built-in template prestored, calculating Image scaling coefficients and the anglec of rotation; Then the Image scaling coefficients and the anglec of rotation that calculate is preserved;

According to described Image scaling coefficients and the anglec of rotation, reverse convergent-divergent and reverse rotation are carried out to subsequent frame image, realize carrying out geometric correction to vision signal FW.

Watermark head detecting unit (112), for detecting watermark header WH in the vision signal through geometric correction; Sub-watermark detection unit (113) specifically for:

Read the two field picture pixel data through the vision signal of geometric correction, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Analyze the DFT coefficient of NS sub-watermark frequency, extract sub-watermark WS (I), I ∈ 0,1 ..., M-1};

From sub-watermark WS (I), extract W (I), K (I) and C (I), obtain segment number I corresponding to sub-watermark according to K (I);

Checking C (I) check code, if check code is incorrect, then deletes this sub-watermark; Otherwise, if check code is correct, then sub-watermark information WS (I) successfully detected;

Sub-watermark detection unit (113), for detecting sub-watermark information WS (I) in the vision signal through geometric correction;

Video watermark integrated unit (114), for merging watermark header WH and sub-watermark information WS (I), generates complete video watermark information W.Video watermark integrated unit (114) specifically for:

From sub-watermark information WS (0) ~ WS (M-1), extract localized watermark information W (0) ~ W (M-1) respectively, the localized watermark information W (0) extracted ~ W (M-1) is connected to become video watermark information W;

Checking C check code, if check code is incorrect, then deletes this video watermark information W, again detects watermark header and sub-watermark information; Otherwise, if check code is correct, then video watermark information W successfully detected.

Video digital watermark method disclosed in this invention may be used for embedding large-capacity video digital watermarking in video, has the following advantages:

(1) watermark capacity is large: by video watermark being resolved into 1 watermark head and some sub-watermarks, increased substantially the capacity of video watermark;

(2) effectively spatial domain geometric attack is resisted: by embedding built-in template in the video frame, template matches is carried out during watermark detection, correct geometric deformation, effectively can resist spatial domain geometric attack, lossy compression method, frame of video is deleted, frame of video is inserted, frame of video is average and the attack such as video cutting.

(3) video distortion degree is low: be embedded in FH frame of video group by built-in template and less first of the watermark of data volume, by sub-watermark embedment larger for data volume in FS (0) ~ FS (M-1) frame of video group, the distortion of all frame of video has good consistency, effectively prevents the flicker that video watermark causes;

(4) can resistance time synchronous distortion: video watermark just detects in single frames, and then is fused into complete watermark, and therefore watermark detection does not rely on the time sequencing between frame of video, can effectively resistance time synchronous distortion.

The above is only the preferred embodiment of the present invention; it should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications also should look protection scope of the present invention.

Claims (15)

1. large-capacity video embedding algorithm and a detection method, is characterized in that, comprises the following steps:

S1, is decomposed into 1 watermark header WH and M sub-watermark information WS (I) by video watermark information W, I ∈ 0,1 ..., M-1}, M are integer, and M >=1; Wherein, M sub-watermark information WS (I) is designated as WS (0) ~ WS (M-1);

All frame of video F of raw video signal are divided into M+1 group frame of video group by video playback time sequencing, obtain 1 group of frame of video group FH and M group frame of video group FS (I) respectively, I ∈ 0,1 ..., M-1}, M are integer, and M >=1; Wherein, M group frame of video group FS (I) is designated as FS (0) ~ FS (M-1);

S2, is embedded into built-in template in frame of video group FH, obtains frame of video group FHT;

Watermark header WH is embedded in frame of video group FHT, obtains frame of video group FHW;

M sub-watermark information WS (I) is embedded in M group frame of video group FS (I) respectively, obtains M group frame of video group FSW (I); Wherein, M group frame of video group FSW (I) is designated as FSW (0) ~ FSW (M-1);

S3, composite video frame group FHW and M group frame of video group FSW (0) ~ FSW (M-1), obtains watermarked vision signal FW in order;

S4, utilizes described built-in template to carry out geometric correction to vision signal FW;

S5, detects watermark header WH in the vision signal through geometric correction;

And, in the vision signal through geometric correction, detect sub-watermark information WS (I);

S6, merges watermark header WH and sub-watermark information WS (I), generates complete video watermark information W.

2. method according to claim 1, is characterized in that, S1, video watermark information W is decomposed into 1 watermark header WH and M sub-watermark information WS (I) and specifically comprises the following steps:

Watermark header WH acquisition methods is:

S1-1, is equally divided into M section by the video watermark information W of L bit, obtains M localized watermark information W (I); Wherein, M localized watermark information W (I) is designated as W (0) ~ W (M-1); The then length N1 bit=L bit/M-bit of each localized watermark information W (I);

S1-2, carries out binary coding to total number M of local watermark information, obtains binary number K; Wherein, the number of bits N2 of binary number K is for being greater than the integer of log (M-1)/log (2);

Based on L and N3, calculate the CRC check code of video watermark information W, be designated as video watermark check code C; Wherein, N3 is set point, is the number of bits of video watermark check code C;

Combine K and C sequentially, obtain watermark header WH; Wherein, the number of bits of watermark header WH is N2+N3;

M sub-watermark information WS (I) acquisition methods is:

Respectively binary coding is carried out to the segment number I of M localized watermark information W (I), obtain M binary number K (I); Wherein, the number of bits of binary number K (I) is N2; M binary number K (I) is designated as K (0) ~ K (M-1);

Calculate the CRC check code of W (I) and K (I), obtain sub-watermark check code C (I); Wherein, the number of bits of sub-watermark check code C (I) is N4;

Combine W (I), K (I) and C (I) sequentially, obtain the sub-watermark information WS (I) of (N1+N2+N4) bit.

3. method according to claim 1, it is characterized in that, in S1, all frame of video of raw video signal are divided into M+1 group frame of video group by video playback time sequencing, obtain 1 group of frame of video group FH and M group frame of video group FS (I) respectively, specifically comprise the following steps:

Raw video signal is divided into the video-frequency band of fixing frame number;

Each video-frequency band is subdivided into the frame of video group that (M+1) individual frame number is D again, namely obtains 1 group of frame of video group FH and M group frame of video group FS (I).

4. method according to claim 1, is characterized in that, in S2, built-in template is embedded in frame of video group FH, obtains frame of video group FHT, specifically comprise the following steps:

Read the two field picture pixel data of frame of video group FH, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Choose NT template frequency in described DFT territory;

Revise the DFT coefficient of the described template frequency position chosen, then described built-in template is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding built-in template, obtains the frame of video group FHT embedding built-in template.

5. method according to claim 1, is characterized in that, in S2, is embedded in frame of video group FHT by watermark header WH, obtains frame of video group FHW, specifically comprise the following steps:

Described watermark head embeds and comprises the steps:

Read the two field picture pixel data of frame of video group FHT, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Choose NH watermark head frequency in DFT territory;

Revise the DFT coefficient of the described watermark head frequency position chosen, then described watermark header WH is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient of watermarked header, obtains the frame of video group FHW of watermarked header.

6. method according to claim 1, is characterized in that, in S2, M sub-watermark information WS (I) is embedded into respectively in M group frame of video group FS (I), obtains M group frame of video group FSW (I), specifically comprise the following steps:

Read the two field picture pixel data of frame of video group FS (I), overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Choose NS sub-watermark frequency in described DFT territory;

Revise the DFT coefficient of the described sub-watermark frequency position chosen, then described sub-watermark information WS (I) is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding sub-watermark information WS (I), obtains the frame of video group FSW (I) embedding sub-watermark information WS (I).

7. method according to claim 1, is characterized in that, S4, utilizes described built-in template to carry out geometric correction to vision signal FW, specifically comprises the following steps:

Read the two field picture pixel data of vision signal FW, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read;

The DFT coefficient obtained after overall DFT discrete Fourier transform is analyzed, judges whether to comprise built-in template; If built-in template do not detected, then keep raw video image, exit geometric correction operation; If built-in template detected, then contrasting detecting the built-in template obtained with the original built-in template prestored, calculating Image scaling coefficients and the anglec of rotation; Then the Image scaling coefficients and the anglec of rotation that calculate is preserved;

According to described Image scaling coefficients and the anglec of rotation, reverse convergent-divergent and reverse rotation are carried out to subsequent frame image, realize carrying out geometric correction to vision signal FW.

8. method according to claim 1, is characterized in that, S5, detects watermark header WH, specifically comprise the following steps in the vision signal through geometric correction:

Read the two field picture pixel data through the vision signal of geometric correction, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Analyze the DFT coefficient of NH watermark head frequency, extract watermark header WH;

From watermark header WH, extract data K and check code C, obtain sub-watermark sum M by K.

9. method according to claim 1, is characterized in that, detects sub-watermark information WS (I), specifically comprise the following steps in the vision signal through geometric correction:

Read the two field picture pixel data through the vision signal of geometric correction, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Analyze the DFT coefficient of NS sub-watermark frequency, extract sub-watermark WS (I), I ∈ 0,1 ..., M-1};

From sub-watermark WS (I), extract W (I), K (I) and C (I), obtain segment number I corresponding to sub-watermark according to K (I);

Checking C (I) check code, if check code is incorrect, then deletes this sub-watermark; Otherwise, if check code is correct, then sub-watermark information WS (I) successfully detected.

10. method according to claim 1, is characterized in that, S6, merges watermark header WH and sub-watermark information WS (I), generates complete video watermark information W, specifically comprise the following steps:

From sub-watermark information WS (0) ~ WS (M-1), extract localized watermark information W (0) ~ W (M-1) respectively, the localized watermark information W (0) extracted ~ W (M-1) is connected to become video watermark information W;

Checking C check code, if check code is incorrect, then deletes this video watermark information W, again detects watermark header and sub-watermark information; Otherwise, if check code is correct, then video watermark information W successfully detected.

11. 1 kinds of large-capacity video embedding algorithm and detection system, is characterized in that, comprises watermark embedding module (100) and watermark detection module (110);

Described watermark embedding module (100) comprising:

Video frame packet unit (101), for all frame of video F of raw video signal are divided into M+1 group frame of video group by video playback time sequencing, obtains 1 group of frame of video group FH and M group frame of video group FS (I) respectively, I ∈ 0,1 ... M-1}, M are integer, and M >=1; Wherein, M group frame of video group FS (I) is designated as FS (0) ~ FS (M-1);

Video watermark resolving cell (103), for video watermark information W being decomposed into 1 watermark header WH and M sub-watermark information WS (I), I ∈ 0,1 ..., M-1}, M are integer, and M >=1; Wherein, M sub-watermark information WS (I) is designated as WS (0) ~ WS (M-1);

Template embedded unit (102), for built-in template being embedded in frame of video group FH, obtains frame of video group FHT;

Watermark head embedded unit (104), is embedded into watermark header WH in frame of video group FHT, obtains frame of video group FHW;

Sub-watermark embedder unit (105), is embedded into M sub-watermark information WS (I) respectively in M group frame of video group FS (I), obtains M group frame of video group FSW (I); Wherein, M group frame of video group FSW (I) is designated as FSW (0) ~ FSW (M-1);

Frame of video assembled unit (106), for composite video frame group FHW and M group frame of video group FSW (0) ~ FSW (M-1) in order, obtains watermarked vision signal FW;

Watermark detection module (110) comprising:

Geometry correction unit (111), utilizes described built-in template to carry out geometric correction to vision signal FW;

Watermark head detecting unit (112), for detecting watermark header WH in the vision signal through geometric correction;

Sub-watermark detection unit (113), for detecting sub-watermark information WS (I) in the vision signal through geometric correction;

Video watermark integrated unit (114), for merging watermark header WH and sub-watermark information WS (I), generates complete video watermark information W.

12. large-capacity video embedding algorithm according to claim 11 and detection systems, it is characterized in that, described template embedded unit (102) is specifically for the two field picture pixel data that reads frame of video group FH, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Choose NT template frequency in described DFT territory;

Revise the DFT coefficient of the described template frequency position chosen, then described built-in template is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding built-in template, obtains the frame of video group FHT embedding built-in template;

Described watermark head embedded unit (104), specifically for the two field picture pixel data that reads frame of video group FHT, is carried out overall DFT discrete Fourier transform to the described two field picture pixel data read, is obtained DFT territory;

Choose NH watermark head frequency in DFT territory;

Revise the DFT coefficient of the described watermark head frequency position chosen, then described watermark header WH is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient of watermarked header, obtains the frame of video group FHT of watermarked header;

Described sub-watermark embedder unit (105) specifically for: read frame of video group FS (I) two field picture pixel data, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Choose NS sub-watermark frequency in described DFT territory;

Revise the DFT coefficient of the described sub-watermark frequency position chosen, then described sub-watermark information WS (I) is embedded in amended DFT coefficient;

The discrete inversefouriertransform of IDFT is carried out to the described DFT coefficient embedding sub-watermark information WS (I), obtains the frame of video group FSW (I) embedding sub-watermark information WS (I).

13. large-capacity video embedding algorithm according to claim 11 and detection systems, is characterized in that, described frame of video assembled unit (106) specifically for:

Raw video signal is divided into the video-frequency band of fixing frame number;

Each video-frequency band is subdivided into the frame of video group that (M+1) individual frame number is D again, namely obtains 1 group of frame of video FH and M group frame of video FS (I).

14. large-capacity video embedding algorithm according to claim 11 and detection systems, is characterized in that, described geometry correction unit (111) specifically for:

Read the two field picture pixel data of vision signal FW, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read;

The DFT coefficient obtained after overall DFT discrete Fourier transform is analyzed, judges whether to comprise built-in template; If built-in template do not detected, then keep raw video image, exit geometric correction operation; If built-in template detected, then contrasting detecting the built-in template obtained with the original built-in template prestored, calculating Image scaling coefficients and the anglec of rotation; Then the Image scaling coefficients and the anglec of rotation that calculate is preserved;

According to described Image scaling coefficients and the anglec of rotation, reverse convergent-divergent and reverse rotation are carried out to subsequent frame image, realize carrying out geometric correction to vision signal FW.

15. large-capacity video embedding algorithm according to claim 11 and detection systems, is characterized in that, described sub-watermark detection unit (113) specifically for:

Read the two field picture pixel data through the vision signal of geometric correction, overall DFT discrete Fourier transform is carried out to the described two field picture pixel data read, obtains DFT territory;

Analyze the DFT coefficient of NS sub-watermark frequency, extract sub-watermark WS (I), I ∈ 0,1 ..., M-1};

From sub-watermark WS (I), extract W (I), K (I) and C (I), obtain segment number I corresponding to sub-watermark according to K (I);

Checking C (I) check code, if check code is incorrect, then deletes this sub-watermark; Otherwise, if check code is correct, then sub-watermark information WS (I) successfully detected;

Described video watermark integrated unit (114) specifically for:

From sub-watermark information WS (0) ~ WS (M-1), extract localized watermark information W (0) ~ W (M-1) respectively, the localized watermark information W (0) extracted ~ W (M-1) is connected to become video watermark information W;

Checking C check code, if check code is incorrect, then deletes this video watermark information W, again detects watermark header and sub-watermark information; Otherwise, if check code is correct, then video watermark information W successfully detected.