CN102857831B - H.264 video integrality authentication method - Google Patents
H.264 video integrality authentication method Download PDFInfo
- Publication number
- CN102857831B CN102857831B CN201210274393.6A CN201210274393A CN102857831B CN 102857831 B CN102857831 B CN 102857831B CN 201210274393 A CN201210274393 A CN 201210274393A CN 102857831 B CN102857831 B CN 102857831B
- Authority
- CN
- China
- Prior art keywords
- block
- sub
- video
- macro block
- lnz
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
- Editing Of Facsimile Originals (AREA)
- Image Processing (AREA)
Abstract
The invention discloses an H.264 video integrality authentication method which comprises processes of extracting authentication codes, embedding identification information and tamper detecting and positioning. A video sequence is more complex than image information, so that video identification codes can not be extracted simply with the method of image hash value extracting. The invention provides a space energy fluctuation analysis method, according to video frames having different characteristics, an improved hierarchical quantization and vector quantization method is targetedly utilized, and video identification codes can be effectively extracted. In combination with improved H.264 robust watermarking algorithm and H.264 high-capacity information hiding algorithm, an H.264 video integrality authentication system is formed, and researches on tamper detecting and positioning can be conducted.
Description
Technical field
The present invention relates to video authentication field, specifically one H.264 video integrity authentication method.
Background technology
The develop rapidly of Digital Signal Processing and broadband network facility, makes the manufacture of digital content and distribution become more and more convenient.Once only had digital picture/Video editing work that specialist just can complete, nowadays easily can complete by a lot of cheap even free software and hardware.
This facility undoubtedly for most of user brings convenience and material benefit, but also can produce some query simultaneously.First, under not affecting validated user and enjoying the prerequisite of digital multimedia, to such as digital music, the media content of film and books carries out copyright protection, becomes a new problem, anxious to be studied.Next is content authentication aspect, due to easy editor's characteristic of numerical data, people is produced its authenticity and queries, particularly when these data are used as sworn evidence time.
In addition, in order to protect the safety of citizen's property, a lot of occasion is all provided with CCTV camera at present.Conveniently store and transmit, increasing CCTV camera adopts number format store video.The vision data of supervisory control system is very important beyond doubt, and their authenticity needs to keep further, and such digital video can use as reliable and effective evidence beyond suspicion.
Therefore, robust in coding transmission process, efficiently video authentication and the research of distorting repair mechanism become the hot issue of information security field.
Traditional digital information authentication method for first to calculate from digital information cryptographic check and or summary, and complete transmission; Secondly, when receiving terminal obtains information and summary, identical calculating is done to generate a new summary to the information obtained.Finally, the summary received and the new summary calculated are compared, judge whether information is tampered, here, the error of a bit can generate diverse summary, directly affects the result of certification.When we need position and position accurate, traditional authentication techniques show good performance.But the scene of digital picture or video is constantly change, media user pays close attention to the implication of content and unspecific numeral or sample more.Whether they the implication of more focused data are changed, but not through transmission, store and the error of individual bit position that data handling procedure produces.Therefore, when not hindering observer normally to use, allow the data processing operation of such as lossy compression method one class.In addition, when whether certification digital picture/video is tampered, it is also very necessary for judging which part is changed, and can contribute to the motivation determining assailant like this.Generic Authentication method cannot reach this requirement, because they only reflect the whether true of overall data.Like this we redefine authentication image/video be one judge vision content whether complete and report to the police and location malice tampered position process.
Summary of the invention
The invention provides one H.264 video integrity authentication method, have in real time, detect the integrality gathering video exactly, and carry out tampering detection level location.
The technical solution adopted in the present invention is as follows:
One is video integrity authentication method H.264, it is characterized in that, comprises following content:
(1) authentication code generates
Because the length of authentication code can not be excessive, the dimensionality reduction of initial data is very necessary, changes mild smooth block for video image, adopts scalar quantization, and calculating sub block mean carries out sample and simplifies; For the region that video image change is complicated, adopt vector quantization method, carry out sample by going the larger characteristic vector of macro block and simplify; Thus, extract real-time authentication code;
Detailed process is as follows:
A the scalar quantization authentication code of () smooth block generates
For given frame of video F, first extract its luminance component F
lwith dealing with, then by calculating 4 × 4 sub-block average simplify sample;
B the vector quantization authentication code of () edge block generates
Adopt VQ index, be block or vector by original data packet, and use independent index to reproduce each piece, adopt SVD method to reduce the dimension of former data, for transfer sub-block X
m × m, application SVD decomposes as follows to it:
U
i, v
ibe the row of U, V, reproduce XX respectively
tand X
tthe characteristic vector of X, Λ is diagonal matrix, by characteristic vector λ
1>=λ
2>=...>=λ
mform on the diagonal, the L-2 standard of X is λ
i, i=1,2 ..., the root sum square of m, because || u||=||v||=1, will keep front t characteristic vector u for passing through
i, v
i, i=1,2 ..., the approximate sub-block of t composition
realize, then original mean value transfer sub-block X and approximate sub-block
between difference be
T=1 is selected to reduce the dimension of former data;
(c) authentication code dimensionality reduction ciphering process
According to key, pseudorandomly the authentication code that a upper joint generates is divided into a series of subclass { S
1, S
2...,
, each subclass comprises M/2 position, calculate M/2 authentication code of each subclass and to 2 deliverys, and obtain 2 × N number of bit is watermark bit; By this method, authentication code is closed becomes watermark bit, reduce further data volume to be embedded, makes up joint marginate piece an of tool and adopts VQ and increase the operation of authentication code number;
(2) copyright protection/authentication procedures
Digital watermark in video monitoring system is the unique feature of video monitoring camera, and the loading of robust watermarking and extraction are very necessary, based on analysis of complexity watermarked signal in property field, each macro block of each frame of video image all will carry out block-based dct transform, then be determined and whether be used for extracting feature, feature extraction adopts texture energy and kinergety jointly to determine whether alternatively macro block, two threshold values are set corresponding with these energy component respectively by experience, be TH_T(texture energy) and TH_M(kinergety), two energy component as fruit macro block have at least one to be greater than its corresponding threshold value, namely this macro block is judged as feature macro block, can be used for watermarked signal,
For selected macro block (feature macro block), two sub-blocks at diagonal angle are chosen from the central area of 16 sub-blocks, macro block around these two sub-blocks is divided into level and vertical direction two groups, calculates the modulus value of each sub-block motion vectors, embed binary watermarking information according to " cross comparison method ";
Detailed process is as follows:
The analysis of (a) video complexity
If comprise f two field picture in a piece of digital video, each two field picture comprises m × n pixel; F
ibe the i-th frame, i ∈ { 0, f-1}, F
icarry out H.264 encoding operation time be broken down into the macro block that some non-overlapping copies are of a size of 16 × 16, MB
i,jbe the jth macro block in the i-th frame,
, C
i, j, kbe a kth pixel of the jth macro block in the i-th frame, k ∈ { 0, p × q-1}, each macro block MB
i,jbe subdivided into the sub-block b of 16 4 × 4 again
i, j, ncarry out dct transform, b
i, j, nrepresent the n-th sub-block of the jth macro block in the i-th frame, then zigzag scanning is carried out to the sub-block after conversion, in each like this sub-block, just create the DCT coefficient of 7 subbands;
I () image complexity is measured
The image complexity of definition pixel macroblock is the comprehensive consideration of its background luminance and Texture complication, a threshold value is set for pixel macroblock complexity, complexity value exceeds that namely this threshold value constraint represents complicated at this macroblock texture, background luminance is high, is applicable to loading watermark information; Complexity value is less than this threshold value, illustrates that this macroblock area is not suitable for embed watermark information gently, at the theoretical method of decoding end according to Texture complication, determines that the position of watermark embedment is to extract watermarking images;
H.264 the DC coefficient set in macro block being generation after 16 sub-blocks carry out dct transform has been suffered to most of energy of sub-block, and there is compression consistency, therefore, carried out Texture complication tolerance by 16 DC coefficients of interior macroblocks;
Algorithm topography's complexity factors measures the Texture complication of topography, and this topography's complexity factors and bearer documents DC component are constructed compound function jointly:
In formula: α
1, α
2for scale factor, between 0 and 1; Select the pixel macroblock that topography's complexity is larger watermarked, setting video frame size is N
1× N
2, DC
i, j, kfor the i-th two field picture F in video image
ia jth macro block MB
i,jthe DC coefficient of a kth sub-block after carrying out dct transform;
Macro block MB
i,jtexture complication parameter be:
In formula: e
i,jfor the mean value of 16 DC coefficients of interior macroblocks; P (e
i,j) be weight coefficient, β is weighted factor, and span is (0.6,0.7), and the value of image complexity IC (i, j) is with α
1, α
2and the change of β and changing;
(ii) motion feature tolerance
The difference of application video sequence interframe associated macroblock, describes the kinematic parameter of target in frame of video simply; H.264 the macroblock size of processing unit is 16 × 16, each macro block be subdivided into 16 non-overlapping copies 4 × 4 sub-block,
be the pixel value that (m, the n) of a kth sub-block in a jth macro block in the i-th frame puts,
represent the speed of pixel change on time shaft t,
Here, 0≤m, n≤3,0≤n
j≤ 3,
, 0≤k≤15;
In the pixel of sub-block, each point pixel value along the maximum of the absolute value of time axis conversion is
In i-th frame, the speed maximization average of a jth macro block is
In i-th frame, the speed mean square deviation of a jth macro block is
The motion feature of macro block can pass through kinematic parameter
measure;
(iii) video complexity tolerance
Finally, weighting video complexity is adopted
judge whether sub-block has motion feature:
ME (i, j) for feature macro block mark, i be frame number, j is sub-block sequence number, sets two threshold values corresponding with these energy component respectively, be TH_T and TH_M by experience.Two energy component as fruit macro block have at least one to be greater than its corresponding threshold value, and namely this macro block is judged as the complicated macro block of video, can be used for watermarked signal;
Watermarking algorithm in (b) interframe " right-angled intersection " motion vector fields
Robust watermarking embedding method:
(i) judge whether current macro is applicable to according to video complexity measure watermarked, if macro block have height video complexity, carry out next step operation, if the texture of macro block or movement tendency milder, namely forward next macro block to, again judge from this step;
(ii) for selected host's macro block, 12, edge sub-block is divided into two groups, left and right according to the axis of macro block, calculates the two groups of kinematic parameter average V in left and right
rightand V
leftt, select to embed some position according to formula (11):
(iii) select the motion vector embedding point 4 sub-blocks around, be calculated as follows them respectively and embed the modulus value P of motion vector of some sub-block
v,
,
,
,
, and by them by laterally and be vertically divided into 2 groups
With
:
(iv) try to achieve Row respectively according to the following formula
hand Row
vmaximum and minimum value:
(v) draw the relative maximum Amax in each group of coefficient and relative minimum Amin according to the following formula:
(vi) be calculated as follows relative difference D, relatively large difference Dmax and relatively little difference Dmin
(vii) wherein, the scope between Amin<Amax, Amin to Amax can be divided into 4 parts, the regular embed watermark information by following formula:
Robust watermarking extracting rule:
Extracting rule is the inverse process of embedding method, and concrete steps are as follows:
(i) judge whether current macro embedded in watermark according to video complexity measure, if the video complexity with height of macro block, namely represent that this interior macroblocks embedded in watermark information, carry out next step operation and extract watermark information; If the texture of macro block or movement tendency are comparatively mild, namely represent that this macro block does not hide watermark information, namely forward next macro block to, again judge from this step;
(ii) for selected host's macro block, 12, edge sub-block is divided into two groups, left and right according to the axis of macro block, calculates the two groups of kinematic parameter average V in left and right
rightand V
leftt, the embedding point position determined according to the following formula:
(iii), as embedding grammar (iv) ~ (vi), calculate relative difference D, relatively large difference Dmax and relatively little difference Dmin;
(iv) the rule of pressing following formula proposes watermark information:
(3) integrated authentication process
Utilize H.264 redundant information in frame, by space for subsequent use for the coefficient combination creation one having original null value in sub-block to lay in additional information, original LSB(least significant bit) be stored in spare space, fragile watermark is embedded into by a zero valued coefficients in amendment sub-block, interframe utilizes VLC(variable-length encoding) indifference value compensation characteristic, embed fragile watermark by modulation inter-frame forecast mode, in each macro block of each frame of video, complete the embedding of fragile watermark thus;
In decode procedure, regenerate authentication code by authentication code extracting method, carry out the extraction of fragile watermark simultaneously, by authentication code compared with the fragile watermark information extracted, the detection & localization distorted can be realized fast;
Detailed process is as follows:
A () be block-based reversible data concealing method in frame H.264
H.264 the dominant ideas of standard are consistent with existing video encoding and decoding standard---block-based hybrid coding method, each video image is divided into the pixel macroblock of 16 × 16 by cataloged procedure, video image can be processed in units of pixel macroblock, after intraframe coding conducting frame inner estimation mode, determine the size of sub-block, macro block is subdivided into 16 4 × 4 sub-blocks and predicted value and subtracts each other and obtain residual block, carry out dct transform, quantification, Zigzag scanning, through entropy code, finally write code stream;
Because the concentration of energy of H.264 coding standard sub-block is at medium and low frequency subband, the DCT coefficient after most of High-frequency quantization is all 0, and therefore, Zigzag scans 4 × 4 coefficient matrixes and generates the vector that an afterbody comprises continuous zero, and the sequence number of last nonzero coefficient is i
lNZ(Index of last nonzero coefficient), if 16 coefficients are all zero, i in sub-block
lNZ=0;
Watermark embedment rule
(i) for i
lNZthe sub-block of>=2, selects 2 different index i by pseudo random sequence
1and i
2(1≤i
1, i
2≤ i
lNZ-1), according to key, the original least significant bit of two coefficients is replaced with 2 corresponding watermark bit:
Like this, namely 2 watermark bit are hidden into the LSB of two coefficients, in order to increase fail safe, in different sub-blocks, adopt different i
1and i
2, therefore, do not have key, interpolater cannot extract the watermark information of embedding, meanwhile, if i
lNZ≤ 13, according to
with
original LSB revise i
lNZ+ 1 or i
lNZ+ 2, specifically see following formula:
If
original LSB be 0, according to
parity, will
value increase or reduce 1, and
be still 0; If
original LSB be 1, according to
parity, will
value increase or reduce 1, and
be still 0; This represents
with
original LSB pass through
with
association, and
with
absolute value be necessary for 1, by revising the watermark information that a coefficient embeds 2;
(ii) if i
lNZbe 14 or 15, we are basis then
with
original LSB amendment
value, see following formula:
with
original LSB pass through
association, and
absolute value must be greater than 1, through these operation, work as i
lNZwhen being 2, new i
lNZvalue must be more than or equal to 3, the simultaneously C of non-zero
2value constant;
(iii) if i
lNZoriginal value≤1, directly 2 watermark embedment are arrived
or
, and adjust C
0make its absolute value be greater than 1, and keep other coefficient constant,
Please note in the telescopiny of watermark and only change a coefficient, if, w
1be 0, according to w
2and incite somebody to action
increase or reduce by 1, and
be still 0.W
1be 1, according to w
2and incite somebody to action
increase or reduce by 1, and
be still 0, in this case, new i
lNZvalue must be less than or equal to 3, simultaneously original value is the C of zero
2value constant;
(iv) work as i
lNZdo not exist, namely the inner all coefficients of sub-block are all the situation of 0, by 1 watermark embedment to C
0,
Watermark extracting rule
In first stage, attempt extracting watermark data, and recover original sub-block content, for each sub-block in suspect image, receiving terminal scans its 16 coefficients and forms a vector, and definition coefficient vector is [C '
0, C '
1..., C '
15], the call number of last nonzero coefficient is i '
lNZ, watermark data and recovery original image content according to four kinds of situations extraction embeddings below:
(i) situation one: i '
lNZ>3, or i '
lNZ=3 and C '
2when ≠ 0, if
be 1, according to
with
obtain b
1and b
2:
And calculate
If
be greater than 1, can basis
obtain b
1and b
2:
And
Then, adopt same key, pseudorandomly select the index i that two different
1and i
2(1≤i
1, i
2≤
), extract 2 watermarks:
Further, receiving terminal can attempt by following operation the original DCT coefficient recovering sub-block:
With
(ii) if i '
lNZ∈ { 1,2}, or i '
lNZ=3 and C '
2=0, first calculate i '
lNZ, then extract two watermarks
And recover the original contents of sub-block
(iii) if i '
lNZwhen=0
B () interframe is based on the considerable information hide algorithm of information variable-length packets structure
(i) information variable-length packets (Information variance length group)
H.264 7 kinds of inter-frame forecast modes are had, analyze theoretically, if once hide 2 bit-identify information, only need to map the embedding that wherein 4 kinds of patterns can complete identification information, existing algorithm only utilizes 16 × 16,16 × 8,8 × 16,8 × 8 these 4 kinds of Pattern completion identification informations of inter prediction to hide, effectively cannot utilize the average information that inter prediction link produces, will certainly restricted information hide capacity; If but adopt isometric 3 binary code representations, namely 3 bit-identify information are once embedded, there are 000,001,010,011,100,101,110,111 forms in totally 8,7 kinds of inter-frame forecast modes cannot isometric 3 binary codes of Complete Mappings these 8 groups, and wherein binary identification information can be caused complete, to be correctly embedded into and to extract;
(ii) watermarking algorithm
Step 1: read two-value identification image WM
image(M × N), scans this image after simple scramble through Zig-zag and is converted into one group of two-value code stream WM
sequence[i] (i=0,1,2 ..., (M × N-1)), then by WM
sequence[i] is adaptive is divided into the identification information group that length is 2 or 3 bits, forms WM
pair[j]=w
jw
j+1or w
jw
j+1w
j+2(j=0,1,2 ..., (M × N-1));
Step 2: read identification information group WM
pair[j], the table of comparisons 1, inquires about the predictive mode that this information group maps, and the optimum prediction mode replacing current block completes hiding of this group identification information;
Step 3: jump to next macro block and repeat Step 2, go until identification information is hidden in host image completely, complete whole video information hiding process;
(iii) watermark information extraction algorithm
Extracting rule is the inverse process of embedding method, and concrete steps are as follows:
Step 1: the predictive mode reading current macro, question blank obtains its binary identification information group w mapped
jw
j+1or w
jw
j+1w
j+2;
Step 4: jump to next macro block and repeat Step 1, until all identification information groups are extracted completely, by all identification information group combinations extracted, decipher the identification information obtaining extracting, so far completes whole leaching process.
Beneficial effect of the present invention is:
(1) positioning tampering region
Play considerable effect to better illustrate that positioning tampering part is recognized in scheme in multimedia, the present invention is tested by one section of successive frame; The integrity of video Verification System of the present invention's design to the video distorted, can be very sensitive detect the generation of distorting, and tampering location can be carried out accurately;
(2) authentication
Native system detect distort after carry out robust watermarking extraction, carry out authentication, determine the source of this frame of video, copyright; Calculating the similarity index NC extracting watermark and original watermark is 0.938819, reaches the requirement of authentication completely, judges, the meaning telling original mark watermark that the mark watermark extracted also can be correct from human eye vision;
(3) can aggressively analyze
Have detected for assailant, change the complexity of video content when not alert trigger, for the most important characteristics vector of block having edge/texture, can by extracting average and VQ(vector quantization from each fritter) index forms authentication code, in quantizer index, use limited location to carry out reproduced image block, this possibility making assailant still have realization effectively to distort, although SQ is a direct process, but it can detect distorting of major part malice effectively, if suppose that certain assailant can by certain mode by SQ(scalar quantization) check, in this programme, so only there is VQ to carry out verification process, like this, so-called " VQ attack " just may bring some to challenge for us, more specifically illustrate, if VQ index d
ibe used in original sub-block X
i, assailant generates new sub-block X with distorting content
t, this is by reproduced by different indexes.However, assailant may be as much as possible with index d by collecting from graphics/video
ithe relevant sub-block of block, from the angle of vision or the angle of Minimum Mean Square Error index, therefrom select it closest to X
tblock replace current data block.This sub-block is perhaps not only seem and X
tsimilar, even by VQ test, thus, just there is undetected may occurring.In order to reduce the risk that VQ attacks, the present invention is by complicated for the mapping process made at sub-block and corresponding SQ/VQ index.In authentication code, application encipher protection, uses key mapping function, and will be helpful to different frame of video adaptively modifying code books.
Accompanying drawing explanation
Fig. 1 is the structural representation of block and sub-block in frame of video.
Fig. 2 is that watermark data generates example schematic.
Fig. 3 is putting in order of 16 DCT coefficient and energy subband schematic diagram.
Fig. 4 is the schematic diagram that " cross " type embeds scheme.
Fig. 5 is the schematic diagram of watermark embedment position in host's macro block.
Fig. 6 is the schematic diagram being divided into 4 parts between Amin to Amax.
Fig. 7 is the simple example schematic diagram of not matching measurement.
Embodiment
One is video integrity authentication method H.264, comprises following content:
(1) authentication code generates
A scalar quantization authentication code generation that () is smoothly fast
Because the length of authentication code can not be excessive, the dimensionality reduction of initial data is very necessary.For given frame of video F, we first extract its luminance component F
lwith dealing with, then, we by calculating 4 × 4 sub-block average simplify sample.
Let us looks at numerical example, is the CIF image of 352 × 288 for resolution, F
dbe of a size of 88 × 72, this size is still too large as authentication code.We are further by F
dbe divided into some sub-blocks to carry out feature extraction.The size of sub-block again be chosen as 4 × 4, if need less authentication code, can be even 8 × 8.
Fig. 1 is shown as the structure of block and sub-block.Characteristic value will extract as authentication code from sub-block.A direct method is the average of computation partition sub-block.If we reproduce each average with k bit, for the sub-block of 4 × 4, the final size of authentication code can be reduced for 22 × 18 × k bit, and the sub-block for 8 × 8 is 11 × 9 × k bit.Use the bit number of fixed qty can be reproduced as a value very well, this is the process of a SQ.But there is a problem in this method.May occur undetected when processing the average obtained and being similar to original area.More seriously assailant may change content but the average of maintainance block is constant.
B the vector quantization authentication code of () edge block generates
In order to alleviate the problems referred to above.We adopt VQ, and original data packet is block or vector by we in the process, and use independent index to reproduce each piece.SVD method is adopted to reduce the dimension of former data.For transfer sub-block X
m × m, application SVD decomposes as follows to it:
U
i, v
ibe the row of U, V, reproduce XX respectively
tand X
tthe characteristic vector of X, Λ is diagonal matrix, by characteristic vector λ
1>=λ
2>=...>=λ
mform on the diagonal.The L-2 standard that it should be noted that X is λ
i, i=1,2 ..., the root sum square of m, because || u||=||v||=1, will keep front t characteristic vector u for passing through
i, v
i, i=1,2 ..., the approximate sub-block of t composition
realize.Then original mean value transfer sub-block X and approximate sub-block
between difference be
We simply select t=1 to reduce the dimension of former data, and this is due to u
i, v
ibe most important part and they can be reproduced by VQ index.Another reason is, does relevant to interest block like this.As what mention before us, marginate piece can well be processed by VQ.We find that these blocks comprising more information have the λ of a very high values usually
1.
(c) authentication code dimensionality reduction ciphering process
According to key, pseudorandomly the authentication code that a upper joint generates is divided into a series of subclass { S
1, S
2...,
, each subclass comprises M/2 position, calculate M/2 authentication code of each subclass and to 2 deliverys, and obtain 2 × N number of bit is watermark bit.By this method, authentication code is closed becomes watermark bit, reduce further data volume to be embedded, makes up joint marginate piece an of tool and adopts VQ and increase the operation of authentication code number.Fig. 2 describes the simple case that watermark data generates, and sub-block number N is 4, and the number M of the authentication code of each sub-block is 6.Pseudorandom segmentation and calculate after, generate the watermark bit of 8 immediately.
(2) copyright protection/authentication procedures
The analysis of (a) video complexity
If comprise f two field picture in a piece of digital video, each two field picture comprises m × n pixel.F
ibe the i-th frame, i ∈ { 0, f-1}.F
icarry out H.264 encoding operation time be broken down into the macro block that some non-overlapping copies are of a size of 16 × 16, MB
i,jbe the jth macro block in the i-th frame,
.C
i, j, kbe a kth pixel of the jth macro block in the i-th frame, k ∈ { 0, p × q-1}.Each macro block MB
i,jbe subdivided into the sub-block b of 16 4 × 4 again
i, j, ncarry out dct transform, b
i, j, nrepresent the n-th sub-block of the jth macro block in the i-th frame.Then zigzag scanning is carried out to the sub-block after conversion, in each like this sub-block, just create the DCT coefficient of 7 subbands as shown in Figure 3.
I () image complexity is measured
The image complexity that the present invention defines pixel macroblock is the comprehensive consideration of its background luminance and Texture complication.We set a threshold value for pixel macroblock complexity, and complexity value exceeds that namely this threshold value constraint represents complicated at this macroblock texture, background luminance is high, is applicable to loading watermark information; Complexity value is less than this threshold value, illustrates that this macroblock area is not suitable for embed watermark information gently.At the theoretical method of decoding end according to Texture complication, determine that the position of watermark embedment is to extract watermarking images.
H.264 the DC coefficient set in macro block being generation after 16 sub-blocks carry out dct transform has been suffered to most of energy of sub-block, and had compression consistency, therefore, the present invention carries out Texture complication tolerance by 16 DC coefficients of interior macroblocks.
Algorithm topography's complexity factors measures the Texture complication of topography, and this topography's complexity factors and bearer documents DC component are constructed compound function jointly:
In formula: α
1, α
2for scale factor, between 0 and 1.Select the pixel macroblock that topography's complexity is larger watermarked, setting video frame size is N
1× N
2, DC
i, j, kfor the i-th two field picture F in video image
ia jth macro block MB
i,jthe DC coefficient of a kth sub-block after carrying out dct transform.
Macro block MB
i,jtexture complication parameter be:
In formula: e
i,jfor the mean value of 16 DC coefficients of interior macroblocks; P (e
i,j) be weight coefficient, β is weighted factor, and span is (0.6,0.7).The value of image complexity IC (i, j) is with α
1, α
2and the change of β and changing, demonstrate the image texture complexity that each sub-block is different thus, promote the fail safe of watermarking algorithm further.
(ii) motion feature tolerance
The difference of the present invention's application video sequence interframe associated macroblock, describes the kinematic parameter of target in frame of video simply.H.264 the macroblock size of processing unit is 16 × 16, each macro block be subdivided into 16 non-overlapping copies 4 × 4 sub-block,
be the pixel value that (m, the n) of a kth sub-block in a jth macro block in the i-th frame puts,
represent the speed of pixel change on time shaft t,
Here, 0≤m, n≤3,0≤n
j≤ 3,
, 0≤k≤15.
In the pixel of sub-block, each point pixel value along the maximum of the absolute value of time axis conversion is
In i-th frame, the speed maximization average of a jth macro block is
In i-th frame, the speed mean square deviation of a jth macro block is
The motion feature of macro block can pass through kinematic parameter
measure.
(iii) video complexity amount
Finally, the present invention adopts the weighting video complexity shown in formula (9)
judge whether sub-block has motion feature:
ME (i, j) for feature macro block mark, i be frame number, j is sub-block sequence number, sets two threshold values corresponding with these energy component respectively, be TH_T and TH_M by experience.Two energy component as fruit macro block have at least one to be greater than its corresponding threshold value, and namely this macro block is judged as the complicated macro block of video, can be used for watermarked signal.
Watermarking algorithm in (b) interframe " right-angled intersection " motion vector fields
Robust watermarking embedding method:
(i) judge whether current macro is applicable to according to video complexity measure watermarked.If macro block has the video complexity of height, carry out next step operation.If the texture of macro block or movement tendency are comparatively mild, namely forward next macro block to, again judge from this step.
(ii) for selected host's macro block, 12, edge sub-block is divided into two groups, left and right according to the axis of macro block, calculates the two groups of kinematic parameter average V in left and right
rightand V
leftt.Select as the embedding point position in Fig. 4 according to formula (11):
(iii), according to selecting the motion vector embedding point 4 sub-blocks around shown in Fig. 5, being calculated as follows them respectively and embedding the modulus value P of motion vector of some sub-block
v,
,
,
,
.And by them by laterally and be vertically divided into 2 groups
With
。
(iv) try to achieve Row respectively according to the following formula
hand Row
vmaximum and minimum value:
(v) draw the relative maximum Amax in each group of coefficient and relative minimum Amin according to the following formula
(vi) be calculated as follows relative difference D, relatively large difference Dmax and relatively little difference Dmin
(vii) wherein, the scope between Amin<Amax, Amin to Amax can be divided into 4 parts (as shown in Figure 6), by the regular embed watermark information of following formula.
Robust watermarking extracting rule:
Extracting rule is the inverse process of embedding method, and concrete steps are as follows
(i) judge whether current macro embedded in watermark according to video complexity measure.If the video complexity with height of macro block, namely represents that this interior macroblocks embedded in watermark information.Carry out next step operation and extract watermark information.If the texture of macro block or movement tendency are comparatively mild, namely represent that this macro block does not hide watermark information, namely forward next macro block to, again judge from this step.
(ii) for selected host's macro block, 12, edge sub-block is divided into two groups, left and right according to the axis of macro block, calculates the two groups of kinematic parameter average V in left and right
rightand V
leftt.The embedding point position determined according to the following formula:
(iii), as embedding grammar (iv) ~ (vi), calculate relative difference D, relatively large difference Dmax and relatively little difference Dmin
(iv) the rule of pressing following formula proposes watermark information.
(3) integrated authentication process
A () be block-based reversible data concealing method in frame H.264
H.264 the dominant ideas of standard are consistent with existing video encoding and decoding standard---block-based hybrid coding method.Each video image is divided into the pixel macroblock of 16 × 16 by cataloged procedure, and video image can be processed in units of pixel macroblock.After intraframe coding conducting frame inner estimation mode, determine the size of sub-block, macro block is subdivided into 16 4 × 4 sub-blocks and predicted value and subtracts each other and obtain residual block, carries out dct transform, quantification, Zigzag scanning, through entropy code, finally writes code stream.
Because the concentration of energy of H.264 coding standard sub-block is at medium and low frequency subband, the DCT coefficient after most of High-frequency quantization is all 0.Therefore, Zigzag scans 4 × 4 coefficient matrixes and generates the vector that an afterbody comprises continuous zero.The sequence number of last nonzero coefficient is i
lNZ(Index of lastnonzero coefficient).If 16 coefficients are all zero, i in sub-block
lNZ=0.
Watermark embedment rule
(i) for i
lNZthe sub-block of>=2, selects 2 different index i by pseudo random sequence
1and i
2(1≤i
1, i
2≤ i
lNZ-1), according to key, the original least significant bit of two coefficients is replaced with 2 corresponding watermark bit:
Like this, namely 2 watermark bit are hidden into the LSB of two coefficients, and in order to increase fail safe, we are used in different sub-blocks and adopt different i
1and i
2.Therefore, do not have key, interpolater cannot extract the watermark information of embedding.Meanwhile, if i
lNZ≤ 13, according to
with
original LSB revise i
lNZ+ 1 or i
lNZ+ 2, specifically see following formula:
If
original LSB be 0, according to
parity, will
value increase or reduce 1, and
be still 0; If
original LSB be 1, according to
parity, will
value increase or reduce 1, and
be still 0; This represents
with
original LSB pass through
with
association, and
with
absolute value be necessary for 1.Like this.The watermark information that we embed 2 by revising a coefficient.
(ii) if i
lNZbe 14 or 15, we are basis then
with
original LSB amendment
value.See following formula:
In other words,
with
original LSB pass through
association, and
absolute value must be greater than 1.Through these operations, work as i
lNZwhen being 2, new i
lNZvalue must be more than or equal to 3, the simultaneously C of non-zero
2value constant.
(iii) if i
lNZoriginal value≤1, directly 2 watermark embedment are arrived
or
, and adjust C
0make its absolute value be greater than 1, and keep other coefficient constant.
Please note in the telescopiny of watermark and only change a coefficient, in other words, if, w
1be 0, according to w
2and incite somebody to action
increase or reduce by 1, and
be still 0.W
1be 1, according to w
2and incite somebody to action
increase or reduce by 1, and
be still 0.In this case, new
value must be less than or equal to 3, simultaneously original value is the C of zero
2value constant.
(iv) work as i
lNZdo not exist, namely the inner all coefficients of sub-block are all the situation of 0, by 1 watermark embedment to C
0.
Watermark extracting rule
In first stage, trial is extracted watermark data by us, and recovers original sub-block content.For each sub-block in suspect image, receiving terminal scans its 16 coefficients and forms a vector.Definition coefficient vector is [C '
0, C '
1..., C '
15], the call number of last nonzero coefficient is i '
lNZ, according to watermark data and the recovery original image content of four kinds of situations extraction embeddings below.
(i) situation one: i '
lNZ>3, or i '
lNZ=3 and C '
2when ≠ 0, if
be 1, can basis
with
obtain b
1and b
2.
And calculate
If
be greater than 1, can basis
obtain b
1and b
2
And
Then, adopt same key, pseudorandomly select the index i that two different
1and i
2(1≤i
1, i
2≤
), extract 2 watermarks:
Further, receiving terminal can attempt by following operation the original DCT coefficient recovering sub-block:
With
(ii) if i '
lNZ∈ { 1,2}, or i '
lNZ=3 and C '
2=0, first calculate i '
lNZ, then extract two watermarks
And recover the original contents of sub-block
(iii) if i '
lNZwhen=0
B () interframe is based on the considerable information hide algorithm of information variable-length packets structure
(i) information variable-length packets (Information variance length group)
H.264 7 kinds of inter-frame forecast modes are had, analyze theoretically, if once hide 2 bit-identify information, only need to map the embedding that wherein 4 kinds of patterns can complete identification information, existing algorithm only utilizes 16 × 16,16 × 8,8 × 16,8 × 8 these 4 kinds of Pattern completion identification informations of inter prediction to hide, effectively cannot utilize the average information that inter prediction link produces, will certainly restricted information hide capacity.If but adopt isometric 3 binary code representations, namely 3 bit-identify information are once embedded, there are 000,001,010,011,100,101,110,111 forms in totally 8,7 kinds of inter-frame forecast modes cannot isometric 3 binary codes of Complete Mappings these 8 groups, and wherein binary identification information can be caused complete, to be correctly embedded into and to extract.
Based on the problems referred to above, the present invention is directed to 7 kinds of inter-frame forecast modes H.264, propose binary message variable-length packets concept.The binaryzation information that each piece of size is given and and information group mapping result as shown in table 1:
The codeword structure of the Exp-Golomb that table 1 improves and block type mapping ruler
Predictive mode | Code | M | INFO | Information group |
16×16 | 0 | 2 | 3 | 11 |
16×8 | 1 | 3 | 0 | 000 |
8×16 | 2 | 3 | 1 | 001 |
8×8 | 3 | 3 | 2 | 010 |
8×4 | 4 | 3 | 3 | 011 |
4×8 | 5 | 3 | 4 | 100 |
4×4 | 6 | 3 | 5 | 101 |
Employing information variable-length packets method can be fully effective utilization H.264 inter prediction produce inter-frame information, each macro block on average can hide 2.86 bits identification informations, and than existing algorithm, algorithm capacity promotes greatly.
(ii) watermarking algorithm
Step 1: read two-value identification image WM
image(M × N), scans this image after simple scramble through Zig-zag and is converted into one group of two-value code stream WM
sequence[i] (i=0,1,2 ..., (M × N-1)), then by WM
sequence[i] is adaptive is divided into the identification information group that length is 2 or 3 bits, forms WM
pair[j]=w
jw
j+1or w
jw
j+1w
j+2(j=0,1,2 ..., (M × N-1)).
Step 2: read identification information group WM
pair[j], the table of comparisons 1, inquires about the predictive mode that this information group maps, and the optimum prediction mode replacing current block completes hiding of this group identification information.Such as, when information group to be embedded is " 11 ", if the predictive mode of current macro is 16 × 16 patterns just, does not then make and change; If the predictive mode of current macro is not 16 × 16 patterns, namely change current prediction mode into 16 × 16 patterns artificially, preserve and write code stream.
Step 3: jump to next macro block and repeat Step 2, go until identification information is hidden in host image completely, complete whole video information hiding process.
(iii) watermark information extraction algorithm
Extracting rule is the inverse process of embedding method, and concrete steps are as follows
Step 1: the predictive mode reading current macro, the table of comparisons 1, inquiry obtains its binary identification information group w mapped
jw
j+1or w
jw
j+1w
j+2.
Step 4: jump to next macro block and repeat Step 1, until all identification information groups are extracted completely, by all identification information group combinations extracted, decipher the identification information obtaining extracting, so far completes whole leaching process.
(4) tampering detection, position fixing process
Hostile content replaces the watermark extracting and content recovery that may lead to errors.We will determine the sub-block that the matching relationship between those hash position and the watermark bit extracted is destroyed now.Although the matching relationship of other sub-blocks be correctly recovered may be affected owing to distorting, as long as rate of distorting is less, according to not mating classification, those sub-blocks be really tampered can be distinguished from numerous unmatched situation.Determine that the concrete grammar distorting sub-block specifically describes hereinafter.
64 desorption coefficients inputting each sub-block to same hash function calculate M hash position, are " calculating hash position ".Then, according to identical key by N × M position calculate Hash position be divided into 2 × N number of subclass, and calculate M/2 Hash position of each subclass with and to 2 deliverys.The result obtained is " calculating watermark bit ".To the sub-block that each is determined, there is M to calculate watermark bit and be distributed in M subclass, and the calculating watermark bit of M sub-set associative M.This illustrates that each sub-block maps M and calculates watermark bit.On the other hand, each watermark bit maps M/2 sub-block.Use t
n(1≤n≤N) represents the number inconsistent between watermark bit and the watermark bit extracted mapped in each sub-block.In fact, parametric t
nfor weighing the not matching degree of decoding sub-block content and corresponding extraction watermark.
Fig. 7 is the simple case of not matching measurement.Suppose that second sub-block content is tampered and the sub-block of decoding is incorrect, the hash position of sub-block is 011000, and the original hash position shown in Fig. 1 is 001011.Use same subclass segmentation, obtain and calculate watermark bit.We suppose that the 3rd watermark bit is embedded in second sub-block, and owing to being distorted, the watermark bit extracted is 0, but not original value 1.By calculating and extract the comparison of watermark, coupling and do not mate submeter " o " and "×" and represent.Then, we calculate the t of each sub-block
n.Illustrate, first sub-block maps the 4th, the 2nd, the 3rd, the 1st, the 2nd and the 7th for calculating watermark bit, wherein has 2 and does not mate, therefore, and t
1=2.Second sub-block maps the 3rd, the 1st, the 7th, the 3rd, the 6th and the 8th and calculates watermark bit, because the 3rd watermark is calculated 2 times, and produces 5 and does not mate, be i.e. t
2=5.The mapping relations of second sub-block highlight with Bold arrows as shown in Figure 7.In like manner, be not difficult to draw, t
3=2, t
4=2.Obviously, by the t distorting sub-block
nmore much bigger than other sub-blocks.In fact, we will from t
nvalue confirm to distort sub-block.
Claims (1)
1. a H.264 video integrity authentication method, is characterized in that, comprise following content:
(1) authentication code generates
Because the length of authentication code can not be excessive, the dimensionality reduction of initial data is very necessary, changes mild smooth block for video image, adopts scalar quantization, and calculating sub block mean carries out sample and simplifies; For the region that video image change is complicated, adopt vector quantization method, carry out sample by going the larger characteristic vector of macro block and simplify; Thus, extract real-time authentication code;
(2) copyright protection/authentication procedures
Digital watermark in video monitoring system is the unique feature of video monitoring camera, and the loading of robust watermarking and extraction are very necessary, based on analysis of complexity watermarked signal in property field, each macro block of each frame of video image all will carry out block-based dct transform, then be determined and whether be used for extracting feature, feature extraction adopts texture energy and kinergety jointly to determine whether alternatively macro block, two threshold values are set corresponding with these energy component respectively by experience, described two threshold values are texture energy threshold and kinergety threshold value, wherein texture energy threshold and kinergety threshold value are respectively with symbol TH_T and TH_M statement, two energy component as fruit macro block have at least one to be greater than its corresponding threshold value, namely this macro block is judged as feature macro block, can be used for watermarked signal,
For selected feature macro block, two sub-blocks at diagonal angle are chosen from the central area of 16 sub-blocks, macro block around these two sub-blocks is divided into level and vertical direction two groups, calculates the modulus value of each sub-block motion vectors, embed binary watermarking information according to " cross comparison method ";
(3) integrated authentication process
Utilize H.264 redundant information in frame, by space for subsequent use for the coefficient combination creation one having original null value in sub-block to lay in additional information, original LSB(least significant bit) be stored in spare space, fragile watermark is embedded into by a zero valued coefficients in amendment sub-block, interframe utilizes VLC(variable-length encoding) indifference value compensation characteristic, embed fragile watermark by modulation inter-frame forecast mode, in each macro block of each frame of video, complete the embedding of fragile watermark thus;
In decode procedure, regenerate authentication code by authentication code extracting method, carry out the extraction of fragile watermark simultaneously, by authentication code compared with the fragile watermark information extracted, the detection & localization distorted can be realized fast.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210274393.6A CN102857831B (en) | 2012-08-01 | 2012-08-01 | H.264 video integrality authentication method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210274393.6A CN102857831B (en) | 2012-08-01 | 2012-08-01 | H.264 video integrality authentication method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102857831A CN102857831A (en) | 2013-01-02 |
CN102857831B true CN102857831B (en) | 2015-06-24 |
Family
ID=47403957
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210274393.6A Expired - Fee Related CN102857831B (en) | 2012-08-01 | 2012-08-01 | H.264 video integrality authentication method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102857831B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104683811B (en) * | 2013-11-29 | 2017-12-29 | 中国科学院声学研究所 | A kind of Information hiding and extracting method based on the modulation of integer DCT coefficient |
CN105915912B (en) * | 2016-04-30 | 2018-11-23 | 安徽天虹数码科技股份有限公司 | A kind of insertion of MPEG-2 compressed domain video fragile watermark and detection method |
CN106658021B (en) * | 2016-11-16 | 2019-05-14 | 佛山科学技术学院 | There are two types of the insertions and detection method of watermark MPEG video for a kind of tool |
CN109462763B (en) * | 2018-10-25 | 2021-04-06 | 深圳北斗应用技术研究院有限公司 | Video watermark processing method and system |
CN109660814B (en) * | 2019-01-07 | 2021-04-27 | 福州大学 | Method for detecting deletion tampering of video foreground |
CN112862654B (en) * | 2021-01-26 | 2022-08-30 | 合肥工业大学 | Blind watermark embedding and extracting method based on combination of relational interpolation and linear interpolation |
CN114465826B (en) * | 2022-04-11 | 2022-07-15 | 深圳市天兴诚科技有限公司 | Data encryption method, system and storage medium of coding technology |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101252683A (en) * | 2008-04-02 | 2008-08-27 | 中山大学 | Multifunctional watermark video camera |
CN101668185A (en) * | 2008-09-05 | 2010-03-10 | 广东电子工业研究院有限公司 | Video monitoring system and video data authentication method thereof |
CN102223540A (en) * | 2011-07-01 | 2011-10-19 | 宁波大学 | Information hiding method facing to H.264/AVC (automatic volume control) video |
CN102364949A (en) * | 2011-11-10 | 2012-02-29 | 中山大学深圳研究院 | H.264_advanced video coding (AVC)-video-stream-integrity-based fragile watermark authentication method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7596239B2 (en) * | 2005-08-02 | 2009-09-29 | Lsi Corporation | Method and/or apparatus for video watermarking and steganography using simulated film grain |
-
2012
- 2012-08-01 CN CN201210274393.6A patent/CN102857831B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101252683A (en) * | 2008-04-02 | 2008-08-27 | 中山大学 | Multifunctional watermark video camera |
CN101668185A (en) * | 2008-09-05 | 2010-03-10 | 广东电子工业研究院有限公司 | Video monitoring system and video data authentication method thereof |
CN102223540A (en) * | 2011-07-01 | 2011-10-19 | 宁波大学 | Information hiding method facing to H.264/AVC (automatic volume control) video |
CN102364949A (en) * | 2011-11-10 | 2012-02-29 | 中山大学深圳研究院 | H.264_advanced video coding (AVC)-video-stream-integrity-based fragile watermark authentication method |
Also Published As
Publication number | Publication date |
---|---|
CN102857831A (en) | 2013-01-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102857831B (en) | H.264 video integrality authentication method | |
Yu et al. | New fragile watermarking method for stereo image authentication with localization and recovery | |
CN101681499B (en) | Modifying a coded bitstream | |
Wei et al. | Distortion function based on residual blocks for JPEG steganography | |
He et al. | Reversible data hiding in JPEG images based on negative influence models | |
CN106162195A (en) | A kind of 3D HEVC deep video information concealing method based on single depth frame internal schema | |
CN101527850B (en) | Video self-recovery watermark method based on motion estimation | |
CN102685500B (en) | H.264 large-capacity information hiding method based on information variable-length packet structure | |
CN103533458A (en) | Video watermark embedding method and video watermark extracting method | |
CN101835049A (en) | Generating and authenticating method of self-embedded digital watermark of JPEG (Joint Photographic Experts Group) image | |
CN101572819B (en) | Reversible image watermark method based on quantized DCT coefficient zero values index | |
CN101419702A (en) | Be used for color image tamper location and the semi-fragile watermarking method that recovers | |
CN106127667A (en) | A kind of jpeg image self adaptation steganography method | |
CN101841700A (en) | H.264 compression bit stream oriented video watermark method | |
Yao et al. | Motion vector modification distortion analysis-based payload allocation for video steganography | |
Hu et al. | Tamper detection and image recovery for BTC-compressed images | |
CN109819260A (en) | Video steganography method and device based on the fusion of multi-embedding domain | |
Chu et al. | A digital image watermarking method based on labeled bisecting clustering algorithm | |
Kunhu et al. | A reversible watermarking scheme for ownership protection and authentication of medical images | |
CN105915916A (en) | Video steganalysis method based on motion vector rate-distortion performance estimation | |
Hammami et al. | Blind Semi-fragile Hybrid Domain-Based Dual Watermarking System for Video Authentication and Tampering Localization | |
CN106101713A (en) | A kind of video steganalysis method based on the calibration of window optimum | |
Zhang et al. | Double Reversible Watermarking Algorithm for Image Tamper Detection. | |
CN103325084B (en) | A kind of stereo-picture vulnerable watermark method just can perceiveing distortion based on associating | |
CN111583086B (en) | AMBTC-based self-adaptive digital image watermark and restoration method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150624 Termination date: 20210801 |