[go: up one dir, main page]

CN106055632B - Video authentication method based on scene frame fingerprints - Google Patents

Video authentication method based on scene frame fingerprints Download PDF

Info

Publication number
CN106055632B
CN106055632B CN201610367884.3A CN201610367884A CN106055632B CN 106055632 B CN106055632 B CN 106055632B CN 201610367884 A CN201610367884 A CN 201610367884A CN 106055632 B CN106055632 B CN 106055632B
Authority
CN
China
Prior art keywords
video
fingerprint
record
frame
word
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.)
Active
Application number
CN201610367884.3A
Other languages
Chinese (zh)
Other versions
CN106055632A (en
Inventor
毛家发
张明国
钟丹虹
高飞
肖刚
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN201610367884.3A priority Critical patent/CN106055632B/en
Publication of CN106055632A publication Critical patent/CN106055632A/en
Application granted granted Critical
Publication of CN106055632B publication Critical patent/CN106055632B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
    • G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/783—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content

Landscapes

  • Engineering & Computer Science (AREA)
  • Library & Information Science (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Collating Specific Patterns (AREA)

Abstract

基于场景帧指纹的视频认证方法,首先通过场景帧指纹判定方法，提取出视频片断中5个连续不同的场景帧指纹，组成视频指纹。然后与视频本身的ID信息组成元指纹数据，指纹信息采用Bag‑words形式存储，节省了75％的存储空间。在查找认证过程中，采用倒排文折半技术提高了匹配速度。经仿真实验，我们提出的视频认证方法具有良好的检测性能，平均准确率达到98％以上，在Matlab软环境下查找认证速度每部视频在12秒左右，能够实现网络环境下的实时检测。

The video authentication method based on scene frame fingerprint first extracts 5 consecutive different scene frame fingerprints in the video clip through the scene frame fingerprint determination method to form a video fingerprint. Then, the meta-fingerprint data is formed with the ID information of the video itself. The fingerprint information is stored in the form of Bag‑words, saving 75% of the storage space. In the search and authentication process, the inverted text halving technology is used to improve the matching speed. Through simulation experiments, the video authentication method we proposed has good detection performance, with an average accuracy of more than 98%. In the Matlab soft environment, the search and authentication speed for each video is about 12 seconds, which can achieve real-time detection in a network environment.

Description

Video authentication method based on scene frame fingerprint
Technical field
The invention belongs to video authentication technology fields, disclose one kind and carry out video authentication under new media environment, hit Pirate new method.
Background technique
Most current digital video works use data ciphering method, and digital video content is encrypted, is only awarded The key that power user can just be decrypted.However, data encryption technology faces the problem of being stolen in cipher key transmitting process, once It is stolen that digital video is unable to get protection.The appearance of digital watermark technology can solve the problem of key is lost.Digital watermarking Technology is that hidden label is embedded in digital content, is extracted and is matched by detection instrument, realizes copyright protection purpose.But Digital watermarking product is not strong in the intentional or unintentional attacking ability of resistance at present, and robustness is not firm, greatly restricts number The application development of word digital watermark.
Fingerprint technique can make up the deficiency of encryption technology and digital watermark technology.Video finger print, which refers to, can represent one section The digital signature of vision signal very important visual feature, main purpose are to establish a kind of effective mechanism to compare two video datas Perceived quality.Pay attention to video data itself directly relatively usually very not bigger here, it is corresponding usually smaller to compare it Digital finger-print.
What video finger print technology was particular about is accuracy, robustness, fingerprint size, granularity, certification speed and versatility.Accurately Property includes correct recognition rata, false alarm rate, false dismissed rate；Robustness refers to that unknown video can be subjected to more serious video frequency signal processing After still be able to be identified；Fingerprint size largely determines the inherent capacity of fingerprint database；Granularity be one according to The parameter of Lai Yu application, that is, need unknown video clips how long to identify whole video；The video of practical, commercial is referred to For line system, certification speed is a crucial parameter；Versatility is to refer to carry out recognition capability to different video format. Around these characteristics, numerous scholars set about in terms of the time-space domain of video, airspace, time domain and color space, expand video and refer to The research of line technology achieves gratifying research achievement.In recent years, fingerprint technique is in copyright authentication, copy monitoring, multimedia inspection Rope and tracing pirate etc. are widely used, and vast Study on Fingerprint person proposes many video fingerprinting algorithms, summarize Existing video fingerprinting algorithms can be summarized as 4 classes: color space (color-space-based), time domain (temporal), sky Domain (spatial) and time-space domain (spatio-temporal).
Color space fingerprint extraction method is dependent on the color histogram in video time-space domain.Utilize the color of video clips Statistical property carries out video finger print extraction.But the present vedio color overwhelming majority is 24 true color, statistical magnitude is excessively It is huge, hinder the speed of fingerprint extraction.And different video formats its color can generate apparent change, and it is still more colored Space fingerprint extraction is not applied for black and white video, therefore this method is not applied widely.
Time domain fingerprint extraction method mainly from video sequence from extract time domain specification.This method needs longer video Sequence is not suitable for video clips in short-term.But short-time video is fairly common on webpage now, therefore time domain fingerprint is not It is adapted to online (online) application.
Airspace fingerprint method is to extract feature from each frame or key frame, these methods are similar to finger image method. Airspace fingerprint is divided into global fingerprint and local fingerprint again, and global fingerprint forgives global property, such as image histogram statistical property. The main local feature for extracting image of local fingerprint, such as the partial interest point in frame image, these points of interest are usually answered Use the target retrieval in multimedia.But extract point of interest and need to pre-process image, and video frame enormous amount, This will expend a large amount of calculator memory, therefore this fingerprint extraction method is rarely applied to video field.
Time-space domain fingerprint has forgiven the time domain and spatial information (si) of video, therefore time-space domain fingerprint performance is better than time domain and sky Domain.Mainly there are 3D-DCT, TIRI-DCT, 3D-STIP currently based on time-space domain fingerprint extraction method.These comprehensive video finger prints are calculated Method, they can be reasonably resistant to some common attacks to a certain extent, such as resolution ratio reduce, frame per second reduce, plus make an uproar, Brightness change, contrast change etc., but their certification energy to recodification, reacquisition plus the attacks such as Logo/Text, picture-in-picture Power is limited.
Summary of the invention
The present invention will overcome the disadvantages mentioned above of the prior art, provide a kind of video authentication method based on scene frame fingerprint.
Video authentication method of the present invention based on scene frame fingerprint, comprising the following steps:
1), to the pretreatment of the frame of video；
(1.1) color space conversion is carried out to the color framing in video, takes its luminance component, obtains gray level image；
(1.2) video frame surrounding is sheared, video frame central part is retained；It is scaled to again with fixed dimension (W × H picture Element)；
(1.3) video frame is filtered with 3 × 3 sizes, the Gaussian low-pass filtering that standard deviation is 0.95；
(1.4) by image scaling at 3/4QCIF size (QCIF (144 × 176 pixel)).
2), fingerprint extraction is carried out to by pretreated video frame, comprising the following steps:
(2.1) to pretreated video frame is passed through, piecemeal is carried out, in one 9 × 11 region, a to h is local pixel Be averaged；So frame element extraction method are as follows: (1) the mean value element of entire 9 × 11 subregion；(2) four difference element a-b, C-d, e-f and g-h；720 frame elements are always obtained, wherein 144 mean value elements, are denoted as element A, 576 difference elements are denoted as D element；
(2.2) four weight values are quantized into element A；For the element A of 1-144 dimension, if AiFor element A value, using formula (1) These element As are quantized into four weight values xi:
(2.3) threshold value ThA is dynamically sought, including the following steps:
(2.3.1) takes ai=abs (Ai- 128), abs () is the operator that takes absolute value, by aiA is arranged in by ascending orderk= {a1,a2,…,ak,…,aN}；Here index i and index k be not identical；
(2.3.2) threshold value ThA=ak, k=floor (0.25*N), N=144, floor are to be rounded downwards here；
(2.4) to D Quantification of elements at four weight values；For the D element D of 145-720 dimensioni, they are quantified using formula (2) At four weight values xi:
(2.5) threshold value ThD is dynamically sought, including the following steps:
(2.5.1) takes di=abs (Di), by diD is arranged in by ascending orderk={ d1,d2,…,dk,…,dN}；Here index i It is not identical as index k；
(2.5.2) threshold value ThD=dk, k=floor (0.25*N), N=576, floor are to be rounded downwards here；
(2.6) the 4 heavy element X={ x extracted are stored with binary coded form1,x2,…,x720}
If wordi, i=1,2 ..., 180, which are defined as every 4- dimension element, accounts for 1 coding unit, and this coding mode is using such as Lower formula is calculated:
wordi=43*x(i-1)*4+1+42*x(i-1)*4+2+4*x(i-1)*4+3+x(i-1)*4+4 (3)
(2.7) extraction algorithm of scene frame fingerprint, comprising the following steps:
(2.7.1) whether be blank screen judgement；Applying equation (4) carries out blank screen judgement；
mean(F)<ThBS (4)
Mean (F) is the mean value for indicating image pixel, ThBSIt is blank screen threshold value；
(2.7,2) whether be scene frame judgement；Assuming that the fingerprint of previous scenario frame is SFi-1, the fingerprint of present frame is Fi, i=2 ..., 5；If (5) set up, decide that present frame is another scene frame, otherwise present frame or previous scenario Frame；
d(SFi-1,Fi)≥ThSF, i=2 ..., 5 (5)
Here d (SFi-1,Fi) indicate present frame fingerprint FiPrevious scenario frame fingerprint SFi-1The distance between, ThSFTo determine Threshold value；
3) foundation in video finger print library；The user information, product information and finger print information of copyright authentication video will be needed to tie up It is scheduled on a record, generates metadata (meta data), collection of metadata constitutes metadatabase, it is advised by the row's of pushing down text It is then ranked up and stores；
4) our fingerprint feature: four weight values (Quaternion value) is combined, the invention proposes the row's of falling texts to reduce by half It searches for matching algorithm (inverted file&binary-based Search Matching), its step are as follows:
(4.1) 3600 dimension fingerprint vectors are combined into 900 word, as Bag-Words, each word value by formula (3) Range is 0-255；
(4.2) the literary queue of the row of falling is established；Each video finger print is sequentially inserted into down from small to large by first word size It arranges in literary queue, such as first word is identical, that is so continued on, by the value ascending order arrangement of second word until all Original video fingerprint be inserted into down in the literary queue of row；First fingerprint is constituted with the video finger print and video information of the literary rule compositor of the row of falling Database；
(4.3) Binary searches matching process；Assuming that the Bag-Words sequence of uncertified video fingerprint is AuBWi, i=1, 2,…,900；Specific compromise search step is as follows:
(4.3.1): the record in all metadatabases is put on and does not look into label；
(4.3.2): taking its first word is AuBW1, AuBW is searched in compromise in the literary queue of the row of falling1, the result of lookup can Three kinds of situations can be will appear:
A1) there was only a record；The Bag-Words in the record is so reduced into four weight values fingerprint MeFi, reduction side Method is that each word removes 4 remainders；It is asked to normalize Hamming distance from d by formula (6):
Here i=1,2 ..., L, L are fingerprint length, and AuF is the fingerprint for authenticating video；Then it is asked by (7) formula Value；
As T=0, poll-final shows that the corresponding video of this yuan record is exactly the video for needing to authenticate；Work as T=1 When, the position for writing down the metadata and Hamming distance by the record from and putting on and looked into label；As T=2, only by the record It puts on and has looked into label；
A2) there is a plurality of record；The Hamming distances of all these records is calculated from while marking these records by (6) formula On looked into label；Take minimum Hamming distance from by (7) formula progress evaluation, as T=0, poll-final shows that this yuan records institute Corresponding video is exactly the video for needing to authenticate；The position for writing down the metadata as T=1 and Hamming distance are from working as T=2 When, with no treatment, it is directly entered in next step；
A3 it) does not record；With no treatment, it is directly entered in next step；
(4.3.3): taking its i-th of word is AuBWi, i=2,3 ..., K；AuBW is searched in compromise in the literary queue of the row of fallingi, The result of lookup is it is possible that four kinds of situations；It should be noted that K here is a unknown number, but centainly meet K≤L/m；m For the length of word, m=4 herein；
B1 several) indicate the record for having looked into label；Such case is directly entered in next step；
B2) only have one and do not indicate the record for having looked into label；Such case is pressed and the processing of A1 in (4.3.2)) situation；
B3) there is a plurality of record for not indicating and having looked into label；Such case is pressed and the processing of A2 in (4.3.2)) situation；
B4 it) does not record；In this case by A3 in (4.3.2)) situation processing；
Repeat (4.3.3), until occur T=0 or all record all put on looked into label until；
(4.3.4): if the first two step is that T=0 situation do not occur, only two kinds of situations occur:
C1) at least one record meets T=1；Such case takes the smallest Hamming distance to record from that member, this The video that member record exactly needs to authenticate；Poll-final；
C2) meet T=1 without a record；Such case shows that the video of certification not in metadatabase, issues refusal Information；Poll-final.
Poll-final.
The invention has the advantages that
A. select the intermediate region of video frame as the object to take the fingerprint, this is characterized with the finger print using the mankind The theory of different people it is consistent, while doing so the data operation quantity that can reduce fingerprint extraction process, improve fingerprint Extraction rate.
B. we characterize the difference in video frame region using four weight values, more smart than with two-value Hash, three weight values characterization Carefully, more rationally, to also improve certification discrimination.
C. we store fingerprint metadata library using Bag-words form, save 75% memory space.
D. using the literary binary search algorithm of the row of falling, lookup matching speed is improved.
Detailed description of the invention
Fig. 1 is image block schematic diagram of the invention.
Fig. 2 is that video finger print of the invention extracts flow chart schematic diagram.
Fig. 3 is that the present invention works as ThSFWhen=0.426, five in video display " 28 Weeks Later " segment are continuous different Scene frame.
Fig. 4 aThSFAcquired five different scenes frame when=0.40.Fig. 4 b is ThSFIt is acquired when=0.412 Five different scenes frames.Fig. 4 c is ThSFAcquired five different scenes frame when=0.44.Fig. 4 d is ThSFWhen=0.452 Acquired five different scenes frame.
Fig. 5 is that video finger print of the invention matches architecture diagram.
Specific embodiment
The present invention is further illustrated with reference to the accompanying drawing.
Video authentication method based on scene frame fingerprint of the invention, comprising the following steps:
1), to the pretreatment of the frame of video；
(1.1) color space conversion is carried out to the color framing in video, takes its luminance component, obtains gray level image；
(1.2) video frame surrounding is sheared, video frame central part is retained；It is scaled to again with fixed dimension (W × H picture Element)；
(1.3) video frame is filtered with 3 × 3 sizes, the Gaussian low-pass filtering that standard deviation is 0.95；
(1.4) by image scaling at 3/4QCIF size (QCIF (144 × 176 pixel)).
2), to by pretreated video frame carry out fingerprint extraction, process as shown in Fig. 2 in Figure of description, including Following steps:
(2.1) as shown in Figure of description 1, to pretreated video frame is passed through, piecemeal is carried out, in one 9 × 11 area In domain, a to h is being averaged for local pixel；So frame element extraction method are as follows: (1) the mean value element of entire 9 × 11 subregion； (2) four difference elements a-b, c-d, e-f and g-h；720 frame elements are always obtained, wherein 144 mean value elements, are denoted as A member Element, 576 difference elements, is denoted as D element；
(2.2) four weight values are quantized into element A；For the element A of 1-144 dimension, if AiFor element A value, using formula (1) These element As are quantized into four weight values xi:
(2.3) threshold value ThA is dynamically sought, including the following steps:
(2.3.1) takes ai=abs (Ai- 128), abs () is the operator that takes absolute value, by aiA is arranged in by ascending orderk={ a1, a2,…,ak,…,aN}；Here index i and index k be not identical；
(2.3.2) threshold value ThA=ak, k=floor (0.25*N), N=144, floor are to be rounded downwards here；
(2.4) to D Quantification of elements at four weight values；For the D element D of 145-720 dimensioni, they are quantified using formula (2) At four weight values xi:
(2.5) threshold value ThD is dynamically sought, including the following steps:
(2.5.1) takes di=abs (Di), by diD is arranged in by ascending orderk={ d1,d2,…,dk,…,dN}；Here index i It is not identical as index k；
(2.5.2) threshold value ThD=dk, k=floor (0.25*N), N=576, floor are to be rounded downwards here；
(2.6) the 4 heavy element X={ x extracted are stored with binary coded form1,x2,…,x720}
If wordi, i=1,2 ..., 180, which are defined as every 4- dimension element, accounts for 1 coding unit, and this coding mode is using such as Lower formula is calculated:
wordi=43*x(i-1)*4+1+42*x(i-1)*4+2+4*x(i-1)*4+3+x(i-1)*4+4 (3)
(2.7) extraction algorithm of scene frame fingerprint, comprising the following steps:
(2.7.1) whether be blank screen judgement；Applying equation (4) carries out blank screen judgement；
mean(F)<ThBS (4)
Mean (F) is the mean value for indicating image pixel, ThBSIt is blank screen threshold value；
(2.7,2) whether be scene frame judgement；Assuming that the fingerprint of previous scenario frame is SFi-1, the fingerprint of present frame is Fi, i=2 ..., 5；If (5) set up, decide that present frame is another scene frame, otherwise present frame or previous scenario Frame；
d(SFi-1,Fi)≥ThSF, i=2 ..., 5 (5)
Here d (SFi-1,Fi) indicate present frame fingerprint FiPrevious scenario frame fingerprint SFi-1The distance between, ThSFTo determine Threshold value；
It is to work as Th as shown in attached drawing 3 in specificationSFWhen=0.426, five in video display " 28 Weeks Later " segment Continuous different scene frame.When taking different decision thresholds, the differentiation of scene frame difference, as shown in Figure of description 4, Fig. 4 a is as threshold value ThSFAcquired five different scenes frame when=0.40.Fig. 4 b is as threshold value ThSFInstitute when=0.412 The five different scenes frames obtained.Fig. 4 c is as threshold value ThSFAcquired five different scenes frame when=0.44.Fig. 4 d is to work as Threshold value ThSFAcquired five different scenes frame when=0.452.
3) foundation in video finger print library；The user information, product information and finger print information of copyright authentication video will be needed to tie up It is scheduled on a record, generates metadata (meta data), collection of metadata constitutes metadatabase, it is advised by the row's of pushing down text It is then ranked up and stores, the Meta Fingerprint Database in Figure of description 5 is that the video that we are established refers to Line library；
4) our fingerprint feature: four weight values (Quaternion value) is combined, the invention proposes the row's of falling texts to reduce by half It searches for matching algorithm (inverted file&binary-based Search Matching), if Figure of description 5 is video Fingerprint matching architecture diagram, the figure illustrate macroscopical matching process of fingerprint matching, and its step are as follows:
(4.1) 3600 dimension fingerprint vectors are combined into 900 word, as Bag-Words, each word value by formula (3) Range is 0-255；
(4.2) the literary queue of the row of falling is established；Each video finger print is sequentially inserted into down from small to large by first word size It arranges in literary queue, such as first word is identical, that is so continued on, by the value ascending order arrangement of second word until all Original video fingerprint be inserted into down in the literary queue of row；First fingerprint is constituted with the video finger print and video information of the literary rule compositor of the row of falling Database；
(4.3) Binary searches matching process；Assuming that the Bag-Words sequence of uncertified video fingerprint is AuBWi, i=1, 2,…,900；Specific compromise search step is as follows:
(4.3.1): the record in all metadatabases is put on and does not look into label；
(4.3.2): taking its first word is AuBW1, AuBW is searched in compromise in the literary queue of the row of falling1, the result of lookup can Three kinds of situations can be will appear:
A1) there was only a record；The Bag-Words in the record is so reduced into four weight values fingerprint MeFi, reduction side Method is that each word removes 4 remainders；It is asked to normalize Hamming distance from d by formula (6):
Here i=1,2 ..., L, L are fingerprint length, and AuF is the fingerprint for authenticating video；Then it is asked by (7) formula Value；
As T=0, poll-final shows that the corresponding video of this yuan record is exactly the video for needing to authenticate；Work as T=1 When, the position for writing down the metadata and Hamming distance by the record from and putting on and looked into label；As T=2, only by the record It puts on and has looked into label；
A2) there is a plurality of record；The Hamming distances of all these records is calculated from while marking these records by (6) formula On looked into label；Take minimum Hamming distance from by (7) formula progress evaluation, as T=0, poll-final shows that this yuan records institute Corresponding video is exactly the video for needing to authenticate；The position for writing down the metadata as T=1 and Hamming distance are from working as T=2 When, with no treatment, it is directly entered in next step；
A3 it) does not record；With no treatment, it is directly entered in next step；
(4.3.3): taking its i-th of word is AuBWi, i=2,3 ..., K；AuBW is searched in compromise in the literary queue of the row of fallingi, The result of lookup is it is possible that four kinds of situations；It should be noted that K here is a unknown number, but centainly meet K≤L/m； M is the length of word, herein middle m=4；
B1 several) indicate the record for having looked into label；Such case is directly entered in next step；
B2) only have one and do not indicate the record for having looked into label；Such case is pressed and the processing of A1 in (4.3.2)) situation；
B3) there is a plurality of record for not indicating and having looked into label；Such case is pressed and the processing of A2 in (4.3.2)) situation；
B4 it) does not record；In this case by A3 in (4.3.2)) situation processing；
Repeat (4.3.3), until occur T=0 or all record all put on looked into label until；
(4.3.4): if the first two step is that T=0 situation do not occur, only two kinds of situations occur:
C1) at least one record meets T=1；Such case takes the smallest Hamming distance to record from that member, this The video that member record exactly needs to authenticate；Poll-final；
C2) meet T=1 without a record；Such case shows that the video of certification not in metadatabase, issues refusal Information；Poll-final.

Claims (1)

1.基于场景帧指纹的视频认证方法，包括以下步骤：1. A video authentication method based on scene frame fingerprints, including the following steps: 1)、对视频的帧的预处理；1), the preprocessing of the frame of the video; (1.1)对视频中的彩色帧进行颜色空间转换，取其亮度分量，得到灰度图像；(1.1) Perform color space conversion on the color frame in the video, take its luminance component, and obtain a grayscale image; (1.2)剪切视频帧四周，保留视频帧中心部分；再缩放成具有W×H像素大小的固定尺寸；(1.2) Cut around the video frame, keep the center part of the video frame; rescale to a fixed size with W×H pixel size; (1.3)用3×3大小、标准差为0.95的Gaussian低通滤波对视频帧进行滤波；(1.3) Filter the video frames with Gaussian low-pass filtering with a size of 3×3 and a standard deviation of 0.95; (1.4)将图像缩放成3/4QCIF大小，QCIF为尺寸为144像素×176像素的图像；(1.4) Scale the image to 3/4QCIF size, and QCIF is an image with a size of 144 pixels × 176 pixels; 2)、对经过预处理的视频帧进行指纹提取，包括以下步骤：2), perform fingerprint extraction on the preprocessed video frame, including the following steps: (2.1)对经过预处理的视频帧，进行分块，在一个9×11的区域内，a至h是局部像素的平均；那么帧元素提取方法为:(1)整个9×11子区域的均值元素；(2)四个差分元素a-b、c-d、e-f和g-h；总共得到720帧元素，其中144个均值元素，记为A元素，576个差分元素，记为D元素；(2.1) Divide the preprocessed video frame into blocks. In a 9×11 area, a to h are the average of local pixels; then the frame element extraction method is: (1) The whole 9×11 sub-area Mean element; (2) Four differential elements a-b, c-d, e-f and g-h; a total of 720 frame elements are obtained, of which 144 mean elements are denoted as A elements, and 576 differential elements are denoted as D elements; (2.2)对A元素量化成四重值；对于1-144维的A元素，设Ai为A元素值，应用公式(1)把这些A元素量化成四重值xi:(2.2) A element is quantized into a quadruple value; for the A element of 1-144 dimensions, let A i be the A element value, and formula (1) is used to quantize these A elements into a quadruple value x i : (2.3)动态地求取阈值ThA，包括以下几个步骤：(2.3) Dynamically obtain the threshold ThA, including the following steps: (2.3.1)取ai＝abs(Ai-128)，abs(·)为取绝对值算子，将ai按升序排列成ak＝{a1,a2,…,ak,…,aN}；这里的索引i与索引k不相同；(2.3.1) Take a i =abs(A i -128), abs(·) is the absolute value operator, and arrange a i in ascending order into a k ={a 1 ,a 2 ,..., ak , ...,a N }; here the index i is not the same as the index k; (2.3.2)阈值ThA＝ak，这里k＝floor(0.25*N)，N＝144，floor为向下取整；(2.3.2) Threshold ThA= ak , where k=floor(0.25*N), N=144, floor is rounded down; (2.4)对D元素量化成四重值；对于145-720维的D元素Di，应用公式(2)把它们量化成四重值xi：(2.4) Quantize D elements into quartet values; for D elements D i of dimensions 145-720, apply formula (2) to quantize them into quartet values xi : (2.5)动态地求取阈值ThA，包括以下几个步骤：(2.5) Dynamically obtain the threshold ThA, including the following steps: (2.5.1)取di＝abs(Di)，将di按升序排列成dk＝{d1,d2,…,dk,…,dN}；这里的索引i与索引k不相同；(2.5.1) Take d i =abs(D i ), and arrange d i in ascending order into d k ={d 1 ,d 2 ,...,d k ,...,d N }; here index i and index k Are not the same; (2.5.2)阈值ThD＝dk，这里k＝floor(0.25*N)，N＝576，floor为向下取整；(2.5.2) Threshold ThD=d k , where k=floor(0.25*N), N=576, floor is rounded down; (2.6)用二进制编码形式来存储提取出来的4重元素X＝{x1,x2,…,x720}(2.6) Store the extracted quadruple element X={x 1 ,x 2 ,...,x 720 } in binary coding form 设wordi，i＝1,2,…,180定义为每4-维元素占1个编码单元，这种编码方式采用如下公式计算得到：Suppose word i , i=1,2,...,180 is defined as one coding unit per 4-dimensional element. This coding method is calculated by the following formula: wordi＝43*x(i-1)*4+1+42*x(i-1)*4+2+4*x(i-1)*4+3+x(i-1)*4+4 (3)word i = 4 3 *x (i-1)*4+1 +4 2 *x (i-1)*4+2 +4*x (i-1)*4+3 +x (i-1) *4+4 (3) (2.7)场景帧指纹的提取算法，包括以下步骤：(2.7) The extraction algorithm of scene frame fingerprint, including the following steps: (2.7.1)是否为黑屏的判断；应用式(4)进行黑屏判断；(2.7.1) Judgment of whether it is a black screen; use formula (4) to judge the black screen; mean(F)＜ThBS (4)mean(F) < Th BS (4) mean(F)是表示图像像素的均值，ThBS是黑屏阈值；mean(F) is the mean value representing the image pixels, Th BS is the black screen threshold; (2.7.2)是否为场景帧的判断；假设前一场景帧的指纹为SFi-1，当前帧的指纹为Fi，i＝2,…,5；如果式(5)成立，那么就判定当前帧为另一场景帧，否则当前帧还是前一场景帧；(2.7.2) Judgment of whether it is a scene frame; Suppose the fingerprint of the previous scene frame is SF i-1 , and the fingerprint of the current frame is F i , i=2,...,5; if the formula (5) holds, then Determine that the current frame is another scene frame, otherwise the current frame is the previous scene frame; d(SFi-1,Fi)≥ThSF,i＝2,…,5 (5)d(SF i-1 ,F i )≥Th SF ,i=2,...,5 (5) 这里d(SFi-1,Fi)表示当前帧指纹Fi前一场景帧指纹SFi-1之间的距离，ThSF为判定阈值；Here d(SF i -1 , Fi ) represents the distance between the current frame fingerprint Fi and the previous scene frame fingerprint SF i -1 , and Th SF is the judgment threshold; 3)视频指纹库的建立；将需要版权认证视频的用户信息、产品信息和指纹信息绑定在一条记录上，生成元数据meta data，元数据集合构成元数据库，将其按按倒排文规则进行排序并存储；3) The establishment of a video fingerprint database; bind the user information, product information and fingerprint information of the video that needs copyright authentication to a record, generate metadata meta data, and the metadata collection constitutes a metadata database, which is arranged according to the rules of inverted text. sort and store; 4)结合指纹特点：四重值Quaternion value，提出了倒排文折半搜索匹配算法inverted file&binary-based Search Matching，其步骤如下：4) Combined with fingerprint characteristics: Quaternion value, an inverted file & binary-based Search Matching algorithm is proposed. The steps are as follows: (4.1)按式(3)将3600维指纹向量组合成900个word，即为Bag-Words,每个word值范围为0-255；(4.1) Combine the 3600-dimensional fingerprint vector into 900 words according to formula (3), namely Bag-Words, and the value of each word ranges from 0 to 255; (4.2)建立倒排文队列；每个视频指纹按第一个word大小从小到大顺序插入到倒排文队列中，如第一个word相同，那按第二个word的值升序排列，如此连续下去，直到所有的原视频指纹插入到倒排文队列中；以倒排文规则排序的视频指纹及视频信息构成元指纹数据库；(4.2) Establish an inverted text queue; each video fingerprint is inserted into the inverted text queue according to the size of the first word from small to large. If the first word is the same, it is sorted in ascending order by the value of the second word, so Continue until all the original video fingerprints are inserted into the inverted text queue; the video fingerprints and video information sorted by the inverted text rules constitute a meta-fingerprint database; (4.3)折半搜索匹配方法；假设待认证视频指纹的Bag-Words序列为AuBWi，i＝1,2,…,900；具体的折中搜索步骤如下：(4.3) Half-fold search and matching method; Assume that the Bag-Words sequence of the video fingerprint to be authenticated is AuBW i , i=1,2,...,900; the specific compromise search steps are as follows: (4.3.1):将所有的元数据库中的记录标上未查标记；(4.3.1): mark all records in the metadata database with unchecked marks; (4.3.2):取其第一个word为AuBW1，在倒排文队列中折中查找AuBW1，查找的结果可能会出现三种情况：(4.3.2): Take the first word as AuBW 1 , and search for AuBW 1 in the inverted text queue. There may be three situations in the search result: A1)只有一条记录；那么将该记录中的Bag-Words还原成四重值指纹MeFi，还原方法为每个word除4取余；按式(6)求其归一化Hamming距离d：A1) There is only one record; then the Bag-Words in the record are restored to quadruple-valued fingerprint MeF i , and the restoration method is to divide each word by 4 and take the remainder; according to formula (6), find its normalized Hamming distance d: 这里的i＝1,2,…,L，L为指纹长度，AuF为认证视频的指纹；然后按(7)式进行求值；Here i=1,2,...,L, L is the length of the fingerprint, and AuF is the fingerprint of the authentication video; then evaluate according to formula (7); 当T＝0时，查询结束，表明该记录所对应的视频就是需要认证的视频；当T＝1时，记下该记录的位置和Hamming距离，并将该记录标上已查标记；当T＝2时，仅将该记录标上已查标记；When T=0, the query ends, indicating that the video corresponding to the record is the video that needs to be authenticated; when T=1, record the location and Hamming distance of the record, and mark the record as checked; when T=1 When = 2, only mark the record as checked; A2)有多条记录；按(6)式计算出所有这些记录的Hamming距离，同时将这些记录标上已查标记；取最小Hamming距离，按(7)式进行求值，当T＝0时，查询结束，表明该记录所对应的视频就是需要认证的视频；当T＝1时，记下该记录的位置和Hamming距离，当T＝2时，不作任何处理，直接进入下一步；A2) There are multiple records; calculate the Hamming distance of all these records according to formula (6), and mark these records with checked marks; take the minimum Hamming distance, and evaluate according to formula (7), when T=0 , the query ends, indicating that the video corresponding to the record is the video that needs to be authenticated; when T=1, write down the location of the record and the Hamming distance, when T=2, do not do any processing, and go directly to the next step; A3)没有记录；不作任何处理，直接进入下一步；A3) There is no record; do not do any processing, go directly to the next step; (4.3.3):取其第i个word为AuBWi,i＝2,3,…,K；在倒排文队列中折中查找AuBWi，查找的结果可能会出现四种情况；需要注意的是这里的K是个未知数，但一定满足K≤L/m；在此处m＝4；(4.3.3): Take the i-th word as AuBW i , i=2,3,...,K; search for AuBW i in the inverted text queue, and the result of the search may have four situations; need to pay attention to The point is that K here is an unknown number, but it must satisfy K≤L/m; here m=4; B1)有若干条已标有已查标记的记录；这种情况直接进入下一步；B1) There are several records marked with checked marks; in this case, go directly to the next step; B2)仅有一条未标有已查标记的记录；这种情况按与(4.3.2)中A1)情况处理；B2) There is only one record that is not marked with the checked mark; this case is handled as in the case of A1) in (4.3.2); B3)有多条未标有已查标记的记录；这种情况按与(4.3.2)中A2)情况处理；B3) There are multiple records that are not marked with the checked mark; this situation shall be handled as in the case of A2) in (4.3.2); B4)没有记录；这种情况下按(4.3.2)中A3)情况处理；B4) There is no record; in this case, it is handled as in A3) in (4.3.2); 重复(4.3.3)，直到出现T＝0或所有记录都标上已查标记为止；Repeat (4.3.3) until T=0 or all records are marked as checked; (4.3.4):如果前二步是没有出现T＝0情况，那么只有二种情况出现：(4.3.4): If there is no T=0 situation in the first two steps, then there are only two situations: C1)至少有一条记录满足T＝1；这种情况取最小的Hamming距离那条元记录，这条元记录就是需要认证的视频；查询结束；C1) At least one record satisfies T=1; in this case, take the meta record with the smallest Hamming distance, and this meta record is the video that needs to be authenticated; the query ends; C2)没有一条记录满足T＝1；这种情况表明认证的视频不在元数据库中，发出拒绝信息；查询结束。C2) No record satisfies T=1; this situation indicates that the authenticated video is not in the metadata database, and a rejection message is issued; the query ends.
CN201610367884.3A 2016-05-27 2016-05-27 Video authentication method based on scene frame fingerprints Active CN106055632B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610367884.3A CN106055632B (en) 2016-05-27 2016-05-27 Video authentication method based on scene frame fingerprints

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610367884.3A CN106055632B (en) 2016-05-27 2016-05-27 Video authentication method based on scene frame fingerprints

Publications (2)

Publication Number Publication Date
CN106055632A CN106055632A (en) 2016-10-26
CN106055632B true CN106055632B (en) 2019-06-14

Family

ID=57175798

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610367884.3A Active CN106055632B (en) 2016-05-27 2016-05-27 Video authentication method based on scene frame fingerprints

Country Status (1)

Country Link
CN (1) CN106055632B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109815652B (en) * 2018-12-18 2020-12-25 浙江工业大学 Spark-based real-time active picture tracking protection method
CN111008301B (en) * 2019-12-19 2023-08-15 新华智云科技有限公司 Method for searching video by using graph
CN116821416A (en) * 2022-03-22 2023-09-29 腾讯科技（深圳）有限公司 Video fingerprint processing method and device, storage medium and electronic equipment
CN116582282B (en) * 2023-07-13 2023-09-19 深圳市美力高集团有限公司 Anti-tampering encryption storage method for vehicle-mounted video

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2008045139A3 (en) * 2006-05-19 2008-07-24 Univ New York State Res Found Determining whether or not a digital image has been tampered with
CN101894156A (en) * 2010-07-12 2010-11-24 清华大学 A Video Fingerprint Matching Method Based on Bipartite Graph
CN104036280A (en) * 2014-06-23 2014-09-10 国家广播电影电视总局广播科学研究院 Video fingerprinting method based on region of interest and cluster combination
CN104205865A (en) * 2012-03-29 2014-12-10 阿尔卡特朗讯公司 Method and apparatus for authenticating video content

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2008045139A3 (en) * 2006-05-19 2008-07-24 Univ New York State Res Found Determining whether or not a digital image has been tampered with
CN101894156A (en) * 2010-07-12 2010-11-24 清华大学 A Video Fingerprint Matching Method Based on Bipartite Graph
CN104205865A (en) * 2012-03-29 2014-12-10 阿尔卡特朗讯公司 Method and apparatus for authenticating video content
CN104036280A (en) * 2014-06-23 2014-09-10 国家广播电影电视总局广播科学研究院 Video fingerprinting method based on region of interest and cluster combination

Also Published As

Publication number Publication date
CN106055632A (en) 2016-10-26

Similar Documents

Publication Publication Date Title
Duan et al. High-capacity image steganography based on improved FC-DenseNet
Daniel et al. Texture and quality analysis for face spoofing detection
CN106055632B (en) Video authentication method based on scene frame fingerprints
CN118587568B (en) Depth forging active evidence obtaining method based on separable perception hash enhancement
CN119379524B (en) Image counterfeiting active defense method based on multiple watermark fusion and cross-domain learning
CN104091303A (en) Robust image hashing method and device based on Radon transformation and invariant features
Wan et al. JND-guided perceptually color image watermarking in spatial domain
CN114036467B (en) A short video copyright protection method based on blockchain
CN104268823A (en) Digital watermark algorithm based on image content
CN102693522A (en) Method for detecting region duplication and forgery of color image
Kim et al. Multimodal biometric image watermarking using two-stage integrity verification
CN111861846A (en) Electronic document digital watermark processing method and system
Dogan et al. A robust color image watermarking with Singular Value Decomposition method
CN105227964B (en) Video-frequency identifying method and system
Mao et al. A method for video authenticity based on the fingerprint of scene frame
CN108491913B (en) Method for authenticating QR (quick response) code with secret beautification function
Hu et al. Effective forgery detection using DCT+ SVD-based watermarking for region of interest in key frames of vision-based surveillance
Madake et al. Image tampering detection using error level analysis and metadata analysis
Sharma et al. A review of passive forensic techniques for detection of copy-move attacks on digital videos
CN105741222A (en) Steganographic information positioning method based on pixel subset embedding rate estimation
Lu et al. A robust coverless image steganography method based on face recognition and camouflage image
Rana et al. Image steganography method based on kohonen neural network
CN113989526B (en) Image-aware hashing method based on discrete wavelet transform
Patil et al. Image hashing using AQ-CSLBP with double bit quantization
Sethi et al. Video security against deepfakes and other forgeries

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant