CN107122745A - The method and device of personage track in a kind of identification video - Google Patents
The method and device of personage track in a kind of identification video Download PDFInfo
- Publication number
- CN107122745A CN107122745A CN201710293791.5A CN201710293791A CN107122745A CN 107122745 A CN107122745 A CN 107122745A CN 201710293791 A CN201710293791 A CN 201710293791A CN 107122745 A CN107122745 A CN 107122745A
- Authority
- CN
- China
- Prior art keywords
- face
- tensor
- data
- colour picture
- video
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/48—Matching video sequences
Abstract
The invention discloses a kind of method and device for recognizing personage track in video, this method includes：All video segments comprising facial image in video to be identified are converted into multiple colour pictures for including facial image；The face location data in colour picture are obtained, face tetradic model is set up；Tensor computation is carried out, by the multiple face characteristic tensor data and benchmark face characteristic tensor data, norm calculation summation is carried out；Corresponding face in the colour picture is classified, obtain each corresponding colour picture of same face, each corresponding colour picture of the same face is restored in the video to be identified, personage track of the face in the video to be identified is obtained.Realize applied widely and reduce cost and the purpose of operand.
Description
Technical field
The present invention relates to technical field of face recognition, more particularly to a kind of face identification method and dress based on tensor
Put.
Background technology
Along with science and technology development, traditional personal verification means such as mode such as password, certificate and IC-card, due to
With the separability of holder, forgery is caused, usurp or the phenomenon such as decodes happen occasionally, can not meet modern social economy
The need for activity and social safety are taken precautions against.Therefore, biometrics identification technology arises at the historic moment, and in biometrics identification technology
Due to face recognition technology have convenient collection be easy to be received and should not forge the characteristics of so that face recognition technology is always
It is study hotspot in biometrics identification technology.
Existing face recognition technology can be summarized as：After detecting face and positioning facial key feature, mainly
Human face region be tailored out then by pretreatment after, then by recognizer complete face characteristic extraction, and with
Known face is contrasted in database, to complete final face classification.But, along with increasing for Video Applications, carry
Information in video is taken to become more and more valuable, the track of the someone in corresponding detection video is also a significant work
Make, due to different faces occurring in video, it is necessary to identical face is divided into a class, different faces is divided into different classes
To realize the track for following the trail of someone in video.As can be seen here, if studied using existing face recognition technology in video
Someone track, then need to set up the known face of stock and training sample, and needs the known face of stock to greatly reduce
Although the scope of application of face recognition technology is, it is necessary to a large amount of training samples can improve discrimination but add cost,
And contrasted with the known face of stock also increase+added operand.Obviously, by existing face recognition technology
To realize that this purpose is irrational.
The content of the invention
Above mentioned problem is directed to, the present invention provides a kind of method and device for recognizing personage track in video, realized suitable
It is wide and reduce cost and the purpose of operand with scope.
To achieve these goals, personage track in video is recognized there is provided a kind of according to the first aspect of the invention
Method, this method includes：
All video segments comprising facial image in video to be identified are converted into multiple colours for including facial image
Picture；
The face location data in the colour picture comprising facial image are obtained, are built according to the face location data
The face tetradic model of the vertical colour picture comprising facial image；
Tensor computation is carried out in the face tetradic model, the cromogram that each includes facial image is obtained
Multiple face characteristic tensor data corresponding to piece；
By the multiple face characteristic tensor data and benchmark face characteristic tensor data, norm calculation summation is carried out, its
In, the benchmark face characteristic tensor data are that some the face tensor chosen in the multiple face characteristic tensor data is special
Levy data；
Sampling threshold σ, foundation are set for the sample size of colour picture according to the Video Quality Metric to be identifiedCorresponding face in the colour picture is classified, each corresponding coloured silk of same face is obtained
Chromatic graph piece, wherein, S is the corresponding face location width number of face characteristic tensor data on the basis of norm calculation summed result, m
According to corresponding n benchmark face characteristic tensor data are face position height data, and r=3 represents reddish yellow blue three-color；
Each corresponding colour picture of the same face is restored in the video to be identified, the face is recorded and exists
The positional information occurred in the video to be identified, obtains personage track of the face in the video to be identified.
It is preferred that, the face location data in the colour picture comprising facial image described in the acquisition, according to the people
Face position data sets up the face tetradic model of the colour picture comprising facial image, including：
The colour picture comprising facial image is converted into corresponding black and white picture, and extracts the black and white picture
The position data of middle face, wherein, the position data includes initial coordinate, width data and altitude information；
According to the position data of face in the black and white picture, colour picture corresponding to the black and white picture is got
Face location data；
Width data m and altitude information n in the face location data of the colour picture, set up the cromogram
The rank tensor A [1 of face three in piece：M, 1：N, 1：R], wherein, r=3 represents reddish yellow blue three-color；
Translation tensor data m/3 is set according to the width data m, translation tensor B [i are obtained：M*2/3+i, i：n-m*
2/3+i*2,1：R], wherein, i represents the coefficient of relationship of the tensor and former tensor translated out, and value is 1≤i≤m/3；
Translationization processing is carried out to the rank tensor of face three according to the translation tensor, the face tetradic is obtained
MODEL C [1:m/3,1:m*2/3,1:N-m*2/3,1：r].
It is preferred that, it is described that tensor computation is carried out in the face tetradic model, obtain described each and include face
Multiple face characteristic tensor data corresponding to the colour picture of image, including：
Set up the series C of the face tetradic modelkCalculating function, wherein,
Ck=0.75*Ck-1+0.25*Bk, in formula, 1≤k≤m/3；
According to the translation tensor B, calculate and obtain the translation tensor BkValue；
The face characteristic tensor data C is calculated, wherein, C=Ck, in formula, k=m/3.
According to the second aspect of the invention there is provided the device of personage track in a species video, the device includes：
Conversion module, for whether video to be identified to be included into facial image according to video segment, is converted into multiple include
The colour picture of facial image；
Module is set up, for obtaining the face location data in the colour picture comprising facial image, according to described
Face location data set up the face tetradic model of the colour picture comprising facial image；
First computing module, for carrying out tensor computation in the face tetradic model, obtains each described bag
Multiple face characteristic tensor data corresponding to colour picture containing facial image；
Second computing module, for by the multiple face characteristic tensor data and benchmark face characteristic tensor data, entering
Row norm calculation is summed, wherein, the benchmark face characteristic tensor data are to be selected in the multiple face characteristic tensor data
Some the face tensor property data taken；
Sort module, for setting sampling threshold σ according to the Video Quality Metric to be identified for the sample size of colour picture,
FoundationCorresponding face in the colour picture is classified, same face is obtained corresponding each
Individual colour picture, wherein, S is the corresponding face location width of face characteristic tensor data on the basis of norm calculation summed result, m
Data, corresponding n benchmark face characteristic tensor data are face position height data, and r=3 represents reddish yellow blue three-color；
Track identification module, for each corresponding colour picture of the same face to be restored into the video to be identified
In, the positional information that the face occurs in the video to be identified is recorded, the face is obtained in the video to be identified
In personage track.
It is preferred that, the module of setting up includes：
First extraction unit, for the colour picture comprising facial image to be converted into corresponding black and white picture, and
The position data of face in the black and white picture is extracted, wherein, the position data includes initial coordinate, width data and height
Degrees of data；
Second extraction unit, for the position data according to face in the black and white picture, gets the black and white picture
The face location data of corresponding colour picture；
First sets up unit, for the width data m and the high number of degrees in the face location data according to the colour picture
According to n, the rank tensor A [1 of face three set up in the colour picture：M, 1：N, 1：R], wherein, r=3 represents the blue three kinds of face of reddish yellow
Color；
Acquiring unit, for setting translation tensor data m/3 according to the width data m, obtains translation tensor B [i：m*
2/3+i, i：N-m*2/3+i*2,1：R], wherein, i represents the coefficient of relationship of the tensor and former tensor translated out, and value is 1≤i
≦m/3；
Second sets up unit, for carrying out translationization processing to the rank tensor of face three according to the translation tensor, obtains
Obtain the face tetradic MODEL C [1:m/3,1:m*2/3,1:N-m*2/3,1：r].
It is preferred that, first computing module includes：
Function sets up unit, the series C for setting up the face tetradic modelkCalculating function, wherein,
Ck=0.75*Ck-1+0.25*Bk, in formula, 1≤k≤m/3；
Tensor computation unit is translated, the translation tensor B is obtained for according to the translation tensor B, calculatingkValue；
Data Computation Unit, for calculating the face characteristic tensor data C, wherein, C=Ck, in formula, k=m/3.
Compared to prior art, the present invention is converted to all video segments comprising facial image in video to be identified many
The individual colour picture for including facial image, the face location data in colour picture establish face tensor quadravalence model,
What the face location data in thus according to colour picture were set up, thus the face tensor quadravalence model contain it is more
Face information, carry out recognition of face process will be more accurate；Tensor computation is carried out in the face tetradic model,
Obtain multiple face characteristic tensor data corresponding to each described colour picture comprising facial image；By the multiple face
Characteristic tensor data and benchmark face characteristic tensor data, carry out norm calculation summation, sampling threshold are then set, to the coloured silk
Corresponding face is classified in chromatic graph piece, obtains each corresponding colour picture of same face；By the same face correspondence
Each colour picture be restored in the video to be identified, record the position that the face occurs in the video to be identified
Information, obtains personage track of the face in the video to be identified.Due to based on tensor mould in face recognition process
Type does not need training parameter with regard to that can recognize the different photos of same person, can be effectively saved cost；Simultaneously identification is regarded
Someone track is the known face for not needing this person in frequency, therefore methods described is extremely wide using scope.In general, by this
Invention realizes applied widely and reduces cost and the purpose of operand.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the accompanying drawing used required in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
The embodiment of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can also basis
The accompanying drawing of offer obtains other accompanying drawings.
Fig. 1 is the schematic flow sheet of the method for personage track in a kind of identification video that the embodiment of the present invention one is provided；
Fig. 2 is that the flow for setting up tetradic model shown in the corresponding Fig. 1 of the embodiment of the present invention two in S12 steps is shown
It is intended to；
Fig. 3 is the stream of the calculating face characteristic tensor data in S13 steps shown in the corresponding Fig. 1 of the embodiment of the present invention two
Journey schematic diagram；
Fig. 4 is the structural representation of the device of personage track in a kind of identification video that the embodiment of the present invention three is provided.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than whole embodiments.It is based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made
Embodiment, belongs to the scope of protection of the invention.
Term " first " and " second " in description and claims of this specification and above-mentioned accompanying drawing etc. are to be used for area
Not different objects, rather than for describing specific order.In addition term " comprising " and " having " and their any deformations,
It is intended to cover non-exclusive include.For example contain the process of series of steps or unit, method, system, product or set
It is standby not to be set in the step of having listed or unit, but the step of may include not list or unit.
Embodiment one
Referring to the flow signal for the method that Fig. 1 is personage track in a kind of identification video that the embodiment of the present invention one is provided
Figure, this method comprises the following steps：
S11, all video segments comprising facial image in video to be identified are converted to it is multiple comprising facial image
Colour picture；
Specifically, in order to recognize the personage track in video it is necessary to first getting the colour picture for including personage's face.
S12, the face location data obtained in the colour picture comprising facial image, according to the face location number
According to the face tetradic model for setting up the colour picture comprising facial image；
It is understood that the face tetradic that the face location data in thus according to face colour picture are set up
Model, the face information not only included more comprehensively but also reduces the influence that brings of face skew, can effectively improve identification
Rate.
S13, carry out tensor computation in the face tetradic model, obtain the coloured silk that each includes facial image
Multiple face characteristic tensor data corresponding to chromatic graph piece；
S14, by the multiple face characteristic tensor data and benchmark face characteristic tensor data, carry out norm calculation and ask
With, wherein, the benchmark face characteristic tensor data are some face chosen in the multiple face characteristic tensor data
Tensor property data；
S15, according to the Video Quality Metric to be identified for colour picture sample size set sampling threshold σ, foundationCorresponding face in the colour picture is classified, each corresponding coloured silk of same face is obtained
Chromatic graph piece, wherein, S is the corresponding face location width number of face characteristic tensor data on the basis of norm calculation summed result, m
According to corresponding n benchmark face characteristic tensor data are face position height data, and r=3 represents reddish yellow blue three-color；
S16, each corresponding colour picture of the same face is restored in the video to be identified, records the people
The positional information that face occurs in the video to be identified, obtains personage track of the face in the video to be identified.
Specifically, recording the positional information that the face occurs in the video to be identified and being：Record identical people
The information such as the frame number, displacement and the position that occur in video to be identified.
By technical scheme disclosed in the embodiment of the present invention one, all videos of facial image will be included in video to be identified
Fragment is converted to multiple colour pictures for including facial image, and the face location data in colour picture establish face
Quadravalence model is measured, what the face location data in thus according to colour picture were set up, so the face tensor quadravalence mould
Type contains more face informations, and the process for carrying out recognition of face will be more accurate；In the face tetradic model
Tensor computation is carried out, multiple face characteristic tensor data corresponding to each described colour picture comprising facial image are obtained；
By the multiple face characteristic tensor data and benchmark face characteristic tensor data, norm calculation summation is carried out, then sets and adopts
Sample threshold value, classifies to corresponding face in the colour picture, obtains each corresponding colour picture of same face；By institute
State each corresponding colour picture of same face to be restored in the video to be identified, record the face and to be identified regarded described
The positional information occurred in frequency, obtains personage track of the face in the video to be identified.In face recognition process
Due to not needing training parameter with regard to the different photos of same person can be recognized based on tensor model, cost can be effectively saved；
Cause that someone track is the known face for not needing this person in identification video simultaneously, therefore methods described is extremely wide using scope.
In general, realize applied widely by the present invention and reduce cost and the purpose of operand.
Embodiment two
The detailed process of S11 to S16 steps with reference to described in the embodiment of the present invention one and Fig. 1, and reference picture 2 is this
The schematic flow sheet for setting up tetradic model shown in the corresponding Fig. 1 of inventive embodiments two in S12 steps, the acquisition institute
The face location data in the colour picture comprising facial image are stated, face is included according to being set up the face location data
The face tetradic model of the colour picture of image, including：
S121, the colour picture comprising facial image is converted into corresponding black and white picture, and extracted described black
The position data of face in white picture, wherein, the position data includes initial coordinate, width data and altitude information；
Specifically, the main purpose that colour picture first is converted into black and white picture to obtain position data is to reduce computing
Amount, is converted to black and white picture by colour picture and detects possible face location in black and white picture, by detecting skin, eye
The face representative feature such as eyeball, mouth and nose further confirms that whether be face, and records corresponding face location data.
S122, the position data according to face in the black and white picture, get the colour corresponding to the black and white picture
The face location data of picture；
Wherein, the face location data of the colour picture include initial coordinate, width data and altitude information；
S123, according to the width data m and altitude information n in the face location data of the colour picture, set up described
The rank tensor A [1 of face three in colour picture：M, 1：N, 1：R], wherein, r=3 represents reddish yellow blue three-color；
S124, translation tensor data m/3 is set according to the width data m, translation tensor B [i are obtained：M*2/3+i, i：
N-m*2/3+i*2,1：R], wherein, i represents the coefficient of relationship of the tensor and former tensor translated out, and value is 1≤i≤m/3；
Specifically, not being positive face because the face in video is most of, or block, therefore need in the presence of offseting or existing
Further face translationization processing is wanted to construct a tetradic.I represents relation of the flat tensor gone out with former tensor, than
Such as second tensor that i=2 marks are translated out, the tensor takes the 2 of former first index of tensor to arrive m*2/3+2 rows, the second index
2 arrive n-m*2/3+4 rows.
S125, translationization processing, the acquisition face four are carried out to the rank tensor of face three according to the translation tensor
Rank tensor MODEL C [1:m/3,1:m*2/3,1:N-m*2/3,1：r].
Corresponding reference picture 3 is the calculating face characteristic in S13 steps shown in the corresponding Fig. 1 of the embodiment of the present invention two
The schematic flow sheet of data is measured, it is described that tensor computation is carried out in the face tetradic model, obtain described each and include
Multiple face characteristic tensor data corresponding to the colour picture of facial image, including：
S131, the series C for setting up the face tetradic modelkCalculating function, wherein, Ck=0.75*Ck-1+
0.25*Bk, in formula, 1≤k≤m/3；
S132, according to the translation tensor B, calculate and obtain the translation tensor BkValue；
S133, the calculating face characteristic tensor data C, wherein, C=Ck, in formula, k=m/3.
Specifically, being less than 255 because each numerical value in the RGB figures of face is all higher than 0, so tetradic model correspondence
Tensor C be positive tensor, and the characteristic value of the maximum absolute value of positive tensor is not necessarily negative.If arbitrary element increases in positive tensor
The also strict increase therewith of big then eigenvalue of maximum, thus calculated as follows it is contemplated that solving the corresponding eigenvalue of maximum of positive tensor C and obtaining
Method is linear convergence, and derivation algorithm is as follows：
||X0| |=1
Xk=C Xk-1
In above formula, k represents iterations, i.e. kth time iteration.X0Primary iteration value is represented, positive vector is taken, typically takes complete one
It is vectorial and then unitization.XkThe iterative vectorized of kth time is represented, the vector converges to characteristic vector；tkRepresent vectorial correspondence position ratio
Value maximum that number, the number linear convergence to characteristic value；
Although the solution of the characteristic value and characteristic vector of positive tensor is linear convergence, tensor data volume itself compared with
Greatly, to do multiplication computation amount huge for tensor and vector, even if the few operand of iterations is still larger, in order to improve corresponding fortune
Calculate speed, in embodiments of the present invention the method for preferred ordered series of numbers series come replace tensor property value and characteristic vector solution, i.e.,：
Ck=0.75*Ck-1+0.25*Bk
C1=C (1),
Bi+1=C (i+1)
C1The 1st of ordered series of numbers is represented, is the 1st tensor B translated out1；CiI-th of ordered series of numbers is represented, by expression formula iteration
Obtain；Bi+1Exactly translated outi+1Individual tensor, then face tensor C be updated to a series of to embody the new of face characteristic
Data Ck。
Technical scheme disclosed according to embodiments of the present invention two, is converted to black and white picture, Ran Houtong by colour picture first
The face location data that the face location data calculated in black and white picture obtain the colour picture are crossed, can so be reduced in people
Operand in face identification process；Due to consideration that it is not positive face, it is necessary to carry out the translation of face that face in video is most of
Change processing so as to set up the tetradic of face, and reflect that the essence of face is special with positive tensor property value and characteristic vector
Levy, face characteristic tensor data are solved by setting the method for series and effectively reduce amount of calculation, while improving identification
Efficiency.
Embodiment three
Method with personage track in the identification video disclosed in the embodiment of the present invention one and embodiment two is corresponding, this hair
Bright embodiment three additionally provides a kind of device for recognizing personage track in video, is that the embodiment of the present invention three is provided referring to Fig. 4
A kind of identification video in personage track device structural representation, the device includes：
Conversion module 1, for whether video to be identified to be included into facial image according to video segment, is converted into multiple include
The colour picture of facial image；
Module 2 is set up, for obtaining the face location data in the colour picture comprising facial image, according to described
Face location data set up the face tetradic model of the colour picture comprising facial image；
First computing module 3, for carrying out tensor computation in the face tetradic model, obtains each described bag
Multiple face characteristic tensor data corresponding to colour picture containing facial image；
Second computing module 4, for by the multiple face characteristic tensor data and benchmark face characteristic tensor data, entering
Row norm calculation is summed, wherein, the benchmark face characteristic tensor data are to be selected in the multiple face characteristic tensor data
Some the face tensor property data taken；
Sort module 5, for setting sampling threshold according to the Video Quality Metric to be identified for the sample size of colour picture
σ, foundationCorresponding face in the colour picture is classified, same face is obtained corresponding
Each colour picture, wherein, S is norm calculation summed result, and the corresponding face location of face characteristic tensor data is wide on the basis of m
Degrees of data, corresponding n benchmark face characteristic tensor data are face position height data, and r=3 represents reddish yellow blue three-color；
Track identification module 6, for each corresponding colour picture of the same face to be restored into described to be identified regard
In frequency, the positional information that the face occurs in the video to be identified is recorded, obtain the face and to be identified regarded described
Personage track in frequency.
Accordingly, the module 2 of setting up includes：
First extraction unit 21, for the colour picture comprising facial image to be converted into corresponding black and white picture,
And extract the position data of face in the black and white picture, wherein, the position data include initial coordinate, width data and
Altitude information；
Second extraction unit 22, for the position data according to face in the black and white picture, gets the artwork master
The face location data of colour picture corresponding to piece；
First sets up unit 23, for the width data m and height in the face location data according to the colour picture
Data n, the rank tensor A [1 of face three set up in the colour picture：M, 1：N, 1：R], wherein, r=3 represents reddish yellow blue three kinds
Color；
Acquiring unit 24, for setting translation tensor data m/3 according to the width data m, obtains translation tensor B [i：
M*2/3+i, i：N-m*2/3+i*2,1：R], wherein, i represents the coefficient of relationship of the tensor that translates out and former tensor, value is 1≤
i≦m/3；
Second sets up unit 25, for carrying out translationization processing to the rank tensor of face three according to the translation tensor,
Obtain the face tetradic MODEL C [1:m/3,1:m*2/3,1:N-m*2/3,1：r].
Corresponding, first computing module 3 includes：
Function sets up unit 31, the series C for setting up the face tetradic modelkCalculating function, wherein,
Ck=0.75*Ck-1+0.25*Bk, in formula, 1≤k≤m/3；
Tensor computation unit 32 is translated, the translation tensor B is obtained for according to the translation tensor B, calculatingkValue；
Data Computation Unit 33, for calculating the face characteristic tensor data C, wherein, C=Ck, in formula, k=m/3.
In embodiments of the invention three, all videos of facial image will be included in video to be identified in conversion module
Fragment is converted to multiple colour pictures for including facial image, then by setting up face location of the module in colour picture
Data establish face tensor quadravalence model, what the face location data in thus according to colour picture were set up, so
The face tensor quadravalence model contains more face informations, and the process for carrying out recognition of face will be more accurate；In the first meter
Calculate and tensor computation is carried out in module, obtain multiple face characteristics corresponding to each described colour picture comprising facial image
Measure data；By the multiple face characteristic tensor data and benchmark face characteristic tensor data in the second computing module, carry out
Norm calculation is summed, and then sets sampling threshold by sort module, corresponding face in the colour picture is classified,
Obtain each corresponding colour picture of same face；Finally by each corresponding coloured silk of the same face in the identification module of track
Chromatic graph piece is restored in the video to be identified, is recorded the positional information that the face occurs in the video to be identified, is obtained
Obtain personage track of the face in the video to be identified.Due to not needed based on tensor model in face recognition process
Training parameter can be effectively saved cost with regard to that can recognize the different photos of same person；Simultaneously cause identification video in someone
Track be do not need this person known face, therefore methods described using scope it is extremely wide.In general, present invention realization is passed through
It is applied widely and reduce cost and the purpose of operand.
The embodiment of each in this specification is described by the way of progressive, and what each embodiment was stressed is and other
Between the difference of embodiment, each embodiment identical similar portion mutually referring to.For device disclosed in embodiment
For, because it is corresponded to the method disclosed in Example, so description is fairly simple, related part is said referring to method part
It is bright.
The foregoing description of the disclosed embodiments, enables professional and technical personnel in the field to realize or using the present invention.
A variety of modifications to these embodiments will be apparent for those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, it is of the invention
The embodiments shown herein is not intended to be limited to, and is to fit to and principles disclosed herein and features of novelty phase one
The most wide scope caused.
Claims (6)
1. a kind of method for recognizing personage track in video, it is characterised in that this method includes：
All video segments comprising facial image in video to be identified are converted into multiple colour pictures for including facial image；
The face location data in the colour picture comprising facial image are obtained, institute is set up according to the face location data
State the face tetradic model of the colour picture comprising facial image；
Tensor computation is carried out in the face tetradic model, the colour picture institute that each includes facial image is obtained
Corresponding multiple face characteristic tensor data；
By the multiple face characteristic tensor data and benchmark face characteristic tensor data, norm calculation summation is carried out, wherein, institute
It is some the face tensor property number chosen in the multiple face characteristic tensor data to state benchmark face characteristic tensor data
According to；
Sampling threshold σ, foundation are set for the sample size of colour picture according to the Video Quality Metric to be identified
Corresponding face in the colour picture is classified, each corresponding colour picture of same face is obtained, wherein, S is model
Number calculates the corresponding face location width data of face characteristic tensor data on the basis of summed result, m, n benchmark faces feature
It is face position height data to measure data corresponding, and r=3 represents reddish yellow blue three-color；
Each corresponding colour picture of the same face is restored in the video to be identified, the face is recorded described
The positional information occurred in video to be identified, obtains personage track of the face in the video to be identified.
2. according to the method described in claim 1, it is characterised in that in the colour picture comprising facial image described in the acquisition
Face location data, the face quadravalence that the colour picture comprising facial image is set up according to the face location data opens
Model is measured, including：
The colour picture comprising facial image is converted into corresponding black and white picture, and extracts people in the black and white picture
The position data of face, wherein, the position data includes initial coordinate, width data and altitude information；
According to the position data of face in the black and white picture, the face of the colour picture corresponding to the black and white picture is got
Position data；
Width data m and altitude information n in the face location data of the colour picture, set up in the colour picture
The rank tensor A [1 of face three：M, 1：N, 1：R], wherein, r=3 represents reddish yellow blue three-color；
Translation tensor data m/3 is set according to the width data m, translation tensor B [i are obtained：M*2/3+i, i：n-m*2/3+i*
2,1：R], wherein, i represents the coefficient of relationship of the tensor and former tensor translated out, and value is 1≤i≤m/3；
Translationization processing is carried out to the rank tensor of face three according to the translation tensor, the face tetradic model is obtained
C[1:m/3,1:m*2/3,1:N-m*2/3,1：r].
3. according to the method described in claim 1, it is characterised in that described to carry out tensor in the face tetradic model
Calculate, obtain multiple face characteristic tensor data corresponding to each described colour picture comprising facial image, including：
Set up the series C of the face tetradic modelkCalculating function, wherein, Ck=0.75*Ck-1+0.25*Bk, in formula,
1≦k≦m/3；
According to the translation tensor B, calculate and obtain the translation tensor BkValue；
The face characteristic tensor data C is calculated, wherein, C=Ck, in formula, k=m/3.
4. a kind of device for recognizing personage track in video, it is characterised in that the device includes：
Conversion module, for whether video to be identified to be included into facial image according to video segment, is converted into multiple comprising face
The colour picture of image；
Module is set up, for obtaining the face location data in the colour picture comprising facial image, according to the face
Position data sets up the face tetradic model of the colour picture comprising facial image；
First computing module, for carrying out tensor computation in the face tetradic model, obtains described each and includes people
Multiple face characteristic tensor data corresponding to the colour picture of face image；
Second computing module, for by the multiple face characteristic tensor data and benchmark face characteristic tensor data, carrying out model
Number calculates summation, wherein, the benchmark face characteristic tensor data are chosen in the multiple face characteristic tensor data
Some face tensor property data；
Sort module, for setting sampling threshold σ, foundation according to the Video Quality Metric to be identified for the sample size of colour pictureCorresponding face in the colour picture is classified, each corresponding coloured silk of same face is obtained
Chromatic graph piece, wherein, S is the corresponding face location width number of face characteristic tensor data on the basis of norm calculation summed result, m
According to corresponding n benchmark face characteristic tensor data are face position height data, and r=3 represents reddish yellow blue three-color；
Track identification module, for each corresponding colour picture of the same face to be restored in the video to be identified,
The positional information that the face occurs in the video to be identified is recorded, the face is obtained in the video to be identified
Personage track.
5. device according to claim 4, it is characterised in that the module of setting up includes：
First extraction unit, for the colour picture comprising facial image to be converted into corresponding black and white picture, and is extracted
Go out the position data of face in the black and white picture, wherein, the position data includes initial coordinate, width data and the high number of degrees
According to；
Second extraction unit, for the position data according to face in the black and white picture, gets the black and white picture institute right
The face location data for the colour picture answered；
First sets up unit, for the width data m and altitude information n in the face location data according to the colour picture,
The rank tensor A [1 of face three set up in the colour picture：M, 1：N, 1：R], wherein, r=3 represents reddish yellow blue three-color；
Acquiring unit, for setting translation tensor data m/3 according to the width data m, obtains translation tensor B [i：m*2/3+
I, i：N-m*2/3+i*2,1：R], wherein, i represents the coefficient of relationship of the tensor and former tensor translated out, and value is 1≤i≤m/
3；
Second sets up unit, for carrying out translationization processing to the rank tensor of face three according to the translation tensor, obtains institute
State face tetradic MODEL C [1:m/3,1:m*2/3,1:N-m*2/3,1：r].
6. device according to claim 4, it is characterised in that first computing module includes：
Function sets up unit, the series C for setting up the face tetradic modelkCalculating function, wherein,
Ck=0.75*Ck-1+0.25*Bk, in formula, 1≤k≤m/3；
Tensor computation unit is translated, the translation tensor B is obtained for according to the translation tensor B, calculatingkValue；
Data Computation Unit, for calculating the face characteristic tensor data C, wherein, C=Ck, in formula, k=m/3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710293791.5A CN107122745B (en) | 2017-04-28 | 2017-04-28 | Method and device for identifying person track in video |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710293791.5A CN107122745B (en) | 2017-04-28 | 2017-04-28 | Method and device for identifying person track in video |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107122745A true CN107122745A (en) | 2017-09-01 |
| CN107122745B CN107122745B (en) | 2020-10-20 |
Family
ID=59725486
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710293791.5A Active CN107122745B (en) | 2017-04-28 | 2017-04-28 | Method and device for identifying person track in video |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107122745B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108615007A (en) * | 2018-04-23 | 2018-10-02 | 深圳大学 | Three-dimensional face identification method, device and the storage medium of feature based tensor |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080310687A1 (en) * | 2007-06-15 | 2008-12-18 | Microsoft Corporation | Face Recognition Using Discriminatively Trained Orthogonal Tensor Projections |
| CN101436250A (en) * | 2008-11-19 | 2009-05-20 | 西安电子科技大学 | Multi-view angle human face recognizing method based on non-linear tensor resolution and view angle manifold |
| CN104933428A (en) * | 2015-07-23 | 2015-09-23 | 苏州大学 | Human face recognition method and device based on tensor description |
| CN105956603A (en) * | 2016-04-15 | 2016-09-21 | 天津大学 | Video sequence classifying method based on tensor time domain association model |
-
2017
- 2017-04-28 CN CN201710293791.5A patent/CN107122745B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080310687A1 (en) * | 2007-06-15 | 2008-12-18 | Microsoft Corporation | Face Recognition Using Discriminatively Trained Orthogonal Tensor Projections |
| CN101436250A (en) * | 2008-11-19 | 2009-05-20 | 西安电子科技大学 | Multi-view angle human face recognizing method based on non-linear tensor resolution and view angle manifold |
| CN104933428A (en) * | 2015-07-23 | 2015-09-23 | 苏州大学 | Human face recognition method and device based on tensor description |
| CN105956603A (en) * | 2016-04-15 | 2016-09-21 | 天津大学 | Video sequence classifying method based on tensor time domain association model |
Non-Patent Citations (4)
| Title |
|---|
| SUJING WANG ET AL: "Face recognition and micro-expression recognition based on discriminant tensor subspace analysis plus extreme learning machine", 《NEURAL PROCESS LETT》 * |
| ZIYU GUAN ET AL: "Efficient face recognition using tensor subspace regression", 《NEUROCOMPUTING》 * |
| 史加荣等: "张量补全算法及其在人脸识别中的应用", 《模式识别与人工智能》 * |
| 李勇周等: "张量局部判别投影的人脸识别", 《电子学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108615007A (en) * | 2018-04-23 | 2018-10-02 | 深圳大学 | Three-dimensional face identification method, device and the storage medium of feature based tensor |
| CN108615007B (en) * | 2018-04-23 | 2019-07-19 | 深圳大学 | Three-dimensional face identification method, device and storage medium based on characteristic tensor |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107122745B (en) | 2020-10-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105894047B (en) | A kind of face classification system based on three-dimensional data | |
| CN103824059A (en) | Facial expression recognition method based on video image sequence | |
| CN111027378B (en) | Pedestrian re-identification method, device, terminal and storage medium | |
| CN109670429A (en) | A kind of the monitor video multiple target method for detecting human face and system of Case-based Reasoning segmentation | |
| CN110084156A (en) | A kind of gait feature abstracting method and pedestrian's personal identification method based on gait feature | |
| CN105678231A (en) | Pedestrian image detection method based on sparse coding and neural network | |
| CN106845397A (en) | A kind of confirming face method based on measuring similarity | |
| CN105447532A (en) | Identity authentication method and device | |
| CN104766324B (en) | Based on the colored blood cell image leucocyte dividing method of super-pixel and abnormality detection | |
| CN111126240B (en) | Three-channel feature fusion face recognition method | |
| CN109902590A (en) | Pedestrian's recognition methods again of depth multiple view characteristic distance study | |
| CN104318219A (en) | Face recognition method based on combination of local features and global features | |
| CN104484658A (en) | Face gender recognition method and device based on multi-channel convolution neural network | |
| Hajizadeh et al. | Classification of age groups from facial image using histograms of oriented gradients | |
| CN103839042B (en) | Face identification method and face identification system | |
| CN102902986A (en) | Automatic gender identification system and method | |
| CN102637251A (en) | Face recognition method based on reference features | |
| CN108629336A (en) | Face value calculating method based on human face characteristic point identification | |
| CN111950525B (en) | Fine-grained image classification method based on destructive reconstruction learning and GoogLeNet | |
| CN105117708A (en) | Facial expression recognition method and apparatus | |
| CN106845328A (en) | A kind of Intelligent human-face recognition methods and system based on dual camera | |
| CN106096517A (en) | A kind of face identification method based on low-rank matrix Yu eigenface | |
| CN106920256A (en) | A kind of effective missing child searching system | |
| CN104376611A (en) | Method and device for attendance of persons descending well on basis of face recognition | |
| CN108875907A (en) | A kind of fingerprint identification method and device based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |