CN107292228A - A kind of method for accelerating face recognition search speed - Google Patents
A kind of method for accelerating face recognition search speed Download PDFInfo
- Publication number
- CN107292228A CN107292228A CN201710313903.9A CN201710313903A CN107292228A CN 107292228 A CN107292228 A CN 107292228A CN 201710313903 A CN201710313903 A CN 201710313903A CN 107292228 A CN107292228 A CN 107292228A
- Authority
- CN
- China
- Prior art keywords
- face
- concordance list
- period
- database
- probability
- 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.)
- Pending
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
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/164—Detection; Localisation; Normalisation using holistic features
-
- 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
- G06V40/169—Holistic features and representations, i.e. based on the facial image taken as a whole
-
- 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
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of method for accelerating face recognition search speed, face database including multiple human face datas that are stored with and the face concordance list for determining the human face data sorted order in face database, human face data sorted order on the face concordance list is arranged according to the identification number of times of human face data, recognize that sorted order of the more human face datas of number of times on face concordance list is more forward, it is further comprising the steps of, gather facial image, current face's view data is contrasted with the human face data in face database successively by the sorted order on face concordance list, until finding the human face data of matching, then the number of times being identified according to human face data adjusts the sorted order on face concordance list.Successful probability can be contrasted with the face contrast in concordance list in preceding contrast several times to greatly promote, greatly reduce the generation of useless contrast, be effectively reduced the burden of system, improve the reaction speed of system the invention enables current face.
Description
Technical field
The present invention relates to a kind of face recognition search method, particularly a kind of method for accelerating face recognition search speed.
Background technology
Recognition of face, refers in particular to carry out the computer technology of identity discriminating using com-parison and analysis face visual signature information.People
Face identification is to gather image or video flowing containing face, and automatic detect and track in the picture with video camera or camera
Face, and then a series of correlation techniques of face are carried out to the face that detects, generally also referred to as Identification of Images, face recognition.
But current recognition of face is typically to contrast the face in the face and database on image one by one, due in database
Face be random distribution, carry out face contrast when, it is often necessary to that of contrast can just be found many times by being contrasted
Individual face, has carried out many useless contrasts, has aggravated the burden of system, reduced the reaction speed of system.
The content of the invention
To solve the above problems, it is an object of the invention to provide a kind of quickening face knowledge that can reduce useless contrast
The method of other search speed.
The present invention solves the technical scheme that is used of its problem：A kind of method for accelerating face recognition search speed, bag
Include the face database for multiple human face datas that are stored with and the people for determining the human face data sorted order in face database
Human face data sorted order on face concordance list, the face concordance list is arranged according to the identification number of times of human face data, is known
Sorted order of the more human face datas of other number of times on face concordance list is more forward, further comprising the steps of, gathers face
Image, by current face's view data by the sorted order on face concordance list successively with the human face data in face database
Contrasted, until finding the human face data of matching, the number of times being then identified according to human face data is adjusted on face concordance list
Sorted order.
Specifically, IMAQ is carried out by camera when gathering image, picture format is RGB or YUV.
Specifically, feature extraction is carried out to the face in facial image after collection facial image, to the face in image
Whether there is face in first detection image when carrying out feature extraction, when face has been detected whether, the colour of skin is carried out first
Model Face datection, complexion model Face datection refers to draw face complexion in Y-Cr-Cb color spaces by statistical learning
The complexion model of distribution probability, then judges whether the point on image belongs to human face region, complexion model by complexion model
Viola-Jones Face datections are carried out after Face datection again.
Specifically, feature extraction is carried out to the face in image, using LBP characteristic vector pickups method or depth
Practise characteristic vector pickup method.
Specifically, LBP characteristic vector pickups method comprises the following steps is divided into 16*16's by detection window first
Zonule, gray scale is carried out using a pixel in each zonule as central pixel point eight pixels adjacent thereto
Compare, if the gray value of the pixel is more than the gray value of central pixel point, the position of the pixel is denoted as one, such as
Really the gray value of the pixel is less than the gray value of central pixel point, then the position of the pixel is denoted as in zero, 3*3 fields
Eight points compare generation eight bit, that is, obtain the LBP values of the window center pixel, then calculate each zonule
Histogram, i.e., then the frequency that each decimal system LBP values occur be normalized to the histogram, will finally obtain
The statistic histogram of each zonule be connected to become a characteristic vector, that is, whole accompanying drawing LBP texture feature vectors.
Specifically, deep learning characteristic vector pickup method refers to generate ten regions on facial image, is respectively
Region centered on five global areas and five characteristic points, to each three kinds of yardsticks of Area generation, ultimately forms 30
Patch, is separately input in convolutional neural networks generate 30 160 dimensional features, face finally takes 160* using 30 patch
30=4800 latitude features.
Specifically, multiple periods were divided into by one day, each period corresponds to a face concordance list, extracts
Period after face characteristic according to where current face's recognition time loads the face concordance list of corresponding period.
Specifically, when current face is contrasted with the face in the face concordance list of loading, by face
Order on concordance list carries out face alignment retrieval, and comparison method is calculating face characteristic value to be measured and people in face database
The Euclidean distance of face characteristic value, Euclidean distance calculation formula is εK=| | Ω-ΩK||2, the wherein Ω representatives face to be compared,
ΩKSome face in face database is represented, is both represented by the weight of characteristic value, above-mentioned formula is pair
Both seek Euclidean distance, illustrate that k-th of face in the face and face database to be differentiated is same when distance is less than threshold value
People's.When current face is contrasted successfully with the face in the face concordance list of loading, the successful face rope of storage identification
Draw value and storage identification successful time point.
Current face contrasts retrieval when successfully to face concordance list with the face in the face concordance list of loading
Order is adjusted, and adjusts the specific method of face concordance list sorted order to load the identification of nearest 60 days from record
Success is recorded, and will be divided into 24 periods daily in units of hour, will recognize that the face in successfully record is incorporated into two
In 14 periods, a face in face database is chosen, its face index value is A, is successfully recorded according to recognizing, is united
The frequency n that meter A occurs in period a, the then probability that face A occurs in period a is that (A, a)=n/60 repeat people to M
The calculation procedure for the probability that face A occurs in period a can obtain probability Ms (A, 0) of the face A in 24 periods, M (A,
1),M(A, 2)...M(A,23)。
Further according to face A, all people's face carries out probability meter in the method for calculating probability of 24 periods is to database
Calculate, the probability that all faces occur within each period in face database can be obtained：
M(1,0),M(1,1),M(1,2)...M(1,23)
M(2,0),M(2,1),M(2,2)...M(2,23)
M(3,0),M(3,1),M(3,2)...M(3,23)
......
M(N,0),M(N,1),M(N,2)...M(N,23)
The new face of arrangement generation is carried out to the face in each period finally according to the probability M calculated to index
Table, face is higher in the probability M of the period, then the face is more forward in the arrangement of the face concordance list of the period, i.e., should
Face is more forward in the index value of the face concordance list of the period, and face is lower in the probability M of the period, then the face
The period face concordance list arrangement more rearward, i.e., the face is got in the index value of the face concordance list of the period
Rearward.
The beneficial effects of the invention are as follows：The present invention is a kind of method for accelerating face recognition search speed, of the invention by people
Human face data sorted order on face concordance list is arranged according to the identification number of times of human face data, the more people of identification number of times
Sorted order of the face data on face concordance list is more forward, gathers facial image, and current face's view data is pressed into face rope
Draw the sorted order on table to be contrasted with the human face data in face database successively, until finding the face number of matching
According to the number of times being then identified according to human face data adjusts the sorted order on face concordance list so that current face and index
Face contrast in table can contrast successful probability in preceding contrast several times and greatly promote, and greatly reduce the hair of useless contrast
It is raw, the burden of system is effectively reduced, the human face data improved in the reaction speed of system, database is huger, the present invention
Advantage it is more obvious, meet the development trend of recognition of face.Of the invention recognize every time successfully can all be recorded and be updated
Face concordance list, it is ensured that the real-time optimization of face concordance list, it is not necessary to arrange professional to carry out the progress of face concordance list on time
Optimization, not only reduces the cost safeguarded and runed, and be easy to routine use.
Brief description of the drawings
The invention will be further described with example below in conjunction with the accompanying drawings.
Fig. 1 is the process chart of the present invention.
Embodiment
It is a kind of accelerate face recognition search speed method, including multiple human face datas that are stored with face database and
Face number on face concordance list for determining the human face data sorted order in face database, the face concordance list
Arranged according to sorted order according to the identification number of times of human face data, the more human face datas of identification number of times are in face concordance list
On sorted order it is more forward, it is further comprising the steps of, gather facial image, by current face's view data by face index
Sorted order on table is contrasted with the human face data in face database successively, until finding the human face data of matching, so
The number of times being identified afterwards according to human face data adjusts the sorted order on face concordance list.Pass through shooting when collection image
Head carries out IMAQ, and picture format is RGB or YUV.Collect facial image and feature is carried out to the face in facial image later
Extract, whether have face in first detection image when carrying out feature extraction to the face in image, detecting whether someone
When face, complexion model Face datection is carried out first, and complexion model Face datection refers to draw face skin by statistical learning
The complexion model of color distribution probability in Y-Cr-Cb color spaces, then judges whether the point on image belongs to by complexion model
Carry out Viola-Jones Face datections again after human face region, complexion model Face datection.The present invention is by face concordance list
On human face data sorted order arranged according to the identification number of times of human face data, the more human face data of identification number of times exists
Sorted order on face concordance list is more forward, gathers facial image, by current face's view data by face concordance list
Sorted order is contrasted with the human face data in face database successively, until finding the human face data of matching, Ran Hougen
The sorted order on number of times adjustment face concordance list being identified according to human face data so that current face and the people in concordance list
Face contrast can contrast successful probability in preceding contrast several times and greatly promote, and greatly reduce the generation of useless contrast, effectively subtract
The light burden of system, the human face data improved in the reaction speed of system, database is huger, and advantage of the invention is got over
Substantially, the development trend of recognition of face is met.Of the invention recognize every time successfully can all be recorded and be updated face index
Table, it is ensured that the real-time optimization of face concordance list, it is not necessary to arrange professional to carry out face concordance list on time and optimize, no
But the cost safeguarded and runed is reduced, and is easy to routine use.
It is preferred that, the present invention carries out feature extraction to the face in image, using LBP characteristic vector pickups method or depth
Spend the vectorial extracting method of learning characteristic.LBP characteristic vector pickup methods, which comprise the following steps, is first divided into detection window
16*16 zonule, is clicked through a pixel in each zonule as central pixel point eight pixels adjacent thereto
Row gray scale compares, if the gray value of the pixel is more than the gray value of central pixel point, the position of the pixel is marked
For one, if the gray value of the pixel is less than the gray value of central pixel point, the position of the pixel is denoted as zero, 3*3
Eight points in field compare generation eight bit, that is, obtain the LBP values of the window center pixel, then calculate every
The histogram of individual zonule, i.e., the frequency that each decimal system LBP values occur, then the histogram is normalized,
The statistic histogram of obtained each zonule is finally connected to become a characteristic vector, that is, whole accompanying drawing LBP textures
Characteristic vector.Deep learning characteristic vector pickup method refers to generate ten regions, respectively five overall situations on facial image
Region centered on region and five characteristic points, to each three kinds of yardsticks of Area generation, ultimately forms 30 patch, uses 30
Individual patch is separately input in convolutional neural networks generate 30 160 dimensional features, and face finally takes 160*30=4800 latitudes special
Levy.
It is preferred that, the present invention was divided into multiple periods by one day, and each period corresponds to a face concordance list,
Extract the face concordance list of period loading corresponding period of the face characteristic later according to where current face's recognition time.
When current face is contrasted with the face in the face concordance list of loading, enter by the order on face concordance list
Row face alignment is retrieved, comparison method for calculate face characteristic value to be measured with face database face characteristic value it is European away from
From Euclidean distance calculation formula is εK=| | Ω-ΩK||2, the wherein Ω representatives face to be compared, ΩKRepresent face database
In some face, both represented by the weight of characteristic value, above-mentioned formula is to seek Euclidean distance to both, when
Distance illustrates that k-th of face in the face and face database to be differentiated is same person when being less than threshold value.Current face with
When face in the face concordance list of loading is contrasted successfully, the successful face index value of storage identification and storage are identified as
The time point of work(.Current face is contrasted when successfully to face concordance list with the face in the face concordance list of loading
Sorted order is adjusted, and the specific method of adjustment face concordance list sorted order is to be loaded from record nearest 60 days
Recognize and successfully record, will be divided into 24 periods daily in units of hour, will recognize that the face in successfully record is drawn
It is grouped into 24 periods, chooses a face in face database, its face index value is A, is successfully remembered according to recognizing
Record, the frequency n that statistics A occurs in period a, the then probability that face A occurs in period a is M (A, a)=n/60, weight
The calculation procedure for the probability that multiple face A occurs in period a can obtain face A 24 periods probability M (A,
0),M(A,1),M(A,2)...M(A,23)。
Further according to face A, all people's face carries out probability meter in the method for calculating probability of 24 periods is to database
Calculate, the probability that all faces occur within each period in face database can be obtained：
M(1,0),M(1,1),M(1,2)...M(1,23)
M(2,0),M(2,1),M(2,2)...M(2,23)
M(3,0),M(3,1),M(3,2)...M(3,23)
......
M(N,0),M(N,1),M(N,2)...M(N,23)
The new face of arrangement generation is carried out to the face in each period finally according to the probability M calculated to index
Table, face is higher in the probability M of the period, then the face is more forward in the arrangement of the face concordance list of the period, i.e., should
Face is more forward in the index value of the face concordance list of the period, and face is lower in the probability M of the period, then the face
The period face concordance list arrangement more rearward, i.e., the face is got in the index value of the face concordance list of the period
Rearward.
Fig. 1 is the process chart of the present invention, as shown in figure 1, a kind of method for accelerating face recognition search speed, bag
Include following steps：
S1, facial image is gathered by camera；
Whether there is face in S2, detection facial image, be to go to step S3, otherwise go to step S 1；
S3, face characteristic extraction is carried out to the face in facial image；
S4, face concordance list is loaded according to the face characteristic extracted；
S5, face inspection is carried out by current face's characteristic according to the sorted order on face concordance list in database
Rope；
S6, if identical face is retrieved in face database, is, goes to step S7, otherwise goes to step S1；
S7, preserves the successful face retrieval of identification and records and update face concordance list, go to step S1.
Update face concordance list and refer to that loading nearest recognition of face successfully records, successful face will be recognized according to knowledge
The successful face of identification in period, each period Cheng Gong be divided the time new according to the probability progress arrangement form of appearance
Face concordance list, face occur probability it is higher, putting in order in face concordance list is more forward, face occur it is general
Rate is lower, putting in order more rearward in face concordance list.Although the present invention is also according to people when face is contrasted
Face concordance list is contrasted one by one, but the larger face of probability of occurrence in face database can be entered to row major contrast, greatly
The generation of useless contrast is reduced greatly, the burden of system is effectively reduced, and is improved in the reaction speed of system, database
Human face data is huger, and advantage of the invention is more obvious, meets the development trend of recognition of face.The present invention is recognized successfully every time
It will be recorded and update face concordance list, it is ensured that the real-time optimization of face concordance list, it is not necessary to arrange professional
Face concordance list is carried out on time to optimize, and not only reduces the cost safeguarded and runed, and be easy to routine use.
It is preferred that, step S1 of the invention carries out IMAQ, picture format when image is gathered by camera
It is RGB or YUV.The step S2 of the present invention carries out complexion model Face datection first when face has been detected whether,
Complexion model Face datection refers to the skin that face complexion distribution probability in Y-Cr-Cb color spaces is drawn by statistical learning
Color model, then judges whether the point on image belongs to after human face region, complexion model Face datection by complexion model
Viola-Jones Face datections are carried out again.The step S3 of the present invention is when face characteristic is extracted, using LBP characteristic vectors
Extracting method or deep learning characteristic vector pickup method.
It is preferred that, LBP characteristic vector pickup methods comprise the following steps is divided into the small of 16*16 by detection window first
Region, gray scale ratio is carried out using a pixel in each zonule as central pixel point eight pixels adjacent thereto
Compared with, if the gray value of the pixel is more than the gray value of central pixel point, the position of the pixel is denoted as one, if
The gray value of the pixel is less than the gray value of central pixel point, then the position of the pixel is denoted as in zero, 3*3 fields
Eight points compare generation eight bit, that is, obtain the L BP values of the window center pixel, then calculate each zonule
Histogram, i.e., then the frequency that each decimal system LBP values occur be normalized to the histogram, will finally obtain
The statistic histogram of each zonule be connected to become a characteristic vector, that is, whole accompanying drawing LBP texture feature vectors.
Deep learning characteristic vector pickup method refers to generate ten regions, respectively five global areas and five on facial image
Region centered on individual characteristic point, to each three kinds of yardsticks of Area generation, ultimately forms 30 patch, uses 30 patch
It is separately input in convolutional neural networks generate 30 160 dimensional features, face finally takes 160*30=4800 latitude features.
It is preferred that, step S4 of the invention selects face rope when face concordance list is loaded according to current slot
Draw table.The step S5 of the present invention carries out face alignment retrieval, comparison method when face retrieval by face concordance list order
To calculate the Euclidean distance of face characteristic value to be measured and face characteristic value in face database, Euclidean distance calculation formula is εK
=| | Ω-ΩK||2, the wherein Ω representatives face to be compared, ΩKSome face in face database is represented, both
Represented by the weight of characteristic value, above-mentioned formula is to seek Euclidean distance to both, when distance is less than threshold value, explanation will be sentenced
K-th of face in other face and face database is same person.
It is preferred that, keeping records of the invention refers to that the successful face index value of storage identification and storage identification are successful
Time point.Identification of the specific method of the renewal face concordance list of the present invention to load nearest 60 days from record is successfully remembered
Record, will be divided into 24 periods daily in units of hour, will recognize that the face in successfully record is incorporated into 24
In period, a face in face database is chosen, its face index value is A, is successfully recorded according to recognizing, statistics A exists
The frequency n occurred in period a, the then probability that face A occurs in period a is that (A, a)=n/60 repeat face A and existed M
The calculation procedure of the probability occurred in period a can obtain probability Ms (A, 0) of the face A in 24 periods, M (A, 1), M
(A,2)...M(A, 23).Further according to face A, all people's face enters in the method for calculating probability of 24 periods is to database
Row probability calculation, can obtain the probability that all faces occur within each period in face database：
M(1,0),M(1,1),M(1,2)...M(1,23)
M(2,0),M(2,1),M(2,2)...M(2,23)
M(3,0),M(3,1),M(3,2)...M(3,23)
......
M(N,0),M(N,1),M(N,2)...M(N,23)
The new face of arrangement generation is carried out to the face in each period finally according to the probability M calculated to index
Table, face is higher in the probability M of the period, then the face is more forward in the arrangement of the face concordance list of the period, i.e., should
Face is more forward in the index value of the face concordance list of the period, and face is lower in the probability M of the period, then the face
The period face concordance list arrangement more rearward, i.e., the face is got in the index value of the face concordance list of the period
Rearward.
It is described above, simply presently preferred embodiments of the present invention, the invention is not limited in above-mentioned embodiment, only
Want it to reach the technique effect of the present invention with identical means, should all belong to protection scope of the present invention.
Claims (10)
1. a kind of method for accelerating face recognition search speed, it is characterised in that：Include the face for multiple human face datas that are stored with
On database and the face concordance list for determining the human face data sorted order in face database, the face concordance list
Human face data sorted order is arranged according to the identification number of times of human face data, and the more human face datas of identification number of times are in face rope
The sorted order drawn on table is more forward, further comprising the steps of, gathers facial image, and current face's view data is pressed into face rope
Draw the sorted order on table to be contrasted with the human face data in face database successively, until the human face data of matching is found,
Then the number of times being identified according to human face data adjusts the sorted order on face concordance list.
2. a kind of method for accelerating face recognition search speed according to claim 1, it is characterised in that：Gather image
When by camera carry out IMAQ, picture format is RGB or YUV.
3. a kind of method for accelerating face recognition search speed according to claim 1, it is characterised in that：Gather face figure
As carrying out feature extraction to the face in facial image later, figure is first detected when carrying out feature extraction to the face in image
Whether there is face as in, when face has been detected whether, complexion model Face datection, the inspection of complexion model face are carried out first
The complexion model for referring to draw face complexion distribution probability in Y-Cr-Cb color spaces by statistical learning is surveyed, is then passed through
Complexion model judges the point on image carries out Viola-Jones again after whether belonging to human face region, complexion model Face datection
Face datection.
4. a kind of method for accelerating face recognition search speed according to claim 3, it is characterised in that：To in image
Face carries out feature extraction, using LBP characteristic vector pickups method or deep learning characteristic vector pickup method.
5. a kind of method for accelerating face recognition search speed according to claim 4, it is characterised in that：LBP features to
Amount extracting method comprises the following steps the zonule that detection window is divided into 16*16 first, by one in each zonule
Pixel carries out gray scale comparison as central pixel point eight pixels adjacent thereto, if the gray value of the pixel is more than
The gray value of central pixel point, then the position of the pixel is denoted as one, if the gray value of the pixel is less than center pixel
The gray value of point, then the position of the pixel be denoted as zero, so produce eight points in 3*3 fields and compare and produce eight two and enter
Number processed, that is, obtain the LBP values of the window center pixel, then calculate the histogram of each zonule, i.e., each decimal system LBP
It is worth the frequency occurred, then the histogram is normalized, finally by the statistic histogram of obtained each zonule
Be connected to become a characteristic vector, that is, whole accompanying drawing LBP texture feature vectors.
6. a kind of method for accelerating face recognition search speed according to claim 4, it is characterised in that：Deep learning is special
Levy vectorial extracting method to refer to generate ten regions on facial image, during respectively five global areas and five characteristic points are
The region of the heart, to each three kinds of yardsticks of Area generation, ultimately forms 30 patch, convolution is separately input to using 30 patch
30 160 dimensional features are generated in neutral net, face finally takes 160*30=4800 latitude features.
7. a kind of method for accelerating face recognition search speed according to claim 4, it is characterised in that：One day is divided
For multiple periods, each period corresponds to a face concordance list, extracts face characteristic and is recognized later according to current face
Period where time loads the face concordance list of corresponding period.
8. a kind of method for accelerating face recognition search speed according to claim 7, it is characterised in that：By current people
When face is contrasted with the face in the face concordance list of loading, face alignment inspection is carried out by the order on face concordance list
Rope, comparison method is calculating face characteristic value to be measured and the Euclidean distance of face characteristic value in face database, Euclidean distance meter
Calculation formula is εK=| | Ω-ΩK||2, the wherein Ω representatives face to be compared, ΩKSome face in face database is represented,
Both represented by the weight of characteristic value, above-mentioned formula is to seek Euclidean distance to both, when distance is less than threshold value
Illustrate that k-th of face in the face and face database to be differentiated is same person.
9. a kind of method for accelerating face recognition search speed according to claim 8, it is characterised in that：Current face
When contrast successfully with the face in the face concordance list of loading, the successful face index value of storage identification and storage are identified as
The time point of work(.
10. a kind of method for accelerating face recognition search speed according to claim 9, it is characterised in that：Current people
Face contrasts with the face in the face concordance list loaded and the sorted order of face concordance list is adjusted when successfully, adjusts
Identification of the specific method of face concordance list sorted order to load nearest 60 days from record is successfully recorded, using hour to be single
Position will be divided into 24 periods daily, will recognize that the face in successfully record was incorporated into 24 periods, chooses
A face in face database, its face index value is A, is successfully recorded according to recognizing, statistics A occurs in period a
Frequency n, the then probability that face A occurs in period a is that (A, a)=n/60, it is general that repetition face A occurs M in period a
The calculation procedure of rate can obtain probability of the face A in 24 periods, according to face A 24 periods probability calculation
Method carries out probability calculation to all people's face in database, can obtain in face database all faces within each period
The probability of appearance, the new face concordance list of arrangement generation is carried out to the face in each period according to the probability M calculated,
Face is higher in the probability M of the period, then the face is more forward in the arrangement of the face concordance list of the period, i.e. the face
More forward in the index value of the face concordance list of the period, face is lower in the probability M of the period, then the face is at this
Between section face concordance list arrangement more rearward, i.e., the face the face concordance list of the period index value more rearward.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710313903.9A CN107292228A (en) | 2017-05-05 | 2017-05-05 | A kind of method for accelerating face recognition search speed |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710313903.9A CN107292228A (en) | 2017-05-05 | 2017-05-05 | A kind of method for accelerating face recognition search speed |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107292228A true CN107292228A (en) | 2017-10-24 |
Family
ID=60094553
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710313903.9A Pending CN107292228A (en) | 2017-05-05 | 2017-05-05 | A kind of method for accelerating face recognition search speed |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107292228A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110363180A (en) * | 2019-07-24 | 2019-10-22 | 厦门云上未来人工智能研究院有限公司 | A kind of method and apparatus and equipment that statistics stranger's face repeats |
| CN110580754A (en) * | 2018-06-11 | 2019-12-17 | 杭州海康威视系统技术有限公司 | Face authentication method, device and system |
| CN111898559A (en) * | 2020-08-03 | 2020-11-06 | 南京奥拓电子科技有限公司 | Method and device for improving face recognition speed and electronic equipment |
| CN112215250A (en) * | 2019-07-12 | 2021-01-12 | 杭州海康威视数字技术股份有限公司 | Method and device for improving data characteristic comparison efficiency |
| CN112488078A (en) * | 2020-12-23 | 2021-03-12 | 浙江大华技术股份有限公司 | Face comparison method and device and readable storage medium |
| CN112487222A (en) * | 2020-11-30 | 2021-03-12 | 江苏正赫通信息科技有限公司 | Method for quickly searching and effectively storing similar human faces |
| CN112712569A (en) * | 2020-12-25 | 2021-04-27 | 百果园技术(新加坡)有限公司 | Skin color detection method, device, mobile terminal and storage medium |
| CN113344132A (en) * | 2021-06-30 | 2021-09-03 | 成都商汤科技有限公司 | Identity recognition method, system, device, computer equipment and storage medium |
| CN113343004A (en) * | 2021-06-10 | 2021-09-03 | 浙江大华技术股份有限公司 | Object recognition method and device, storage medium and electronic device |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008107713A1 (en) * | 2007-03-07 | 2008-09-12 | Aurora Computer Services Limited | Controlled high resolution sub-image capture with time domain multiplexed high speed full field of view reference video stream for image based biometric applications |
| CN102265612A (en) * | 2008-12-15 | 2011-11-30 | 坦德伯格电信公司 | Method for speeding up face detection |
| CN102831408A (en) * | 2012-08-29 | 2012-12-19 | 华南理工大学 | Human face recognition method |
| CN103353940A (en) * | 2013-05-15 | 2013-10-16 | 吴玉平 | Identification method and system for dynamically adjusting comparison sequence based on probability of occurrence |
| CN103455790A (en) * | 2013-06-24 | 2013-12-18 | 厦门美图网科技有限公司 | Skin identification method based on skin color model |
| CN105469033A (en) * | 2015-11-13 | 2016-04-06 | 广东欧珀移动通信有限公司 | Fingerprint identification method, fingerprint identification device and terminal equipment |
| CN105488475A (en) * | 2015-11-30 | 2016-04-13 | 西安闻泰电子科技有限公司 | Method for detecting human face in mobile phone |
| CN105824559A (en) * | 2016-02-29 | 2016-08-03 | 维沃移动通信有限公司 | Unintended activation recognizing and treating method and electronic equipment |
-
2017
- 2017-05-05 CN CN201710313903.9A patent/CN107292228A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008107713A1 (en) * | 2007-03-07 | 2008-09-12 | Aurora Computer Services Limited | Controlled high resolution sub-image capture with time domain multiplexed high speed full field of view reference video stream for image based biometric applications |
| CN102265612A (en) * | 2008-12-15 | 2011-11-30 | 坦德伯格电信公司 | Method for speeding up face detection |
| CN102831408A (en) * | 2012-08-29 | 2012-12-19 | 华南理工大学 | Human face recognition method |
| CN103353940A (en) * | 2013-05-15 | 2013-10-16 | 吴玉平 | Identification method and system for dynamically adjusting comparison sequence based on probability of occurrence |
| CN103455790A (en) * | 2013-06-24 | 2013-12-18 | 厦门美图网科技有限公司 | Skin identification method based on skin color model |
| CN105469033A (en) * | 2015-11-13 | 2016-04-06 | 广东欧珀移动通信有限公司 | Fingerprint identification method, fingerprint identification device and terminal equipment |
| CN105488475A (en) * | 2015-11-30 | 2016-04-13 | 西安闻泰电子科技有限公司 | Method for detecting human face in mobile phone |
| CN105824559A (en) * | 2016-02-29 | 2016-08-03 | 维沃移动通信有限公司 | Unintended activation recognizing and treating method and electronic equipment |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110580754A (en) * | 2018-06-11 | 2019-12-17 | 杭州海康威视系统技术有限公司 | Face authentication method, device and system |
| CN112215250A (en) * | 2019-07-12 | 2021-01-12 | 杭州海康威视数字技术股份有限公司 | Method and device for improving data characteristic comparison efficiency |
| CN112215250B (en) * | 2019-07-12 | 2024-03-08 | 杭州海康威视数字技术股份有限公司 | Method and device for improving data feature comparison efficiency |
| CN110363180A (en) * | 2019-07-24 | 2019-10-22 | 厦门云上未来人工智能研究院有限公司 | A kind of method and apparatus and equipment that statistics stranger's face repeats |
| CN111898559A (en) * | 2020-08-03 | 2020-11-06 | 南京奥拓电子科技有限公司 | Method and device for improving face recognition speed and electronic equipment |
| CN111898559B (en) * | 2020-08-03 | 2024-04-26 | 南京奥拓电子科技有限公司 | Method and device for improving face recognition speed and electronic equipment |
| CN112487222B (en) * | 2020-11-30 | 2021-11-30 | 江苏正赫通信息科技有限公司 | Method for quickly searching and effectively storing similar human faces |
| CN112487222A (en) * | 2020-11-30 | 2021-03-12 | 江苏正赫通信息科技有限公司 | Method for quickly searching and effectively storing similar human faces |
| CN112488078A (en) * | 2020-12-23 | 2021-03-12 | 浙江大华技术股份有限公司 | Face comparison method and device and readable storage medium |
| CN112712569A (en) * | 2020-12-25 | 2021-04-27 | 百果园技术(新加坡)有限公司 | Skin color detection method, device, mobile terminal and storage medium |
| CN112712569B (en) * | 2020-12-25 | 2023-12-12 | 百果园技术(新加坡)有限公司 | Skin color detection method and device, mobile terminal and storage medium |
| CN113343004A (en) * | 2021-06-10 | 2021-09-03 | 浙江大华技术股份有限公司 | Object recognition method and device, storage medium and electronic device |
| CN113344132A (en) * | 2021-06-30 | 2021-09-03 | 成都商汤科技有限公司 | Identity recognition method, system, device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107292228A (en) | A kind of method for accelerating face recognition search speed | |
| CN109583342B (en) | Human face living body detection method based on transfer learning | |
| WO2020151489A1 (en) | Living body detection method based on facial recognition, and electronic device and storage medium | |
| CN109522853B (en) | Face datection and searching method towards monitor video | |
| CN109543526B (en) | True and false facial paralysis recognition system based on depth difference characteristics | |
| CN107273796A (en) | A kind of fast face recognition and searching method based on face characteristic | |
| CN110728225B (en) | High-speed face searching method for attendance checking | |
| CN105740758A (en) | Internet video face recognition method based on deep learning | |
| CN108960047B (en) | Face duplication removing method in video monitoring based on depth secondary tree | |
| CN103996046A (en) | Personnel recognition method based on multi-visual-feature fusion | |
| CN107633226A (en) | A kind of human action Tracking Recognition method and system | |
| CN104881671B (en) | A kind of high score remote sensing image Local Feature Extraction based on 2D Gabor | |
| CN106778786A (en) | Apple disease recognition methods based on log-spectral domain laminated gradient direction histogram | |
| CN111209818A (en) | Video individual identification method, system, equipment and readable storage medium | |
| CN111863232B (en) | Remote disease intelligent diagnosis system based on block chain and medical image | |
| Chandy | RGBD analysis for finding the different stages of maturity of fruits in farming | |
| CN107230267A (en) | Intelligence In Baogang Kindergarten based on face recognition algorithms is registered method | |
| CN110070024B (en) | Method and system for identifying skin pressure injury thermal imaging image and mobile phone | |
| CN110008793A (en) | Face identification method, device and equipment | |
| CN106529441B (en) | Depth motion figure Human bodys' response method based on smeared out boundary fragment | |
| CN112150692A (en) | Access control method and system based on artificial intelligence | |
| CN111832405A (en) | Face recognition method based on HOG and depth residual error network | |
| CN109190456A (en) | Pedestrian detection method is overlooked based on the multiple features fusion of converging channels feature and gray level co-occurrence matrixes | |
| CN109508755A (en) | A kind of Psychological Evaluation method based on image cognition | |
| CN113112498A (en) | Grape leaf scab identification method based on fine-grained countermeasure generation network |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20171024 |