CN105590089A - Face identification method and device - Google Patents
Face identification method and device Download PDFInfo
- Publication number
- CN105590089A CN105590089A CN201510696203.3A CN201510696203A CN105590089A CN 105590089 A CN105590089 A CN 105590089A CN 201510696203 A CN201510696203 A CN 201510696203A CN 105590089 A CN105590089 A CN 105590089A
- Authority
- CN
- China
- Prior art keywords
- user
- compression
- vector
- facial image
- characteristic vector
- 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
- G06V40/168—Feature extraction; Face representation
-
- 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
Abstract
The present invention discloses a face identification method. The method comprises the following steps: receiving the face image of a user to be identified, and extracting a characteristic vector from the face image; matching the characteristic vector and at least two pre-stored compression vectors, and calculating the similarity of the characteristic vector and each compression vector, wherein the compression vector is obtained by compressing at least two corresponding characteristic vectors extracted from at least two face images of the same user; obtaining the maximum value in the at least two calculated similarities; and outputting the user information corresponding to the compression vector corresponding to the maximum value. The invention further discloses a face identification device. The face identification method and device are able to improve the speed and the accuracy of face identification.
Description
Technical field
The present invention relates to recognition of face field, relate in particular to a kind of face identification method and device.
Background technology
Recognition of face is widely used in the fields such as security monitoring, secure payment and clock in and out. Face at present
Identification is generally that the face characteristic by storing in the face characteristic of coupling input picture and database is known
Not.
Under prior art, in order to improve the precision of identification, sometimes need to gather various faces for everyone and shine
Sheet, to obtain more detailed facial information. It is individual that but the diversity of individual specimen can make in overall sample
Between the property distinguished variation, thereby cause in identifying False Rate higher, and the photo of collecting when everyone
While increasing, corresponding sample characteristics quantity also can become greatly thereupon, thereby causes traveling through in face recognition process
The time of coupling lengthens, and affects recognition speed.
Summary of the invention
For the problems referred to above, the object of the present invention is to provide a kind of characteristic processing method for recognition of face,
Realize recognition of face fast and accurately, met instructions for use.
The embodiment of the present invention provides a kind of face identification method, comprises the steps:
Receive user's to be identified facial image, and extract characteristic vector from described facial image;
Described characteristic vector is mated with at least two compression vectors that prestore, calculate described characteristic vector
With the vectorial similarity of each compression; Wherein, described compression vector passes through same user's at least two
At least two corresponding characteristic vectors that facial image extracts are compressed acquisition;
Obtain the maximum at least two similarities that calculate;
Export the user's that the compression vector corresponding with described maximum is corresponding information.
As the improvement of such scheme, described reception user's to be identified facial image, and from described face figure
In picture, extract characteristic vector, be specially:
Receive user's to be identified facial image, and pass through degree of deep learning network algorithm from described facial image
Extract characteristic vector.
As the improvement of such scheme, at described reception user's to be identified facial image, and from described face
Before extracting characteristic vector in image, also comprise:
The image that gathers N user, wherein, each user gathers M and opens image, N, M is greater than 1
Integer;
Utilize described degree of deep learning network algorithm to extract the characteristic vector V of every facial imageij, wherein, VijTable
Show the characteristic vector that i user's j opens facial image, and 1≤i≤N, 1≤j≤M;
Same user's M is opened after M characteristic vector that facial image extracts compress, generate and N
N the compression vector that individual user is corresponding; Wherein, i user's compression vector is labeled as Wi, and
As the improvement of such scheme, described described characteristic vector and prestore at least two compression vectors are entered
Row coupling, calculates the vectorial similarity of described characteristic vector and each compression; Wherein, described compression vector is
At least two corresponding characteristic vectors that at least two facial images of a user are extracted are compressed and are obtained
, be specially:
The user's to be identified who extracts characteristic vector F is mated with at least two compression vectors that prestore,
Obtain and describedly compress vectorial W to measure feature F and eachiSimilarity S (F, Wi), wherein,
FkFor k the feature of described characteristic vector F, WikFor the vectorial W of described compressioni
K feature, and 1≤k≤L, L is greater than 1 integer.
As the improvement of such scheme, each user gathers M and opens image, is specially:
Gather the M of each user under different light rays intensity and open image; Or,
Gather the M of each user under different shooting angles and open image; Or,
Gather the M of each user under different light rays intensity and different shooting angles and open image.
As the improvement of such scheme, after the described maximum of obtaining in N the similarity calculating,
Also comprise:
Judge whether described maximum is greater than a default threshold value;
If so, export user corresponding to compression vector corresponding with described maximum;
If not, export the information of recognition failures.
The present invention also provides a kind of face identification device, comprising:
First Characteristic extraction unit, for receiving user's to be identified facial image, and from described facial image
In extract characteristic vector;
Matching unit, for described characteristic vector is mated with at least two compression vectors that prestore, meter
Calculate the vectorial similarity of described characteristic vector and each compression; Wherein, described compression vector passes through same
At least two corresponding characteristic vectors that at least two facial images of user extract are compressed acquisition;
Maximum acquiring unit, for obtaining the maximum of at least two similarities that calculate;
Output unit, for exporting the user's that the compression vector corresponding with described maximum is corresponding information.
As the improvement of such scheme, described face identification device also comprises:
Collecting unit, for gathering N user's image, wherein, each user gathers M and opens image, N,
M is greater than 1 integer;
Second Characteristic extraction unit, for utilizing degree of deep learning network algorithm to extract the feature of every facial image
Vector Vij, wherein, VijRepresent the characteristic vector that i user's j opens facial image, and 1≤i≤N,
1≤j≤M；
Compression unit, presses for same user's M being opened to M the characteristic vector that facial image extracts
After contracting, generate and N N the compression vector that user is corresponding; Wherein, i user's compression vector mark
Be designated as Wi, and
As the improvement of such scheme, described matching unit specifically for, by the user's to be identified spy who extracts
Levy vectorial F with described N compression vector mate, obtain described to measure feature F with each compress to
Amount WiSimilarity S (F, Wi), wherein,FkFor the k of described characteristic vector F
Individual feature, WikFor the vectorial W of described compressioniK feature, and 1≤k≤L, L is greater than 1 integer.
As the improvement of such scheme, described face identification device also comprises judging unit, wherein:
Described judging unit, for judging whether described maximum is greater than a default threshold value;
If so, notify described output unit output user corresponding to compression vector corresponding with described maximum;
If not, notify the information of described output unit output recognition failures.
The face identification method that the embodiment of the present invention provides and device, by the people from receiving user to be identified
In face image, extract characteristic vector, and by least two pressures in described characteristic vector and described sample database
Contracting vector mates, and obtains face recognition result, has improved the property distinguished between individual specimen, has shortened
The characteristic matching time in recognition of face, thus realize recognition of face fast and accurately.
Brief description of the drawings
In order to be illustrated more clearly in technical scheme of the present invention, below by required use in embodiment
Accompanying drawing is briefly described, and apparently, the accompanying drawing in the following describes is only enforcements more of the present invention
Mode, for those of ordinary skill in the art, not paying under the prerequisite of creative work, all right
Obtain other accompanying drawing according to these accompanying drawings.
Fig. 1 is the schematic flow sheet of the face identification method that provides of first embodiment of the invention.
Fig. 2 be the embodiment of the present invention provide for carrying out the degree of deep learning network algorithm of characteristic vector extraction
Schematic diagram.
Fig. 3 is the schematic flow sheet of the face identification method that provides of second embodiment of the invention.
Fig. 4 is the schematic flow sheet of the face identification method that provides of third embodiment of the invention.
Fig. 5 is the structural representation of the face identification device that provides of fourth embodiment of the invention.
Fig. 6 is the structural representation of the face identification device that provides of fifth embodiment of the invention.
Fig. 7 is the structural representation of the face identification device that provides of sixth embodiment of the invention.
Detailed description of the invention
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 clearly
Chu, intactly description, obviously, described embodiment is only the present invention's part embodiment, instead of
Whole embodiment. Based on the embodiment in the present invention, those of ordinary skill in the art are not making creation
The every other embodiment obtaining under property work prerequisite, belongs to the scope of protection of the invention.
A kind of face identification method and device that the embodiment of the present invention provides, for having improved the speed of recognition of face
Degree and the degree of accuracy. Be described in detail respectively below.
Refer to Fig. 1, Fig. 1 is the schematic flow sheet of the face identification method that provides of first embodiment of the invention.
Described face identification method can be carried out by face identification device, and comprises the steps:
S101, receives user's to be identified facial image, and extract characteristic vector from described facial image.
In embodiments of the present invention, described face identification device can pass through camera or other IMAQ instruments
Gather user's to be identified facial image, and the feature that proposes described facial image from described facial image to
Amount F.
See also Fig. 2, in one embodiment of the invention, described face identification device can pass through the degree of depth
Learning network algorithm extracts characteristic vector from described facial image. Particularly, described face identification device
Gathering after described user's to be identified facial image, can extract at random size to described facial image and be
The image of 224*224, tri-Color Channel inputs of point RGB, wherein, the main part of original facial image
Divide and be included in training image. The basic structure of described degree of deep learning network algorithm has 8 layers, its
In, first 5 layers is convolutional layer, latter 3 layers is full articulamentum. In first convolutional layer, 96 are adopted greatly
Little be the convolution mask of 11*11 on 3 passages, with the sample frequency that is spaced apart 4 pixels, input is schemed
Picture carries out convolution operation, obtains the picture that 96 sizes are 55*55, after convolution operation, carries out once
After the conversion of ReLU (relu1) and Norm (norm1), then carry out max-pooling (pool1)
Operation, is delivered to lower one deck as output; In second convolutional layer, having adopted 256 sizes is 5*5
Convolution mask to input picture carry out convolution operation, obtain the picture that 256 sizes are 27*27, then enter
After the conversion of a ReLU of row (relu2) and Norm (norm2), then carry out max-pooling (pool2)
Operation, is delivered to lower one deck as output; The generative process of the 3rd convolutional layer and the class of second convolutional layer
Seemingly, different is that this one deck is to have adopted the convolution mask that 384 sizes are 3*3 to obtain 384 sizes
For the picture of 13*13; The 4th convolutional layer is that the 3rd convolutional layer carries out after a ReLU (relu3), straight
Connect the picture that adopts convolution mask that 384 sizes are 3*3 to obtain 384 big or small 13*13; The 5th volume
The generative process of lamination and the 4th convolutional layer similar, has just carried out a ReLU (relu4) on upper strata
After, directly adopt the convolution mask that 256 sizes are 3*3 to obtain the picture that 256 sizes are 13*13,
Then carry out a max-pooling (pool5) operation, output to first full articulamentum and carry out the full behaviour of connection
Do, obtain 4096 nodes; Second full articulamentum is that a upper full articulamentum carries out ReLU (relu6)
After, then carry out carrying out full attended operation after dropout (drop6) operation, obtain 4096 nodes; The
The generative process of three full articulamentums and second full articulamentum similar, obtains 1000 nodes. Last
What layer was exported is the classification results of picture, and common layer second from the bottom can well be described the overall situation spy of picture
Levy, therefore they are used as the feature of extracting in original facial image, and to form a length be 4096
Characteristic vector F, wherein, k the feature of described characteristic vector F can be labeled as Fk, and 1≤k≤4096.
S102, mates described characteristic vector with at least two compression vectors that prestore, calculate described spy
Levy the vectorial similarity of vector and each compression; Wherein, described compression vector passes through to same user extremely
At least two corresponding characteristic vectors that few two facial images extract are compressed acquisition.
In embodiments of the present invention, described face identification device is extracting described user's to be identified facial image
Characteristic vector F after, described characteristic vector F is mated one by one with prestore at least two compression vectors,
Obtain and describedly compress vectorial W to measure feature F and eachiSimilarity S (F, Wi), wherein, described compression to
Amount is that at least two corresponding characteristic vectors that at least two facial images of a user are extracted are pressed
Contracting obtains. For example, suppose that described compression vector has N (N is greater than 1 integer), i compress to
Amount is labeled as Wi, and the individual W that is characterized as of the k of i compression vectorik. Similarity
(i=1,2 ... N), due to a total N compression vector, so the similarity calculating also has N.
S103, obtains the maximum at least two similarities that calculate.
In embodiments of the present invention, described face identification device sorts to the N a calculating similarity
After, obtain N the maximum in similarity.
S104, exports the user's that the compression corresponding with described maximum vector is corresponding information.
In embodiments of the present invention, the described face identification device output compression vector corresponding with described maximum
Corresponding user's information, as output user's name, ID, job number or directly output are by information such as checkings.
The face identification method that the embodiment of the present invention provides, by the facial image from receiving user to be identified
In extract characteristic vector, and described characteristic vector is mated with prestore at least two compression vectors,
Obtain face recognition result, improved the property distinguished between individual specimen, shortened the feature in recognition of face
Match time, thus realize recognition of face fast and accurately.
See also Fig. 3, Fig. 3 is the flow process signal of the face identification method that provides of second embodiment of the invention
Figure. It at least comprises the steps:
S201, N user's of collection image, wherein, each user gathers M and opens image, N, M is large
In 1 integer.
S202, utilizes degree of deep learning network to extract the characteristic vector V of every facial imageij, wherein, VijRepresent
The characteristic vector that i user's j opens facial image, and 1≤i≤N, 1≤j≤M. Vij
S203, the characteristic vector that same user's M is opened to facial image generates N compression after compressing
Vector; Wherein, i user's compression vector is labeled as Wi, and
In embodiments of the present invention, described face identification device can first gather N user's facial image, its
In, in order to obtain the details of user face, need to gather M to each user and open facial image, here
N can be total number of persons.
Particularly, described face identification device, in order to obtain the details of user face, can gather each use
The M of family under different light rays intensity opens image; Or, gather the M of each user under different shooting angles and open
Image; Or, gather the M of each user under different light rays intensity and different shooting angles and open image; Wherein,
M is greater than 1 integer.
In embodiments of the present invention, preferably, described face identification device can gather 10 people to each user
Face image. Be understandable that, described M also can value be 3,6 or be greater than arbitrarily 1 integer, and these can
Arrange according to the actual needs, the present invention does not do concrete restriction.
In embodiments of the present invention, being extracted to M, each user opens after facial image described recognition of face dress
Put and first utilize the above-mentioned degree of deep learning network of mentioning to extract the characteristic vector V of every facial imageij, wherein, VijTable
Show the characteristic vector that i user's j opens facial image, and 1≤i≤N, 1≤j≤M. As mentioned above,
Described characteristic vector VijComprise the individual feature of L (4096), and k signature is Vijk. Then,
The characteristic vector that face identification device is opened facial image to each user's M generates N pressure after compressing
Contracting vector. That is:
In embodiments of the present invention, described face identification device can be stored in one by described N compression vector
In individual sample database, wherein, in described sample database, each compression vector and corresponding user's
Information association, for example, described compression vector can be associated with a user's name, or user's ID, job number
Association, or simultaneously associated with a user's name, ID etc.
S204, receives user's to be identified facial image, and extract characteristic vector from described facial image.
S205, mates described characteristic vector with at least two compression vectors that prestore, calculate described spy
Levy the vectorial similarity of vector and each compression; Wherein, described compression vector passes through to same user extremely
At least two corresponding characteristic vectors that few two facial images extract are compressed acquisition.
S206, obtains the maximum at least two similarities that calculate.
S207, exports the user's that the compression corresponding with described maximum vector is corresponding information.
In the embodiment of the present invention, because compression vector is the feature by each user's M being opened to facial image
Vector compresses and obtains, thereby has improved the discrimination of identification, simultaneously due to each user only corresponding to
A compression vector, has also improved the speed of identification.
See also Fig. 4, Fig. 4 is the flow process signal of the face identification method that provides of third embodiment of the invention
Figure. It at least comprises the steps:
S301, receives user's to be identified facial image, and extract characteristic vector from described facial image.
S302, mates described characteristic vector with at least two compression vectors that prestore, calculate described spy
Levy the vectorial similarity of vector and each compression; Wherein, described compression vector passes through to same user extremely
At least two corresponding characteristic vectors that few two facial images extract are compressed acquisition.
S303, obtains the maximum at least two similarities that calculate.
S304, judges whether described maximum is greater than a default threshold value; If so, carry out S305, if not,
Carry out S306.
S305, exports the user's that the compression corresponding with described maximum vector is corresponding information.
S306, the information of output recognition failures.
Particularly, in the time of identification, possibility user's to be identified compression vector does not also deposit described sample database in
In, or do not exist in sample database. Now, described face identification device need first judge described in extremely
Whether the maximum in few two similarities is greater than a default threshold value (for example, described threshold value can be set to 0.8),
If so, explanation is identified successfully, exports the user's that the compression vector corresponding with described maximum is corresponding information
And be user's to be identified information, otherwise, export the information of recognition failures, user's to be identified letter
Breath is not present in described sample database.
In the embodiment of the present invention, by described threshold value is set, can prevent that uncorrelated personnel from, by identification, strengthening
The safety and reliability of identification.
See also Fig. 5, Fig. 5 is the structural representation of the face identification device that provides of fourth embodiment of the invention
Figure. Described face identification device 400 comprises:
First Characteristic extraction unit 410, for receiving user's to be identified facial image, and from described face figure
In picture, extract characteristic vector.
In embodiments of the present invention, described First Characteristic extraction unit 410 can pass through camera or other images
Sampling instrument gathers user's to be identified facial image, and passes through degree of deep learning network algorithm from described face figure
The characteristic vector of described facial image is proposed in picture.
Matching unit 420, for described characteristic vector is mated with at least two compression vectors that prestore,
Calculate the vectorial similarity of described characteristic vector and each compression; Wherein, described compression vector passes through same
At least two corresponding characteristic vectors that at least two facial images of individual user extract are compressed acquisition.
In embodiments of the present invention, described First Characteristic extraction unit 410 is extracting described user's to be identified
After the characteristic vector F of facial image, described matching unit 420 by described characteristic vector F with prestore at least
Two compression vectors mate one by one, obtain and describedly compress vectorial W to measure feature F and eachiSimilar
Degree S (F, Wi), wherein, described compression vector is right for what at least two facial images of a user were extracted
At least two characteristic vectors of answering are compressed acquisition, in embodiments of the present invention, suppose described compression vector
Number be N, i is compressed vector and is labeled as Wi, and the individual W that is characterized as of the k of i compression vectorik。
SimilarityDue to a total N compression vector, so meter
The similarity obtaining also has N.
Maximum acquiring unit 430, for obtaining the maximum of at least two similarities that calculate.
For example, described maximum acquiring unit 430 can sort to the N a calculating similarity, with
Obtain the maximum in N similarity.
Output unit 440, for exporting the user's that the compression vector corresponding with described maximum is corresponding information.
In embodiments of the present invention, the described output unit 440 output compression vector corresponding with described maximum
Corresponding user's information, as output user's name, ID, job number or directly output are by information such as checkings.
The face identification device 400 that the embodiment of the present invention provides, by described First Characteristic extraction unit 410 from
Receive in user's to be identified facial image and extract characteristic vector, and will by described matching unit 420
Described characteristic vector F mates with at least two compression vectors that prestore, and obtains face recognition result, carries
The property distinguished between high individual specimen, has shortened the characteristic matching time in recognition of face, thereby has realized
Recognition of face fast and accurately.
See also Fig. 6, Fig. 6 is the structural representation of the face identification device that provides of fifth embodiment of the invention
Figure. Described face identification device 500 comprises:
Collecting unit 510, for gathering N user's image, wherein, each user gathers M and opens image,
N, M is greater than 1 integer.
Second Characteristic extraction unit 520, for utilizing degree of deep learning network algorithm to extract the spy of every facial image
Levy vectorial Vij, wherein, VijRepresent the characteristic vector that i user's j opens facial image, and 1≤i≤N,
1≤j≤M。
Compression unit 530, for same user's M being opened after the characteristic vector of facial image compresses,
Generate N compression vector; Wherein, i user's compression vector is labeled as Wi, and
First Characteristic extraction unit 540, for receiving user's to be identified facial image, and from described face figure
In picture, extract characteristic vector.
Matching unit 550, for described characteristic vector is mated with at least two compression vectors that prestore,
Calculate the vectorial similarity of described characteristic vector and each compression; Wherein, described compression vector passes through same
At least two corresponding characteristic vectors that at least two facial images of individual user extract are compressed acquisition.
Maximum acquiring unit 560, for obtaining the maximum of at least two similarities that calculate.
Output unit 570, for exporting the user's that the compression vector corresponding with described maximum is corresponding information.
In the embodiment of the present invention, because compression vector is by described compression unit 530 by each user extremely
The characteristic vector of few two facial images is compressed acquisition, thereby has improved the discrimination of identification, simultaneously
Because each user is only corresponding to a compression vector, also improve the speed of identification.
See also Fig. 7, Fig. 7 is the structural representation of the face identification device that provides of sixth embodiment of the invention
Figure. Described face identification device 600 comprises First Characteristic extraction unit 610, matching unit 620, maximum
Acquiring unit 630, judging unit 640 and output unit 650, wherein:
Described First Characteristic extraction unit 610, for receiving user's to be identified facial image, and from described people
In face image, extract characteristic vector.
Described matching unit 620, for carrying out described characteristic vector and prestore at least two compression vectors
Join, calculate the vectorial similarity of described characteristic vector and each compression; Wherein, described compression vector is by right
At least two corresponding characteristic vectors that at least two facial images of same user extract are compressed and are obtained
.
Described maximum acquiring unit 630, for obtaining the maximum of at least two similarities that calculate
Value.
Described judging unit 640, for judging whether described maximum is greater than a default threshold value;
If so, notify the described output unit 650 outputs compression vector corresponding with described maximum corresponding
User;
If not, notify described output unit 650 to export the information of recognition failures.
Particularly, in the time of identification, possibility user's to be identified compression vector does not also deposit described sample database in
In, or do not exist in sample database. Now, described judging unit 640 need first judge described in extremely
Whether the maximum in few two similarities is greater than a default threshold value (for example, described threshold value can be set to 0.8),
If so, explanation is identified successfully, the described output unit 650 output compression vector corresponding with described maximum
Corresponding user's information and be user's to be identified information, otherwise, described output unit 650 output knowledges
Not failed information, user's to be identified information is not present in sample database.
In the embodiment of the present invention, by described threshold value is set, can prevent that uncorrelated personnel from, by identification, strengthening
The safety and reliability of identification.
In the description of this description, reference term " embodiment ", " some embodiment ", " example ",
The description of " concrete example " or " some examples " etc. means the concrete spy in conjunction with this embodiment or example description
Levy, structure, material or feature be contained at least one embodiment of the present invention or example. In this explanation
In book, to the schematic statement of above-mentioned term not must for be identical embodiment or example. And,
Specific features, structure, material or the feature of describing can one or more embodiment in office or example in
Suitable mode combination. In addition,, not conflicting in the situation that, those skilled in the art can be by this
The feature of the different embodiment that describe in description or example and different embodiment or example is carried out combination and group
Close.
In addition, term " first ", " second " be only for describing object, and can not be interpreted as instruction or hint
Relative importance or the implicit quantity that indicates indicated technical characterictic. Thus, be limited with " first ", "
Two " at least one this feature can be expressed or impliedly be comprised to feature. In description of the invention, " many
Individual " implication be at least two, for example two, three etc., unless otherwise expressly limited specifically.
Any process of otherwise describing in flow chart or at this or method are described and can be understood to, table
Show and comprise that one or more is for realizing the code of executable instruction of step of specific logical function or process
Module, fragment or part, and the scope of the preferred embodiment of the present invention comprises other realization, its
In can be not according to order shown or that discuss, comprise according to related function by the mode of basic while or
By contrary order, carry out function, this should be by embodiments of the invention person of ordinary skill in the field
Institute is understood.
The logic and/or the step that in flow chart, represent or otherwise describe at this, for example, can be recognized
For being the sequencing list of the executable instruction for realizing logic function, may be embodied in any computer
In computer-readable recording medium, for instruction execution system, device or equipment (as computer based system, comprise place
The reason system of device or other can and be carried out the system of instruction from instruction execution system, device or equipment instruction fetch)
Use, or use in conjunction with these instruction execution systems, device or equipment. With regard to this description, " calculate
Machine computer-readable recording medium " can be anyly can comprise, storage, communication, propagation or transmitting software carry out for instruction
System, device or equipment or the device using in conjunction with these instruction execution systems, device or equipment. Calculate
The example more specifically (non-exhaustive list) of machine computer-readable recording medium comprises following: have one or more wirings
Electrical connection section (electronic installation), portable computer diskette box (magnetic device), random access memory (RAM),
Read-only storage (ROM), erasable read-only storage (EPROM or flash memory), the light edited
Fine device, and portable optic disk read-only storage (CDROM). In addition, computer-readable medium even
Can be paper or other the suitable media that can print described software thereon, because can for example pass through paper
Or other media carry out optical scanner, then edit, decipher or carry out with other suitable methods if desired
Process and obtain described software in electronics mode, be then stored in computer storage.
In the above-described embodiment, multiple steps or method can be with being stored in memory and by suitable finger
Order is held and be should be appreciated that each several part of the present invention can realize with hardware, software, firmware or their combination.
Software or firmware that row system is carried out are realized. For example, if realized with hardware, and in another enforcement side
The same in formula, can realize by any one in following technology well known in the art or their combination: have
For data-signal being realized to the discrete logic of the logic gates of logic function, there is suitable combination
The special IC of logic gates, programmable gate array (PGA), field programmable gate array (FPGA)
Deng.
Those skilled in the art are appreciated that the whole or portion that above-described embodiment method is carried that realizes
Be can carry out the hardware that instruction is relevant by software to complete step by step, described software can be stored in a kind of meter
In calculation machine readable storage medium storing program for executing, this software, in the time carrying out, comprises step of embodiment of the method one or a combination set of.
In addition, the each functional unit in each embodiment of the present invention can be integrated in a processing module,
Also can be that the independent physics of unit exists, also can be integrated in a module in two or more unit
In. Above-mentioned integrated module both can adopt the form of hardware to realize, and also can adopt software function module
Form realizes. If described integrated module realizes and as product independently using the form of software function module
When selling or using, also can be stored in a computer read/write memory medium.
The above-mentioned storage medium of mentioning can be read-only storage, disk or CD etc. Although show above
Go out and described embodiments of the invention, being understandable that, above-described embodiment is exemplary, Bu Nengli
Separate as limitation of the present invention, those of ordinary skill in the art within the scope of the invention can be to above-mentioned reality
Execute that example changes, amendment, replacement and modification.
Claims (10)
1. a face identification method, is characterized in that, comprises the steps:
Receive user's to be identified facial image, and extract characteristic vector from described facial image;
Described characteristic vector is mated with at least two compression vectors that prestore, calculate described characteristic vector
With the vectorial similarity of each compression; Wherein, described compression vector passes through same user's at least two
At least two corresponding characteristic vectors that facial image extracts are compressed acquisition;
Obtain the maximum at least two similarities that calculate;
Export the user's that the compression vector corresponding with described maximum is corresponding information.
2. face identification method according to claim 1, is characterized in that, described reception use to be identified
The facial image at family, and extract characteristic vector from described facial image, be specially:
Receive user's to be identified facial image, and pass through degree of deep learning network algorithm from described facial image
Extract characteristic vector.
3. face identification method according to claim 2, is characterized in that, to be identified in described reception
User's facial image, and extract characteristic vector from described facial image before, also comprise:
The image that gathers N user, wherein, each user gathers M and opens image, N, M is greater than 1
Integer;
Utilize described degree of deep learning network algorithm to extract the characteristic vector V of every facial imageij, wherein, VijTable
Show the characteristic vector that i user's j opens facial image, and 1≤i≤N, 1≤j≤M;
Same user's M is opened after M characteristic vector that facial image extracts compress, generate and N
N the compression vector that individual user is corresponding; Wherein, i user's compression vector is labeled as Wi, and
4. face identification method according to claim 3, is characterized in that, described by described feature to
Amount is mated with prestore at least two compression vectors, calculate described characteristic vector with each compress vectorial
Similarity; Wherein, the correspondence of described compression vector at least two facial images of a user are extracted
At least two characteristic vectors compress acquisition, be specially:
The user's to be identified who extracts characteristic vector F is mated with at least two compression vectors that prestore,
Obtain and describedly compress vectorial W to measure feature F and eachiSimilarity S (F, Wi), wherein,
FkFor k the feature of described characteristic vector F, WikFor the vectorial W of described compressioni
K feature, and 1≤k≤L, L is greater than 1 integer.
5. face identification method according to claim 3, is characterized in that, each user gathers M and opens
Image, is specially:
Gather the M of each user under different light rays intensity and open image; Or,
Gather the M of each user under different shooting angles and open image; Or,
Gather the M of each user under different light rays intensity and different shooting angles and open image.
6. according to the face identification method described in claim 1 to 5 any one, it is characterized in that, in institute
After stating the maximum of obtaining in N the similarity calculating, also comprise:
Judge whether described maximum is greater than a default threshold value;
If so, export user corresponding to compression vector corresponding with described maximum;
If not, export the information of recognition failures.
7. a face identification device, is characterized in that, comprising:
First Characteristic extraction unit, for receiving user's to be identified facial image, and from described facial image
In extract characteristic vector;
Matching unit, for described characteristic vector is mated with at least two compression vectors that prestore, meter
Calculate the vectorial similarity of described characteristic vector and each compression; Wherein, described compression vector passes through same
At least two corresponding characteristic vectors that at least two facial images of user extract are compressed acquisition;
Maximum acquiring unit, for obtaining the maximum of at least two similarities that calculate;
Output unit, for exporting the user's that the compression vector corresponding with described maximum is corresponding information.
8. face identification device according to claim 7, is characterized in that, described face identification device
Also comprise:
Collecting unit, for gathering N user's image, wherein, each user gathers M and opens image, N,
M is greater than 1 integer;
Second Characteristic extraction unit, for utilizing degree of deep learning network algorithm to extract the feature of every facial image
Vector Vij, wherein, VijRepresent the characteristic vector that i user's j opens facial image, and 1≤i≤N,
1≤j≤M；
Compression unit, opens after M characteristic vector that facial image extracts compress same user's M,
Generate and N N the compression vector that user is corresponding; Wherein, i user's compression vector is labeled as Wi，
And
9. face identification device according to claim 8, is characterized in that, described matching unit is concrete
Be used for, the user's to be identified who extracts characteristic vector F is mated with described N compression vector, obtain
Describedly compress vectorial W to measure feature F and eachiSimilarity S (F, Wi), wherein,
FkFor k the feature of described characteristic vector F, WikFor the vectorial W of described compressioni
K feature, and 1≤k≤L, L is greater than 1 integer.
10. according to the face identification device described in claim 7 to 9 any one, it is characterized in that institute
State face identification device and also comprise judging unit, wherein:
Described judging unit, for judging whether described maximum is greater than a default threshold value;
If so, notify described output unit output user corresponding to compression vector corresponding with described maximum;
If not, notify the information of described output unit output recognition failures.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510696203.3A CN105590089A (en) | 2015-10-22 | 2015-10-22 | Face identification method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510696203.3A CN105590089A (en) | 2015-10-22 | 2015-10-22 | Face identification method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105590089A true CN105590089A (en) | 2016-05-18 |
Family
ID=55929661
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510696203.3A Pending CN105590089A (en) | 2015-10-22 | 2015-10-22 | Face identification method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105590089A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108108696A (en) * | 2017-12-22 | 2018-06-01 | 歌尔科技有限公司 | A kind of safety protecting method, apparatus and system |
| CN108595927A (en) * | 2018-04-04 | 2018-09-28 | 北京市商汤科技开发有限公司 | Authentication, unlock and method of payment, device, storage medium, product and equipment |
| CN109241868A (en) * | 2018-08-16 | 2019-01-18 | 平安科技（深圳）有限公司 | Face identification method, device, computer equipment and storage medium |
| CN109408676A (en) * | 2018-01-25 | 2019-03-01 | 维沃移动通信有限公司 | A kind of method and terminal device showing user information |
| CN110659541A (en) * | 2018-06-29 | 2020-01-07 | 深圳云天励飞技术有限公司 | Image recognition method, device and storage medium |
| CN111209862A (en) * | 2020-01-03 | 2020-05-29 | 深圳力维智联技术有限公司 | Face image clustering method, device and medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070160296A1 (en) * | 2006-01-11 | 2007-07-12 | Samsung Electronics Co., Ltd. | Face recognition method and apparatus |
| CN101281598A (en) * | 2008-05-23 | 2008-10-08 | 清华大学 | Method for recognizing human face based on amalgamation of multicomponent and multiple characteristics |
| CN102004899A (en) * | 2010-11-03 | 2011-04-06 | 无锡中星微电子有限公司 | Human face identifying system and method |
| CN102254151A (en) * | 2011-06-16 | 2011-11-23 | 清华大学 | Driver fatigue detection method based on face video analysis |
| CN103218609A (en) * | 2013-04-25 | 2013-07-24 | 中国科学院自动化研究所 | Multi-pose face recognition method based on hidden least square regression and device thereof |
| CN104899579A (en) * | 2015-06-29 | 2015-09-09 | 小米科技有限责任公司 | Face recognition method and face recognition device |
-
2015
- 2015-10-22 CN CN201510696203.3A patent/CN105590089A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070160296A1 (en) * | 2006-01-11 | 2007-07-12 | Samsung Electronics Co., Ltd. | Face recognition method and apparatus |
| CN101281598A (en) * | 2008-05-23 | 2008-10-08 | 清华大学 | Method for recognizing human face based on amalgamation of multicomponent and multiple characteristics |
| CN102004899A (en) * | 2010-11-03 | 2011-04-06 | 无锡中星微电子有限公司 | Human face identifying system and method |
| CN102254151A (en) * | 2011-06-16 | 2011-11-23 | 清华大学 | Driver fatigue detection method based on face video analysis |
| CN103218609A (en) * | 2013-04-25 | 2013-07-24 | 中国科学院自动化研究所 | Multi-pose face recognition method based on hidden least square regression and device thereof |
| CN104899579A (en) * | 2015-06-29 | 2015-09-09 | 小米科技有限责任公司 | Face recognition method and face recognition device |
Non-Patent Citations (1)
| Title |
|---|
| 孙涛等: "基于PCA算法的人脸识别方法研究比较", 《现代电子技术》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108108696A (en) * | 2017-12-22 | 2018-06-01 | 歌尔科技有限公司 | A kind of safety protecting method, apparatus and system |
| CN108108696B (en) * | 2017-12-22 | 2020-11-20 | 歌尔科技有限公司 | Safety protection method, device and system |
| CN109408676A (en) * | 2018-01-25 | 2019-03-01 | 维沃移动通信有限公司 | A kind of method and terminal device showing user information |
| CN108595927A (en) * | 2018-04-04 | 2018-09-28 | 北京市商汤科技开发有限公司 | Authentication, unlock and method of payment, device, storage medium, product and equipment |
| CN108595927B (en) * | 2018-04-04 | 2023-09-19 | 北京市商汤科技开发有限公司 | Identity authentication, unlocking and payment method and device, storage medium, product and equipment |
| CN110659541A (en) * | 2018-06-29 | 2020-01-07 | 深圳云天励飞技术有限公司 | Image recognition method, device and storage medium |
| CN109241868A (en) * | 2018-08-16 | 2019-01-18 | 平安科技（深圳）有限公司 | Face identification method, device, computer equipment and storage medium |
| CN111209862A (en) * | 2020-01-03 | 2020-05-29 | 深圳力维智联技术有限公司 | Face image clustering method, device and medium |
| CN111209862B (en) * | 2020-01-03 | 2023-09-29 | 深圳力维智联技术有限公司 | Clustering method, device and medium for face images |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105590089A (en) | Face identification method and device | |
| CN101271515B (en) | Image detection device capable of recognizing multi-angle objective | |
| CN109145745B (en) | Face recognition method under shielding condition | |
| CN109784186A (en) | A kind of pedestrian recognition methods, device, electronic equipment and computer readable storage medium again | |
| CN102938065B (en) | Face feature extraction method and face identification method based on large-scale image data | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN106530200A (en) | Deep-learning-model-based steganography image detection method and system | |
| CN109871780B (en) | Face quality judgment method and system and face identification method and system | |
| CN110598800A (en) | Garbage classification and identification method based on artificial intelligence | |
| CN109948616A (en) | Image detecting method, device, electronic equipment and computer readable storage medium | |
| CN104240256A (en) | Image salient detecting method based on layering sparse modeling | |
| CN109886222A (en) | Face identification method, neural network training method, device and electronic equipment | |
| US11023714B2 (en) | Suspiciousness degree estimation model generation device | |
| CN111915437A (en) | RNN-based anti-money laundering model training method, device, equipment and medium | |
| CN110490214A (en) | The recognition methods and system of image, storage medium and processor | |
| CN110503099A (en) | Information identifying method and relevant device based on deep learning | |
| CN110263807A (en) | Anomaly detection method based on auto-encoder | |
| CN109446345A (en) | Nuclear power file verification processing method and system | |
| CN110009051A (en) | Feature extraction unit and method, DCNN model, recognition methods and medium | |
| CN111046879A (en) | Certificate image classification method and device, computer equipment and readable storage medium | |
| CN106951819A (en) | The single sample face recognition method screened based on sparse probability distribution and multistage classification | |
| CN108875907A (en) | A kind of fingerprint identification method and device based on deep learning | |
| CN115131580B (en) | Space target small sample identification method based on attention mechanism | |
| CN103310235A (en) | Steganalysis method based on parameter identification and estimation | |
| CN105138886B (en) | Robot biometric identification system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160518 |
|
| RJ01 | Rejection of invention patent application after publication |