CN110298249A - Face identification method, device, terminal and storage medium - Google Patents
Face identification method, device, terminal and storage medium Download PDFInfo
- Publication number
- CN110298249A CN110298249A CN201910457256.8A CN201910457256A CN110298249A CN 110298249 A CN110298249 A CN 110298249A CN 201910457256 A CN201910457256 A CN 201910457256A CN 110298249 A CN110298249 A CN 110298249A
- Authority
- CN
- China
- Prior art keywords
- low
- integer coding
- face
- feature vector
- target
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
-
- 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
Abstract
The present invention provides a kind of face identification method, device, terminal and storage mediums.The face identification method includes: to obtain facial image to be identified, according to the first high dimensional feature vector of preset deep learning model extraction facial image to be identified;Using product quantization algorithm by the first high dimensional feature vector quantization be the first low-dimensional integer coding;First low-dimensional integer coding of facial image to be identified and the second low-dimensional integer coding of the face template image being previously obtained are subjected to similarity mode;It is determining with the most like target low-dimensional integer coding of the first low-dimensional integer coding in the second low-dimensional integer coding according to similarity mode result, using the corresponding face template image of target low-dimensional integer coding as target facial image.The present invention carries out similarity-rough set in face recognition process, using low-dimensional integer coding, to reduce comparison number between the two, greatly accelerates the response speed of system, promotes recognition of face efficiency.
Description
Technical field
The present invention relates to technical field of face recognition more particularly to a kind of face identification method, device, terminal and storage to be situated between
Matter.
Background technique
With the development of face recognition technology, recognition of face be widely used in security monitoring, secure payment, unlocking screen and
The fields such as attendance, terminal device acquire facial image by photographic device in real time, and with pre-stored face template image into
Row matching, after successful match, is then verified.
Specifically, the feature vector of facial image to be identified can be generated first in 1:N recognition of face, then by the spy
The feature vector for levying each face template image in vector and face template image library carries out similarity-rough set, takes most like work
For recognition result.In the process, the speed of comparison two-by-two of feature vector itself be than faster, but when N is very big, such as 1
When hundred million or more, comparing number can also become very huge, the response speed of system can be made to be substantially reduced, to influence recognition of face
Efficiency.
Summary of the invention
The present invention provides a kind of face identification method, device, terminal and storage medium, to solve current 1:N recognition of face
In the process, since face template amount of images is huge, when being compared one by one with facial image to be identified, it is huge to compare number
Greatly, the problem of leading to the response speed of system reduces, influences recognition of face efficiency.
To solve the above problems, the present invention adopts the following technical scheme:
The present invention provides a kind of face identification method, includes the following steps:
Facial image to be identified is obtained, first of the facial image to be identified according to preset deep learning model extraction
High dimensional feature vector;
Using product quantization algorithm by the first high dimensional feature vector quantization be the first low-dimensional integer coding;
By the second low-dimensional of the first low-dimensional integer coding of facial image to be identified and the face template image being previously obtained
Integer coding carries out similarity mode;
It is determined and the first low-dimensional integer coding in the second low-dimensional integer coding according to similarity mode result
Most like target low-dimensional integer coding, using the corresponding face template image of the target low-dimensional integer coding as target face
Image.
In one embodiment, the first of the facial image to be identified according to preset deep learning model extraction is high
Before dimensional feature vector, further includes:
The deep learning model is trained using the face template image being previously obtained, with the determination depth
Practise the optimal weight parameter of the connection in model between each node.
In one embodiment, the face template image that the utilization is previously obtained is trained the deep learning model
The step of, comprising:
Face template image is inputted in deep learning model, the corresponding characteristic information of the face template image is obtained;
Based on preset loss function and the corresponding characteristic information of the face template image, the deep learning mould is calculated
The loss of type;Wherein, the loss function is made of Softmax function and the weighting of L-Softmax function;
When loss is higher than preset value, the weight parameter of the connection in the deep learning model between each node is adjusted,
To deep learning model re -training, until when loss is less than or equal to preset value, optimal weight parameter and its corresponding is obtained
Deep learning model.
In one embodiment, the loss function are as follows:
Wherein, the LSoftmaxFor Softmax function, the LL-SoftmaxFor L-Softmax function, ∝ is regulatory factor,
5≤∝≤8。
In one embodiment, the first low-dimensional integer coding by facial image to be identified and the face mould being previously obtained
Second low-dimensional integer coding of plate image carries out before similarity mode, further includes:
Face template image is obtained, according to the second higher-dimension of face template image described in preset deep learning model extraction
Feature vector;
Using product quantization algorithm by the second high dimensional feature vector quantization be the second low-dimensional integer coding.
In one embodiment, described using the corresponding face template image of the target low-dimensional integer coding as target face
The step of image, comprising:
Target higher-dimension before obtaining the corresponding quantization of the target low-dimensional integer coding in the second high dimensional feature vector
Feature vector;
The first high dimensional feature vector of facial image to be identified is calculated at a distance from target high dimensional feature vector;
When the distance is greater than preset value, using the corresponding face template image of the target high dimensional feature vector as mesh
Mark facial image.
In one embodiment, the first high dimensional feature vector for calculating facial image to be identified and target high dimensional feature to
Amount apart from the step of, comprising:
Calculate the first high dimensional feature vector of facial image to be identified and the Euclidean distance of target high dimensional feature vector and remaining
Chordal distance;
The Euclidean distance and COS distance are weighted combination, obtain the distance.
A kind of face identification device provided by the invention, comprising:
Extraction module, it is to be identified according to preset deep learning model extraction for obtaining facial image to be identified
First high dimensional feature vector of facial image;
Quantization modules are used to using product quantization algorithm be the first low-dimensional integer by the first high dimensional feature vector quantization
Coding;
Distance metric module, for by the first low-dimensional integer coding of facial image to be identified and the face mould that is previously obtained
Second low-dimensional integer coding of plate image carries out similarity mode;
Determining module, for being determined and described first in the second low-dimensional integer coding according to similarity mode result
The most like target low-dimensional integer coding of low-dimensional integer coding, by the corresponding face template image of the target low-dimensional integer coding
As target facial image.
The present invention provides a kind of terminal, including memory and processor, is stored with computer-readable finger in the memory
It enables, when the computer-readable instruction is executed by the processor, so that the processor executes as above described in any item people
The step of face recognition method.
The present invention provides a kind of storage medium, is stored thereon with computer program, and the computer program is held by processor
When row, as above described in any item face identification methods are realized.
Compared with the existing technology, technical solution of the present invention at least has following advantage:
Face identification method provided by the invention, by obtaining facial image to be identified, and according to preset deep learning
First high dimensional feature vector of model extraction facial image to be identified;Then using product quantization algorithm that first higher-dimension is special
Levy the first low-dimensional integer coding that vector quantization is negligible amounts;Again by the first low-dimensional integer coding of facial image to be identified with
Second low-dimensional integer coding of the face template image being previously obtained carries out similarity mode;Finally according to similarity mode result
The determining and most like target low-dimensional integer coding of the first low-dimensional integer coding in the second low-dimensional integer coding, will
The corresponding face template image of the target low-dimensional integer coding is as target facial image, in identification process, due to passing through
Product quantization algorithm processing after, can by the corresponding high dimensional feature of facial image and face template image to be identified of substantial amounts to
It is quantified as the low-dimensional integer coding of negligible amounts, similarity-rough set is carried out using low-dimensional integer coding, to reduce the two
Between comparison number, greatly accelerate the response speed of system, promote recognition of face efficiency.
Detailed description of the invention
Fig. 1 is the implementation environment figure of the face identification method provided in one embodiment of the invention;
Fig. 2 is a kind of embodiment flow chart of the present inventor's face recognition method;
Fig. 3 is a kind of embodiment module frame chart of face identification device of the present invention;
Fig. 4 is the internal structure block diagram of terminal in one embodiment of the invention.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, below in conjunction in the embodiment of the present invention
Attached drawing, technical scheme in the embodiment of the invention is clearly and completely described.
In some processes of the description in description and claims of this specification and above-mentioned attached drawing, contain according to
Multiple operations that particular order occurs, but it should be clearly understood that these operations can not be what appears in this article suitable according to its
Sequence is executed or is executed parallel, and the serial number of operation such as S11, S12 etc. be only used for distinguishing each different operation, serial number
It itself does not represent and any executes sequence.In addition, these processes may include more or fewer operations, and these operations can
To execute or execute parallel in order.It should be noted that the description such as " first " herein, " second ", is for distinguishing not
Same message, equipment, module etc., does not represent sequencing, does not also limit " first " and " second " and be different type.
It will appreciated by the skilled person that unless expressly stated, singular " one " used herein, " one
It is a ", " described " and "the" may also comprise plural form.It is to be further understood that being arranged used in specification of the invention
Diction " comprising " refer to that there are the feature, integer, step, operation, element and/or component, but it is not excluded that in the presence of or addition
Other one or more features, integer, step, operation, element, component and/or their group.It should be understood that when we claim member
Part is " connected " or when " coupled " to another element, it can be directly connected or coupled to other elements, or there may also be
Intermediary element.In addition, " connection " used herein or " coupling " may include being wirelessly connected or wirelessly coupling.It is used herein to arrange
Diction "and/or" includes one or more associated wholes for listing item or any cell and all combinations.
It will appreciated by the skilled person that unless otherwise defined, all terms used herein (including technology art
Language and scientific term), there is meaning identical with the general understanding of those of ordinary skill in fields of the present invention.Should also
Understand, those terms such as defined in the general dictionary, it should be understood that have in the context of the prior art
The consistent meaning of meaning, and unless idealization or meaning too formal otherwise will not be used by specific definitions as here
To explain.
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description in which the same or similar labels are throughly indicated same or similar element or has same or like function
Element.Obviously, described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Based on this
Embodiment in invention, those skilled in the art's every other implementation obtained without creative efforts
Example, shall fall within the protection scope of the present invention.
Fig. 1 is the implementation environment figure of the face identification method provided in one embodiment, as shown in Figure 1, in the implementation ring
In border, including server 110, terminal 120.Terminal 120 is connect by network with server, and the terminal 120 can be used for acquiring
Facial image, and facial image is uploaded to server 110 and carries out recognition of face processing.Wherein, above-mentioned network may include because
Special net, 2G/3G/4G, wi f i etc..
It should be noted that server 110 can be independent physical server or terminal, it is also possible to multiple physics clothes
The server cluster for device composition of being engaged in can be to provide the basic cloud computing service such as Cloud Server, cloud database, cloud storage and CDN
Cloud Server.
Terminal 120 can be smart phone, tablet computer, laptop, desktop computer, intelligent sound box, intelligent hand
Table etc., however, it is not limited to this.
Referring to Fig. 2, the present invention provides a kind of face identification method, to solve in current 1:N face recognition process, by
Huge in face template amount of images, when being compared one by one with facial image to be identified, comparison number is huge, leads to system
Response speed reduce, the problem of influencing recognition of face efficiency.In one of embodiment, the face identification method can be wrapped
Include following steps:
S21, facial image to be identified is obtained, the facial image to be identified according to preset deep learning model extraction
First high dimensional feature vector.
Before carrying out recognition of face, the image comprising face can be acquired by terminal device, face is identified from image
Part obtains facial image to be identified, then special according to the key of preset deep learning model extraction facial image to be identified
Sign point and the corresponding first high dimensional feature vector of each key feature points.Wherein, the key feature points may include people nose,
68 face key feature points such as mouth, eyes, forehead.
When using the first high dimensional feature vector of deep learning model extraction facial image to be identified, face can be first extracted
The characteristic value of the key feature points of central area, to be positioned to facial image, then according to the key of face central area
Characteristic point successively extracts the key feature points in other regions of facial image, last to be clicked through according to all key features extracted
Row geometrical characteristic vector construction, so that corresponding first high dimensional feature vector is mapped to, to improve the first high dimensional feature vector
Extraction rate.
S22, using product quantization algorithm by the first high dimensional feature vector quantization be the first low-dimensional integer coding.
Feature Space Decomposing is the cartesian product of multiple lower-dimensional subspaces, then individually by so-called product quantization
Each lower-dimensional subspace is quantified.In the training stage, each sub-spaces obtain kk quantizer, institute after cluster
The intensive division for thering is the cartesian product of these quantizers to constitute one to the total space, and can guarantee to quantify application condition
It is small.It in the present embodiment, can be respectively by facial image and the face template figure to be identified of substantial amounts by product quantization algorithm
As the low-dimensional integer coding that corresponding high dimensional feature vector quantization is negligible amounts.
S23, by the first low-dimensional integer coding of facial image to be identified be previously obtained the second of face template image
Low-dimensional integer coding carries out similarity mode.
It in the present embodiment, can be according to the first low-dimensional when facial image to be identified being compared with face template image
Integer coding calculates similarity with the corresponding character of the second low-dimensional integer coding, and the identical quantity of correspondence character is more, then matches
It spends higher.For example, when second low-dimensional integer coding is 11011111, two low when the first low-dimensional integer coding is 11101111
The character quantity for tieing up integer coding is all 8, and unmatched character is third position and the 4th, and the character to match has 6, then
The matching degree of the first low-dimensional integer coding and the second low-dimensional integer coding isI.e. 75%.Wherein, the face mould being previously obtained
Second low-dimensional integer coding of plate image is a low-dimensional integer coding set comprising multiple low-dimensional integer codings.
In the present embodiment, due to the limited amount of the low-dimensional integer coding of facial image to be identified and face template image,
And in comparison process, the search time in quantization encoding space is solely dependent upon the number M of quantization encoding, i.e. calculation amount is O
(M), calculation amount can be greatly reduced, so the low-dimensional integer of the low-dimensional integer coding of facial image to be identified and face template image
The distance between coding calculation amount can greatly reduce.
S24, it is determined and the first low-dimensional integer in the second low-dimensional integer coding according to similarity mode result
Most like target low-dimensional integer coding is encoded, using the corresponding face template image of the target low-dimensional integer coding as target
Facial image.
In the present embodiment, when the first low-dimensional integer coding of facial image to be identified and the face template figure being previously obtained
When the matching degree highest of the second low-dimensional integer coding of picture, then the highest second low-dimensional integer coding of the matching degree is target low-dimensional
Integer coding, and the corresponding face template image of target low-dimensional integer coding is the target most like with facial image to be identified
Facial image, to complete recognition of face.
Face identification method provided by the invention, by obtaining facial image to be known, and according to preset deep learning mould
Type extracts the first high dimensional feature vector of facial image to be identified;Then utilize product quantization algorithm by first high dimensional feature
Vector quantization is the first low-dimensional integer coding of negligible amounts;Again by the first low-dimensional integer coding of facial image to be identified and in advance
Second low-dimensional integer coding of the face template image first obtained carries out similarity mode;Finally existed according to similarity mode result
The determining and most like target low-dimensional integer coding of the first low-dimensional integer coding in the second low-dimensional integer coding, by institute
The corresponding face template image of target low-dimensional integer coding is stated as target facial image, in identification process, due to by multiplying
It, can be by the corresponding high dimensional feature vector of facial image and face template image to be identified of substantial amounts after accumulated amount algorithm process
It is quantified as the low-dimensional integer coding of negligible amounts, carries out similarity-rough set using low-dimensional integer coding, to reduce between the two
Comparison number, greatly accelerate the response speed of system, promote recognition of face efficiency.
In one embodiment, the facial image to be identified according to preset deep learning model extraction of step S21
Before first high dimensional feature vector, it may also include that
S20, the deep learning model is trained using the face template image being previously obtained, with the determination depth
Spend the optimal weight parameter of the connection in learning model between each node.
In the present embodiment, the deep learning model can be convolutional neural networks model, each of which base all includes
Several nodes, the node between base and base is in a kind of state connected entirely, and the connection between node usually has
One weight parameter.Before deep learning model training, the weight parameter between node is the parameter value being arbitrarily arranged.Right
When deep learning model is trained, the face template image of magnanimity can be inputted in deep learning model, according to depth
The output of model is practised as a result, constantly adjusting to the weight parameter of the connection between node in each base of deep learning model
It is whole, until optimal weight parameter is obtained, to obtain the qualified deep learning model of training.
In one embodiment, the face template image that is previously obtained of utilization of step S20 to the deep learning model into
The step of row training, may particularly include:
S201, face template image is inputted in deep learning model, obtains the corresponding feature of the face template image
Information.
In the present embodiment, when being trained to deep learning model, face template image can be inputted deep learning mould
In type, first face template image is positioned, all characteristic informations are then extracted from face template image.Wherein, institute
Stating characteristic information includes such as eyes, nose, the key feature points of ear and corresponding characteristic value.
In one embodiment, before face template image is inputted deep learning model, further includes:
BoxBlur Fuzzy Processing is carried out to face template image.
The present embodiment can be by carrying out BoxBlur Fuzzy Processing for face template image, to filter out in face template image
High frequency section and only retain low frequency part, thus filtering environmental factor, with preferable keeping characteristics information.For example, illumination, screening
The environmental factors such as gear, angle will affect the acquisition of eye information, between the characteristic information and its characteristic value and true value made
There are relatively large deviations.
S202, it is based on preset loss function and the corresponding characteristic information of the face template image, calculates the depth
The loss of learning model;Wherein, the loss function is made of Softmax function and the weighting of L-Softmax function.
There can be an objective function in each algorithm of usual machine learning, the solution procedure of algorithm is to this
The process that objective function is continued to optimize.In classification or regression problem, usually using loss function as its objective function.Damage
Predicted value and the different degree of true value that function is used to evaluate deep learning model are lost, loss function is better, then depth
The performance for practising model is also better.The loss function of the present embodiment can be made of Softmax function and the weighting of L-Softmax function,
To improve the accuracy of deep learning model costing bio disturbance.
In one embodiment, the loss function are as follows:
Wherein, the LSoftmaxFor Softmax function, the LL-SoftmaxFor L-Softmax function, ∝ is regulatory factor,
5≤∝≤8.To reach optimal loss meter by the specific gravity that Softmax function and L-Softmax function setup is certain
Calculate result.
S203, when loss be higher than preset value when, adjust the weight of the connection in the deep learning model between each node
Parameter until when loss is less than or equal to preset value, obtains optimal weight parameter and its right to deep learning model re -training
The deep learning model answered.
After the loss for calculating deep learning model, judge whether its loss is greater than preset value, if so, adjusting the depth
Then the weight parameter of connection in learning model between each node calculates its loss, directly to deep learning model re -training
When to loss less than or equal to preset value, the optimal weight parameter of the connection in deep learning model between each node is obtained, from
And the corresponding deep learning model of optimal weight parameter is obtained to get the deep learning model qualified to training.Wherein, institute
Stating preset value can flexibly set as needed, and default settings obtain smaller, then the effect of deep learning model that training obtains is got over
It is good.
In one embodiment, in step S23, the first low-dimensional integer coding by facial image to be identified and in advance
Before second low-dimensional integer coding of obtained face template image carries out similarity mode, it may also include that
Face template image is obtained, according to the second higher-dimension of face template image described in preset deep learning model extraction
Feature vector.
Using product quantization algorithm by the second high dimensional feature vector quantization be the second low-dimensional integer coding.
The present embodiment also needs to obtain the face template image of magnanimity, saves it in face template image library, and according to
Preset deep learning model extracts the second high dimensional feature vector of each face template image from face template image library, so
It is afterwards the second low-dimensional integer coding by the second high dimensional feature vector quantization using product quantization algorithm, in order to subsequent progress
The similarity mode of first low-dimensional integer coding and the second low-dimensional integer coding.Wherein, it is stored in advance in face template image library
There is the corresponding second low-dimensional integer coding of each face template image, is used for subsequent similarity mode.
In one embodiment, often relatively high by lookup frequency due to target facial image, it, can in similarity mode
Facial image to be identified is preferentially matched with the higher face template image of frequency is queried, so that matching efficiency is improved,
It is quickly found out most like target facial image.
In one embodiment, described to make the corresponding face template image of the target low-dimensional integer coding in step S24
It the step of for target facial image, may particularly include:
S241, the target before the corresponding quantization of the target low-dimensional integer coding is obtained in the second high dimensional feature vector
High dimensional feature vector.
S242, the first high dimensional feature vector of facial image to be identified is calculated at a distance from target high dimensional feature vector.
When the distance is greater than preset value, using the corresponding face template image of the target high dimensional feature vector as mesh
Mark facial image.
In the present embodiment, it after obtaining most like target low-dimensional integer coding, is obtained in the second high dimensional feature vector
Target high dimensional feature vector before taking the target low-dimensional integer coding is corresponding to quantify, then calculate facial image to be identified first are high
Dimensional feature vector is at a distance from the target high dimensional feature vector, to carry out accurate 1:N comparison, to verify the target higher-dimension
Whether the corresponding face template image of feature vector is the target facial image to match.When distance is greater than preset value, then table
Show that the corresponding face template image of target high dimensional feature vector is target facial image, thus the case where accelerating recognition speed
The lower accuracy for guaranteeing identification.The precise alignment time at this time depends on the vector number L and vector dimension of target high dimensional feature vector
Number V, i.e. calculation amount are O (VL).
Specifically, it is assumed that the first high dimensional feature vector dimension of facial image to be identified is V dimension, face template image
Quantity is N, using in existing 1:N face identification method, extracted from facial image to be identified first high dimensional feature to
After amount, with face template image each in face template image library extract the second high dimensional feature vector directly ask two-by-two away from
From finally finding apart from nearest face template image as comparison result, calculation amount needed for identifying is O (VN).And pass through
Using product quantization algorithm high dimensional feature vector quantization be low-dimensional integer coding after, due to low-dimensional integer coding quantity compared with
Few, after the distance between high dimensional feature vector is calculated the comparison be converted between low-dimensional integer coding, calculation amount is specifically included that
Facial image to be identified be quantified as the calculation amount O (UV) of the first low-dimensional integer coding, from face template image library search with to
It identifies the calculation amount O (M) for the second low-dimensional integer coding that the first low-dimensional integer coding of facial image matches and calculates wait know
Calculation amount of the first high dimensional feature vector of others' face image at a distance from target high dimensional feature vector, therefore, the amount of calculation can
It reduces as O (UV)+O (M)+O (VL), compared to the calculation amount O (VN) before being not optimised, due to N > > U, M, L, so O (VN) > > O
(UV)+O (M)+O (VL), therefore computation amount, are remarkably improved recognition of face efficiency.
In one embodiment, in step S242, the first high dimensional feature vector and mesh for calculating facial image to be identified
Absolute altitude dimensional feature vector apart from the step of, may particularly include:
Calculate the first high dimensional feature vector of facial image to be identified and the Euclidean distance of target high dimensional feature vector and remaining
Chordal distance.
The Euclidean distance and COS distance are weighted combination, obtain the distance.
Euclidean distance is also referred to as euclidean metric, Euclidean distance, it is true between two points in m-dimensional space
Actual distance is from being exactly the straightway distance between two o'clock in the Euclidean distance in two-dimensional space.What Euclidean distance was measured is that space is each
The absolute distance of point, it is directly related with the position coordinates where each point;And COS distance, also referred to as cosine similarity, it is to use
Measurement of two vectorial angle cosine values as the size for measuring two inter-individual differences in vector space, what is measured is empty
Between vector angle, the difference being more embodied on direction, rather than position.Therefore, the present embodiment is by calculating separately wait know
First high dimensional feature vector of others' face image and the Euclidean distance and COS distance of target high dimensional feature vector, and two are counted
It calculates result and carries out comprehensive analysis, for example, Euclidean distance and COS distance are arranged respectively different reference ratios, be based on the reference
The Euclidean distance and COS distance are weighted combination by ratio, to obtain the higher distance of accuracy, then extracting should
Apart from corresponding target high dimensional feature vector, and the corresponding face template image of target high dimensional feature vector is searched, completes people
Face identification, to improve the precision of recognition of face.
Referring to FIG. 3, the embodiment of the present invention also provides a kind of face identification device, in a kind of the present embodiment, including mention
Modulus block 31, quantization modules 32, matching module 33 and determining module 34.Wherein,
Extraction module 31, for obtaining facial image to be identified, wait know according to preset deep learning model extraction
First high dimensional feature vector of others' face image.
Before carrying out recognition of face, the image comprising face can be acquired by terminal device, face is identified from image
Part obtains facial image to be identified, then special according to the key of preset deep learning model extraction facial image to be identified
Sign point and the corresponding first high dimensional feature vector of each key feature points.Wherein, the key feature points may include people nose,
68 face key feature points such as mouth, eyes, forehead.
When using the first high dimensional feature vector of deep learning model extraction facial image to be identified, face can be first extracted
The characteristic value of the key feature points of central area, to be positioned to facial image, then according to the key of face central area
Characteristic point successively extracts the key feature points in other regions of facial image, last to be clicked through according to all key features extracted
Row geometrical characteristic vector construction, so that corresponding first high dimensional feature vector is mapped to, to improve the first high dimensional feature vector
Extraction rate.
Quantization modules 32, for using product quantization algorithm that the first high dimensional feature vector quantization is whole for the first low-dimensional
Number encoder.
Feature Space Decomposing is the cartesian product of multiple lower-dimensional subspaces, then individually by so-called product quantization
Each lower-dimensional subspace is quantified.In the training stage, each sub-spaces obtain kk quantizer, institute after cluster
The intensive division for thering is the cartesian product of these quantizers to constitute one to the total space, and can guarantee to quantify application condition
It is small.It in the present embodiment, can be respectively by facial image and the face template figure to be identified of substantial amounts by product quantization algorithm
As the low-dimensional integer coding that corresponding high dimensional feature vector quantization is negligible amounts.
Matching module 33, for by the first low-dimensional integer coding of facial image to be identified and the face template that is previously obtained
First low-dimensional integer coding of image carries out similarity mode.
It in the present embodiment, can be according to the first low-dimensional when facial image to be identified being compared with face template image
Integer coding calculates similarity with the corresponding character of the second low-dimensional integer coding, and the identical quantity of correspondence character is more, then matches
It spends higher.For example, when second low-dimensional integer coding is 11011111, two low when the first low-dimensional integer coding is 11101111
The character quantity for tieing up integer coding is all 8, and unmatched character is third position and the 4th, and the character to match has 6, then
The matching degree of the first low-dimensional integer coding and the second low-dimensional integer coding isI.e. 75%.Wherein, the face mould being previously obtained
Second low-dimensional integer coding of plate image is a low-dimensional integer coding set comprising multiple low-dimensional integer codings.
In the present embodiment, due to the limited amount of the low-dimensional integer coding of facial image to be identified and face template image,
And in comparison process, the search time in quantization encoding space is solely dependent upon the number M of quantization encoding, i.e. calculation amount is O
(M), calculation amount can be greatly reduced, so the low-dimensional integer of the low-dimensional integer coding of facial image to be identified and face template image
The distance between coding calculation amount can greatly reduce.
Determining module 34, for determining with described the in the second low-dimensional integer coding according to similarity mode result
The most like target low-dimensional integer coding of one low-dimensional integer coding, by the corresponding face template figure of the target low-dimensional integer coding
As being used as target facial image.
In the present embodiment, when the first low-dimensional integer coding of facial image to be identified and the face template figure being previously obtained
When the matching degree highest of the second low-dimensional integer coding of picture, then the highest second low-dimensional integer coding of the matching degree is target low-dimensional
Integer coding, and the corresponding face template image of target low-dimensional integer coding is target person similar with facial image to be identified
Face image, to complete recognition of face.
Face identification device provided by the invention, by obtaining facial image to be identified, and according to preset deep learning
First high dimensional feature vector of model extraction facial image to be identified;Then using product quantization algorithm that first higher-dimension is special
Levy the first low-dimensional integer coding that vector quantization is negligible amounts;Again by the first low-dimensional integer coding of facial image to be identified with
Second low-dimensional integer coding of the face template image being previously obtained carries out similarity mode;Finally according to similarity mode result
The determining and most like target low-dimensional integer coding of the first low-dimensional integer coding in the second low-dimensional integer coding, will
The corresponding face template image of the target low-dimensional integer coding is as target facial image, in identification process, due to passing through
Product quantization algorithm processing after, can by the corresponding high dimensional feature of facial image and face template image to be identified of substantial amounts to
It is quantified as the low-dimensional integer coding of negligible amounts, similarity-rough set is carried out using low-dimensional integer coding, to reduce the two
Between comparison number, greatly accelerate the response speed of system, promote recognition of face efficiency.
In one embodiment, the extraction module 31 is also configured to
The deep learning model is trained using the face template image being previously obtained, with the determination depth
Practise the optimal weight parameter of the connection in model between each node.
In one embodiment, the extraction module 31 is also configured to
Face template image is inputted in deep learning model, the corresponding characteristic information of the face template image is obtained.
Based on preset loss function and the corresponding characteristic information of the face template image, the deep learning mould is calculated
The loss of type;Wherein, the loss function is made of Softmax function and the weighting of L-Softmax function.
When loss is higher than preset value, the weight parameter of the connection in the deep learning model between each node is adjusted,
To deep learning model re -training, until when loss is less than or equal to preset value, optimal weight parameter and its corresponding is obtained
Deep learning model.
In one embodiment, the loss function are as follows:
Wherein, the LSoftmaxFor Softmax function, the LL-SoftmaxFor L-Softmax function, ∝ is regulatory factor,
5≤∝≤8。
In one embodiment, further includes:
Face template image collection module, for obtaining face template image, according to preset deep learning model extraction
Second high dimensional feature vector of the face template image;
Second high dimensional feature vector quantization module, for utilizing product quantization algorithm by the second high dimensional feature vector quantity
Turn to the second low-dimensional integer coding.
In one embodiment, the determining module 34 is also configured to
Target higher-dimension before obtaining the corresponding quantization of the target low-dimensional integer coding in the second high dimensional feature vector
Feature vector.
The first high dimensional feature vector of facial image to be identified is calculated at a distance from target high dimensional feature vector.
When the distance is greater than preset value, using the corresponding face template image of the target high dimensional feature vector as mesh
Mark facial image.
In one embodiment, the determining module 34 is also configured to
Calculate the first high dimensional feature vector of facial image to be identified and the Euclidean distance of target high dimensional feature vector and remaining
Chordal distance.
The Euclidean distance and COS distance are weighted combination, obtain the distance.
About the device in above-described embodiment, wherein modules execute the concrete mode of operation in related this method
Embodiment in be described in detail, no detailed explanation will be given here.
A kind of terminal provided by the invention, including memory and processor are stored in the memory computer-readable
Instruction, when the computer-readable instruction is executed by the processor, so that processor execution is as above described in any item
The step of face identification method.
In one embodiment, the terminal is a kind of computer equipment, as shown in Figure 4.Computer described in the present embodiment
Equipment can be the equipment such as server, personal computer and the network equipment.The computer equipment includes processor 402, deposits
The devices such as reservoir 403, input unit 404 and display unit 405.It will be understood by those skilled in the art that the equipment shown in Fig. 4
Structure devices do not constitute the restriction to all devices, may include components more more or fewer than diagram, or combine certain
Component.Memory 403 can be used for storing computer program 401 and each functional module, and the operation of processor 402 is stored in memory
403 computer program 401, thereby executing the various function application and data processing of equipment.Memory can be interior storage
Device or external memory, or including both built-in storage and external memory.Built-in storage may include read-only memory (ROM),
Programming ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory or
Random access memory.External memory may include hard disk, floppy disk, ZIP disk, USB flash disk, tape etc..Memory packet disclosed in this invention
Include but be not limited to the memory of these types.Memory disclosed in this invention is only used as example rather than as restriction.
Input unit 404 is used to receive the input of signal, and receives the keyword of user's input.Input unit 404 can
Including touch panel and other input equipments.Touch panel collects the touch operation of user on it or nearby and (for example uses
Family uses the operations of any suitable object or attachment on touch panel or near touch panel such as finger, stylus), and root
According to the corresponding attachment device of preset driven by program;Other input equipments can include but is not limited to physical keyboard, function
One of key (such as broadcasting control button, switch key etc.), trace ball, mouse, operating stick etc. are a variety of.Display unit
405 can be used for showing the information of user's input or be supplied to the information of user and the various menus of computer equipment.Display is single
The forms such as liquid crystal display, Organic Light Emitting Diode can be used in member 405.Processor 402 is the control centre of computer equipment, benefit
With the various pieces of various interfaces and the entire computer of connection, by running or executing the software being stored in memory 402
Program and/or module, and the data being stored in memory are called, perform various functions and handle data.
As one embodiment, the computer equipment includes: one or more processors 402, memory 403, and one
Or multiple computer programs 401, wherein one or more of computer programs 401 are stored in memory 403 and are matched
It is set to and is executed by one or more of processors 402, one or more of computer programs 401 are configured to carry out above
Face identification method described in embodiment.
In one embodiment, the invention also provides a kind of storage medium for being stored with computer-readable instruction, the meters
When calculation machine readable instruction is executed by one or more processors, so that one or more processors execute above-mentioned recognition of face side
Method.It is deposited for example, the storage medium can be ROM, random access memory (RAM), CD-ROM, tape, floppy disk and light data
Store up equipment etc..
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, which can be stored in a storage medium, the program
When being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, storage medium above-mentioned can for magnetic disk, CD, only
Read non-volatile memory mediums or random access memory (Random such as storage memory (Read-Only Memory, ROM)
Access Memory, RAM) etc..
Based on the above embodiments it is found that the maximum beneficial effect of the present invention is:
Face identification method, device, terminal and storage medium provided by the invention, by obtaining facial image to be identified,
And according to the first high dimensional feature vector of preset deep learning model extraction facial image to be identified;Then quantified using product
The first high dimensional feature vector quantization is the first low-dimensional integer coding of negligible amounts by algorithm;Again by facial image to be identified
The first low-dimensional integer coding carry out similarity mode with the second low-dimensional integer coding of face template image being previously obtained;Most
According to similarity mode result, determination is most like with the first low-dimensional integer coding in the second low-dimensional integer coding afterwards
Target low-dimensional integer coding, using the corresponding face template image of the target low-dimensional integer coding as target facial image,
In identification process, due to by product quantization algorithm processing after, can be by facial image and the face mould to be identified of substantial amounts
The corresponding high dimensional feature vector quantization of plate image is the low-dimensional integer coding of negligible amounts, is carried out using low-dimensional integer coding similar
Degree compares, to reduce comparison number between the two, greatly accelerates the response speed of system, promotes recognition of face efficiency.
Each technical characteristic of embodiment described above can be combined arbitrarily, for simplicity of description, not to above-mentioned reality
It applies all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, all should be considered as described in this specification.
The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously
Limitations on the scope of the patent of the present invention therefore cannot be interpreted as.It should be pointed out that for those of ordinary skill in the art
For, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to guarantor of the invention
Protect range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (10)
1. a kind of face identification method characterized by comprising
Facial image to be identified is obtained, the first higher-dimension of the facial image to be identified according to preset deep learning model extraction
Feature vector;
Using product quantization algorithm by the first high dimensional feature vector quantization be the first low-dimensional integer coding;
By the second low-dimensional integer of the first low-dimensional integer coding of facial image to be identified and the face template image being previously obtained
Coding carries out similarity mode;
It is determined and the first low-dimensional integer coding most phase in the second low-dimensional integer coding according to similarity mode result
As target low-dimensional integer coding, using the corresponding face template image of the target low-dimensional integer coding as target face figure
Picture.
2. face identification method according to claim 1, which is characterized in that described to be mentioned according to preset deep learning model
Before taking the first high dimensional feature vector of the facial image to be identified, further includes:
The deep learning model is trained using the face template image being previously obtained, with the determination deep learning mould
The optimal weight parameter of connection in type between each node.
3. face identification method according to claim 2, which is characterized in that the face template figure that the utilization is previously obtained
As the step of being trained to the deep learning model, comprising:
Face template image is inputted in deep learning model, the corresponding characteristic information of the face template image is obtained;
Based on preset loss function and the corresponding characteristic information of the face template image, the deep learning model is calculated
Loss;Wherein, the loss function is made of Softmax function and the weighting of L-Softmax function;
When loss is higher than preset value, the weight parameter of the connection in the deep learning model between each node is adjusted, to depth
Learning model re -training is spent, until obtaining optimal weight parameter and its corresponding depth when loss is less than or equal to preset value
Learning model.
4. face identification method according to claim 3, which is characterized in that the loss function are as follows:
Wherein, the LSoftmaxFor Softmax function, the LL-SoftmaxFor L-Softmax function, ∝ is regulatory factor, 5≤∝
≤8。
5. face identification method according to claim 1, which is characterized in that described low by the first of facial image to be identified
Before dimension integer coding and the second low-dimensional integer coding of the face template image being previously obtained carry out similarity mode, also wrap
It includes:
Face template image is obtained, according to the second high dimensional feature of face template image described in preset deep learning model extraction
Vector;
Using product quantization algorithm by the second high dimensional feature vector quantization be the second low-dimensional integer coding.
6. face identification method according to claim 5, which is characterized in that described by the target low-dimensional integer coding pair
The step of face template image answered is as target facial image, comprising:
Target high dimensional feature before obtaining the corresponding quantization of the target low-dimensional integer coding in the second high dimensional feature vector
Vector;
The first high dimensional feature vector of facial image to be identified is calculated at a distance from target high dimensional feature vector;
When the distance is greater than preset value, using the corresponding face template image of the target high dimensional feature vector as target person
Face image.
7. face identification method according to claim 6, which is characterized in that described to calculate the first of facial image to be identified
The step of high dimensional feature vector is at a distance from target high dimensional feature vector, comprising:
Calculate the first high dimensional feature vector of facial image to be identified and the Euclidean distance of target high dimensional feature vector and cosine away from
From;
The Euclidean distance and COS distance are weighted combination, obtain the distance.
8. a kind of face identification device characterized by comprising
Extraction module, for obtaining facial image to be identified, according to face to be identified described in preset deep learning model extraction
First high dimensional feature vector of image;
Quantization modules are used to using product quantization algorithm be the first low-dimensional integer volume by the first high dimensional feature vector quantization
Code;
Matching module, for by the first low-dimensional integer coding of facial image to be identified and the face template image that is previously obtained
Second low-dimensional integer coding carries out similarity mode;
Determining module, for being determined and first low-dimensional in the second low-dimensional integer coding according to similarity mode result
The most like target low-dimensional integer coding of integer coding, using the corresponding face template image of the target low-dimensional integer coding as
Target facial image.
9. a kind of terminal, which is characterized in that including memory and processor, computer-readable finger is stored in the memory
It enables, when the computer-readable instruction is executed by the processor, so that the processor is executed as appointed in claim 1 to 7
The step of face identification method described in one.
10. a kind of storage medium, is stored thereon with computer program, which is characterized in that the computer program is held by processor
When row, the face identification method as described in any one of claims 1 to 7 is realized.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910457256.8A CN110298249A (en) | 2019-05-29 | 2019-05-29 | Face identification method, device, terminal and storage medium |
PCT/CN2019/103645 WO2020237887A1 (en) | 2019-05-29 | 2019-08-30 | Face recognition method, apparatus, terminal, and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910457256.8A CN110298249A (en) | 2019-05-29 | 2019-05-29 | Face identification method, device, terminal and storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110298249A true CN110298249A (en) | 2019-10-01 |
Family
ID=68027274
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910457256.8A Pending CN110298249A (en) | 2019-05-29 | 2019-05-29 | Face identification method, device, terminal and storage medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN110298249A (en) |
WO (1) | WO2020237887A1 (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111091080A (en) * | 2019-12-06 | 2020-05-01 | 贵州电网有限责任公司 | Face recognition method and system |
CN111198963A (en) * | 2019-12-11 | 2020-05-26 | 智慧眼科技股份有限公司 | Target retrieval method and device based on average characteristics and related equipment thereof |
CN111222485A (en) * | 2020-01-15 | 2020-06-02 | 北京明略软件系统有限公司 | 3D face recognition method and device, electronic equipment and storage medium |
CN111325156A (en) * | 2020-02-24 | 2020-06-23 | 北京沃东天骏信息技术有限公司 | Face recognition method, device, equipment and storage medium |
CN111368644A (en) * | 2020-02-14 | 2020-07-03 | 深圳市商汤科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
CN111968152A (en) * | 2020-07-15 | 2020-11-20 | 桂林远望智能通信科技有限公司 | Dynamic identity recognition method and device |
CN112560635A (en) * | 2020-12-10 | 2021-03-26 | 深圳云天励飞技术股份有限公司 | Face matching acceleration method and device, electronic equipment and storage medium |
CN113221820A (en) * | 2021-05-28 | 2021-08-06 | 网易(杭州)网络有限公司 | Object identification method, device, equipment and medium |
CN113496222A (en) * | 2021-09-09 | 2021-10-12 | 南方电网数字电网研究院有限公司 | Mode identification method and device based on label-free chart data and computer equipment |
CN113743239A (en) * | 2021-08-12 | 2021-12-03 | 青岛图灵科技有限公司 | Pedestrian re-identification method and device and electronic equipment |
CN113807164A (en) * | 2021-07-29 | 2021-12-17 | 四川天翼网络服务有限公司 | Face recognition method based on cosine loss function |
CN113971805A (en) * | 2021-12-22 | 2022-01-25 | 深圳市迪博企业风险管理技术有限公司 | Intelligent marking and scoring method combining machine vision and semantic analysis |
CN117558057A (en) * | 2024-01-12 | 2024-02-13 | 清华大学深圳国际研究生院 | Face recognition method |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112488062B (en) * | 2020-12-18 | 2023-04-07 | 浙江大华技术股份有限公司 | Image identification method, device, equipment and medium |
CN112668632B (en) * | 2020-12-25 | 2022-04-08 | 浙江大华技术股份有限公司 | Data processing method and device, computer equipment and storage medium |
CN113408348B (en) * | 2021-05-14 | 2022-08-19 | 桂林电子科技大学 | Video-based face recognition method and device and storage medium |
CN114842544B (en) * | 2022-07-04 | 2022-09-06 | 江苏布罗信息技术有限公司 | Intelligent face recognition method and system suitable for facial paralysis patient |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103745235A (en) * | 2013-12-18 | 2014-04-23 | 小米科技有限责任公司 | Human face identification method, device and terminal device |
US20150154229A1 (en) * | 2013-11-29 | 2015-06-04 | Canon Kabushiki Kaisha | Scalable attribute-driven image retrieval and re-ranking |
WO2017181769A1 (en) * | 2016-04-21 | 2017-10-26 | 腾讯科技(深圳)有限公司 | Facial recognition method, apparatus and system, device, and storage medium |
US20180341805A1 (en) * | 2015-11-06 | 2018-11-29 | Thomson Licensing | Method and Apparatus for Generating Codebooks for Efficient Search |
CN109190505A (en) * | 2018-08-11 | 2019-01-11 | 石修英 | The image-recognizing method that view-based access control model understands |
CN109426781A (en) * | 2017-08-29 | 2019-03-05 | 阿里巴巴集团控股有限公司 | Construction method, face identification method, device and the equipment of face recognition database |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120213419A1 (en) * | 2011-02-22 | 2012-08-23 | Postech Academy-Industry Foundation | Pattern recognition method and apparatus using local binary pattern codes, and recording medium thereof |
CN103150561A (en) * | 2013-03-19 | 2013-06-12 | 华为技术有限公司 | Face recognition method and equipment |
-
2019
- 2019-05-29 CN CN201910457256.8A patent/CN110298249A/en active Pending
- 2019-08-30 WO PCT/CN2019/103645 patent/WO2020237887A1/en active Application Filing
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150154229A1 (en) * | 2013-11-29 | 2015-06-04 | Canon Kabushiki Kaisha | Scalable attribute-driven image retrieval and re-ranking |
CN103745235A (en) * | 2013-12-18 | 2014-04-23 | 小米科技有限责任公司 | Human face identification method, device and terminal device |
US20180341805A1 (en) * | 2015-11-06 | 2018-11-29 | Thomson Licensing | Method and Apparatus for Generating Codebooks for Efficient Search |
WO2017181769A1 (en) * | 2016-04-21 | 2017-10-26 | 腾讯科技(深圳)有限公司 | Facial recognition method, apparatus and system, device, and storage medium |
CN109426781A (en) * | 2017-08-29 | 2019-03-05 | 阿里巴巴集团控股有限公司 | Construction method, face identification method, device and the equipment of face recognition database |
CN109190505A (en) * | 2018-08-11 | 2019-01-11 | 石修英 | The image-recognizing method that view-based access control model understands |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111091080A (en) * | 2019-12-06 | 2020-05-01 | 贵州电网有限责任公司 | Face recognition method and system |
CN111198963A (en) * | 2019-12-11 | 2020-05-26 | 智慧眼科技股份有限公司 | Target retrieval method and device based on average characteristics and related equipment thereof |
CN111222485A (en) * | 2020-01-15 | 2020-06-02 | 北京明略软件系统有限公司 | 3D face recognition method and device, electronic equipment and storage medium |
CN111368644B (en) * | 2020-02-14 | 2024-01-05 | 深圳市商汤科技有限公司 | Image processing method, device, electronic equipment and storage medium |
CN111368644A (en) * | 2020-02-14 | 2020-07-03 | 深圳市商汤科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
CN111325156B (en) * | 2020-02-24 | 2023-08-11 | 北京沃东天骏信息技术有限公司 | Face recognition method, device, equipment and storage medium |
CN111325156A (en) * | 2020-02-24 | 2020-06-23 | 北京沃东天骏信息技术有限公司 | Face recognition method, device, equipment and storage medium |
CN111968152A (en) * | 2020-07-15 | 2020-11-20 | 桂林远望智能通信科技有限公司 | Dynamic identity recognition method and device |
CN111968152B (en) * | 2020-07-15 | 2023-10-17 | 桂林远望智能通信科技有限公司 | Dynamic identity recognition method and device |
CN112560635A (en) * | 2020-12-10 | 2021-03-26 | 深圳云天励飞技术股份有限公司 | Face matching acceleration method and device, electronic equipment and storage medium |
CN112560635B (en) * | 2020-12-10 | 2024-03-26 | 深圳云天励飞技术股份有限公司 | Face matching acceleration method and device, electronic equipment and storage medium |
CN113221820A (en) * | 2021-05-28 | 2021-08-06 | 网易(杭州)网络有限公司 | Object identification method, device, equipment and medium |
CN113807164A (en) * | 2021-07-29 | 2021-12-17 | 四川天翼网络服务有限公司 | Face recognition method based on cosine loss function |
CN113743239A (en) * | 2021-08-12 | 2021-12-03 | 青岛图灵科技有限公司 | Pedestrian re-identification method and device and electronic equipment |
CN113496222A (en) * | 2021-09-09 | 2021-10-12 | 南方电网数字电网研究院有限公司 | Mode identification method and device based on label-free chart data and computer equipment |
CN113971805A (en) * | 2021-12-22 | 2022-01-25 | 深圳市迪博企业风险管理技术有限公司 | Intelligent marking and scoring method combining machine vision and semantic analysis |
CN117558057A (en) * | 2024-01-12 | 2024-02-13 | 清华大学深圳国际研究生院 | Face recognition method |
CN117558057B (en) * | 2024-01-12 | 2024-04-16 | 清华大学深圳国际研究生院 | Face recognition method |
Also Published As
Publication number | Publication date |
---|---|
WO2020237887A1 (en) | 2020-12-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110298249A (en) | Face identification method, device, terminal and storage medium | |
CN107944020B (en) | Face image searching method and device, computer device and storage medium | |
CN108399386A (en) | Information extracting method in pie chart and device | |
CN110766080B (en) | Method, device and equipment for determining labeled sample and storage medium | |
CN108875487B (en) | Training of pedestrian re-recognition network and pedestrian re-recognition based on training | |
CN111008640A (en) | Image recognition model training and image recognition method, device, terminal and medium | |
CN110084165A (en) | The intelligent recognition and method for early warning of anomalous event under the open scene of power domain based on edge calculations | |
CN109598234A (en) | Critical point detection method and apparatus | |
CN105354595A (en) | Robust visual image classification method and system | |
CN105740808B (en) | Face identification method and device | |
CN113128478B (en) | Model training method, pedestrian analysis method, device, equipment and storage medium | |
CN109711228A (en) | A kind of image processing method that realizing image recognition and device, electronic equipment | |
CN112784778A (en) | Method, apparatus, device and medium for generating model and identifying age and gender | |
CN109740567A (en) | Key point location model training method, localization method, device and equipment | |
CN107251048A (en) | Reliable finger tip and palm detection | |
CN110070106A (en) | Smog detection method, device and electronic equipment | |
CN111178537A (en) | Feature extraction model training method and device | |
CN110362648A (en) | Update method and device, storage medium, the computer equipment of questionnaire survey topic | |
US20180247105A1 (en) | High speed reference point independent database filtering for fingerprint identification | |
CN109753577A (en) | A kind of method and relevant apparatus for searching for face | |
CN110309774A (en) | Iris segmentation method, apparatus, storage medium and electronic equipment | |
CN110287938A (en) | Event recognition method, system, equipment and medium based on critical segment detection | |
CN108875445B (en) | Pedestrian re-identification method and device | |
CN104572930B (en) | Data classification method and device | |
CN109961103A (en) | The training method of Feature Selection Model, the extracting method of characteristics of image and device |
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 |