CN110263768A - A kind of face identification method based on depth residual error network - Google Patents
A kind of face identification method based on depth residual error network Download PDFInfo
- Publication number
- CN110263768A CN110263768A CN201910652309.1A CN201910652309A CN110263768A CN 110263768 A CN110263768 A CN 110263768A CN 201910652309 A CN201910652309 A CN 201910652309A CN 110263768 A CN110263768 A CN 110263768A
- Authority
- CN
- China
- Prior art keywords
- convolution
- network
- face
- branch
- residual error
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Abstract
The invention discloses a kind of face identification methods based on depth residual error network comprising the steps of: A, image acquisition device acquisition facial image are simultaneously stored into memory；B, face detection module detects human face region, while the key point of locating human face；C, human face region image is cut out, and is aligned according to the key point of face；D, human face region image is trained, calculates the average facial image in training set, every secondary human face region image in facial image training set is subtracted and carries out the training of network parameter after average facial image and obtains the model of convolutional neural networks；E, the model feedforward network obtained using training, obtains the feature vector of facial image；F, the distance between feature vector is calculated, classification is completed, completes the identification to face, the present invention can be rapidly performed by recognition of face, and identify that face result is more accurate, and can improve the accuracy in face recognition technology.
Description
Technical field
The present invention relates to technical field of face recognition, specifically a kind of face identification method based on depth residual error network.
Background technique
Recognition of face is a kind of biological identification technology for carrying out identification based on facial feature information of people.With camera shooting
Machine or camera acquire image or video flowing containing face, and automatic detection and tracking face in the picture, and then to detection
The face that arrives carries out a series of the relevant technologies of face recognition, usually also referred to as Identification of Images, face recognition.Face recognition technology
It is the face feature based on people, the facial image or video flowing to input first determine whether that it whether there is face, if there is
Face then further provides the position of each face, the location information of size and each major facial organ.And according to these letters
Breath, further extracts the identity characteristic contained in each face, and it is compared with known face, so that identification is every
The identity of a face.
The research of face identification system starts from the 1960s, with computer technology and optical imagery skill after the eighties
The development of art is improved, and actually enters the primary application stage then 90 year later period, in recent years, with deep neural network
The fast development of technology has further pushed the commercialization of face recognition technology and product.
Since deep neural network can block posture, expression, decoration, illumination etc. has preferable robustness, therefore,
Deep neural network achieves excellent as a result, face identification system under substantially increasing complex environment in image recognition
Accuracy.But meanwhile deep learning algorithm is larger to training sample set scale requirements, trained fast convergence is more difficult.
Summary of the invention
The purpose of the present invention is to provide a kind of face identification methods based on depth residual error network, to solve above-mentioned background
The problem of being proposed in technology.
To achieve the above object, the invention provides the following technical scheme:
A kind of face identification method based on depth residual error network comprising the steps of:
A, image acquisition device acquires facial image and stores into memory；
B, face detection module detects human face region, while the key point of locating human face；
C, human face region image is cut out, and is aligned according to the key point of face；
D, human face region image is trained, calculates the average facial image in training set, facial image training
The every secondary human face region image concentrated, which subtracts, to carry out the training of network parameter after average facial image and obtains convolutional neural networks
Model；
E, the model feedforward network obtained using training, obtains the feature vector of facial image；
F, the distance between feature vector is calculated, classification is completed, completes the identification to face.
As further technical solution of the present invention: method for detecting human face in the step B the following steps are included: 1) for
Given input picture carries out image pyramid operation, the row of selected digital image, the minimum value of column, with the minimum frame width of 12/ detection
Obtained number is spent as scaling number, is selected 0.709 and is used as zoom factor, obtains the input figure of several different zoom ratios
Picture, the purpose of this step are that the detection of candidate frame can be carried out for different size face；2) the full convolution net of 3 3x3 is used
Network, the maximum pond layer of 3x3, candidate window and frame regression vector are generated using the network after the convolution of first 3x3；3) make
Candidate frame is obtained with threshold value 0.6, for these candidate frames by maximum restrainable algorithms, given threshold is 0.5 merging candidate frame；4) will
The point of these candidate circles is mapped to original without by having obtained the original coordinates of candidate frame in the figure of scale；5) make again
Candidate frame is merged with maximum restrainable algorithms, selecting threshold value is 0.7, finally obtains the candidate frame output of the first step；6)
Previous step is obtained all candidate frames to extract, and the regular size to 24*24, be input in convolutional neural networks, and root
According to the value of face classification, further candidate is screened, be similar to 3), 4), 5) in operation, target frame position is sieved
Choosing；7) it is similar to the operation of the 6) step, further target frame is screened, and obtains final output as a result, i.e. face frame,
The confidence level and face key point coordinate of face frame.
As further technical solution of the present invention: the method that face is aligned in the step C is the following steps are included: 1) root
Facial image is intercepted out according to the coordinate of Face datection；It 2) the use of affine transformation is new coordinate by former coordinate mapping transformation；3) basis
Coordinate transformation method, original image are transformed to new images, it is as regular after image.
As further technical solution of the present invention: the method for training pattern is the following steps are included: 1) right in the step D
Facial image is normalized and as trained input, image pixel value/127.5-1, and normalization section is [- 1,1], and 2) on
The output of one step is sent into core network and carries out convolution, and backbone network network structure connects the maximum pond of a 3x3 by the convolution of 3 3x3
Change layer, connect the convolution of a 1x1, connect a 3x3 convolution, connects a 3x3 convolution composition；3) residual error net is sent into the output of previous step
Network A carries out convolution, 4) output of previous step is sent into dimensionality reduction network A and carries out convolution and pond；5) residual error is sent into the output of previous step
Network B carries out convolution, 6) output of previous step is sent into dimensionality reduction network B and carries out convolution pond, 7) residual error is sent into the output of previous step
Network C carries out convolution, and 0 network of branch is 1x1 convolution, and 1 network of branch is 1x1 convolution, connects the convolution of a 1x3, meets a 3x1
Convolution, Liang Ge branch merges, then does residual error after connecing a 1x1 convolution；8) result output layer, including average pond layer, Dropout
Layer, full linking layer, the value that average pond layer output is 8 × 8 × 1792, Dropout layers of keep_probe is 0.8, full linking layer
Output is batch_size × 128.
As further technical solution of the present invention: the step F is specifically: 1) 128 dimensional feature vectors calculate cosine away from
From；2) COS distance value is mapped as [0,100] section；3) threshold value comparison of the value and setting that map completes identification.
As further technical solution of the present invention: the residual error network A structure is as follows: 0 network of branch is each 1x1 volumes
Product, 1 network of branch is 1x1 convolution, connects a 3x3 convolution, and 2 network of branch is 1x1 convolution, connects a 3x3 convolution, then connect one
3x3 convolution, 1x1 convolution after three branches merge do residual error after convolution.
As further technical solution of the present invention: the dimensionality reduction network A structure is as follows: 0 network of branch is 3x3, step-length
For 2 convolution, 1 network of branch is 1x1 convolution, connects a 3x3 convolution, connects a 3x3, the convolution that step-length is 2,2 network of branch
It is 3x3, the convolution that step-length is 2, three branches' merging.
As further technical solution of the present invention: the residual error network B structure is as follows: 0 network of branch is 1x1 convolution,
1 network of branch is 1x1 convolution, connects a 1x7 convolution, connects a 7x1 convolution, and 2 network of branch is 3x3, the convolution that step-length is 2,
Liang Ge branch merges, and does residual error after connecing a 1x1 convolution.
As further technical solution of the present invention: the dimensionality reduction network B structure is as follows: 0 network of branch is 3x3, step-length
For 2 maximum pond, 1 network of branch is 1x1 convolution, meets a 3x3, and the convolution that step-length is 2,2 network of branch is 1x1 convolution,
A 3x3 is met, the convolution that step-length is 2,3 network of branch is 1x1 convolution, connects a 3x3 convolution, meets a 3x3, and step-length is 2
Convolution, four branches merge.
As further technical solution of the present invention: the residual error network C structure is as follows: 0 network of branch is 1x1 convolution,
1 network of branch is 1x1 convolution, connects the convolution of a 1x3, connects a 3x1 convolution, Liang Ge branch merges, then connects a 1x1 convolution
After do residual error.
Compared with prior art, the invention has the following advantages that the present invention can be rapidly performed by recognition of face, and identify
Face result is more accurate, and can improve the accuracy in face recognition technology.
Detailed description of the invention
Fig. 1 is the whole flow chart of the present invention.
Specific embodiment
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, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Embodiment 1: referring to Fig. 1, a kind of face identification method based on depth residual error network comprising the steps of:
A, image acquisition device acquires facial image and stores into memory；
B, face detection module detects human face region, while the key point of locating human face；The following steps are included: 1) for giving
Fixed input picture carries out image pyramid operation, the row of selected digital image, the minimum value of column, with the minimum width of frame of 12/ detection
Obtained number selectes 0.709 and is used as zoom factor, obtain the input picture of several different zoom ratios as scaling number,
The purpose of this step is that the detection of candidate frame can be carried out for different size face；2) the full convolutional network of 3 3X3 is used,
The maximum pond layer of 3x3, candidate window and frame regression vector are generated using the network after the convolution of first 3x3；3) it uses
Threshold value 0.6 obtains candidate frame, and for these candidate frames by maximum restrainable algorithms, given threshold is 0.5 merging candidate frame；4) by this
The point of a little candidate's circles is mapped to original without by having obtained the original coordinates of candidate frame in the figure of scale；5) it reuses
Maximum restrainable algorithms merge candidate frame, and selecting threshold value is 0.7, finally obtain the candidate frame output of the first step；6) will
Previous step obtains all candidate frames and extracts, and the regular size to 24*24, is input in convolutional neural networks, and according to
The value of face classification, further screens candidate, be similar to 3), 4), 5) in operation, target frame position is sieved
Choosing；7) it is similar to the operation of the 6) step, further target frame is screened, and obtains final output as a result, i.e. face frame,
The confidence level and face key point coordinate of face frame.
C, human face region image is cut out, and is aligned according to the key point of face；The following steps are included: 1) according to people
The coordinate of face detection intercepts out facial image；It 2) the use of affine transformation is new coordinate by former coordinate mapping transformation；3) according to coordinate
Transform method, original image are transformed to new images, it is as regular after image.
D, human face region image is trained, calculates the average facial image in training set, facial image training
The every secondary human face region image concentrated, which subtracts, to carry out the training of network parameter after average facial image and obtains convolutional neural networks
Model；The following steps are included: 1) facial image is normalized and as trained input, image pixel value/127.5-1,
Normalizing section is [- 1,1], 2) output of previous step is sent into core network and carries out convolution, and backbone network network structure is by 3 3x3
Convolution, connect the maximum pond layer of a 3x3, connect the convolution of a 1x1, connect a 3x3 convolution, connect 3x3 convolution composition；
3) output of previous step is sent into residual error network A and carries out convolution, the 4) output of previous step is sent into dimensionality reduction network A and carries out convolution and pond
Change；5) output of previous step is sent into residual error network B and carries out convolution, the 6) output of previous step is sent into dimensionality reduction network B and carries out convolution pond
Change, 7) output of previous step is sent into residual error network C and carries out convolution, and 0 network of branch is 1x1 convolution, and 1 network of branch is 1x1 convolution,
The convolution for meeting a 1x3 connects a 3x1 convolution, and Liang Ge branch merges, then does residual error after connecing a 1x1 convolution；8) result exports
Layer, including average pond layer, Dropout layers, full linking layer, average pond layer output is 8 × 8 × 1792, Dropout layers
The value of keep_probe is 0.8, and full linking layer output is batch_size × 128.
E, the model feedforward network obtained using training, obtains the feature vector of facial image；
F, the distance between feature vector is calculated, classification is completed, completes the identification to face.Detailed process is as follows: 1) 128 dimension
Feature vector calculates COS distance；2) COS distance value is mapped as [0,100] section；3) the threshold value ratio of the value and setting that map
Compared with completion identification.
Embodiment 2, on the basis of embodiment 1, residual error network A structure are as follows: 0 network of branch is each 1x1 convolution, branch
1 network is 1x1 convolution, connects a 3x3 convolution, and 2 network of branch is 1x1 convolution, connects a 3x3 convolution, then connect one 3x3 volumes
It accumulates, 1x1 convolution after three branches merge, does residual error after convolution.Dimensionality reduction network A structure is as follows: 0 network of branch is 3x3, and step-length is
2 convolution, 1 network of branch are 1x1 convolution, connect a 3x3 convolution, meet a 3x3, the convolution that step-length is 2, and 2 network of branch is
3x3, the convolution that step-length is 2, three branches merge.Residual error network B structure is as follows: 0 network of branch is 1x1 convolution, 1 network of branch
It is 1x1 convolution, connects a 1x7 convolution, connect a 7x1 convolution, 2 network of branch is 3x3, and the convolution that step-length is 2, Liang Ge branch closes
And residual error is done after connecing a 1x1 convolution.Dimensionality reduction network B structure is as follows: 0 network of branch is 3x3, the maximum pond that step-length is 2,
1 network of branch is 1x1 convolution, meets a 3x3, the convolution that step-length is 2,2 network of branch is 1x1 convolution, connects a 3x3, step-length
For 2 convolution, 3 network of branch is 1x1 convolution, connects a 3x3 convolution, meets a 3x3, the convolution that step-length is 2, four branches close
And.Residual error network C structure is as follows: 0 network of branch is 1x1 convolution, and 1 network of branch is 1x1 convolution, connects the convolution of a 1x3, connects
One 3x1 convolution, Liang Ge branch merges, then does residual error after connecing a 1x1 convolution.
It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie
In the case where without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power
Benefit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent elements of the claims
Variation is included within the present invention.Any reference signs in the claims should not be construed as limiting the involved claims.
In addition, it should be understood that although this specification is described in terms of embodiments, but not each embodiment is only wrapped
Containing an independent technical solution, this description of the specification is merely for the sake of clarity, and those skilled in the art should
It considers the specification as a whole, the technical solutions in the various embodiments may also be suitably combined, forms those skilled in the art
The other embodiments being understood that.
Claims (10)
1. a kind of face identification method based on depth residual error network, which is characterized in that comprise the steps of:
A, image acquisition device acquires facial image and stores into memory；
B, face detection module detects human face region, while the key point of locating human face；
C, human face region image is cut out, and is aligned according to the key point of face；
D, human face region image is trained, the average facial image in training set is calculated, in facial image training set
Every secondary human face region image subtract and carry out the training of network parameter after average facial image and obtain the model of convolutional neural networks；
E, the model feedforward network obtained using training, obtains the feature vector of facial image；
F, the distance between feature vector is calculated, classification is completed, completes the identification to face.
2. a kind of face identification method based on depth residual error network according to claim 1, which is characterized in that the step
Method for detecting human face in rapid B carries out image pyramid operation the following steps are included: 1) for given input picture, selectes figure
The row of picture, the minimum value of column select 0.709 as scaling using the number that the minimum width of frame of 12/ detection obtains as scaling number
The factor, obtains the input picture of several different zoom ratios, and the purpose of this step is can to carry out for different size face
The detection of candidate frame；2) the full convolutional network of 3 3x3 is used, the maximum pond layer of 3x3, uses this after the convolution of first 3x3
Network generates candidate window and frame regression vector；3) candidate frame is obtained using threshold value 0.6, these candidate frames are pressed down by maximum
Algorithm processed, given threshold are 0.5 merging candidate frame；4) point of these candidate circles is mapped to original without by scale's
In figure, the original coordinates of candidate frame have been obtained；5) maximum restrainable algorithms are reused to merge candidate frame, select threshold value
It is 0.7, finally obtains the candidate frame output of the first step；6) previous step is obtained all candidate frames to extract, and regular arrived
The size of 24x24, is input in convolutional neural networks, and according to the value of face classification, further screens to candidate, similar
In 3), 4) operation in, 5), screens target frame position；7) be similar to the operation of the 6) step, further to target frame into
Row screening, and final output is obtained as a result, i.e. face frame, the confidence level and face key point coordinate of face frame.
3. a kind of face identification method based on depth residual error network according to claim 1, which is characterized in that the step
In rapid C 1) method of face alignment is the following steps are included: intercept out facial image according to the coordinate of Face datection；2) using affine
Former coordinate mapping transformation is new coordinate by transformation；3) according to coordinate transformation method, original image is transformed to new images, after as regular
Image.
4. a kind of face identification method based on depth residual error network according to claim 1, which is characterized in that the step
In rapid D 1) method of training pattern is the following steps are included: be normalized facial image and as trained input, image slices
Element value/127.5-1, normalization section are [- 1,1], 2) output of previous step is sent into core network and carries out convolution, backbone network network
Structure is connect the maximum pond layer of a 3x3, connects the convolution of a 1x1, connect a 3x3 convolution, connect one by the convolution of 3 3x3
3x3 convolution composition；3) output of previous step is sent into residual error network A and carries out convolution, 4) output of previous step be sent into dimensionality reduction network A into
Row convolution and pond；5) dimensionality reduction network B is sent into the output of previous step is sent into residual error network B and carries out convolution, the 6) output of previous step
Carry out convolution pond, 7) output of previous step is sent into residual error network C and carries out convolution, and 0 network of branch is 1x1 convolution, 1 network of branch
It is 1x1 convolution, connects the convolution of a 1x3, connects a 3x1 convolution, Liang Ge branch merges, then does residual error after connecing a 1x1 convolution；
8) result output layer, including average pond layer, Dropout layers, full linking layer, average pond layer output is 8 × 8 × 1792,
The value of Dropout layers of keep_probe is 0.8, and full linking layer output is batch_size × 128.
5. a kind of face identification method based on depth residual error network according to claim 1, which is characterized in that the step
Rapid F is specifically: 1) 128 dimensional feature vectors calculate COS distance；2) COS distance value is mapped as [0,100] section；3) it maps
Value and setting threshold value comparison, complete identification.
6. a kind of face identification method based on depth residual error network according to claim 4, which is characterized in that described residual
Poor network A structure is as follows: 0 network of branch is 1x1 convolution, and 1 network of branch is 1x1 convolution, connects a 3x3 convolution, 2 network of branch
It is 1x1 convolution, connects a 3x3 convolution, then connect a 3x3 convolution, 1x1 convolution after three branches merge does residual error after convolution.
7. a kind of face identification method based on depth residual error network according to claim 4, which is characterized in that the drop
Dimension network A structure is as follows: 0 network of branch is 3x3, and the convolution that step-length is 2,1 network of branch is 1x1 convolution, connects one 3x3 volumes
Product, meets a 3x3, the convolution that step-length is 2, and 2 network of branch is 3x3, the convolution that step-length is 2, three branches' merging.
8. a kind of face identification method based on depth residual error network according to claim 4, which is characterized in that described residual
Poor network B structure is as follows: 0 network of branch is 1x1 convolution, and 1 network of branch is 1x1 convolution, connects a 1x7 convolution, meets a 7x1
Convolution, 2 network of branch are 3x3, and the convolution that step-length is 2, Liang Ge branch merges, and do residual error after connecing a 1x1 convolution.
9. a kind of face identification method based on depth residual error network according to claim 4, which is characterized in that the drop
Dimension network B structure is as follows: 0 network of branch is 3x3, and the maximum pond that step-length is 2,1 network of branch is 1x1 convolution, connects one
3x3, the convolution that step-length is 2,2 network of branch is 1x1 convolution, meets a 3x3, the convolution that step-length is 2,3 network of branch is 1x1 volumes
Product, connects a 3x3 convolution, meets a 3x3, the convolution that step-length is 2, and four branches merge.
10. a kind of face identification method based on depth residual error network according to claim 4, which is characterized in that described
Residual error network C structure is as follows: 0 network of branch is 1x1 convolution, and 1 network of branch is 1x1 convolution, connects the convolution of a 1x3, connects one
A 3x1 convolution, Liang Ge branch merges, then does residual error after connecing a 1x1 convolution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910652309.1A CN110263768A (en) | 2019-07-19 | 2019-07-19 | A kind of face identification method based on depth residual error network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910652309.1A CN110263768A (en) | 2019-07-19 | 2019-07-19 | A kind of face identification method based on depth residual error network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110263768A true CN110263768A (en) | 2019-09-20 |
Family
ID=67927131
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910652309.1A Pending CN110263768A (en) | 2019-07-19 | 2019-07-19 | A kind of face identification method based on depth residual error network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110263768A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110781784A (en) * | 2019-10-18 | 2020-02-11 | 高新兴科技集团股份有限公司 | Face recognition method, device and equipment based on double-path attention mechanism |
| CN111241998A (en) * | 2020-01-09 | 2020-06-05 | 中移（杭州）信息技术有限公司 | Face recognition method and device, electronic equipment and storage medium |
| CN111400535A (en) * | 2020-03-11 | 2020-07-10 | 广东宜教通教育有限公司 | Lightweight face recognition method, system, computer device and storage medium |
| CN111488811A (en) * | 2020-03-31 | 2020-08-04 | 长沙千视通智能科技有限公司 | Face recognition method and device, terminal equipment and computer readable medium |
| CN112304435A (en) * | 2020-10-10 | 2021-02-02 | 广州中大数字家庭工程技术研究中心有限公司 | Human body thermal imaging temperature measurement method combining face recognition |
| CN112766167A (en) * | 2021-01-20 | 2021-05-07 | 宁夏计算机软件与技术服务有限公司 | Face recognition method based on face feature vector |
| CN113239858A (en) * | 2021-05-28 | 2021-08-10 | 西安建筑科技大学 | Face detection model training method, face recognition method, terminal and storage medium |
| CN113239728A (en) * | 2021-04-03 | 2021-08-10 | 国家计算机网络与信息安全管理中心 | Method for accelerating internet video identification algorithm |
| CN116524327A (en) * | 2023-06-25 | 2023-08-01 | 云账户技术(天津)有限公司 | Training method and device of face recognition model, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106203395A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | Face character recognition methods based on the study of the multitask degree of depth |
| CN106557743A (en) * | 2016-10-26 | 2017-04-05 | 桂林电子科技大学 | A kind of face characteristic extraction system and method based on FECNN |
| CN107832667A (en) * | 2017-10-11 | 2018-03-23 | 哈尔滨理工大学 | A kind of face identification method based on deep learning |
| US20180137388A1 (en) * | 2016-11-14 | 2018-05-17 | Samsung Electronics Co., Ltd. | Method and apparatus for analyzing facial image |
| CN109784219A (en) * | 2018-12-28 | 2019-05-21 | 广州海昇计算机科技有限公司 | A kind of face identification method, system and device based on concentration cooperated learning |
| CN109948573A (en) * | 2019-03-27 | 2019-06-28 | 厦门大学 | A kind of noise robustness face identification method based on cascade deep convolutional neural networks |
-
2019
- 2019-07-19 CN CN201910652309.1A patent/CN110263768A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106203395A (en) * | 2016-07-26 | 2016-12-07 | 厦门大学 | Face character recognition methods based on the study of the multitask degree of depth |
| CN106557743A (en) * | 2016-10-26 | 2017-04-05 | 桂林电子科技大学 | A kind of face characteristic extraction system and method based on FECNN |
| US20180137388A1 (en) * | 2016-11-14 | 2018-05-17 | Samsung Electronics Co., Ltd. | Method and apparatus for analyzing facial image |
| CN107832667A (en) * | 2017-10-11 | 2018-03-23 | 哈尔滨理工大学 | A kind of face identification method based on deep learning |
| CN109784219A (en) * | 2018-12-28 | 2019-05-21 | 广州海昇计算机科技有限公司 | A kind of face identification method, system and device based on concentration cooperated learning |
| CN109948573A (en) * | 2019-03-27 | 2019-06-28 | 厦门大学 | A kind of noise robustness face identification method based on cascade deep convolutional neural networks |
Non-Patent Citations (3)
| Title |
|---|
| CHRISTIAN SZEGEDY等: "Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning", 《HTTPS://ARXIV.ORG/PDF/1602.07261.PDF》 * |
| CHRISTIAN SZEGEDY等: "Rethinking the Inception Architecture for Computer Vision", 《HTTPS://ARXIV.ORG/PDF/1512.00567V3.PDF》 * |
| KAIPENG ZHANG等: "Joint Face Detection and Alignment Using Multitask Cascaded Convolutional Networks", 《 IEEE SIGNAL PROCESSING LETTERS》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110781784A (en) * | 2019-10-18 | 2020-02-11 | 高新兴科技集团股份有限公司 | Face recognition method, device and equipment based on double-path attention mechanism |
| CN111241998A (en) * | 2020-01-09 | 2020-06-05 | 中移（杭州）信息技术有限公司 | Face recognition method and device, electronic equipment and storage medium |
| CN111241998B (en) * | 2020-01-09 | 2023-04-28 | 中移（杭州）信息技术有限公司 | Face recognition method, device, electronic equipment and storage medium |
| CN111400535A (en) * | 2020-03-11 | 2020-07-10 | 广东宜教通教育有限公司 | Lightweight face recognition method, system, computer device and storage medium |
| CN111488811A (en) * | 2020-03-31 | 2020-08-04 | 长沙千视通智能科技有限公司 | Face recognition method and device, terminal equipment and computer readable medium |
| CN111488811B (en) * | 2020-03-31 | 2023-08-22 | 长沙千视通智能科技有限公司 | Face recognition method, device, terminal equipment and computer readable medium |
| CN112304435A (en) * | 2020-10-10 | 2021-02-02 | 广州中大数字家庭工程技术研究中心有限公司 | Human body thermal imaging temperature measurement method combining face recognition |
| CN112766167A (en) * | 2021-01-20 | 2021-05-07 | 宁夏计算机软件与技术服务有限公司 | Face recognition method based on face feature vector |
| CN113239728A (en) * | 2021-04-03 | 2021-08-10 | 国家计算机网络与信息安全管理中心 | Method for accelerating internet video identification algorithm |
| CN113239858A (en) * | 2021-05-28 | 2021-08-10 | 西安建筑科技大学 | Face detection model training method, face recognition method, terminal and storage medium |
| CN116524327A (en) * | 2023-06-25 | 2023-08-01 | 云账户技术(天津)有限公司 | Training method and device of face recognition model, electronic equipment and storage medium |
| CN116524327B (en) * | 2023-06-25 | 2023-08-25 | 云账户技术(天津)有限公司 | Training method and device of face recognition model, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110263768A (en) | A kind of face identification method based on depth residual error network | |
| CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
| CN112766160B (en) | Face replacement method based on multi-stage attribute encoder and attention mechanism | |
| CN110909690B (en) | Method for detecting occluded face image based on region generation | |
| CN110147721B (en) | Three-dimensional face recognition method, model training method and device | |
| CN108717531B (en) | Human body posture estimation method based on Faster R-CNN | |
| CN112241731B (en) | Attitude determination method, device, equipment and storage medium | |
| CN105740780B (en) | Method and device for detecting living human face | |
| CN104601964B (en) | Pedestrian target tracking and system in non-overlapping across the video camera room of the ken | |
| CN107852533A (en) | Three-dimensional content generating means and its three-dimensional content generation method | |
| CN110532897A (en) | The method and apparatus of components image recognition | |
| CN108960404B (en) | Image-based crowd counting method and device | |
| JP2004086891A (en) | Object detection method in digital image | |
| CN111814661A (en) | Human behavior identification method based on residual error-recurrent neural network | |
| CN113240691A (en) | Medical image segmentation method based on U-shaped network | |
| CN109740572A (en) | A kind of human face in-vivo detection method based on partial color textural characteristics | |
| CN107766864B (en) | Method and device for extracting features and method and device for object recognition | |
| CN110363770A (en) | A kind of training method and device of the infrared semantic segmentation model of margin guide formula | |
| CN106485186A (en) | Image characteristic extracting method, device, terminal device and system | |
| CN109886153A (en) | A kind of real-time face detection method based on depth convolutional neural networks | |
| CN112836625A (en) | Face living body detection method and device and electronic equipment | |
| CN109117753A (en) | Position recognition methods, device, terminal and storage medium | |
| CN109325408A (en) | A kind of gesture judging method and storage medium | |
| CN110276831A (en) | Constructing method and device, equipment, the computer readable storage medium of threedimensional model | |
| CN108229432A (en) | Face calibration method 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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190920 |