CN107766811A - A kind of face identification method and system based on complicated flow structure - Google Patents
A kind of face identification method and system based on complicated flow structure Download PDFInfo
- Publication number
- CN107766811A CN107766811A CN201710935978.0A CN201710935978A CN107766811A CN 107766811 A CN107766811 A CN 107766811A CN 201710935978 A CN201710935978 A CN 201710935978A CN 107766811 A CN107766811 A CN 107766811A
- Authority
- CN
- China
- Prior art keywords
- face
- convolutional neural
- neural networks
- feat
- flow structure
- 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
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/94—Hardware or software architectures specially adapted for image or video understanding
-
- 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
Abstract
The invention discloses a kind of face identification method and system based on complicated flow structure, this method includes the steps such as target Face datection, the feature extraction of target face and facial feature database are established, face characteristic compares and judges.Picture to be detected is the picture with redundancy of a higher-dimension, detect target face and obtain accurate face location and key point using improved MTCNN, convolutional neural networks based on a kind of training of flow structure, the face detected is mapped to comparison basis feature of the characteristic vector as this face for the low-dimensional that Euclidean space can divide, enter when there is new face information, the European L2 distances of the feature and existing reference characteristic of new face are compared, judge face ownership.The present invention departs from the convolutional neural networks framework of server end, based on C++ related libraries, Armadillo matrixes storehouse, face identification system is constructed on RK3288 embedded platforms, optimized rear system frame per second reaches as high as 15FPS, and the use that convolutional neural networks are promoted for reality is made that certain contribution.
Description
Technical field
The present invention relates to machine learning, deep neural network scientific research field, while the application of industrial quarters is also paid close attention to, especially related to
And a kind of face identification method and system based on complicated flow structure.
Background technology
Classical Face datection algorithm is the problems such as being primarily present missing inspection, false retrieval based on the artificial characteristic Design such as Harr,
Multi-pose Face Detection results are worse.Achieved in recent years using the method for deep neural network in the performance of Face datection greatly
The raising of amplitude.Deep neural network thinks that sample data is formed by the combinations of features of level, only utilizes k sample
Can makees O (2^k) individual division to sample space, and model complexity is higher.Utilize the powerful feature representation abilities of CNN, structure
Multilayer convolutional neural networks, successively extract the low-level image features such as edge, topography's block, the face component with obvious semantic feature
Deng.In order to improve speed, first image is done down-sampled, reduce candidate face number rapidly using the image of low resolution；Then
Continuous improvement divides ratio, further excludes empty inspection, finally obtains testing result.
The content of the invention
It is an object of the invention to for there is the structure of the higher-dimension human face data of High redundancy and manifold possessed by it, carry
A kind of face identification method based on complicated flow structure has been supplied, and the method is realized on the hardware of miniaturization.
To realize above-mentioned technical purpose, the technical scheme specifically used is as follows：A kind of face based on complicated flow structure
Recognition methods, this method comprise the following steps：
(1) picture to be identified is inputted, IMG is designated as and is sent into human-face detector；
(2) target Face datection：Face datection is carried out to IMG using improved MTCNN algorithms, if not detected in IMG
To face, then the next width images to be recognized of return to step (1) input, if detecting face A from IMG, marks its face
Frame Rect (x, y, h, w), x, y, h, w are the horizontal stroke of the anchor point (can be any one angle point or central point) of face frame respectively
(key point can be selected from the right side for the length and width of ordinate and face frame, and key point position Point (p1 ..., pm), wherein m >=2
Eye, left eye, nose, the right corners of the mouth, left corners of the mouth position etc.)；Then face frame Rect is intercepted out, obtains face Face1；It is described
Improved MTCNN algorithms depart from the convolutional neural networks framework (such as Caffe, TensorFlow etc.) of server end, based on C++
Related libraries and Armadillo matrixes accelerate storehouse to realize the feedforward network of convolutional neural networks, more fast and effeciently realize face
Critical point detection；
(3) target face characteristic and facial feature database are established：By key point Point and the face Face1 mono- of interception
Rise and be sent into the convolutional neural networks based on flow structure, supervisory signals e-learning can divide in Euclidean space by classifying
Low-dimensional (can select 320 dimensions, 1024 dimension etc.) face feature vector Feat (1)；Face feature vector Feat (1) is labeled as
The benchmark for belonging to A compares feature, is stored in facial feature database, final to establish the face spy for having some face feature vectors
Database Feat (1 ... n) is levied, and verification threshold Threshold is set；The convolutional neural networks of the flow structure are in ResNet
On the basis of use model compression technology, reduce the parameter of network, realize the acceleration of network；
(4) face characteristic is compared and judged：Image to be verified is read, according to the method in step (2), if detecting people
Face, then key point Point corresponding to the face and the face Face2 intercepted by face frame are sent into the convolution of same step (3)
Neutral net, obtain and step (3) dimension identical face feature vector Feat (n+1)；Directly compare Feat (n+1) and face
The European L2 distances of all face feature vectors in property data base Feat (1 ... n)；Take one that contrast conting result is minimum
, if this is less than verification threshold Threshold, then it is assumed that Face2 is registered in facial feature database, by verifying,
Otherwise refuse inspection of books.
Further, in the step (1), the acquisition modes of image are：Read from video flowing, from existing face number
According to being read in storehouse, or gathered by camera；In the step (4), before comparison, in addition to face feature vector
Feat (n+1) and facial feature database Feat (1 ... n) carries out the step of dimensionality reduction, for example drops to 32 dimensions or 64 dimensions, with further
Reduce amount of calculation.
A kind of face identification system based on complicated flow structure, the neutral net detection of the high-accuracy of complexity is used
And recognition methods, and the feedforward network after accelerating is realized, can not possibly be small before being realized on low profile edge platform
The algorithm realized in type hardware plate；The system is realized on RK3288 series embedded platforms, realizes that GUI interacts boundary using QT
Face；RK3288 series embedded platform is under 1.61GHz dominant frequency, and highest frame per second is 15FPS, it is necessary to carry out face inspection when idle
Highest frame per second is 10FPS during survey, meets recognition of face demand in reality scene；The system includes image collection module, face is examined
Survey module, face characteristic acquisition module, face alignment module and facial feature database：
The image input face characteristic of collection is obtained mould by described image acquisition module by camera collection image IMG
Block；
The face detection module carries out Face datection using improved MTCNN algorithms to IMG, if not detected in IMG
To face, image collection module is re-called, if detecting face A, marks its face frame Rect and key point Point,
The face Face1 intercepted by face frame Rect is input to face characteristic acquisition module；The improved MTCNN algorithms depart from
The convolutional neural networks framework of server end, break away from and relied on using numerous and diverse software of deep learning framework, based on C++ related libraries
Accelerate storehouse to realize the feedforward network of convolutional neural networks with Armadillo matrixes, and do not need the calculating of backpropagation, can put
The files of framework bulk redundancy such as de- Caffe and the internal memory taken；Secondly, by reducing MTCNN decision thresholds, fortune can be reduced
Calculation amount, more fast and effeciently realize face critical point detection；
Face A key points Point and face Face1 are sent into based on flow structure by the face characteristic acquisition module together
Convolutional neural networks, by supervisory signals e-learning of classifying to the low-dimensional face feature vector that can divide in Euclidean space
Feat(1)；If performing registration operation, face feature vector is stored in facial feature database, if performing verification operation,
Call face alignment module；
The face alignment module is calculated in face feature vector and facial feature database corresponding to image to be verified
The European L2 distances of all people's face characteristic vector, one apart from result of calculation minimum is taken, if this is less than verification threshold
Threshold, then it is assumed that the image is registered in facial feature database, by checking, otherwise refuses inspection of books；
The face feature vector of all registrations is stored in the facial feature database, and verification threshold is set
Threshold。
The beneficial effects of the invention are as follows：The inventive method includes images to be recognized input, target Face datection, target face
The steps such as feature extraction and facial feature database are established, face characteristic compares and judges.Picture to be detected is a higher-dimension
Picture with redundancy, detection target face obtain accurate face location and key point, target using improved MTCNN
Convolutional neural networks of the face characteristic extraction based on a kind of training of flow structure, the face detected is mapped into Euclidean space can
The characteristic vector for the low-dimensional divided, as the comparison basis feature of this face, finally when there is new face information to enter, compare new person
The European L2 distances of the feature of face and existing reference characteristic, judge face ownership.Present invention uses the people of mass data training
Face detects and the deep neural network of recognition of face, and realizes the feedforward network of acceleration again；The nerve net of complexity is used
Network method carries out the detection and identification of face, while uses model compression technology and related C++ related libraries to cause the standard of algorithm
True rate is improved significantly.The present invention departs from convolutional neural networks framework Caffe, TensorFlow of server end etc., base
In C++ related libraries, Armadillo matrixes storehouse, a people is constructed on RK3288 embedded platforms using above-mentioned network model
Face identifying system, the algorithm that can not possibly be realized before being realized on low profile edge platform in hardware plate is minimized, through excellent
System frame per second reaches as high as 15FPS after change, and the use that convolutional neural networks are promoted for reality is made that certain contribution.
Brief description of the drawings
Fig. 1 is extraction face characteristic flow chart；
Fig. 2 is that the similarity score of two sample images of same person is calculated compared with；
Fig. 3 is face identification system schematic diagram；
Fig. 4 is the actual software interface of face identification system；
Fig. 5 is the database interface of face identification system；
Fig. 6 is the recognition effect displaying of face identification system.
Embodiment
The present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings.
A kind of face identification method based on complicated flow structure of the present invention, is mainly included the following steps that：
(1) picture to be identified is inputted, IMG is designated as and is sent into human-face detector；
(2) target Face datection：Face datection is carried out to IMG using improved MTCNN algorithms, if not detected in IMG
To face, then the next width images to be recognized of return to step (1) input, if detecting face A from IMG, marks its face
Frame Rect (x, y, h, w), x, y, h, w are the horizontal stroke of the anchor point (can be any one angle point or central point) of face frame respectively
(key point can be selected from the right side for the length and width of ordinate and face frame, and key point position Point (p1 ..., pm), wherein m >=2
Eye, left eye, nose, the right corners of the mouth, left corners of the mouth position etc.)；Then face frame Rect is intercepted out, obtains face Face1；It is described
Improved MTCNN algorithms depart from the convolutional neural networks framework (such as Caffe, TensorFlow etc.) of server end, based on C++
Related libraries and Armadillo matrixes accelerate storehouse to realize the feedforward network of convolutional neural networks, more fast and effeciently realize face
Critical point detection；
(3) target face characteristic and facial feature database are established：By key point Point and the face Face1 mono- of interception
Rise and be sent into the convolutional neural networks based on flow structure, supervisory signals e-learning can divide in Euclidean space by classifying
Low-dimensional (can select 320 dimensions, 1024 dimension etc.) face feature vector Feat (1)；Face feature vector Feat (1) is labeled as
The benchmark for belonging to A compares feature, is stored in facial feature database, final to establish the face spy for having some face feature vectors
Database Feat (1 ... n) is levied, and verification threshold Threshold is set；The convolutional neural networks of the flow structure are in ResNet
On the basis of use model compression technology, reduce the parameter of network, realize the acceleration of network；
(4) face characteristic is compared and judged：Image to be verified is read, according to the method in step (2), if detecting people
Face, then key point Point corresponding to the face and the face Face2 intercepted by face frame are sent into the convolution of same step (3)
Neutral net, obtain and step (3) dimension identical face feature vector Feat (n+1)；Directly compare Feat (n+1) and face
The European L2 distances of all face feature vectors in property data base Feat (1 ... n)：
Wherein, xiRepresent the characteristic vector that the frame currently in internal memory extracts, yiBelong to everyone in representation database
Face feature vector, n is characteristic vector length.In order to reduce amount of calculation, first carry out PCA dimensionality reductions and feature is dropped into 32 dimensions, then
Distance is calculated, one of contrast conting result minimum is finally taken, compared with the verification threshold Threshold of setting：
Think that Face2 is registered in database if this is less than verification threshold, and show corresponding during its registration
Name, complete to log in, if being higher than verification threshold, then it is assumed that Face2 does not have registration in database, refuses certification.
As shown in figure 4, being real-time video picture on the left of the interface of face identification system, when detecting face, can show
The position of face frame and 5 face key points.Now user presses " registration ", then the face FaceX in rectangle frame is individually cut
Go out, and the convolutional neural networks acquisition Euclidean space based on manifold structure is sent into key point position and interception face in the lump to divide
320 dimensional feature FeatX, preserve into database (see Fig. 5).It is registered in meeting preloading data storehouse when reloading program
All people's face data, then when user presses " login ", face FaceY is obtained again by cutting operation, is sent into convolution god
FeatY is obtained through network, finally by PCA dimensionality reductions, FeatY and all registered faces in database distance is calculated, takes meter
One of calculation result minimum is compared with verification threshold, if being less than verification threshold, by verifying and showing name (see figure
6), if being higher than verification threshold, refuse inspection of books.
Claims (3)
1. a kind of face identification method based on complicated flow structure, it is characterised in that this method comprises the following steps：
(1) picture to be identified is inputted, IMG is designated as and is sent into human-face detector；
(2) target Face datection：Face datection is carried out to IMG using improved MTCNN algorithms, if not detecting people in IMG
Face, then the next width images to be recognized of return to step (1) input, if detecting face A from IMG, marks its face frame
Rect (x, y, h, w), x, y, h, w are the transverse and longitudinal coordinate of the anchor point of face frame and the length and width of face frame respectively, and key point
Position Point (p1 ..., pm), wherein m >=2；Then face frame Rect is intercepted out, obtains face Face1；The improvement
MTCNN algorithms depart from the convolutional neural networks framework of server end, storehouse is accelerated based on C++ related libraries and Armadillo matrixes
The feedforward network of convolutional neural networks is realized, more fast and effeciently realizes face critical point detection；
(3) target face characteristic and facial feature database are established：Key point Point and interception face Face1 are sent together
Enter the convolutional neural networks based on flow structure, it is low to that can divide in Euclidean space by supervisory signals e-learning of classifying
Tie up face feature vector Feat (1)；The benchmark that face feature vector Feat (1) is labeled as belonging to A compares feature, is stored to face
It is final to establish the facial feature database Feat (1 ... n) for having some face feature vectors in property data base, and set and test
Demonstrate,prove threshold value Threshold；The convolutional neural networks of the flow structure use model compression technology on the basis of ResNet, reduce
The parameter of network, realize the acceleration of network；
(4) face characteristic is compared and judged：Image to be verified is read, according to the method in step (2), if detecting face,
Key point Point corresponding to the face and the face Face2 intercepted by face frame are then sent into the convolutional Neural of same step (3)
Network, obtain and step (3) dimension identical face feature vector Feat (n+1)；Directly compare Feat (n+1) and face characteristic
The European L2 distances of all face feature vectors in database Feat (1 ... n)；One that contrast conting result is minimum is taken, if
This is less than verification threshold Threshold, then it is assumed that Face2 is registered in facial feature database, by checking, otherwise refuses
Checking absolutely.
A kind of 2. face identification method based on complicated flow structure according to claim 1, it is characterised in that the step
Suddenly in (1), the acquisition modes of image are：Read from video flowing, read from existing face database, or pass through shooting
Head collection；In the step (4), before comparison, in addition to face feature vector Feat (n+1) and facial feature database
Feat (1 ... n) carries out the step of dimensionality reduction, further to reduce amount of calculation.
3. a kind of face identification system based on complicated flow structure, it is characterised in that the system is embedded in RK3288 series
Realized on platform, GUI interactive interfaces are realized using QT；RK3288 series embedded platform is under 1.61GHz dominant frequency, when idle
Highest frame per second is 15FPS, it is necessary to highest frame per second is 10FPS when carrying out Face datection, meets recognition of face demand in reality scene；
The system includes image collection module, face detection module, face characteristic acquisition module, face alignment module and face characteristic number
According to storehouse：
The image of collection is inputted face characteristic acquisition module by described image acquisition module by camera collection image IMG；
The face detection module carries out Face datection using improved MTCNN algorithms to IMG, if not detecting people in IMG
Face, image collection module is re-called, if detecting face A, mark its face frame Rect and key point Point, will be logical
The face Face1 for crossing face frame Rect interceptions is input to face characteristic acquisition module；The improved MTCNN algorithms depart from service
The convolutional neural networks framework at device end, break away from and relied on using the numerous and diverse software of deep learning framework, based on C++ related libraries and
Realize the feedforward network of convolutional neural networks in Armadillo matrixes acceleration storehouse, it is not necessary to the calculating of backpropagation；Pass through reduction
MTCNN decision thresholds reduce operand, more fast and effeciently realize face critical point detection；
Face A key points Point and face Face1 are sent into the volume based on flow structure by the face characteristic acquisition module together
Product neutral net, by supervisory signals e-learning of classifying to the low-dimensional face feature vector Feat that can divide in Euclidean space
(1)；If performing registration operation, face feature vector is stored in facial feature database, if performing verification operation, called
Face alignment module；
The face alignment module calculates face feature vector corresponding to image to be verified with owning in facial feature database
Face feature vector European L2 distances, take apart from minimum one of result of calculation, if this is less than verification threshold
Threshold, then it is assumed that the image is registered in facial feature database, by checking, otherwise refuses inspection of books；
The face feature vector of all registrations is stored in the facial feature database, and verification threshold Threshold is set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710935978.0A CN107766811A (en) | 2017-10-10 | 2017-10-10 | A kind of face identification method and system based on complicated flow structure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710935978.0A CN107766811A (en) | 2017-10-10 | 2017-10-10 | A kind of face identification method and system based on complicated flow structure |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107766811A true CN107766811A (en) | 2018-03-06 |
Family
ID=61267530
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710935978.0A Pending CN107766811A (en) | 2017-10-10 | 2017-10-10 | A kind of face identification method and system based on complicated flow structure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107766811A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109472289A (en) * | 2018-10-09 | 2019-03-15 | 北京陌上花科技有限公司 | Critical point detection method and apparatus |
| CN109472193A (en) * | 2018-09-21 | 2019-03-15 | 北京飞搜科技有限公司 | Method for detecting human face and device |
| CN109685044A (en) * | 2019-02-18 | 2019-04-26 | 上海德拓信息技术股份有限公司 | Recognition of face search method based on k means clustering algorithm |
| CN110287955A (en) * | 2019-06-05 | 2019-09-27 | 北京字节跳动网络技术有限公司 | Target area determines model training method, device and computer readable storage medium |
| CN110503059A (en) * | 2019-08-27 | 2019-11-26 | 国网电子商务有限公司 | A kind of face identification method and system |
| CN110598627A (en) * | 2019-09-11 | 2019-12-20 | 旭辉卓越健康信息科技有限公司 | GUI demonstration tool based on face recognition |
| CN110659570A (en) * | 2019-08-21 | 2020-01-07 | 北京地平线信息技术有限公司 | Target object posture tracking method, and neural network training method and device |
| TWI705377B (en) * | 2019-02-01 | 2020-09-21 | 緯創資通股份有限公司 | Hardware boost method and hardware boost system |
| CN112236777A (en) * | 2018-06-11 | 2021-01-15 | 思科技术公司 | Vector-based object recognition in hybrid clouds |
| CN116723292A (en) * | 2023-05-12 | 2023-09-08 | 北京宝通利达科技有限公司 | Monitoring system based on Internet of things |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101329722A (en) * | 2007-06-21 | 2008-12-24 | 上海北控智能科技有限公司 | Human face recognition method for performing recognition algorithm based on neural network |
| CN101526997A (en) * | 2009-04-22 | 2009-09-09 | 无锡名鹰科技发展有限公司 | Embedded infrared face image identifying method and identifying device |
| CN105809125A (en) * | 2016-03-06 | 2016-07-27 | 北京工业大学 | Multi-core ARM platform based human face recognition system |
| CN106022317A (en) * | 2016-06-27 | 2016-10-12 | 北京小米移动软件有限公司 | Face identification method and apparatus |
| CN106650694A (en) * | 2016-12-30 | 2017-05-10 | 江苏四点灵机器人有限公司 | Human face recognition method taking convolutional neural network as feature extractor |
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
-
2017
- 2017-10-10 CN CN201710935978.0A patent/CN107766811A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101329722A (en) * | 2007-06-21 | 2008-12-24 | 上海北控智能科技有限公司 | Human face recognition method for performing recognition algorithm based on neural network |
| CN101526997A (en) * | 2009-04-22 | 2009-09-09 | 无锡名鹰科技发展有限公司 | Embedded infrared face image identifying method and identifying device |
| CN105809125A (en) * | 2016-03-06 | 2016-07-27 | 北京工业大学 | Multi-core ARM platform based human face recognition system |
| CN106022317A (en) * | 2016-06-27 | 2016-10-12 | 北京小米移动软件有限公司 | Face identification method and apparatus |
| CN106650694A (en) * | 2016-12-30 | 2017-05-10 | 江苏四点灵机器人有限公司 | Human face recognition method taking convolutional neural network as feature extractor |
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
Non-Patent Citations (7)
| Title |
|---|
| GOODLUCKCWL: "学习Caffe（三）Caffe卷积实现解析并手写前馈网络Armadillo+Opencv", 《HTTP：//BLOG.CSDN.NET/U014230646/ARTICLE/DETAILS/52664915》 * |
| HIROKI: "C++矩阵运算库Armadillo配置笔记", 《HTTPS://WWW.BBSMAX.COM/A/VX5M0NPLJN/》 * |
| KAIPENG ZHANG 等: "Joint Face Detection and Aligment using Multi-task Cascaded Convolutional Networks", 《IEEE SIGNAL PROCESSING LETTERS》 * |
| WHRSUNNY: "C++下媲美MATLAB矩阵运算的Armadillo库", 《BLOG.SCIENCENET.CN/BLOG-579537-898250.HTML》 * |
| 姜媚: "干货｜CNN 模型压缩与加速算法综述", 《HTTPS://WWW.SOHU.COM/A/167703589_642762》 * |
| 清水汪汪: "深度学习网络压缩模型方法总结(model compression)", 《HTTPS://WWW.CNBLOGS.COM/ZHONGHUASONG/P/7493475.HTML》 * |
| 百度NLP: "百度NLP | 神经网络模型压缩技术", 《HTTPS://WWW.SOHU.COM/A/152691907_465975》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112236777A (en) * | 2018-06-11 | 2021-01-15 | 思科技术公司 | Vector-based object recognition in hybrid clouds |
| CN109472193A (en) * | 2018-09-21 | 2019-03-15 | 北京飞搜科技有限公司 | Method for detecting human face and device |
| CN109472289A (en) * | 2018-10-09 | 2019-03-15 | 北京陌上花科技有限公司 | Critical point detection method and apparatus |
| CN109472289B (en) * | 2018-10-09 | 2022-03-29 | 北京陌上花科技有限公司 | Key point detection method and device |
| TWI705377B (en) * | 2019-02-01 | 2020-09-21 | 緯創資通股份有限公司 | Hardware boost method and hardware boost system |
| CN109685044A (en) * | 2019-02-18 | 2019-04-26 | 上海德拓信息技术股份有限公司 | Recognition of face search method based on k means clustering algorithm |
| CN109685044B (en) * | 2019-02-18 | 2023-06-06 | 上海德拓信息技术股份有限公司 | Face recognition retrieval method based on k-means clustering algorithm |
| CN110287955A (en) * | 2019-06-05 | 2019-09-27 | 北京字节跳动网络技术有限公司 | Target area determines model training method, device and computer readable storage medium |
| CN110659570A (en) * | 2019-08-21 | 2020-01-07 | 北京地平线信息技术有限公司 | Target object posture tracking method, and neural network training method and device |
| CN110503059A (en) * | 2019-08-27 | 2019-11-26 | 国网电子商务有限公司 | A kind of face identification method and system |
| CN110598627A (en) * | 2019-09-11 | 2019-12-20 | 旭辉卓越健康信息科技有限公司 | GUI demonstration tool based on face recognition |
| CN116723292A (en) * | 2023-05-12 | 2023-09-08 | 北京宝通利达科技有限公司 | Monitoring system based on Internet of things |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107766811A (en) | A kind of face identification method and system based on complicated flow structure | |
| US11195037B2 (en) | Living body detection method and system, computer-readable storage medium | |
| KR102596897B1 (en) | Method of motion vector and feature vector based fake face detection and apparatus for the same | |
| US20190034746A1 (en) | System and method for identifying re-photographed images | |
| CN106897675B (en) | Face living body detection method combining binocular vision depth characteristic and apparent characteristic | |
| US8600121B2 (en) | Face recognition system and method | |
| CN104933414B (en) | A kind of living body faces detection method based on WLD-TOP | |
| CN108416250A (en) | Demographic method and device | |
| WO2021017610A1 (en) | Certificate authenticity verification method and apparatus, computer device and storage medium | |
| CN111241975B (en) | Face recognition detection method and system based on mobile terminal edge calculation | |
| CN109871780B (en) | Face quality judgment method and system and face identification method and system | |
| CN103605972A (en) | Non-restricted environment face verification method based on block depth neural network | |
| CN109858375A (en) | Living body faces detection method, terminal and computer readable storage medium | |
| CN108491794A (en) | The method and apparatus of face recognition | |
| CN107316029A (en) | A kind of live body verification method and equipment | |
| CN112541434B (en) | Face recognition method based on central point tracking model | |
| CN108280421A (en) | Human bodys' response method based on multiple features Depth Motion figure | |
| WO2021082562A1 (en) | Spoofing detection method and apparatus, electronic device, storage medium and program product | |
| CN112966574A (en) | Human body three-dimensional key point prediction method and device and electronic equipment | |
| Zuo et al. | Face liveness detection algorithm based on livenesslight network | |
| CN112488072A (en) | Method, system and equipment for acquiring face sample set | |
| CN116229528A (en) | Living body palm vein detection method, device, equipment and storage medium | |
| CN113468954B (en) | Face counterfeiting detection method based on local area features under multiple channels | |
| Zhang et al. | YOLO based intelligent tracking system for curling sport | |
| CN112818728B (en) | Age identification method and related products |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180306 |