CN109977887A - A kind of face identification method of anti-age interference - Google Patents
A kind of face identification method of anti-age interference Download PDFInfo
- Publication number
- CN109977887A CN109977887A CN201910250058.4A CN201910250058A CN109977887A CN 109977887 A CN109977887 A CN 109977887A CN 201910250058 A CN201910250058 A CN 201910250058A CN 109977887 A CN109977887 A CN 109977887A
- Authority
- CN
- China
- Prior art keywords
- face
- age
- picture
- cascaded
- identification method
- 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
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000012549 training Methods 0.000 claims abstract description 57
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 28
- 238000000605 extraction Methods 0.000 claims abstract description 19
- 238000012360 testing method Methods 0.000 claims abstract description 18
- 230000008859 change Effects 0.000 claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims abstract description 8
- 239000013598 vector Substances 0.000 claims description 34
- 208000016615 Central areolar choroidal dystrophy Diseases 0.000 claims description 19
- 238000010606 normalization Methods 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 8
- 238000012804 iterative process Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 4
- 238000002360 preparation method Methods 0.000 claims description 4
- 230000008901 benefit Effects 0.000 abstract description 8
- 230000001815 facial effect Effects 0.000 abstract description 4
- 238000005286 illumination Methods 0.000 abstract description 4
- 238000012937 correction Methods 0.000 description 8
- 239000000523 sample Substances 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 7
- 238000012795 verification Methods 0.000 description 6
- 230000001537 neural effect Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000003475 lamination Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000009647 facial growth Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000003704 image resize Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000002203 pretreatment Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
- G06V40/169—Holistic features and representations, i.e. based on the facial image taken as a whole
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/178—Human faces, e.g. facial parts, sketches or expressions estimating age from face image; using age information for improving recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of face identification methods of anti-age interference, feature extraction and recognition of face are carried out using picture of the non-cascaded constructional depth convolutional neural networks end to end to same person's all ages and classes stage, include: that picture is obtained by across age face database, forms training set and test set；Data expand using the method for increasing data set and form training image；It establishes and connects layer and one softmax layers of constructional depth convolutional neural networks non-cascaded end to end entirely containing 7 convolutional layers, 3 maximum value pond layers, 1 and network training is carried out to it；Recognition of face is carried out by way of abstract feature of the deep neural network to extract face profound level；Two advantages of time and performance can not only be preferably had both, also to facial angle, intensity of illumination and well adapting to property of coverage extent, effectively overcoming change of age influences recognition of face bring, improves the precision across age recognition of face.
Description
Technical field
The present invention relates to computer vision field (Computer Vision) and deep learning field (Deep
It Learning), specifically, is a kind of face identification method of anti-age interference.
Background technique
Face identification system is an emerging biological identification technology, is the current world using face recognition technology as core
The high-quality precision and sophisticated technology of sciemtifec and technical sphere tackling key problem.Regional characteristics analysis method is widely used in it, has merged computer image processing technology
With biostatistics principle in one, portrait characteristic point is extracted from video using computer image processing technology, utilizes biology
Statistical principle carries out analysis founding mathematical models, has vast potential for future development.2006, the U.S. required to have with it
The country for entering and leaving visa-free agreement must use the electronic passport system for combining recognition of face before October 26, to 2006
The end of the year has had a country more than 50 to realize such system.In April, 2012, railway department announce that station security inspection area will be installed
High-tech safe examination system face identification system for identification；Face light and shade can be detected, adjust automatically dynamic exposure
Compensation, face tracking detecting, adjust automatically image zoom.
Face is widely used in the identification of various occasions as one most significant region of people of identification.Generally
For, the recognition methods of face includes four steps: picture collection and detection, picture pretreatment, picture feature extraction, face
With with verifying.Usually using some feature describing words manually set, such as LBP, SIFT and Gabor etc., to indicate face number
According to, the similarity of a pair of of image is measured using COS distance, thus realize judgement verifying.
But with advancing age, the face of people can be inevitably generated variation (as shown in Figure 1).In some fields
Close, the only photo of people's different age group, such as the photo only before the more than ten years, need by the head portrait of alternative personnel with
Verifying is compared in some clues, and to achieve the goal, this requires carry out across age face verification.So-called across age face is tested
Card, exactly gives the picture of some different age groups, determines whether these pictures belong to the same person.If face verification method
The variation that generates with advancing age that copes with face, in archive management system, security authentication systems, public security system
The fields such as criminal's identification, the monitoring of bank and customs, will have broad application prospects.
In order to realize across age verification, most of traditional methods were modeled to the age, passed through design face growth
Model carries out the face verification across the age.However, such methods generally require rely on priori, such as individual practical year
Age, and not all data set can provide age information.
Deep learning method simulates the Gradation processing structure of human brain, and the inherence of data rich is portrayed with succinct expression way
Information, it is a kind of model of nonlinearity, and data capability of fitting and learning ability with super strength, ability to express is stronger,
The internal information of data rich can more be portrayed.Depth network can learn from data to feature, this mode unsupervisedly
The feature practised also complies with the mechanism in the human perception world, and often has one by the feature that deep learning method learns
Fixed semantic feature.
For the step feature extraction of most critical in face verification, it is primarily present two problems at present:
1, the monotonicity of picture.The a large amount of human face datas being currently known are concentrated, and picture is often more dull, and mesh
Preceding most methods are done on single scale, and the feature extracted in this way is often not abundant enough, are not enough to characterize face.
2, the problem of another merits attention is exactly the acquisition of feature.What traditional face verification used is all hand-designed
Feature, this feature specific aim is relatively high, but typically low-level feature, does not often include semantic information, and extensive
It is indifferent.With the arrival of big data era, data volume is also increasing, and how automatically to obtain feature is worth as one
The project of research.
Summary of the invention
The purpose of the present invention is to provide a kind of face identification methods of anti-age interference, can not only preferably have both the time
With two advantages of performance (time shorter/performance stronger), also facial angle, intensity of illumination and coverage extent are well adapted to
Property, also effectively overcoming change of age influences recognition of face bring, solves the prior art to the identification across age face
The poor problem of ability, improves the precision across age recognition of face.
The present invention is achieved through the following technical solutions: a kind of face identification method of anti-age interference, using end to end
Non-cascaded constructional depth convolutional neural networks carry out feature extraction and recognition of face to the picture in same person's all ages and classes stage.
Further is that the present invention is better achieved, and especially using following set-up modes: the face identification method includes
Following steps:
1) data preparation: picture is obtained by across age face database, forms training set and test set；
2) data expand using the method for increasing data set and form training image；
3) it establishes and complete connects the end-to-end of layer and one softmax layers containing 7 convolutional layers, 3 maximum value pond layers, 1
Non-cascaded constructional depth convolutional neural networks and network instruction is carried out to non-cascaded constructional depth convolutional neural networks end to end
Practice；
4) recognition of face is carried out by way of the abstract feature deep neural network to extract face profound level.
Further is that the present invention is better achieved, and especially use following set-up modes: the step 1) includes following tool
Body step:
1.1) picture being obtained from CACD database and forming training set, picture is obtained from MORPH database and forms test
Collection；
1.2) training set is divided into different age groups, and using each of CACD database people as a class
Not, class label file is generated, is recorded in txt file；
1.3) after step 1.2), (face extraction, face correction, figure are pre-processed to several pictures in training set
As size is fixed), picture is cut out by More General Form, and zoom to unified size 128x128.Pretreated purpose exists
In: since original image is the picture with extended background, in order to reduce interference, by main body-face of picture in picture
Face, which is cut out, to be come.
It is further for the present invention is better achieved, especially use following set-up modes: the age group with every 5 years old one
Span.
Further is that the present invention is better achieved, and especially use following set-up modes: the step 2) includes following step
It is rapid:
2.1) increase new data by celebFaces, formed after new data set, face extraction, people are carried out to it
Face correction, picture size are fixed to 128 × 128；New data set refers to expand training set capacity and adds celebFaces
New training set is formed by after to training set；
2.2) after step 2.1), 5 random croppings are carried out to all pictures, random cropping is after the completion by gained picture
Size is fixed as 128 × 128；
2.3) all pictures of the step 2.2) after size is fixed are subjected to random brightness adjusting or/and contrast tune
Section forms training image.
Further is that the present invention is better achieved, and especially uses following set-up modes: the knot non-cascaded end to end
Structure depth convolutional neural networks carry out solving optimization using the algorithm of SGD type when establishing, and the basic learning rate of setting is
0.001, basic learning rate is adjusted in an iterative process by way of step；Configuration network structure and solver text
After the completion of parameter in part, network training is carried out using caffe.exe.
Further is that the present invention is better achieved, and especially uses following set-up modes: the specific step of the network training
Suddenly are as follows:
3.1) training image of 128 × 128 × 3 sizes is inputted into non-cascaded constructional depth convolutional Neural net end to end
Network obtains 64 × 64 × 64 spy after the operation of three-layer coil product, normalization operation, nonlinear activation operation and pond layer operation
Levy matrix；Wherein, 128 × 128 length and width for referring to training image, subsequent 3 represent port number, common picture
It is all RGB color triple channel figure；Grayscale image and only one channel of artwork master, i.e. width*height*1；
3.2) by the eigenmatrix of step 3.1) resulting 64 × 64 × 64 by three-layer coil product operation, normalization operation,
32 × 32 × 128 eigenmatrix is obtained after nonlinear activation operation and pond layer operation；
3.3) obtained by step 3.2) 32 × 32 × 128 eigenmatrix is grasped by one layer of convolution operation, nonlinear activation
Make, pond layer operation obtains 16 × 16 × 256 eigenmatrix；
3.4) by the eigenmatrix of step 3.3) resulting 16 × 16 × 256, by FC, (Full Connected connects entirely
Connect layer) operation, softmax processing, export N-dimensional vector.
Further is that the present invention is better achieved, and especially use following set-up modes: the step 4) includes following tool
Body step:
4.1) picture in two test sets is sent into trained constructional depth convolutional Neural net non-cascaded end to end
Network；
4.2) the softmax layer of trained constructional depth convolutional neural networks non-cascaded end to end is removed, then
The picture being sent into step 4.1) carries out feature extraction；
4.3) 256 dimensional feature vectors extracted after step 4.2) are calculated by Euclidean distance；
4.4) will judge whether two pictures are same after Euclidean distance calculates acquired results by threshold value comparison
People.
Further is that the present invention is better achieved, and especially uses following set-up modes: carrying out the Euclidean distance meter
When calculation, first 256 dimensional feature vectors are normalized, are mapped to the section for being just distributed very much N (0,1), then use Euclidean distance
FormulaCalculate the distance between vector.
Further is that the present invention is better achieved, and especially use following set-up modes: 256 dimensional feature vectors carry out normalizing
Change processing, be mapped to just too be distributed N (0,1) section after data be also held in local disk.
Compared with prior art, the present invention have the following advantages that and the utility model has the advantages that
(1) present invention is using non-cascaded constructional depth convolutional neural networks end to end, the non-cascaded structure end to end
Depth convolutional neural networks contain only 7 convolutional layers, and 3 maximum value pond layers, 1 connects layer entirely, 1 softmax layers, have structure
Make simple, the simple advantage of realization.Since non-cascaded constructional depth convolutional neural networks are full convolutional networks end to end, be free of
The layer that other pairs of input sizes require, so that input can be arbitrary dimension, can be gray scale can be colour.
(2) present invention devises the feature vector of 256 dimensions to indicate a face, one to improve accuracy of identification
The face of 128 × 128 × 3 dimensions by trained network mapping at the vector of 256 dimensions, in this mapping process effectively
Got rid of most of redundancy and the low feature of contribution rate, effectively increase being accurate to for identification；Enhance network structure mould
The anti-interference that type converts external environment；Show that 256 features can effectively indicate a face by testing test
Details, so that the face feature vector difference of different people is big, the face characteristic difference of the same person is small.
(3) present invention uses depth to separate convolution, this is a kind of convolution mode that model parameter amount can be effectively reduced.
Network parameter amount is greatly reduced, further such that the computation amount in training, test process, space needed for storing
It reduces therewith.
(4) feature of the present invention due to having used 256 dimensions, so that non-cascaded constructional depth convolutional Neural net end to end
Network can preferably cope with several scenes in life, and light change, black-and-white photograph, small area block, expression shape change etc..
(5) present invention uses two standard databases (MORPH, CACD) across age field of face identification, by CACD
Database is used as training, and MORPH database is used as test.
(6) present invention is also equipped with feature and reads preservation function, and a face picture is deep by non-cascaded structure end to end
After spending convolutional neural networks, 256 dimensional feature vectors of generation can be saved to local, when next time is with comparing, so that it may directly
Read this feature vector file, rather than again using network extract feature, do so greatly reduce data storage at
Originally, model calculates cost；The 1:1 for accelerating face is compared and 1:N is compared.
Detailed description of the invention
Fig. 1 is the picture of same person's all ages and classes in FGNET database.
Fig. 2 is work flow diagram of the invention.
Fig. 3 is picture after CACD database preprocessing.
Fig. 4 is the model structure schematic diagram of network structure model of the present invention.
Fig. 5 is the tester A all ages and classes stage, the 256 dimension face feature vectors extracted.
Fig. 6 is tested personnel and its face feature vector figure.
Specific embodiment
The present invention is described in further detail below with reference to embodiment, embodiments of the present invention are not limited thereto.
To keep the purposes, technical schemes and advantages of embodiment of the present invention clearer, implement below in conjunction with the present invention
The technical solution in embodiment of the present invention is clearly and completely described in attached drawing in mode, it is clear that described reality
The mode of applying is some embodiments of the invention, rather than whole embodiments.Based on the embodiment in the present invention, ability
Domain those of ordinary skill every other embodiment obtained without creative efforts, belongs to the present invention
The range of protection.Therefore, the detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit below and is wanted
The scope of the present invention of protection is sought, but is merely representative of selected embodiment of the invention.Based on the embodiment in the present invention,
Every other embodiment obtained by those of ordinary skill in the art without making creative efforts belongs to this
Invent the range of protection.
The size is all as unit of pixel herein.
Embodiment 1:
The present invention designs a kind of face identification method of anti-age interference, can not only preferably have both time and performance two
A advantage (time shorter/performance stronger), also to facial angle, intensity of illumination and well adapting to property of coverage extent, also
Effect, which overcomes change of age, influences recognition of face bring, and it is poor to the recognition capability across age face to solve the prior art
The problem of, the precision across age recognition of face is improved, following set-up modes are especially used: using non-cascaded structure end to end
Depth convolutional neural networks carry out feature extraction and recognition of face to the picture in same person's all ages and classes stage.
Embodiment 2:
The present embodiment is further optimized based on the above embodiments, further for the present invention is better achieved,
Especially use following set-up modes: the face identification method the following steps are included:
1) data preparation: picture is obtained by across age face database, forms training set and test set；
When in use, general across age face database obtains picture, due to include in across age face database according to
Multiple picture groups of identity characteristic and the age characteristics classification of face.According to the body of face in across age face database
Part feature and age characteristics have been classified multiple picture groups.Wherein, the identity characteristic of picture is that the image of face representated by picture is special
Sign, different faces, which has, is marked as different identity categories, is divided according to the identity statistical information of face identity characteristic
Group, the age then according to face institute in the different stages, such as baby, teenager, youth, middle age, old age, therefore according to the year of face
Age statistical information is grouped age characteristics.For really across the picture at age in across the age face database preferably used
Sample, it is more flexible compared to using hand-designed feature as training sample using across age face database.In the present invention
Across the age face database used mainly has MORPH database and CACD database.MORPH database scholarly edition has 55134
Picture, 13618 people, average everyone only has four pictures, everyone age range is relatively small, and includes biggish
Posture, expression interference；CACD database data amount is maximum, 2000 people, 163446 pictures, and average everyone has 82 figures
The age range of piece, similar sample is relatively small.
Two databases (MORPH database and CACD database) preferably use CACD database when technology is implemented
Make training (building training set), MORPH database is used as test (building test set).All training sets are divided into the different ages
Group, every 5 years old age range, while will be every in database (MORPH database and CACD database) as an age group
One classification of personal accomplishment one generates class label file, is recorded in txt file.Immediately to several pictures in training set
(face extraction, face correction, picture size are fixed) is pre-processed, picture is cut out by More General Form, and zoom to
Unified size 128x128.If picture is not ideal enough, such as face key point is not aligned or the size disunity of picture, also
Need to be implemented pretreatment.Treated, and face effect picture is as shown in Figure 3.
2) data expand using the method for increasing data set and form training image；
3) it establishes and complete connects the end-to-end of layer and one softmax layers containing 7 convolutional layers, 3 maximum value pond layers, 1
Non-cascaded constructional depth convolutional neural networks and network instruction is carried out to non-cascaded constructional depth convolutional neural networks end to end
Practice；
4) recognition of face is carried out by way of the abstract feature deep neural network to extract face profound level.
Embodiment 3:
The present embodiment is to advanced optimize based on any of the above embodiments, and further is that this hair is better achieved
It is bright, especially use following set-up modes: the step 1) comprising the following specific steps
1.1) picture being obtained from CACD database and forming training set, picture is obtained from MORPH database and forms test
Collection；
1.2) training set is divided into different age groups, and using each of CACD database people as a class
Not, class label file is generated, is recorded in txt file；
1.3) after step 1.2), (face extraction, face correction, figure are pre-processed to several pictures in training set
As size is fixed), picture is cut out by More General Form, and zoom to unified size 128x128；If picture is not ideal enough,
Such as face key point is not aligned or the size disunity of picture, it is also necessary to execute pretreatment.
Embodiment 4:
The present embodiment is to advanced optimize based on any of the above embodiments, and further is that this hair is better achieved
Bright, especially use following set-up modes: the age group is carrying out the progress age to training set with every 5 years old age range
Group division when, use every 5 years for an age range.
Embodiment 5:
The present embodiment is to advanced optimize based on any of the above embodiments, and further is that this hair is better achieved
It is bright, especially use following set-up modes: the step 2) the following steps are included:
2.1) increase new data by celebFaces, formed after new data set, face extraction, people are carried out to it
Face correction, picture size are fixed to 128 × 128；New data set refers to expand training set capacity and adds celebFaces
New training set is formed by after to training set；
2.2) 5 random sanctions are carried out to the picture of (face extraction, face correction, picture size are fixed) after all pretreatments
It cuts, the size of gained picture is fixed as 128 × 128 after the completion of random cropping；Preferably to all pretreated photos with
96 × 96 size carries out 5 random croppings, is successively upper left, upper right, lower-left, bottom right, center, solid again after the completion of cutting
Determine to 128 × 128 sizes.
2.3) all pictures of the step 2.2) after size is fixed are subjected to random brightness adjusting or/and contrast tune
Section forms training image.
Embodiment 6:
The present embodiment is to advanced optimize based on any of the above embodiments, and further is that this hair is better achieved
Bright, especially use following set-up modes: the constructional depth convolutional neural networks non-cascaded end to end use SGD when establishing
The algorithm of type carrys out solving optimization, and the basic learning rate of setting is 0.001, in an iterative process to basis by way of step
Learning rate is adjusted；After the completion of parameter in Configuration network structure and solver file, network is carried out using caffe.exe
Training, the structure chart of non-cascaded constructional depth convolutional neural networks is as shown in Figure 4 end to end.
Embodiment 7:
The present embodiment is to advanced optimize based on any of the above embodiments, and further is that this hair is better achieved
It is bright, especially use following set-up modes: the specific steps of the network training are as follows:
3.1) training image of 128 × 128 × 3 sizes is inputted into non-cascaded constructional depth convolutional Neural net end to end
Network obtains 64 × 64 × 64 spy after the operation of three-layer coil product, normalization operation, nonlinear activation operation and pond layer operation
Levy matrix；Wherein, 128 × 128 length and width for referring to training image, subsequent 3 represent port number, common picture
It is all RGB color triple channel figure；Grayscale image and only one channel of artwork master, i.e. width*height*1；
As the scheme that is preferable to provide, the specific steps of step 3.1) are as follows: the picture of 128 × 128 × 3 sizes is inputted net
Network first passes around three-layer coil product (respectively conv1_1, conv1_2, conv1_3, they are collectively referred to as conv1).Conv1_1 volumes
It is 3 × 3 convolution kernel that lamination, which has disposed 64 sizes, and fixed filling padding is 1；Conv1_3 convolutional layer and conv1_
1 setting is identical；It is 1 × 1 convolution kernel that conv1_2 convolutional layer, which has disposed 32 sizes, and fixed filling padding is 0,
The feature that size is 128 × 128 × 64 is obtained by this three layers；Then pass through normalization operation (batch
Normalization pond layer max pool1, max pool1) and after nonlinear activation (relu) is input to 2 × 2 × 64
Convolution kernel, step-length 2 obtains 64 × 64 × 64 eigenmatrix.
3.2) by the eigenmatrix of step 3.1) resulting 64 × 64 × 64 by three-layer coil product operation, normalization operation,
32 × 32 × 128 eigenmatrix is obtained after nonlinear activation operation and pond layer operation；
As the scheme that is preferable to provide, the specific steps of step 3.2) are as follows: by step 3.1) resulting 64 × 64 × 64
Eigenmatrix is by three-layer coil product (respectively conv2_1, conv2_2, conv2_3, they are collectively referred to as conv2).Conv2_1 volumes
It is 3 × 3 convolution kernel that lamination, which has disposed 128 sizes, and fixed filling padding is 1；Conv2_3 convolutional layer and
The setting of conv2_1 is identical；It is 1 × 1 convolution kernel, and fixed filling that conv2_2 convolutional layer, which has disposed 64 sizes,
Padding is 0, obtains the feature that size is 64 × 64 × 128 by this three layers.Then pass through normalization operation (batch
Normalization) and after nonlinear activation (relu) be input to pond layer max pool2, max pool2 with 2 × 2 ×
128 convolution kernel, step-length 2 obtain 32 × 32 × 128 eigenmatrix.
3.3) obtained by step 3.2) 32 × 32 × 128 eigenmatrix is grasped by one layer of convolution operation, nonlinear activation
Make, pond layer operation obtains 16 × 16 × 256 eigenmatrix；
As the scheme that is preferable to provide, the specific steps of step 3.3) are as follows: by obtained by step 3.2) 32 × 32 × 128 spy
Input matrix is levied to the convolutional layer conv3 with 256 3 × 3 convolution kernels, obtains 32 × 32 × 256 eigenmatrix；
3.4) by the eigenmatrix of step 3.3) resulting 16 × 16 × 256, by FC, (Full Connected connects entirely
Connect layer) operation, softmax processing, N-dimensional vector is exported, the quantity that parameter N depends on people different in training set (is trained
The face classification number of concentration).
The eigenmatrix of step 3.3) resulting 16 × 16 × 256 is used into a FC, obtains the feature square of 1x1x256
Battle array；Softmax is finally reached, output number is set as the face classification number in face database in total.
Embodiment 8:
The present embodiment is to advanced optimize based on any of the above embodiments, and further is that this hair is better achieved
It is bright, especially use following set-up modes: the step 4) comprising the following specific steps
4.1) two faces (picture in test set) is sent into trained constructional depth convolution non-cascaded end to end
Neural network；
4.2) the softmax layer of trained constructional depth convolutional neural networks non-cascaded end to end is removed, then
The picture being sent into step 4.1) carries out feature extraction；
4.3) 256 dimensional feature vectors extracted after step 4.2) are calculated by Euclidean distance；
4.4) will judge whether two pictures are same after Euclidean distance calculates acquired results by threshold value comparison
People.
Embodiment 9:
The present embodiment is to advanced optimize based on any of the above embodiments, and further is that this hair is better achieved
It is bright, it especially uses following set-up modes: when carrying out Euclidean distance calculating, first 256 dimensional feature vectors being normalized
Processing is mapped to the section for being just distributed very much N (0,1), then with Euclidean distance formulaCalculate vector
The distance between.
Embodiment 10:
The present embodiment is to advanced optimize based on any of the above embodiments, and further is that this hair is better achieved
Bright, especially use following set-up modes: 256 dimensional feature vectors are normalized, and are mapped to the area for being just distributed very much N (0,1)
Between after data be also held in local disk.
Embodiment 11:
Across age recognition of face is the extremely challenging international problem of one in field of face identification.It is well known that
The picture in all ages and classes stage of the same person has very big difference, these differences can seriously affect across age face and know
Other precision.So far, deep learning has been widely used recognition of face, and achieves extraordinary performance.But
It is, for across age recognition of face problem, since there is non-between multiple faces under all ages and classes stage by the same person
Normal significant difference, this seriously affects the performance of existing depth human face recognition model.According to the obtained feature of study come into
Row identification is the research direction risen recently.Depth convolutional network is trained by existing face database, into
And the picture feature learnt is more flexible than manual designs feature.
In order to overcome this huge age differences, the invention proposes a kind of face identification method of anti-age interference,
Picture progress the method use a kind of neural network of structure non-cascaded end to end to same person's all ages and classes stage
Feature extraction, and then identify, it can not only preferably have both two advantages of time and performance (time shorter/performance stronger), it is also right
Facial angle, intensity of illumination and well adapting to property of coverage extent also effectively overcome change of age and bring to recognition of face
Influence, solve the problems, such as that the prior art is poor to the recognition capability across age face, improves across age recognition of face
Precision.Fig. 2 illustrates the workflow entirely invented.Itself comprising the following specific steps
1, data preparation
Original image (A/B) step: it obtains general across age face database and obtains picture, due to across age human face data
It include the multiple picture groups classified according to the identity characteristic and age characteristics of face in library.In across age face database
Classify multiple picture groups according to the identity characteristic and age characteristics of face.Wherein, the identity characteristic of picture is representated by picture
The characteristics of image of face, different faces, which has, is marked as different identity categories, according to the identity statistical information of face to body
Part feature is grouped, the age then according to face institute in the different stages, such as baby, teenager, youth, middle age, old age, therefore
Age characteristics is grouped according to the age statistical information of face.To be true in across the age face database preferably used
Picture sample across the age, it is more clever compared to using hand-designed feature as training sample using across age face database
It is living.Across age face database used in the present invention mainly has MORPH database and CACD database.MORPH database
Art version has 55134 pictures, and 13618 people, average everyone only has four pictures, everyone age range is relatively small,
It and include biggish posture, expression interference；CACD database data amount is maximum, 2000 people, 163446 pictures, average each
People has 82 pictures, and the age range of similar sample is relatively small.
Two databases (MORPH database and CACD database) preferably use CACD database when technology is implemented
Make training (building training set), MORPH database is used as test (building test set).All training sets are divided into the different ages
Group, every 5 years old age range, while will be every in database (MORPH database and CACD database) as an age group
One classification of personal accomplishment one generates class label file, is recorded in txt file.Image flame detection alignment/image RESIZE step
It is rapid: (face extraction, face correction, picture size are fixed) to be pre-processed to several pictures in training set immediately, by picture
It is cut out to come by More General Form, and zooms to unified size 128x128.If picture is not ideal enough, such as face key point does not have
The size disunity of alignment or picture, it is also necessary to execute pretreatment.Treated, and face effect picture is as shown in Figure 3.
2, data extending
Better effect in order to obtain, the present invention is further using the method for increasing data set, first is that increase new data,
After increasing data set, face extraction, face correction, figure are carried out to it by celebFaces (87628 pictures, 5436 people)
As size is fixed；Second is that sequence of operations: random cropping is carried out again to all pretreated photos, it will
The picture of 128x128 carries out 5 random croppings with the size of 96x96, is successively upper left, upper right, lower-left, bottom right, center, cuts
After the completion, it is again fixed to 128x128 size.
Random brightness adjusting, contrast adjustment finally are carried out to the training image of all 128x128.
3, model foundation
In the constructional depth convolutional neural networks non-cascaded end to end that the present invention uses, 7 convolution are contained altogether
Layer, 3 maximum value pond layers, 1 connects layer entirely, and 1 softmax layers, net structure is simple, realizes simple.Implement in the present invention
It is provided in example and carrys out solving optimization using the algorithm of SGD type.Secondly being provided with basic learning rate is 0.001, is then passed through
The mode of step is in an iterative process adjusted basic learning rate.Ginseng in Configuration network structure and solver file
After the completion of number, network training is carried out using caffe.exe.The specific structure of network structure model is as shown in Figure 4.
The picture of 128 × 128 × 3 sizes is inputted into non-cascaded constructional depth convolutional neural networks end to end, is passed through first
Three-layer coil product, respectively conv1_1, conv1_2, conv1_3 are crossed, they are collectively referred to as conv1.Conv1_1 convolutional layer disposes
The convolution kernel that 64 sizes are 3 × 3, and fixed filling padding is 1；The setting phase of conv1_3 convolutional layer and conv1_1
Together；It is 1 × 1 convolution kernel that conv1_2 convolutional layer, which has disposed 32 sizes, and fixed filling padding is 0, by this three
Layer obtains the feature that size is 128 × 128 × 64.By normalization operation (batch normalization) and it is non-linear swash
The convolution kernel that pond layer max pool1, max pool1 has 2 × 2 × 64 is input to after (relu) living, step-length 2 obtains 64
× 64 × 64 eigenmatrix.
Using three-layer coil product, respectively conv2_1, conv2_2, conv2_3, they are collectively referred to as conv2.Conv2_1
It is 3 × 3 convolution kernel that convolutional layer, which has disposed 128 sizes, and fixed filling padding is 1；Conv2_3 convolutional layer and
The setting of conv2_1 is identical；It is 1 × 1 convolution kernel, and fixed filling that conv2_2 convolutional layer, which has disposed 64 sizes,
Padding is 0, obtains the feature that size is 64 × 64 × 128 by this three layers.By normalization operation (batch
Normalization it) and after nonlinear activation (relu) is input to pond layer max pool2, pool2 is with 2 × 2 × 128
Convolution kernel, step-length 2 obtain 32 × 32 × 128 eigenmatrix.
It is then inputted into the convolutional layer conv3 with 256 3 × 3 convolution kernels, obtains 32 × 32 × 256 eigenmatrix.
Relu3, max pool3 are also passed through, the eigenmatrix that size is 16 × 16 × 256 is obtained.Then a FC is used, is obtained
The eigenmatrix of 1x1x256.Softmax is finally reached, output number is set as the face classification number in face database in total.
4, recognition of face
After Face datection, face alignment, whether we will start to judge two faces actually to be same face.
The method that the present invention uses briefly, exactly extracts the abstract feature of face profound level by deep neural network, this
Kind feature abstraction is classified after extracting feature, then with classifier to that can distinguish two different faces.
Two faces are sent into trained network structure model, remove network the last layer softmax, feature is carried out and mentions
It takes, (opens whether face is that the probability of the same person is sentenced carrying out two with the 256 dimensional vector combination Euclidean distance calculation formula extracted
When disconnected, the numberical range of the result is between 0~1, and 0 to represent be exactly a people completely, and 1 to represent be not centainly a people.
If more similar, obtained feature vector with regard to closer, then the distance between feature vector just it is smaller, just closer to
0) (similarity) threshold value (preferred given threshold as 0.22, less than 0.22 it is determined that the same person), is set, determines 2 people
Whether face is the same person.Euclidean distance is originated from vector x 1 in N-dimensional Euclidean space, the range formula of x2:
In order to guarantee value data order-of-magnitude agreement, vector is first carried out normalizing using standardization Euclidean distance by us
Change, be mapped to the section for being just distributed very much N (0,1), reuse distance formula calculates the distance between vector.
Specifically, the face picture a 128x128x3 dimension is deep by trained structure non-cascaded end to end
Degree convolutional neural networks are mapped to the vector of 256 dimensions, have effectively got rid of most of redundancy and tribute in this mapping process
The low feature of rate is offered, being accurate to for identification is effectively increased；Enhance non-cascaded constructional depth convolutional neural networks end to end
To the anti-interference of external environment transformation；Reduce data carrying cost, end to end non-cascaded constructional depth convolutional Neural net
Network calculates cost；The 1:1 for accelerating face is compared and 1:N is compared.As shown in figure 5, we are by people's all ages and classes stage
The feature vector of 256 dimensions is adjusted to 16x16, then shows, has respectively corresponded primary school, junior middle school, senior middle school, university, research
Raw five stages.Fig. 6 is the face feature vector of other two people.It will be seen that the same person, all ages and classes stage
Feature vector is much like, but the feature vector between different people is widely different.
5, result saves
The present invention also provides features to save load function, and a face picture is by trained non-cascaded end to end
After constructional depth convolutional neural networks, 256 dimensional feature vectors of generation can be saved to local after standard normal, next time
When with comparing, so that it may directly read this feature vector file, rather than be sent into network using picture again and extract spy
Sign does so and greatly reduces data carrying cost, model calculates cost；Accelerate the progress of follow-up work.
The present invention enhances training set, and a variety of different types of data enhancements are utilized to construct high quality number
According to including luminance transformation, contrast variation, fuzzy operation, noise addition etc., (these operations are all the Pixel-levels on picture
Operation.By changing the numerical values recited of the pixel of picture, to reach luminance transformation, contrast variation, fuzzy operation, noise addition
Purpose.The purpose for doing these operations is provided to that network architecture is allowed to enhance stability, allows it to have and preferably adapts to energy
Power.In daily life, environment locating for our monitoring probes be all it is complicated and changeable, cause us to obtain from monitoring probe
Picture is also complicated.), by these data enhancement methods, training set complexity is increased manually, effectively improves model
To the adaptability and stability of real scene.
The above is only presently preferred embodiments of the present invention, not does limitation in any form to the present invention, it is all according to
According to technical spirit any simple modification to the above embodiments of the invention, equivalent variations, protection of the invention is each fallen within
Within the scope of.
Claims (10)
1. a kind of face identification method of anti-age interference, it is characterised in that: using non-cascaded constructional depth convolution end to end
Neural network carries out feature extraction and recognition of face to the picture in same person's all ages and classes stage.
2. a kind of face identification method of anti-age interference according to claim 1, it is characterised in that: the recognition of face
Method the following steps are included:
1) data preparation: picture is obtained by across age face database, forms training set and test set；
2) data expand using the method for increasing data set and form training image；
3) it establishes and complete connects the non-end to end of layer and one softmax layers containing 7 convolutional layers, 3 maximum value pond layers, 1
Cascade structure depth convolutional neural networks, and network training is carried out to non-cascaded constructional depth convolutional neural networks end to end；
4) recognition of face is carried out by way of the abstract feature deep neural network to extract face profound level.
3. a kind of face identification method of anti-age interference according to claim 2, it is characterised in that: the network training
Specific steps are as follows:
3.1) training image of 128 × 128 × 3 sizes is inputted into non-cascaded constructional depth convolutional neural networks end to end, warp
64 × 64 × 64 feature square is obtained after crossing the operation of three-layer coil product, normalization operation, nonlinear activation operation and pond layer operation
Battle array；
3.2) by the eigenmatrix of step 3.1) resulting 64 × 64 × 64 by three-layer coil product operation, normalization operation, non-thread
Property activation operation and pond layer operation after obtain 32 × 32 × 128 eigenmatrix；
3.3) by obtained by step 3.2) 32 × 32 × 128 eigenmatrix by one layer of convolution operation, nonlinear activation operation, pond
Change layer operation and obtains 16 × 16 × 256 eigenmatrix；
3.4) eigenmatrix of step 3.3) resulting 16 × 16 × 256 is exported into N-dimensional by FC operation, softmax processing
Vector.
4. a kind of face identification method of anti-age interference according to claim 2 or 3, it is characterised in that: the end is arrived
The non-cascaded constructional depth convolutional neural networks at end come solving optimization, the basis of setting using the algorithm of SGD type when establishing
Learning rate is 0.001, is adjusted in an iterative process to basic learning rate by way of step；Configuration network structure and
After the completion of parameter in solver file, network training is carried out using caffe.exe.
5. a kind of face identification method of anti-age interference according to claim 2 or 3, it is characterised in that: the step
4) comprising the following specific steps
4.1) picture in two test sets is sent into trained constructional depth convolutional neural networks non-cascaded end to end；
4.2) the softmax layer of trained constructional depth convolutional neural networks non-cascaded end to end is removed, then to step
The rapid picture 4.1) being sent into carries out feature extraction；
4.3) 256 dimensional feature vectors extracted after step 4.2) are calculated by Euclidean distance；
4.4) will judge whether two pictures are the same person after Euclidean distance calculates acquired results by threshold value comparison.
6. a kind of face identification method of anti-age interference according to claim 5, it is characterised in that: carrying out the Europe
When formula distance calculates, first 256 dimensional feature vectors are normalized, are mapped to the section for being just distributed very much N (0,1), then use
Euclidean distance formulaCalculate the distance between vector.
7. a kind of face identification method of anti-age interference according to claim 6, it is characterised in that: 256 dimensional features to
Amount is normalized, be mapped to just too be distributed N (0,1) section after data be also held in local disk.
8. a kind of face identification method of anti-age interference according to Claims 2 or 3 or 6 or 7, it is characterised in that: institute
State step 1) comprising the following specific steps
1.1) picture being obtained from CACD database and forming training set, picture is obtained from MORPH database and forms test set；
1.2) training set is divided into different age groups, and using each of CACD database people as a classification, it is raw
At class label file, it is recorded in txt file；
1.3) after step 1.2), several pictures in training set are pre-processed, picture is cut out by More General Form,
And zoom to unified size 128x128.
9. a kind of face identification method of anti-age interference according to claim 8, it is characterised in that: the age group with
Every 5 years old span.
10. a kind of face identification method of anti-age interference, feature according to Claims 2 or 3 or 6 or 7 or 9 exist
In: the step 2) the following steps are included:
2.1) increase new data by celebFaces, formed after new data set, face extraction is carried out to it, face is rectified
Just, picture size is fixed；
2.2) after step 2.1), 5 random croppings are carried out to all pictures, random cropping is after the completion by the size of gained picture
It is fixed as 128 × 128；
2.3) all pictures of the step 2.2) after size is fixed are subjected to random brightness adjusting or/and contrast adjustment, most
Become satisfactory training image eventually.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910089331 | 2019-01-30 | ||
| CN201910089331X | 2019-01-30 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109977887A true CN109977887A (en) | 2019-07-05 |
Family
ID=67081688
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910250058.4A Pending CN109977887A (en) | 2019-01-30 | 2019-03-29 | A kind of face identification method of anti-age interference |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109977887A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390290A (en) * | 2019-07-17 | 2019-10-29 | 北京天正聚合科技有限公司 | A kind of method, apparatus, electronic equipment and storage medium identifying target object |
| CN110598535A (en) * | 2019-07-31 | 2019-12-20 | 广西大学 | Face recognition analysis method used in monitoring video data |
| CN111611877A (en) * | 2020-04-30 | 2020-09-01 | 杭州电子科技大学 | Age interference resistant face recognition method based on multi-temporal-spatial information fusion |
| CN111680597A (en) * | 2020-05-29 | 2020-09-18 | 北京百度网讯科技有限公司 | Face recognition model processing method, device, equipment and storage medium |
| CN111881721A (en) * | 2020-06-10 | 2020-11-03 | 广东芯盾微电子科技有限公司 | Cross-age face recognition method, system, device and storage medium |
| CN112070744A (en) * | 2020-09-08 | 2020-12-11 | 济南博观智能科技有限公司 | Face recognition method, system, device and readable storage medium |
| CN112214283A (en) * | 2020-10-16 | 2021-01-12 | 苏州浪潮智能科技有限公司 | Method and device for predicting duration of Kubernetes mounted volume |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107480575A (en) * | 2016-06-07 | 2017-12-15 | 深圳市商汤科技有限公司 | The training method of model, across age face identification method and corresponding device |
| CN107679451A (en) * | 2017-08-25 | 2018-02-09 | 百度在线网络技术（北京）有限公司 | Establish the method, apparatus, equipment and computer-readable storage medium of human face recognition model |
| CN109117864A (en) * | 2018-07-13 | 2019-01-01 | 华南理工大学 | Coronary heart disease risk prediction technique, model and system based on heterogeneous characteristic fusion |
| CN109117744A (en) * | 2018-07-20 | 2019-01-01 | 杭州电子科技大学 | A kind of twin neural network training method for face verification |
| CN109255289A (en) * | 2018-07-27 | 2019-01-22 | 电子科技大学 | A kind of across aging face identification method generating model based on unified formula |
-
2019
- 2019-03-29 CN CN201910250058.4A patent/CN109977887A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107480575A (en) * | 2016-06-07 | 2017-12-15 | 深圳市商汤科技有限公司 | The training method of model, across age face identification method and corresponding device |
| CN107679451A (en) * | 2017-08-25 | 2018-02-09 | 百度在线网络技术（北京）有限公司 | Establish the method, apparatus, equipment and computer-readable storage medium of human face recognition model |
| CN109117864A (en) * | 2018-07-13 | 2019-01-01 | 华南理工大学 | Coronary heart disease risk prediction technique, model and system based on heterogeneous characteristic fusion |
| CN109117744A (en) * | 2018-07-20 | 2019-01-01 | 杭州电子科技大学 | A kind of twin neural network training method for face verification |
| CN109255289A (en) * | 2018-07-27 | 2019-01-22 | 电子科技大学 | A kind of across aging face identification method generating model based on unified formula |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390290A (en) * | 2019-07-17 | 2019-10-29 | 北京天正聚合科技有限公司 | A kind of method, apparatus, electronic equipment and storage medium identifying target object |
| CN110598535A (en) * | 2019-07-31 | 2019-12-20 | 广西大学 | Face recognition analysis method used in monitoring video data |
| US11263435B2 (en) | 2019-07-31 | 2022-03-01 | Guangxi University | Method for recognizing face from monitoring video data |
| CN111611877A (en) * | 2020-04-30 | 2020-09-01 | 杭州电子科技大学 | Age interference resistant face recognition method based on multi-temporal-spatial information fusion |
| CN111680597A (en) * | 2020-05-29 | 2020-09-18 | 北京百度网讯科技有限公司 | Face recognition model processing method, device, equipment and storage medium |
| CN111680597B (en) * | 2020-05-29 | 2023-09-01 | 北京百度网讯科技有限公司 | Face recognition model processing method, device, equipment and storage medium |
| CN111881721A (en) * | 2020-06-10 | 2020-11-03 | 广东芯盾微电子科技有限公司 | Cross-age face recognition method, system, device and storage medium |
| CN112070744A (en) * | 2020-09-08 | 2020-12-11 | 济南博观智能科技有限公司 | Face recognition method, system, device and readable storage medium |
| CN112070744B (en) * | 2020-09-08 | 2022-11-22 | 济南博观智能科技有限公司 | Face recognition method, system, device and readable storage medium |
| CN112214283A (en) * | 2020-10-16 | 2021-01-12 | 苏州浪潮智能科技有限公司 | Method and device for predicting duration of Kubernetes mounted volume |
| CN112214283B (en) * | 2020-10-16 | 2022-05-31 | 苏州浪潮智能科技有限公司 | Method and device for predicting duration of Kubernetes mounted volume |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977887A (en) | A kind of face identification method of anti-age interference | |
| CN108537743B (en) | Face image enhancement method based on generation countermeasure network | |
| CN107886064B (en) | Face recognition scene adaptation method based on convolutional neural network | |
| CN104866829B (en) | A kind of across age face verification method based on feature learning | |
| CN112215180B (en) | Living body detection method and device | |
| CN104050471B (en) | Natural scene character detection method and system | |
| CN106934359B (en) | Multi-view gait recognition method and system based on high-order tensor subspace learning | |
| CN108334848A (en) | A kind of small face identification method based on generation confrontation network | |
| CN110059586B (en) | Iris positioning and segmenting system based on cavity residual error attention structure | |
| CN108491797A (en) | A kind of vehicle image precise search method based on big data | |
| CN106845328B (en) | A kind of Intelligent human-face recognition methods and system based on dual camera | |
| CN106650693A (en) | Multi-feature fusion identification algorithm used for human face comparison | |
| CN109255289B (en) | Cross-aging face recognition method based on unified generation model | |
| CN104504362A (en) | Face detection method based on convolutional neural network | |
| CN111126240B (en) | Three-channel feature fusion face recognition method | |
| CN110348416A (en) | Multi-task face recognition method based on multi-scale feature fusion convolutional neural network | |
| CN107220598B (en) | Iris image classification method based on deep learning features and Fisher Vector coding model | |
| CN109086660A (en) | Training method, equipment and the storage medium of multi-task learning depth network | |
| CN113221655B (en) | Face spoofing detection method based on feature space constraint | |
| CN106778512A (en) | Face identification method under the conditions of a kind of unrestricted based on LBP and depth school | |
| CN112052772A (en) | Face shielding detection algorithm | |
| CN109886153A (en) | A kind of real-time face detection method based on depth convolutional neural networks | |
| CN109977830A (en) | Face fusion detection method based on color and vein binary channels convolutional neural networks and Recognition with Recurrent Neural Network | |
| CN112070158A (en) | Facial flaw detection method based on convolutional neural network and bilateral filtering | |
| CN106778489A (en) | The method for building up and equipment of face 3D characteristic identity information banks |
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 |
Application publication date: 20190705 |
|
| RJ01 | Rejection of invention patent application after publication |