CN106056088B - The single sample face recognition method of criterion is generated based on adaptive virtual sample - Google Patents
The single sample face recognition method of criterion is generated based on adaptive virtual sample Download PDFInfo
- Publication number
- CN106056088B CN106056088B CN201610390003.XA CN201610390003A CN106056088B CN 106056088 B CN106056088 B CN 106056088B CN 201610390003 A CN201610390003 A CN 201610390003A CN 106056088 B CN106056088 B CN 106056088B
- Authority
- CN
- China
- Prior art keywords
- image
- block
- sample
- training sample
- test sample
- 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.)
- Active
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
- G06V40/172—Classification, e.g. identification
Abstract
The invention discloses a kind of single sample face recognition methods that criterion is generated based on adaptive virtual sample, mainly solve the problems, such as that prior art face identification rate is low.Implementation step are as follows: 1. selection facial images simultaneously divide trained and test sample collection；2. pair training sample carries out singular value decomposition, according to the new training sample image of the base image reconstruction after decomposition；3. combined training sample image and new reconstructed image construct virtual training sample image, and to training sample image and virtual training sample image piecemeal, constitute block training sample set；4. utilizing the optimal projector space of these block training samples training；5. being projected into optimal spatial to test sample piecemeal with same method, block sample characteristics are obtained；6. classifying according to block sample characteristics to block test sample, final recognition result is obtained with maximum ballot criterion.Present invention decreases the missings of authentication information in recognition of face, improve face identification rate, can be used for the identification of identity card, driving license and passport.
Description
Technical field
The invention belongs to technical field of image processing, in particular to a kind of face identification method can be used for identity card, drive
According to and passport identification.
Technical background
Recognition of face is the hot topic in the fields such as pattern-recognition and computer vision, methods many in recent years by
It proposes, is widely used in public safety, video monitoring.But it is also simultaneously a difficulty and complicated problem, for example is considered
Sample storage problem and sample acquisition difficulty problem, often face every a kind of training sample only one the case where, this
In the case of, some common face identification methods cannot be applied directly, and needing to design a kind of recognizer can be effectively
Therefore the essential diagnostic characteristics for extracting Different Individual from single training sample design effective one training sample identification side
Method is problem important in face recognition study field in recent years.Currently, being directed to this problem, main solution has general
Learning method, image block method and virtual sample method.
General learning method is to learn authentication information from one group of multisample face database, to solve asking for single sample recognition of face
Topic.Su etc. proposes a kind of general learning method of adaptability to solve the problems, such as single sample recognition of face, due in singly training sample
In the case where this, Scatter Matrix is zero in class when being solved with LDA, and therefore, it passes through every class first many training samples
Public face database solve Scatter Matrix and class scatter matrix in class, then in the class of linear expression one training sample
Scatter Matrix and class scatter matrix.However, disparate databases have very big difference, linear expression cannot sufficiently reflect people
The authentication information of face, while choosing suitable Universal Database is also a problem.
Image block method is the information using image itself, divides the image into equirotal fritter, these fritters are worked as
Feature extraction and identification are carried out at independent sample.The block divided is carried out feature extraction by image block, using FLDA by Chen etc.,
Classified with nearest neighbor classifier, the classification results of final test sample are the maximum ballots of all block sort results.Zhu etc.
A kind of face identification method of the rarefaction representation of piecemeal is proposed, he is that each single sample image is divided into the fritter for having overlapping,
Dictionary is constructed with these fritters, is solved with the method for rarefaction representation.Image block method can make full use of the local message of image, but
It is the time complexity that will increase algorithm.
Virtual sample method is to generate virtual sample by the one training sample of every one kind, from single training sample and
The virtual sample there study diagnostic characteristics that it is generated.Gao etc. proposes the virtual sample production method based on singular value decomposition,
Every class training sample image is decomposed into a set of basic image using the principle of singular value decomposition by it, and it is corresponding to choose biggish singular value
Base image reconstruction virtual sample, one kind every in this way has two samples, single training sample and the virtual sample that it is reconstructed.
Koc etc. is proposed based on QRCP picture breakdown principle and is generated virtual sample, it by single sample image and it transposition image into
Row QRCP is decomposed, and is generated two groups of basic images, is then reconstructed this two groups of basic images respectively to obtain two virtual sample images, as a result
There are three training samples for every one kind.However both methods have one common disadvantage is that restructuring procedure basic image number
It is fixed and invariable, causes part authentication information that can lack, influence the discrimination of face identification system.
Summary of the invention
It is an object of the invention to be directed to the deficiency of above-mentioned prior art, propose a kind of based on the generation of adaptive virtual sample
The single sample face recognition method of criterion improves face identification rate to reduce the missing of authentication information.
The technical scheme is that in conjunction with the singular value decomposition principle of image and the section thinking of image, to every one kind
The adaptive generation virtual training sample of training sample, training sample image and the virtual training sample image of generation are divided into
Equirotal overlapping block；These overlapping blocks are treated as independent training sample, feature is carried out to it with 2D-FLDA method and is mentioned
It takes, is classified with k nearest neighbor classification device, obtain the tag along sort of various pieces overlapping block on facial image, it is quasi- with maximum ballot
Then obtain final classification results.Implementation step includes the following:
(1) the G width facial image of C class sample is obtained from standard faces library, and is chosen piece image in every one kind and made
For training sample image, remaining image is as N width test sample image, composing training sample setAnd test
Sample setWherein G >=2, C >=2, N >=1, XiIndicate i-th of training sample, yiIndicate XiClass label, Zi
Indicate i-th of test sample, viIndicate ZiClass label；
(2) the i-th sub-picture X that training sample is concentratediSingular value decomposition is carried out, is obtainedWherein n is
XiColumns, σjIt is XiSingular value, and σ1≥σ2≥…≥σn, ujAnd vjIt is respectivelyWithJth column,Table
Show that basic image, T indicate transposition, j=1,2 ..., n；
(3) according to the basic image after singular value decompositionThe corresponding base image reconstruction of k maximum singular value before choosing
One new imageWherein k is the quantity for the basic image chosen,Wherein, r Xi's
Order, avg expression take mean value,Integer is removed in expression；
(4) according to the i-th width image XiWith the new image after reconstructObtain virtual training sample image
Wherein, g is a control parameter, value 0.25；
(5) respectively by the i-th width image XiWith virtual training sample imageIt is divided into the identical overlapping block { x of sizei1,…,
xip,…,xilAndWherein, xipIt is image XiP-th of overlapping block,It is virtual training sample image
P-th of overlapping block, p=1,2 ..., l, l indicate overlapping block quantity；
(6) step (2)-(5) are repeated, C virtual training sample image successively are generated to C training sample, and to C width
Training sample image and the virtual training sample image that they are generated carry out piecemeal, obtain the overlapping block of C training sample imageWith the overlapping block of virtual training sample image
(7) overlapping block of all training sample images and virtual training sample image obtained according to step (6) constitutes block
Training sample set
(8) block training sample set is utilizedTrain l optimal projector space { W1,…,Wp,…,Wl, it will
The overlapping block of training sample imageProject to optimal projector space { W1,…,Wp,…,Wl}；
(9) for any one face test sample Z in face test sample collection Φi, piecemeal first is carried out to it, is obtained
L block test sample { z1,…,zp,…,zl, then it is projected into corresponding optimal projector space { W respectively1,…,Wp,…,
Wl, obtain block test sample { z1,…,zp,…,zlFeature
(10) feature obtained according to step (9)With k nearest neighbor classification device to block test sample { z1,…,
zp,…,zlClassify, according to block sort as a result, finding out face test sample Z by maximum ballot criterioniRecognition result；It presses
Successively the N width face test specimens in face test sample collection Φ are identified according to this method, obtain N width face test sample
Final recognition result.
The invention has the following advantages over the prior art:
One training sample is decomposed 1. the present invention is based on singular value decomposition principles, and reconstructs new images, in conjunction with new
Image and training sample image obtain virtual training sample image, and making every one kind, all there are two training samples, to solve 2D-
FLDA is unable to the problem of direct solution list sample recognition of face.
2. the present invention is when being reconstructed the primary image after singular value decomposition, in conjunction with the energy point of different faces
This different principle of cloth situation, the quantity of adaptive selection basic image, and obtained virtually in conjunction with original training sample image
Training sample image reduces the missing of authentication information, improves the discrimination of face identification system.
3. the present invention carries out piecemeal to each training sample image and test sample image, feature is carried out to these blocks and is mentioned
It takes and classifies, final identification classification results are the maximum voting results of these blocks, to take full advantage of the part letter of image
Breath, enhances face identification system to the robustness of expression, posture and illumination variation.
Detailed description of the invention
Fig. 1 is realization general flow chart of the invention；
Fig. 2 is that the present invention uses the sample image in database；
Fig. 3 is under different subspace with the present invention and existing there are two methods to YALE, FERET, UMIST and ORL face
The recognition result figure of image library.
Specific embodiment
Below in conjunction with attached drawing, technical solutions and effects of the present invention is described in further detail.
Referring to Fig.1, implementation steps of the invention are as follows:
The pretreatment of step 1. facial image.
(1a) chooses facial image:
The 165 width face figures that this example selects 15 people to form from Yale face database select 70 people to form from FERET face database
490 width face figures, chosen from UMIST face database 20 people composition 380 width face figures, from ORL face database select 40 people form
400 width facial images.Protoplast's face image sample size is set as 64 × 64,80 × 80,112 × 112 and 256 × 256 respectively；
(1b) composing training sample set and test sample collection:
The G width facial image of C class sample is obtained from every group of face database, and chooses piece image as instruction in every one kind
Practice sample image, remaining image is as N width test sample image, composing training sample setAnd test sample
CollectionWherein G >=2, C >=2, N >=1, XiIndicate i-th of training sample, yiIndicate XiClass label, ZiIt indicates
I-th of test sample, viIndicate ZiClass label.
Step 2. carries out singular value decomposition to training sample image.
The i-th sub-picture X that training sample is concentratediSingular value decomposition is carried out, is obtainedWherein n is Xi's
Columns, σjIt is XiSingular value, and σ1≥σ2≥…≥σn, ujAnd vjIt is respectivelyWithJth column,It indicates
Basic image, T indicate transposition, j=1,2 ..., n.
Step 3. reconstructs new image.
According to the basic image after singular value decompositionThe corresponding base image reconstruction one of k maximum singular value before choosing
A new imageWherein k is the quantity for the basic image chosen,Wherein, r XiOrder,
Avg expression takes mean value,Integer is removed in expression.
Step 4. reconstructs virtual training sample.
According to the i-th width training image XiWith the new image after reconstructObtain virtual training sample image
Wherein, g is a control parameter, value 0.25.
Step 5. carries out piecemeal to training sample.
Respectively by the i-th width image XiWith virtual training sample imageIt is divided into the identical overlapping block { x of sizei1,…,
xip,…,xilAndWherein, xipIt is image XiP-th of overlapping block,It is virtual training sample image
P-th of overlapping block, p=1,2 ..., l, l indicates the quantity of overlapping block, and taking the size of block is 10 × 10, and the part of overlapping is 5
×5。
Step 6. repeats step 2-step 5, successively generates C virtual training sample image to C training sample, and to C
Width training sample image and the virtual training sample image that they are generated carry out piecemeal, obtain the overlapping of C training sample image
BlockWith the overlapping block of virtual training sample image
The overlapping block of all training samples and virtual training sample image that step 7. is obtained according to step 6 constitutes block instruction
Practice sample set
The block training sample set that step 8. is obtained using step 7Train l optimal projector spaces
{W1,…,Wp,…,Wl}。
(8a) defines Scatter Matrix in the class of p-th of overlapping block based on 2D-FLDAWith class scatter matrix
Wherein,Indicate mean value in the class of p-th of overlapping block,Indicate the equal of all training samples of p-th of overlapping block
Value, they are respectively as follows:
(8b) is according to Scatter Matrix in the class of p-th of overlapping blockWith class scatter matrixIt is rightCarry out feature
Value is decomposed, q eigenvalue λ before obtaining1> λ2> ... > λqThe corresponding feature vector η of > 01,η2,…,ηq, constitute p-th of overlapping
The optimal projector space W of blockp, wherein q < C；
(8c) repeats step (8a)-(8b), finds out the optimal projector space { W of l overlapping block1,…,Wp,…,Wl, it will
The overlapping block of training imageProject to optimal projector space { W1,…,Wp,…,Wl}。
Step 9. identifies N width face test specimens.
(9a) is for any one face test sample Z in face test sample collection Φi, piecemeal first is carried out to it, is obtained
To l block test sample { z1,…,zp,…,zl, then it is projected into corresponding optimal projector space { W respectively1,…,
Wp,…,Wl, obtain block test sample { z1,…,zp,…,zlFeatureIt carries out according to the following formula:
Wherein T indicates transposition；
The feature that (9b) is obtained according to (9a)With k nearest neighbor classification device to block test sample { z1,…,
zp,…,zlClassify；
(9c) is according to the block sort of (9b) as a result, finding out face test sample Z by maximum ballot criterioniRecognition result；
(9d) repeats (9c), successively identifies to the N width face test specimens in face test sample collection Φ, obtains N width
The final recognition result of face test sample.
Effect of the invention can be further illustrated by emulation experiment:
1. experiment condition:
Experiment is Core (TM) 3.40GHZ, is carried out using MatlabR2012b in 7 system of memory 4G, WINDOWS in CPU
Emulation.
165 width images of 15 people composition in Yale face database are chosen, everyone includes different illumination, different tables
Feelings are worn or variation of not wearing glasses, such as Fig. 2 a；
490 width images of 70 people composition in FERET face database are chosen, everyone has different posture and expression
Variation, such as Fig. 2 b；
380 width images of 20 people composition in UMIST face database are chosen, every width facial image is different directions rotation
Variation, such as Fig. 2 c；
Choose 400 width images of 40 people composition in ORL face database, everyone espressiove, illumination and angle change
Change, such as Fig. 2 d；
The image sampling in this four groups of face databases is dimensioned to 64 × 64,80 × 80,112 × 112 and 256 respectively ×
256。
2. experiment content:
Experiment 1: being directed to above four face databases, random one sample of selection of every class as training sample, other
Sample as test sample, respectively with existing svd algorithm, QRCP algorithm and the method for the present invention under different subspace number into
Row face recognition experiment, as a result such as Fig. 3.Wherein:
Fig. 3 a is to carry out 30 independent experiments respectively with three kinds of methods on Yale face database and be averaged
The result arrived；
Fig. 3 b is to carry out 30 independent experiments respectively with three kinds of methods on FERET face database and be averaged
Obtained result；
Fig. 3 c is to carry out 30 independent experiments respectively with three kinds of methods on UMIST face database and be averaged
Obtained result；
Fig. 3 d is to carry out 30 independent experiments respectively with three kinds of methods on ORL face database and be averaged
The result arrived.
It can be seen that on Yale, UMIST and ORL face database from Fig. 3 a, Fig. 3 c, Fig. 3 d, method phase of the invention
Have greatly improved for existing svd algorithm and QRCP algorithm discrimination.
It can be seen that on FERET face database from Fig. 3 b, more other two algorithms of method of the invention are being known
It is not improved slightly in rate.
Experiment 2: being directed to above four face databases, carries out face with existing SVD, QRCP and the method for the present invention respectively
Identification, experiment are to carry out 30 obtained average values under the same conditions, compare its discrimination, the results are shown in Table 1:
The discrimination of 1 present invention of table and the other methods of comparison on four face databases
From table 1 it follows that method of the invention all has highest discrimination on four groups of face databases, this master
Be reconstruct virtual sample during basic image number k it is adaptive selection, in conjunction with one training sample reconstruct virtual sample and
The introducing of image block method.Authentication information can be preferably extracted from one training sample in this way, while also adequately being utilized
The local message of image.
Claims (3)
1. a kind of single sample face recognition method for generating criterion based on adaptive virtual sample, comprising:
(1) the G width facial image of C class sample is obtained from standard faces library, and chooses piece image as instruction in every one kind
Practice sample image, remaining image is as N width test sample image, composing training sample setAnd test sample
CollectionWherein G >=2, C >=2, N >=1, XiIndicate i-th of training sample, yiIndicate XiClass label, ZiIt indicates
I-th of test sample, viIndicate ZiClass label；
(2) the i-th sub-picture X that training sample is concentratediSingular value decomposition is carried out, is obtainedWherein n is Xi's
Columns, σjIt is XiSingular value, and σ1≥σ2≥…≥σn, ujAnd vjIt is respectivelyWithJth column,It indicates
Basic image, T indicate transposition, j=1,2 ..., n；
(3) according to the basic image after singular value decompositionK maximum singular value corresponding base image reconstruction one before choosing
New imageWherein k is the quantity for the basic image chosen,Wherein, r XiOrder,
Avg expression takes mean value,Integer is removed in expression；
(4) according to the i-th width image XiWith the new image after reconstructObtain virtual training sample image
Wherein, g is a control parameter, value 0.25；
(5) respectively by the i-th width image XiWith virtual training sample imageIt is divided into the identical overlapping block { x of sizei1,…,xip,…,
xilAndWherein, xipIt is image XiP-th of overlapping block,It is virtual training sample imageP-th
Overlapping block, p=1,2 ..., l, l indicate the quantity of overlapping block；
(6) step (2)-(5) are repeated, C virtual training sample image successively are generated to C training sample, and to the training of C width
Sample image and the virtual training sample image that they are generated carry out piecemeal, obtain the overlapping block of C training sample imageWith the overlapping block of virtual training sample image
(7) overlapping block of all training samples and virtual training sample image obtained according to step (6) constitutes block training sample
Collection
(8) block training sample set is utilizedTrain l optimal projector space { W1,…,Wp,…,Wl, it will train
The overlapping block of sample imageProject to optimal projector space { W1,…,Wp,…,Wl}:
(8a) defines Scatter Matrix in the class of p-th of overlapping block based on 2D-FLDAWith class scatter matrix
Wherein,Indicate mean value in the class of p-th of overlapping block,Indicate the mean value of all training samples of p-th of overlapping block, it
Be respectively as follows:
(8b) is according to Scatter Matrix in the class of p-th of overlapping blockWith class scatter matrixIt is rightCarry out characteristic value point
Solution, q eigenvalue λ before obtaining1> λ2> ... > λqThe corresponding feature vector η of > 01,η2,…,ηq, constitute p-th of overlapping block
Optimal projector space Wp, wherein q < C；
(8c) repeats step (8a)-(8b), finds out the optimal projector space { W of l overlapping block1,…,Wp,…,Wl}；
(9) for any one face test sample Z in face test sample collection Φi, piecemeal first is carried out to it, obtains l block
Test sample { z1,…,zp,…,zl, then it is projected into corresponding optimal projector space { W respectively1,…,Wp,…,Wl,
Obtain block test sample { z1,…,zp,…,zlFeature
(10) feature obtained according to step (9)With k nearest neighbor classification device to block test sample { z1,…,
zp,…,zlClassify, according to block sort as a result, finding out face test sample Z by maximum ballot criterioniRecognition result；It presses
Successively the N width face test specimens in face test sample collection Φ are identified according to this method, obtain N width face test sample
Final recognition result.
2. being to a width according to the method described in claim 1, training sample is wherein divided into overlapping block in the step (5)
Image XiThe size for taking block is 10 × 10, and the part of overlapping is 5 × 5, is finally divided into the block of l overlapping, obtains image XiOverlapping
Block { xi1,…,xip,…,xilAnd virtual training sample imageOverlapping block
3. according to the method described in claim 1, wherein the step (9) is by block test sample { z1,…,zp,…,zlProjection
To optimal projector space { W1,…,Wp,…,Wl, obtain block test sample { z1,…,zp,…,zlFeatureIt carries out according to the following formula:
Wherein T indicates transposition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610390003.XA CN106056088B (en) | 2016-06-03 | 2016-06-03 | The single sample face recognition method of criterion is generated based on adaptive virtual sample |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610390003.XA CN106056088B (en) | 2016-06-03 | 2016-06-03 | The single sample face recognition method of criterion is generated based on adaptive virtual sample |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106056088A CN106056088A (en) | 2016-10-26 |
| CN106056088B true CN106056088B (en) | 2019-03-08 |
Family
ID=57170086
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610390003.XA Active CN106056088B (en) | 2016-06-03 | 2016-06-03 | The single sample face recognition method of criterion is generated based on adaptive virtual sample |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106056088B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107292256B (en) * | 2017-06-14 | 2019-12-24 | 西安电子科技大学 | Auxiliary task-based deep convolution wavelet neural network expression recognition method |
| CN107563334B (en) * | 2017-09-07 | 2020-08-11 | 南京信息工程大学 | Face recognition method based on identification linear representation preserving projection |
| CN109615611B (en) * | 2018-11-19 | 2023-06-27 | 国家电网有限公司 | Inspection image-based insulator self-explosion defect detection method |
| CN109902657B (en) * | 2019-03-12 | 2022-07-08 | 哈尔滨理工大学 | Face recognition method based on block collaborative representation |
| CN110188828A (en) * | 2019-05-31 | 2019-08-30 | 大连理工大学 | A kind of image sources discrimination method based on virtual sample integrated study |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8411910B2 (en) * | 2008-04-17 | 2013-04-02 | Biometricore, Inc. | Computationally efficient feature extraction and matching iris recognition |
| CN103714326B (en) * | 2013-12-26 | 2017-04-12 | 江南大学 | One-sample face identification method |
| CN105023006B (en) * | 2015-08-05 | 2018-05-04 | 西安电子科技大学 | Face identification method based on enhanced nonparametric maximal margin criterion |
-
2016
- 2016-06-03 CN CN201610390003.XA patent/CN106056088B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN106056088A (en) | 2016-10-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106056088B (en) | The single sample face recognition method of criterion is generated based on adaptive virtual sample | |
| Kyperountas et al. | Salient feature and reliable classifier selection for facial expression classification | |
| Zhang et al. | Real-time multi-view face detection | |
| CN100426314C (en) | Feature classification based multiple classifiers combined people face recognition method | |
| CN104408440A (en) | Identification method for human facial expression based on two-step dimensionality reduction and parallel feature fusion | |
| CN104657718A (en) | Face recognition method based on face image feature extreme learning machine | |
| CN105023006B (en) | Face identification method based on enhanced nonparametric maximal margin criterion | |
| CN105740787B (en) | Identify the face identification method of color space based on multicore | |
| CN108564061A (en) | A kind of image-recognizing method and system based on two-dimensional principal component analysis | |
| CN109840567A (en) | A kind of steady differentiation feature extracting method indicated based on optimal collaboration | |
| Hongtao et al. | Face recognition using multi-feature and radial basis function network | |
| Atito et al. | Mc-ssl0. 0: Towards multi-concept self-supervised learning | |
| Xu et al. | On improving the generalization of face recognition in the presence of occlusions | |
| Hsu et al. | Masked face recognition from synthesis to reality | |
| Gou et al. | mom: Mean of moments feature for person re-identification | |
| CN106778522A (en) | A kind of single sample face recognition method extracted based on Gabor characteristic with spatial alternation | |
| Günay et al. | Age estimation based on AAM and 2D-DCT features of facial images | |
| Guo et al. | Face gender recognition using improved appearance-based Average Face Difference and support vector machine | |
| Fernandes et al. | A comparative study on various state of the art face recognition techniques under varying facial expressions. | |
| CN101482917B (en) | Human face recognition system and method based on second-order two-dimension principal component analysis | |
| Xia et al. | Can 3D shape of the face reveal your age? | |
| Del Coco et al. | Assessment of deep learning for gender classification on traditional datasets | |
| Su et al. | Patch-based gabor fisher classifier for face recognition | |
| Sun et al. | Video-based parent-child relationship prediction | |
| CN103116758A (en) | Color face identification method based on RGB (red, green and blue) color feature double identification analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |