CN108805040A - It is a kind of that face recognition algorithms are blocked based on piecemeal - Google Patents
It is a kind of that face recognition algorithms are blocked based on piecemeal Download PDFInfo
- Publication number
- CN108805040A CN108805040A CN201810507166.0A CN201810507166A CN108805040A CN 108805040 A CN108805040 A CN 108805040A CN 201810507166 A CN201810507166 A CN 201810507166A CN 108805040 A CN108805040 A CN 108805040A
- Authority
- CN
- China
- Prior art keywords
- face
- block
- nose
- leye
- reye
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—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
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical fields of image procossing, specially a kind of to block face recognition algorithms based on piecemeal.The present invention is in off-line training step, the image pretreatment operations such as Face datection, face geometrical normalization, unitary of illumination are carried out first, it is then based on pretreated result, four left eye, right eye, nose, face blocks are extracted from facial image, then the network model of each block is trained, and extracts corresponding feature, blocking for each block is then trained to differentiate that differentiation result is blocked in network acquisition, finally differentiate that result merges its feature according to blocking for each block, and builds K-D tree aspect indexings；The online recognition stage extracts to obtain facial image feature using method identical with off-line training step, then carries out characteristic query, the result to be identified by the indexed mode of K-D trees.The experimental results showed that inventive algorithm has better accuracy rate.
Description
Technical field
The invention belongs to the technical fields of image procossing, and in particular to a kind of to block recognition of face calculation based on piecemeal
Method.
Background technology
Face recognition technology has highly important status in biological identification technology, and for a long time, researchers one
The straight research for being dedicated to face recognition algorithms.Over time, face recognition algorithms are constantly suggested, accuracy rate and steady
It is qualitative also to be promoted constantly.It is developed so far, recognition of face is sent out in fields such as authentication, video monitoring, personage's trackings
Huge effect is waved.
The commercialization of face recognition technology, which implies it in certain circumstances, very high accuracy rate, spy here
The acquisition for determining environment representation facial image is needed based on controlled condition, such as illumination is sufficient, expression angle is single, be blocked rate
It is low.Current many face recognition algorithms only can be only achieved expected accuracy rate on the data set for meet above-mentioned condition.However,
In actual scene (such as network picture, monitor video, the picture etc. of mobile phone shooting), what we got is all often illumination
Face atlas inadequate, expression angle is various, sunglasses or mask etc. block.How to allow face recognition technology true at these
Also have the effect of ideal becoming new challenge under real field scape.
In real scene, the quality of human face image that we get is always irregular, blocks even more inevitable.Such as
What overcomes this unpredictable partial occlusion, the disaster used under complicated real scene at face recognition technology
Topic.These occlusion issues all cause tremendous influence during each step of recognition of face：
First, it during Face datection, blocks and destroys the intrinsic structure of face and geometric properties, so many warps
The Face datection algorithm of allusion quotation can not all detect the face that recognition of face step needs.Secondly, face alignment be typically necessary into
The detection of five key points of row：Left eye center, right eye center, nose, the left side corners of the mouth and the right side corners of the mouth, if these key points without
Method is calibrated to come by face alignment algorithm, then subsequent identification then can be by strong influence.Finally, in face recognition process
In, the local feature that shelter is brought will be introduced by blocking, and the capability of influence of these local features is substantially with masking ratio at just
Than.With the rising of facial area masking ratio, the accuracy of face recognition algorithms can significantly decline.
In conclusion eliminating to block is particularly important the influence that recognition of face is brought.Occlusion issue is solved, to face
The development of identification and more extensive from now on thering is far-reaching influence and recognition of face to march toward practicality from controlled environment
An important ring.
Invention content
Face recognition algorithms are blocked based on piecemeal the purpose of the present invention is to provide a kind of, can be applied to real scene
Human face detects the system with identification, to meet system to there is the requirement for blocking recognition of face accuracy.
It is proposed by the present invention that face recognition algorithms are blocked based on piecemeal, it is divided into 2 stages：Off-line training step
With the online recognition stage；Wherein：
In off-line training step, the image preprocessings such as Face datection, face geometrical normalization, unitary of illumination are carried out first
Operation is then based on pretreated as a result, extracting four left eye, right eye, nose, face blocks from facial image, then instructs
Practice the network model of each block, and extract corresponding feature, blocking for each block is then trained to differentiate that differentiation is blocked in network acquisition
As a result, finally differentiating that result merges its feature according to blocking for each block, and build K-D tree aspect indexings；
The online recognition stage extracts to obtain facial image feature, then lead to using method identical with off-line training step
The indexed mode for crossing K-D trees carries out characteristic query, the result to be identified.Main frame is as shown in Figure 1.
One, off-line training step, the specific steps are：
(1) image preprocessing is as follows：
The first step：Face datection.Face datection is carried out to input picture using CFAN methods [1], it is right if detecting face
Each face all marks its position with 5 key points, is left eye (x respectivelyleye,yleye), right eye (xreye,yreye), nose
(xnose,ynose), the left corners of the mouth (xlmouth,ylmouth), the right corners of the mouth (xrmouth,yrmouth)。
Second step：Face geometrical normalization.The five face key points detected according to previous step, to facial image do as
Lower operation：The central point V for first determining image rotates image around V points so that left eye and right eye to the same horizontal position；It puts down again
Moving image makes nose be in image center location.
Third walks：Unitary of illumination.Histogram equalization is done to facial image.
4th step：Training set extends.Facial image is carried out to the overturning of horizontal direction, and to artwork and the image being turned over
6 kinds of conversion process are carried out respectively, are respectively：Brightness increase, brightness reduction, contrast increase, contrast reduce, Gaussian Blur and
Gaussian noise.1 original image spreading is schemed at 14 by aforesaid operations.
(2) face piecemeal
Facial image size is uniformly adjusted to 256*256 first, then extracts left eye, right eye, nose and four, face
Face block (Fig. 2 illustrates the block sample figure of part face), these blocks we use ((x1,y1),(x2,y2)) indicate,
Wherein (x1,y1) indicate block top left co-ordinate, (x2,y2) indicate block bottom right angular coordinate.
It is as follows：
The first step：For right and left eyes block, with the right and left eyes key point (x calibratedleye,yleye)、(xreye,yreye) be
Center cuts out following two pieces of blocks：
((xleye-32,yleye-32),(xleye+32,yleye+32)) (1)
((xreye-32,yreye-32),(xreye+32,yreye+32)) (2)
Second step：For nose block, according to the key point (x of the nasal portion of calibrationnose,ynose), cut out following area
Block：
((xnose-32,yleye),(xnose+32,(ylmouth+yrmouth)/2-32)) (3)
Third walks：For face block, according to the key point (x of the face part calibratedlmouth,ylmouth), (xrmouth,
yrmouth), cut out following block：
((xlmouth-8,ylmouth-32),(xrmouth+8,yrmouth+32)) (4)。
(3) feature extraction
Feature extraction network used in the present invention can substantially be divided into following five parts:First part is by two convolution
Layer, an activation primitive and a pond layer form, and the characteristic pattern size of input is 64*64*1, and output size is after processing
32*32*24；Second part is made of three convolutional layers, an activation primitive and a pond layer, and input size is 32*32*
24, output size is 16*16*48 after processing；Part III is by three convolutional layers, an activation primitive and a pond layer group
At input size is 16*16*48, and output size is 8*8*96 after processing；Part IV is by three convolutional layers, an activation
Function and a pond layer form, and input size is 8*8*96, and output size is 4*4*128 after processing；Part V is by two
A full articulamentum composition, input size are 4*4*128, and output is characterized as 128 dimensions.
The design parameter of network is as shown in table 1, and wherein convx_y indicates y-th of convolutional layer of x-th of part, mfmx tables
Show that the activation primitive of xth part, poolx indicate the pond layer of x-th of part；Convolution nuclear parameter is indicated by taking " 3*3/1,1 " as an example
Convolution kernel size is 3*3, and step-length 1 is filled with 1；For output size by taking " 64*64*24 " as an example, expression image size is 64*64,
Port number is 24.
To each of four blocks of every facial image, the present invention is carried with an above-mentioned feature extraction network
Take its feature.The face characteristic V of four blocks will finally be obtainedleye,Vreye,Vnose,Vmouth, wherein VleyeIndicate left eye block
Feature, VreyeIndicate right eye Block Characteristic, VnoseIndicate nose Block Characteristic, VmouthIndicate face Block Characteristic.
(4) occlusion detection
The present invention is based on Inception V3 networks to carry out fine-tune, and whether training one is for differentiating face block
The depth convolutional neural networks being blocked, i.e. two graders.When doing classification based training collection construction, by left eye block and
The categories combination of right eye block is eyes block, therefore the training set constructed is divided into 4 classes, respectively eyes block, nose region
Block, face block and Background.Wherein, background here includes that four face blocks of extraction are removed in facial image with outside
Point.
The present invention detects whether blocking of face block using the sorter network of above-mentioned training, if blocking, blocks
Testing result output is 0, and result output is 1 if not blocking.
(5) Fusion Features
The feature that each face block learns is normalized into [- 1 ,+1] first, obtains V'leye,V'reye,V'nose,
V'mouth.Then sequential concatenation is carried out to each section feature, i.e., for facial image I, the face characteristic V (I) ultimately generated is such as
Under：
V (I)=[V'leye,V'reye,V'nose,V'mouth] (5)。
(6) structure index
The present invention is based on K-D tree construction features indexes, and the number of nodes of every K-D tree is within 150,000.According to formula
(5), every facial image I can extract 4 Partial Features, and for this four features, the present invention expands 15 features and is used for
K-D trees are constructed, these features are respectively：
[V'leye,0,0,0]
[0,V'reye,0,0]
[0,0,V'nose,0]
[0,0,0,V'mouth]
[V'leye,V'reye,0,0]
[V'leye,0,V'nose,0]
[V'leye,0,0,V'mouth]
[0,V'reye,V'nose,0]
[0,V'reye,0,V'mouth]
[0,0,V'nose,V'mouth]
[0,V'reye,V'nose,V'mouth]
[V'leye,0,Vnose,Vmouth]
[V'leye,V'reye,0,V'mouth]
[V'leye,V'reye,V'nose,0]
[V'leye,V'reye,V'nose,V'mouth]
After features described above extends, a K-D tree can handle 10,000 primitive characters, that is, handle 10,000 face pictures.
Two, the online recognition stage：
This stage the specific steps are：(1) image preprocessing, (2) face piecemeal, (3) feature extraction, (4) occlusion detection,
(5) Fusion Features, (6) online query.Wherein (1)-(5) step specific implementation mode is passed through as off-line phase processing mode
The feature vector of facial image to be identified is obtained after above-mentioned steps.
The flow searched online is specific as follows：
Facial image I and image J distances S in index database to be identified is calculated using formula (6)f(i, j), Sf(i, j) is smaller
Indicate the image to more similar.
Wherein, Sp(Ileye,Jleye),Sp(Ireye,Jreye),Sp(Inose,Jnose),Sp(Imouth,Jmouth) two are indicated respectively
Left eye block similarity, right eye block similarity, nose block similarity and the face block similarity of image are compared, is calculated
Formula is as follows：
Wherein, part indicates face block, CpartIndicate face block occlusion detection as a result, n be feature vector V'part
Dimension,For the value of vector kth dimension.
The final image for choosing similarity minimum is exported as query result.
The present invention is directed to occlusion issue in recognition of face, it is proposed that and it is a kind of that face recognition algorithms are blocked based on piecemeal,
In image pre-processing phase, by carrying out a variety of conversion process to training set so that a facial image is extended to 14, to prevent
Model over-fitting after only training preferably is suitable for actual conditions；In feature extraction phases, improved LCNN networks have
Faster training and test speed, make time and space be advanced optimized；The occlusion detection stage blocks differentiation by being added
Network largely eliminates the influence blocked and brought to face recognition algorithms；In the Fusion Features stage, according to the knot of occlusion detection
Fruit selects different modes in Local Feature Fusion to final feature, will to improve the accuracy for blocking face characteristic inquiry.It is real
Test the result shows that, algorithm proposed by the present invention have better accuracy rate.
Description of the drawings
Fig. 1 is algorithm flow chart.
Fig. 2 is that block cuts sample figure.
Fig. 3 is AR data sets part facial image example.
Specific implementation mode
The present invention using AR data sets block the experiment of recognition of face, and AR data sets include 126 people totally 3276
Facial image (including 70 men and 56 woman), everyone 26 facial images, which includes different faces
Expression (normal, smile, is cold and detached, indignation), different illumination and the case where block (sunglasses, scarf).The facial image difference blocked
It is the picture and three scarf pictures of different lower three wear dark glasses of illumination.AR data sets part facial image example is as shown in Figure 3.
It chooses AR human face datas and concentrates 100 people (50 men and 50 woman), from everyone 14 unobstructed facial images
In randomly select 8 be used as training set, the image of other everyone 2 wear dark glasses and 2 images for wearing scarf are as test set.It surveys
The local facial block of examination process combined training blocks discrimination model progress.If the face block of input is differentiated by two graders
To block, then the local feature is added without in face characteristic comparison process.At the same time, test face is divided into sunglasses to block
Two classes are blocked with scarf to be tested.The contrast experiment of the present invention has used classical SRC, GSRC and RSC algorithm.Experimental result
As shown in table 1.
Table 1 blocks recognition of face Contrast on effect
| Algorithm | SRC | GSRC | RSC | This algorithm |
| Sunglasses block | 87.0% | 93% | 99% | 97.5% |
| Scarf blocks | 59.5% | 79% | 97% | 99% |
As shown in Table 1, in the case where scarf blocks, this algorithm achieves extraordinary recognition effect, but in sunglasses
In the case of blocking, this algorithm accuracy rate is slightly below RSC.This is because in the case where scarf blocks, face part face area
Block is blocked, therefore this Partial Feature can be rejected, but the feature for being left three face blocks still has enough discriminations,
It can be very good one face of characterization.But in the case where sunglasses block, the left eye and right eye block of facial image are blocked,
The feature of the two parts is rejected, and for wearing scarf and only need to give up face Partial Feature, sunglasses block the shadow shone
Ring bigger.But consider, this algorithm remains to obtain good effect in the case where blocking.
AR data sets are chosen again to be tested, and 8 are randomly selected from everyone 14 unobstructed facial images as training
Collection randomly selects 2 as test set from remaining 6, four people is carried out to the facial image in test set using black patch
Face block blocks processing, and recognition of face is carried out (since this algorithm depends on to the image of different circumstance of occlusion using this algorithm
Facial modeling as a result, under high obstruction conditions, certain artificial progress auxiliary positioning work is added).Compare SRC and
RSC algorithms, experimental result are as shown in table 2.
2 large area face occlusion effect of table compares
| Algorithm | SRC | RSC | This algorithm |
| Block a block | 92% | 100% | 100% |
| Block two blocks | 83% | 95% | 96% |
| Block three blocks | 65% | 82% | 94% |
Every a line of 2 result of table counted respectively block one, two, each algorithm identification is accurate in the case of three blocks
The average value of true rate.It can be found that with the increase for blocking number of blocks, the accuracys rate of SRC and RSC algorithms decline all very
Soon, and this algorithm still can ensure certain recognition accuracy.
The block characteristic of this algorithm ensure that when at least one face block is not blocked or is only blocked on a small quantity, calculate
The recognition accuracy of method will not be much affected.
LCNN network architecture parameters after the adjustment of table 3
Note：
[1]Jie Zhang,Shiguang Shan,MeinaKan,Xilin Chen.Coarse-to-Fine Auto-
Encoder Networks(CFAN)for Real-Time Face Alignment.ECCV 2014。
Claims (3)
1. a kind of blocking face recognition algorithms based on piecemeal, which is characterized in that be divided into 2 stages：Off-line training rank
Section and online recognition stage；Wherein：
In off-line training step, the image preprocessings such as Face datection, face geometrical normalization, unitary of illumination behaviour is carried out first
Make, is then based on pretreated as a result, extracting four left eye, right eye, nose, face blocks from facial image, then trains
The network model of each block, and corresponding feature is extracted, then train blocking for each block to differentiate that differentiation knot is blocked in network acquisition
Fruit finally differentiates that result merges its feature according to blocking for each block, and builds K-D tree aspect indexings；
The online recognition stage extracts to obtain facial image feature, then passes through K-D using method identical with off-line training step
The indexed mode of tree carries out characteristic query, the result to be identified.
2. according to claim 1 block face recognition algorithms based on piecemeal, which is characterized in that off-line training rank
Section, the specific steps are：
(1) image preprocessing is as follows：
The first step：Face datection：Face datection is carried out to input picture using CFAN methods, if detecting face, to each
Face is all with 5 key points come to mark its position, 5 key points be left eye (x respectivelyleye,yleye), right eye (xreye,yreye)、
Nose (xnose,ynose), the left corners of the mouth (xlmouth,ylmouth), the right corners of the mouth (xrmouth,yrmouth)；
Second step：Face geometrical normalization：The five face key points detected according to previous step, following behaviour is to facial image
Make：The central point V for first determining image rotates image around V points so that left eye and right eye to the same horizontal position；Figure is translated again
As so that nose is in image center location；
Third walks：Unitary of illumination does histogram equalization to facial image；
4th step：Training set extends：Facial image is carried out to the overturning of horizontal direction, and artwork and the image being turned over are distinguished
6 kinds of conversion process are carried out, are respectively：Brightness increase, brightness reduction, contrast increase, contrast reduction, Gaussian Blur and Gauss
Noise；1 original image spreading is schemed at 14 by aforesaid operations；
(2) face piecemeal：
Facial image size is uniformly adjusted to 256*256 first, then extracts four left eye, right eye, nose and face faces
Block, these blocks ((x1,y1),(x2,y2)) indicate, wherein (x1,y1) indicate block top left co-ordinate, (x2,y2) indicate
Block bottom right angular coordinate；
Face piecemeal is as follows：
The first step：For right and left eyes block, with the right and left eyes key point (x calibratedleye,yleye)、(xreye,yreye) centered on,
Cut out following two pieces of blocks：
((xleye-32,yleye-32),(xleye+32,yleye+32)) (1)
((xreye-32,yreye-32),(xreye+32,yreye+32)) (2)
Second step：For nose block, according to the key point (x of the nasal portion calibratednose,ynose), cut out following area
Block：
((xnose-32,yleye),(xnose+32,(ylmouth+yrmouth)/2-32)) (3)
Third walks：For face block, according to the key point (x of the face part calibratedlmouth,ylmouth), (xrmouth,
yrmouth), cut out following block：
((xlmouth-8,ylmouth-32),(xrmouth+8,yrmouth+32)) (4)
(3) feature extraction
Feature extraction network used is divided into following five parts:First part is by two convolutional layers, an activation primitive and one
The characteristic pattern size of a pond layer composition, input is 64*64*1, and output size is 32*32*24 after processing；Second part by
Three convolutional layers, an activation primitive and a pond layer form, and input size is 32*32*24, and output size is after processing
16*16*48；Part III is made of three convolutional layers, an activation primitive and a pond layer, and input size is 16*16*
48, output size is 8*8*96 after processing；Part IV is by three convolutional layers, an activation primitive and a pond layer group
At input size is 8*8*96, and output size is 4*4*128 after processing；Part V is made of two full articulamentums, input
Size is 4*4*128, and output is characterized as 128 dimensions；
To each of four blocks of every facial image, its feature is extracted with an above-mentioned feature extraction network；
The face characteristic V of four blocks will finally be obtainedleye,Vreye,Vnose,Vmouth, wherein VleyeIndicate left eye Block Characteristic, Vreye
Indicate right eye Block Characteristic, VnoseIndicate nose Block Characteristic, VmouthIndicate face Block Characteristic；
(4) occlusion detection
Fine-tune is carried out based on Inception V3 networks, training one is for differentiating the depth whether face block is blocked
Spend convolutional neural networks, i.e. two graders；When doing classification based training collection construction, by left eye block and right eye block
Categories combination is eyes block, therefore the training set constructed is divided into 4 classes, respectively eyes block, nose block, face block
And Background；Here background includes that four face blocks of extraction are removed in facial image with outer portion；
Whether blocking of face block is detected using the sorter network of above-mentioned training, if blocking, occlusion detection result is defeated
It is 0 to go out, and result output is 1 if not blocking；
(5) Fusion Features
The feature that each face block learns is normalized into [- 1 ,+1] first, obtains V'leye,V'reye,V'nose,
V'mouth；Then sequential concatenation is carried out to each section feature, i.e., for facial image I, the face characteristic V (I) ultimately generated is such as
Under：
V (I)=[V'leye,V'reye,V'nose,V'mouth] (5)
(6) structure index
It is indexed based on K-D tree construction features, the number of nodes of every K-D tree is within 150,000；According to formula (5), every face
Image I extracts 4 Partial Features, for this four features, expands 15 features for constructing K-D trees, these features are distinguished
For：
[V'leye,0,0,0]
[0,V'reye,0,0]
[0,0,V'nose,0]
[0,0,0,V'mouth]
[V'leye,V'reye,0,0]
[V'leye,0,V'nose,0]
[V'leye,0,0,V'mouth]
[0,V'reye,V'nose,0]
[0,V'reye,0,V'mouth]
[0,0,V'nose,V'mouth]
[0,V'reye,V'nose,V'mouth]
[V'leye,0,Vnose,Vmouth]
[V'leye,V'reye,0,V'mouth]
[V'leye,V'reye,V'nose,0]
[V'leye,V'reye,V'nose,V'mouth]
After features described above extends, a K-D tree can handle 10,000 primitive characters, that is, handle 10,000 face pictures.
3. according to claim 2 block face recognition algorithms based on piecemeal, which is characterized in that the online recognition stage
The specific steps are：(1) image preprocessing, (2) face piecemeal, (3) feature extraction, (4) occlusion detection, (5) Fusion Features,
(6) online query；Wherein step (1)-(5) obtain people to be identified as off-line phase processing mode after above-mentioned steps
The feature vector of face image；
The flow searched online is specific as follows：
Facial image I and image J distances S in index database to be identified is calculated using formula (6)f(i, j), Sf(i, j) smaller expression
The image is to more similar；
Wherein, Sp(Ileye,Jleye),Sp(Ireye,Jreye),Sp(Inose,Jnose),Sp(Imouth,Jmouth) two comparisons are indicated respectively
Left eye block similarity, right eye block similarity, nose block similarity and the face block similarity of image, calculation formula
It is as follows：
Wherein, part indicates face block, CpartIndicate face block occlusion detection as a result, n be feature vector V'partDimension
Degree,For the value of vector kth dimension；
The final image for choosing similarity minimum is exported as query result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810507166.0A CN108805040A (en) | 2018-05-24 | 2018-05-24 | It is a kind of that face recognition algorithms are blocked based on piecemeal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810507166.0A CN108805040A (en) | 2018-05-24 | 2018-05-24 | It is a kind of that face recognition algorithms are blocked based on piecemeal |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108805040A true CN108805040A (en) | 2018-11-13 |
Family
ID=64091647
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810507166.0A Pending CN108805040A (en) | 2018-05-24 | 2018-05-24 | It is a kind of that face recognition algorithms are blocked based on piecemeal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108805040A (en) |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109711356A (en) * | 2018-12-28 | 2019-05-03 | 广州海昇教育科技有限责任公司 | A kind of expression recognition method and system |
| CN109766854A (en) * | 2019-01-15 | 2019-05-17 | 济南浪潮高新科技投资发展有限公司 | A kind of robust human face recognizer based on two stages complementary networks |
| CN109784255A (en) * | 2019-01-07 | 2019-05-21 | 深圳市商汤科技有限公司 | Neural network training method and device and recognition methods and device |
| CN109829448A (en) * | 2019-03-07 | 2019-05-31 | 苏州市科远软件技术开发有限公司 | Face identification method, device and storage medium |
| CN109886121A (en) * | 2019-01-23 | 2019-06-14 | 浙江大学 | A kind of face key independent positioning method blocking robust |
| CN109902581A (en) * | 2019-01-28 | 2019-06-18 | 重庆邮电大学 | It is a kind of based on multistep weighting single sample portion block face identification method |
| CN110175529A (en) * | 2019-04-30 | 2019-08-27 | 东南大学 | A kind of three-dimensional face features' independent positioning method based on noise reduction autoencoder network |
| CN110232369A (en) * | 2019-06-20 | 2019-09-13 | 深圳和而泰家居在线网络科技有限公司 | A kind of face identification method and electronic equipment |
| CN110298225A (en) * | 2019-03-28 | 2019-10-01 | 电子科技大学 | A method of blocking the human face five-sense-organ positioning under environment |
| CN110533001A (en) * | 2019-09-06 | 2019-12-03 | 厦门久凌创新科技有限公司 | Big data face identification method based on recognition of face |
| CN110569756A (en) * | 2019-08-26 | 2019-12-13 | 长沙理工大学 | face recognition model construction method, recognition method, device and storage medium |
| CN110728215A (en) * | 2019-09-26 | 2020-01-24 | 杭州艾芯智能科技有限公司 | Face living body detection method and device based on infrared image |
| CN110728330A (en) * | 2019-10-23 | 2020-01-24 | 腾讯科技（深圳）有限公司 | Object identification method, device, equipment and storage medium based on artificial intelligence |
| CN111027504A (en) * | 2019-12-18 | 2020-04-17 | 上海眼控科技股份有限公司 | Face key point detection method, device, equipment and storage medium |
| CN111126307A (en) * | 2019-12-26 | 2020-05-08 | 东南大学 | Small sample face recognition method of joint sparse representation neural network |
| CN111460945A (en) * | 2020-03-25 | 2020-07-28 | 亿匀智行(深圳)科技有限公司 | Algorithm for acquiring 3D expression in RGB video based on artificial intelligence |
| CN111723612A (en) * | 2019-03-20 | 2020-09-29 | 北京市商汤科技开发有限公司 | Face recognition and face recognition network training method and device, and storage medium |
| CN111738043A (en) * | 2019-12-10 | 2020-10-02 | 珠海大横琴科技发展有限公司 | Pedestrian re-identification method and device |
| CN111739070A (en) * | 2020-05-28 | 2020-10-02 | 复旦大学 | Real-time multi-pose face detection algorithm based on progressive calibration type network |
| CN111783670A (en) * | 2020-07-02 | 2020-10-16 | 哈尔滨理工大学 | Face recognition method based on neural network and face composition |
| CN111814571A (en) * | 2020-06-12 | 2020-10-23 | 深圳禾思众成科技有限公司 | Mask face recognition method and system based on background filtering |
| CN111914665A (en) * | 2020-07-07 | 2020-11-10 | 泰康保险集团股份有限公司 | Face shielding detection method, device, equipment and storage medium |
| CN112070015A (en) * | 2020-09-08 | 2020-12-11 | 广州云从博衍智能科技有限公司 | Face recognition method, system, device and medium fusing occlusion scene |
| CN112132011A (en) * | 2020-09-22 | 2020-12-25 | 深圳市捷顺科技实业股份有限公司 | Face recognition method, device, equipment and storage medium |
| CN112418195A (en) * | 2021-01-22 | 2021-02-26 | 电子科技大学中山学院 | Face key point detection method and device, electronic equipment and storage medium |
| CN112990007A (en) * | 2021-03-13 | 2021-06-18 | 山东大学 | Facial expression recognition method and system based on regional grouping and internal association fusion |
| CN113158939A (en) * | 2021-04-29 | 2021-07-23 | 南京甄视智能科技有限公司 | Method and system for identifying human face shielding part |
| CN113688657A (en) * | 2020-05-19 | 2021-11-23 | 深圳云天励飞技术有限公司 | Face recognition method and device, electronic equipment and storage medium |
| CN114360033A (en) * | 2022-03-18 | 2022-04-15 | 武汉大学 | Mask face recognition method, system and equipment based on image convolution fusion network |
| CN114582002A (en) * | 2022-04-18 | 2022-06-03 | 华南理工大学 | Facial expression recognition method combining attention module and second-order pooling mechanism |
| CN115131843A (en) * | 2021-03-24 | 2022-09-30 | 北京君正集成电路股份有限公司 | Method for detecting face shielding based on image segmentation |
| CN115619410A (en) * | 2022-10-19 | 2023-01-17 | 闫雪 | Self-adaptive financial payment platform |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855496A (en) * | 2012-08-24 | 2013-01-02 | 苏州大学 | Method and system for authenticating shielded face |
| CN104008372A (en) * | 2014-05-23 | 2014-08-27 | 南京邮电大学 | Distributed face recognition method in wireless multi-media sensor network |
| CN105488478A (en) * | 2015-12-02 | 2016-04-13 | 深圳市商汤科技有限公司 | Face recognition system and method |
-
2018
- 2018-05-24 CN CN201810507166.0A patent/CN108805040A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855496A (en) * | 2012-08-24 | 2013-01-02 | 苏州大学 | Method and system for authenticating shielded face |
| CN104008372A (en) * | 2014-05-23 | 2014-08-27 | 南京邮电大学 | Distributed face recognition method in wireless multi-media sensor network |
| CN105488478A (en) * | 2015-12-02 | 2016-04-13 | 深圳市商汤科技有限公司 | Face recognition system and method |
Non-Patent Citations (2)
| Title |
|---|
| XIANG WU: "A Lightened CNN for Deep Face Representation", 《ARXIV:1511.02683V1 [CS.CV]》 * |
| 周孝佳，朱允斌，张跃: "基于分块的有遮挡人脸识别算法", 《计算机应用于软件》 * |
Cited By (50)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109711356A (en) * | 2018-12-28 | 2019-05-03 | 广州海昇教育科技有限责任公司 | A kind of expression recognition method and system |
| CN109711356B (en) * | 2018-12-28 | 2023-11-10 | 广州海昇教育科技有限责任公司 | Expression recognition method and system |
| CN109784255A (en) * | 2019-01-07 | 2019-05-21 | 深圳市商汤科技有限公司 | Neural network training method and device and recognition methods and device |
| CN109766854A (en) * | 2019-01-15 | 2019-05-17 | 济南浪潮高新科技投资发展有限公司 | A kind of robust human face recognizer based on two stages complementary networks |
| CN109886121A (en) * | 2019-01-23 | 2019-06-14 | 浙江大学 | A kind of face key independent positioning method blocking robust |
| CN109902581A (en) * | 2019-01-28 | 2019-06-18 | 重庆邮电大学 | It is a kind of based on multistep weighting single sample portion block face identification method |
| CN109829448A (en) * | 2019-03-07 | 2019-05-31 | 苏州市科远软件技术开发有限公司 | Face identification method, device and storage medium |
| CN109829448B (en) * | 2019-03-07 | 2021-05-28 | 苏州市科远软件技术开发有限公司 | Face recognition method, face recognition device and storage medium |
| CN111723612A (en) * | 2019-03-20 | 2020-09-29 | 北京市商汤科技开发有限公司 | Face recognition and face recognition network training method and device, and storage medium |
| CN110298225A (en) * | 2019-03-28 | 2019-10-01 | 电子科技大学 | A method of blocking the human face five-sense-organ positioning under environment |
| CN110175529B (en) * | 2019-04-30 | 2023-05-16 | 东南大学 | Three-dimensional face feature point positioning method based on noise reduction self-coding network |
| CN110175529A (en) * | 2019-04-30 | 2019-08-27 | 东南大学 | A kind of three-dimensional face features' independent positioning method based on noise reduction autoencoder network |
| CN110232369B (en) * | 2019-06-20 | 2021-10-01 | 深圳数联天下智能科技有限公司 | Face recognition method and electronic equipment |
| CN110232369A (en) * | 2019-06-20 | 2019-09-13 | 深圳和而泰家居在线网络科技有限公司 | A kind of face identification method and electronic equipment |
| CN110569756A (en) * | 2019-08-26 | 2019-12-13 | 长沙理工大学 | face recognition model construction method, recognition method, device and storage medium |
| CN110569756B (en) * | 2019-08-26 | 2022-03-22 | 长沙理工大学 | Face recognition model construction method, recognition method, device and storage medium |
| CN110533001A (en) * | 2019-09-06 | 2019-12-03 | 厦门久凌创新科技有限公司 | Big data face identification method based on recognition of face |
| CN110533001B (en) * | 2019-09-06 | 2022-02-08 | 厦门久凌创新科技有限公司 | Big data face recognition method based on face recognition |
| CN110728215A (en) * | 2019-09-26 | 2020-01-24 | 杭州艾芯智能科技有限公司 | Face living body detection method and device based on infrared image |
| US12099577B2 (en) | 2019-10-23 | 2024-09-24 | Tencent Technology (Shenzhen) Company Limited | Object recognition method and apparatus, electronic device, and readable storage medium |
| CN110728330A (en) * | 2019-10-23 | 2020-01-24 | 腾讯科技（深圳）有限公司 | Object identification method, device, equipment and storage medium based on artificial intelligence |
| CN111738043A (en) * | 2019-12-10 | 2020-10-02 | 珠海大横琴科技发展有限公司 | Pedestrian re-identification method and device |
| CN111027504A (en) * | 2019-12-18 | 2020-04-17 | 上海眼控科技股份有限公司 | Face key point detection method, device, equipment and storage medium |
| CN111126307B (en) * | 2019-12-26 | 2023-12-12 | 东南大学 | Small sample face recognition method combining sparse representation neural network |
| CN111126307A (en) * | 2019-12-26 | 2020-05-08 | 东南大学 | Small sample face recognition method of joint sparse representation neural network |
| CN111460945A (en) * | 2020-03-25 | 2020-07-28 | 亿匀智行(深圳)科技有限公司 | Algorithm for acquiring 3D expression in RGB video based on artificial intelligence |
| CN113688657A (en) * | 2020-05-19 | 2021-11-23 | 深圳云天励飞技术有限公司 | Face recognition method and device, electronic equipment and storage medium |
| CN111739070A (en) * | 2020-05-28 | 2020-10-02 | 复旦大学 | Real-time multi-pose face detection algorithm based on progressive calibration type network |
| CN111739070B (en) * | 2020-05-28 | 2022-07-22 | 复旦大学 | Real-time multi-pose face detection algorithm based on progressive calibration type network |
| CN111814571B (en) * | 2020-06-12 | 2024-07-12 | 深圳禾思众成科技有限公司 | Mask face recognition method and system based on background filtering |
| CN111814571A (en) * | 2020-06-12 | 2020-10-23 | 深圳禾思众成科技有限公司 | Mask face recognition method and system based on background filtering |
| CN111783670A (en) * | 2020-07-02 | 2020-10-16 | 哈尔滨理工大学 | Face recognition method based on neural network and face composition |
| CN111914665A (en) * | 2020-07-07 | 2020-11-10 | 泰康保险集团股份有限公司 | Face shielding detection method, device, equipment and storage medium |
| CN111914665B (en) * | 2020-07-07 | 2023-06-20 | 泰康保险集团股份有限公司 | Face shielding detection method, device, equipment and storage medium |
| CN112070015B (en) * | 2020-09-08 | 2021-05-18 | 广州云从博衍智能科技有限公司 | Face recognition method, system, device and medium fusing occlusion scene |
| CN112070015A (en) * | 2020-09-08 | 2020-12-11 | 广州云从博衍智能科技有限公司 | Face recognition method, system, device and medium fusing occlusion scene |
| CN112132011A (en) * | 2020-09-22 | 2020-12-25 | 深圳市捷顺科技实业股份有限公司 | Face recognition method, device, equipment and storage medium |
| CN112132011B (en) * | 2020-09-22 | 2024-04-26 | 深圳市捷顺科技实业股份有限公司 | Face recognition method, device, equipment and storage medium |
| CN112418195A (en) * | 2021-01-22 | 2021-02-26 | 电子科技大学中山学院 | Face key point detection method and device, electronic equipment and storage medium |
| CN112990007A (en) * | 2021-03-13 | 2021-06-18 | 山东大学 | Facial expression recognition method and system based on regional grouping and internal association fusion |
| CN115131843A (en) * | 2021-03-24 | 2022-09-30 | 北京君正集成电路股份有限公司 | Method for detecting face shielding based on image segmentation |
| CN115131843B (en) * | 2021-03-24 | 2024-05-07 | 北京君正集成电路股份有限公司 | Method for detecting face shielding based on image segmentation |
| CN113158939A (en) * | 2021-04-29 | 2021-07-23 | 南京甄视智能科技有限公司 | Method and system for identifying human face shielding part |
| CN113158939B (en) * | 2021-04-29 | 2022-08-23 | 南京甄视智能科技有限公司 | Method and system for identifying human face shielding part |
| CN114360033A (en) * | 2022-03-18 | 2022-04-15 | 武汉大学 | Mask face recognition method, system and equipment based on image convolution fusion network |
| CN114360033B (en) * | 2022-03-18 | 2022-06-14 | 武汉大学 | Mask face recognition method, system and equipment based on image volume fusion network |
| CN114582002A (en) * | 2022-04-18 | 2022-06-03 | 华南理工大学 | Facial expression recognition method combining attention module and second-order pooling mechanism |
| CN114582002B (en) * | 2022-04-18 | 2024-07-09 | 华南理工大学 | Facial expression recognition method combining attention module and second-order pooling mechanism |
| CN115619410B (en) * | 2022-10-19 | 2024-01-26 | 闫雪 | Self-adaptive financial payment platform |
| CN115619410A (en) * | 2022-10-19 | 2023-01-17 | 闫雪 | Self-adaptive financial payment platform |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108805040A (en) | It is a kind of that face recognition algorithms are blocked based on piecemeal | |
| CN109543606B (en) | Human face recognition method with attention mechanism | |
| CN109522853B (en) | Face datection and searching method towards monitor video | |
| CN111445459B (en) | Image defect detection method and system based on depth twin network | |
| CN111274921B (en) | Method for recognizing human body behaviors by using gesture mask | |
| US20210034840A1 (en) | Method for Recognzing Face from Monitoring Video Data | |
| CN104766343B (en) | A kind of visual target tracking method based on rarefaction representation | |
| CN110751098A (en) | Face recognition method for generating confrontation network based on illumination and posture | |
| CN107066969A (en) | A kind of face identification method | |
| CN106570447B (en) | Based on the matched human face photo sunglasses automatic removal method of grey level histogram | |
| CN108537143B (en) | A kind of face identification method and system based on key area aspect ratio pair | |
| CN107330360A (en) | A kind of pedestrian's clothing colour recognition, pedestrian retrieval method and device | |
| CN108960076A (en) | Ear recognition and tracking based on convolutional neural networks | |
| CN106874867A (en) | A kind of face self-adapting detecting and tracking for merging the colour of skin and profile screening | |
| Wang et al. | Fake face images detection and identification of celebrities based on semantic segmentation | |
| Hu et al. | Fast face detection based on skin color segmentation using single chrominance Cr | |
| Duffner et al. | A neural scheme for robust detection of transparent logos in TV programs | |
| CN111881803B (en) | Face recognition method based on improved YOLOv3 | |
| Na et al. | A combination of feature selection and co-occurrence matrix methods for leukocyte recognition system | |
| Borah et al. | ANN based human facial expression recognition in color images | |
| Curran et al. | The use of neural networks in real-time face detection | |
| Sikarwar et al. | An edge based efficient method of face detection and feature extraction | |
| Tang et al. | Research on feature enhancement for small object detection | |
| He et al. | Research on digital image recognition algorithm based on modular intelligent image recognition | |
| Wang | Adaptive preprocessing of character recognition image based on neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181113 |