CN106096538B - Face identification method and device based on sequencing neural network model - Google Patents
Face identification method and device based on sequencing neural network model Download PDFInfo
- Publication number
- CN106096538B CN106096538B CN201610403028.9A CN201610403028A CN106096538B CN 106096538 B CN106096538 B CN 106096538B CN 201610403028 A CN201610403028 A CN 201610403028A CN 106096538 B CN106096538 B CN 106096538B
- Authority
- CN
- China
- Prior art keywords
- training
- face
- neural network
- sequencing
- 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
- 238000012163 sequencing technique Methods 0.000 title claims abstract description 53
- 238000003062 neural network model Methods 0.000 title claims abstract description 40
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 63
- 230000001815 facial effect Effects 0.000 claims abstract description 45
- 238000013528 artificial neural network Methods 0.000 claims abstract description 24
- 238000013527 convolutional neural network Methods 0.000 claims description 11
- 238000004422 calculation algorithm Methods 0.000 claims description 9
- 238000005286 illumination Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000012216 screening Methods 0.000 claims description 2
- 238000013145 classification model Methods 0.000 claims 1
- 210000004218 nerve net Anatomy 0.000 claims 1
- 238000000605 extraction Methods 0.000 abstract 1
- 230000006870 function Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000001537 neural effect Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- G06V40/168—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
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/165—Detection; Localisation; Normalisation using facial parts and geometric relationships
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Physics & Mathematics (AREA)
- Human Computer Interaction (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention discloses a kind of face identification method and device based on sequencing neural network model.This method comprises: the facial image to input carries out pretreatment operation, the angle and expression of facial image are corrected；Facial image/video feature has been corrected using the neural network extraction comprising ordering operation；The similarity between image pair is calculated according to the feature representation of facial image, to learn the identity of special object in input facial image.The present invention is in recognition of face problem, and human face recognition model parameter neural network based is more, the big problem of computing cost, proposes sequencing neural network structure, efficiently reduces network parameter by the sequencing expression between different characteristic, saves and calculate the time；And the problem less for training data, propose the training method based on comparison loss, triple loss.
Description
Technical field
The present invention relates to artificial intelligence, pattern-recognition, the technical fields such as Digital Image Processing, and in particular to one kind is based on fixed
The face identification method and device of sequence neural network model.
Background technique
As one kind of biometrics identification technology, recognition of face since its is untouchable and the convenient feature of acquisition,
With good development and application prospect.Face recognition technology has all played highly important effect in many application scenarios,
Such as airport security, frontier inspection clearance etc..In recent years with the high speed development of internet finance, face recognition technology is in mobile payment
On show great application advantage.The purpose of recognition of face is to learn user's according to the user's facial image or video of acquisition
Identity.Currently, face recognition technology is still unable to satisfy real requirement under outdoor uncontrolled environment, Major Difficulties are illumination
It variation, user's posture expression shape change, age pattern of body form change and blocks.
In recent years, deep learning all achieves the effect to attract people's attention in the various fields of machine vision.Wherein look steadily the most
Purpose model surely belongs to convolutional neural networks, which, can be with abstract image or video counts using multilayer convolutional layer and pond layer
Effective hierarchical feature, realizes stronger non-linear expression in.Convolutional neural networks are in object classification, action recognition, figure
As fields such as segmentation and recognitions of face, the effect for being significantly stronger than conventional method is achieved.In some Low Level Vision problems,
Such as image denoising, image super-resolution enhancing, the problems such as image deblurring in, depth learning technology also all achieves good
Effect.In field of face identification, the face identification method based on neural network and deep learning also due to its excellent performance and
It is concerned, front-runner's face recognizer is mostly based on deep learning model both at home and abroad at present.Face based on deep learning
Recognition methods is generally divided into two steps: calculating a mark sheet using facial image of the neural network model to input first
It reaches；Then facial image is obtained according to the similitude between feature representation.
With the arriving of big data era, the data scale that we need to be handled is often very big, the speed of face recognition algorithms
The high efficiency ever more important of degree.Especially in mobile payment field, the request memory and speed of face recognition algorithms are directly affected
The waiting time of user.Therefore, there is an urgent need to develop a kind of face recognition algorithms at present, can guarantee the same of high discrimination
When, meet the high requirement of lightweight, Generalization Capability.
Summary of the invention
(1) technical problems to be solved
In order to solve to improve the accuracy rate of face recognition algorithms, while guaranteeing recognizer rapidly and efficiently, the present invention mentions
A kind of face identification method based on sequencing neural network model is gone out.Using sequencing neural unit, by keeping different levels
Ordering relationship between feature excavates the validity feature in input picture or video.The spy itself having due to sequencing neural unit
Sign selection characteristic has the characteristics of lightweight so that often parameter amount is smaller for the neural network model comprising sequencing neural unit,
To ensure that the faster calculating speed of face recognition algorithms and lesser storage demand.
(2) technical solution
The invention proposes a kind of face identification methods based on sequencing neural network model, comprising:
Step S1, the images to be recognized of input is read in, face location and key point confidence in images to be recognized are detected
Breath；
Step S2, pretreatment operation is carried out to images to be recognized according to the face location information and key point information；
Step S3, pretreated images to be recognized is input in sequencing neural network model, obtains images to be recognized
Feature representation；
Step S4, the similarity of known facial image feature in the feature representation and database of images to be recognized is calculated, with
Identify images to be recognized.
The invention also provides a kind of face identification devices based on sequencing neural network model characterized by comprising
Input module detects face location and key point in images to be recognized for reading in the images to be recognized of input
Location information；
Preprocessing module, for being pre-processed according to the face location information and key point information to images to be recognized
Operation；
Feature obtains module and obtains for pretreated images to be recognized to be input in sequencing neural network model
The feature representation of images to be recognized；
Identification module, the feature representation for calculating images to be recognized are similar to facial image feature known in database
Degree, to identify images to be recognized.
The present invention is in recognition of face problem, and human face recognition model parameter neural network based is more, and computing cost is big
The problem of, it proposes sequencing neural network structure, network parameter is efficiently reduced by the sequencing expression between different characteristic, saves meter
Evaluation time；And the problem less for training data, propose the training method based on comparison loss, triple loss.This hair
The sequencing neural network model of bright use can be used in the problems such as image/video classification, image retrieval, recognition of face, protect
While demonstrate,proving high-accuracy, has less network parameter compared to existing neural network model, so that carrying cost, calculating
Cost substantially reduces, each task being more adaptive under big data scene.For recognition of face, method used by inventing is not
The parameter amount of neural network model is only effectively reduced, and the generalization ability of face representation can be obviously improved, compares equivalent parameters
Face recognition accuracy rate is greatly improved for amount and the model for calculating the time.
Detailed description of the invention
Fig. 1 is the method flow diagram of the face identification method based on sequencing neural network model in the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with specific example, and referring to detailed
Thin attached drawing, the present invention is described in more detail.But described embodiment is intended merely to facilitate the understanding of the present invention, and right
It does not play any restriction effect.
Fig. 1 is the method flow diagram of the face identification method proposed by the present invention based on sequencing neural network model, such as Fig. 1
It is shown, the face identification method proposed by the present invention based on sequencing neural network model including the following steps:
Step S1, the facial image or video of input are read in, the face location information in input picture or video frame is detected
With key point location information；
In one embodiment, according to the facial image or video of the input, application image recognizer detects face position
Confidence breath, and according to gained face location information, application image recognizer obtains the key point location information of face.Wherein,
The key point of face is predetermined, such as eyes, nose, mouth profile, face week profile.
Step S2, according to the face location information and key point location information to the face in input picture or video frame
Image carries out pretreatment operation.The pretreatment operation includes attitude updating and light correction；
In one embodiment, the attitude updating comprises determining that key point position and the illumination condition of standard face, then root
The face key point position of input picture is aligned to standard face according to the step S1 face location information obtained and key point information
Key point position, to achieve the purpose that correct human face posture；Wherein, key point position and the illumination item of standard face can be pre-defined
Part, or be directly used on training set and average face is calculated as standard face, then determine its key point location information and light
According to condition standard face；
In one embodiment, the light correction includes by image processing algorithm, extremely by the light change of facial image
It is consistent with standard face.
Attitude updating and the number of operations that light corrects are unlimited, and its sequencing is interchangeable.
Step S3, it will be input in sequencing neural network model by pretreated facial image, and obtain facial image
Feature representation；The sequencing neural network model first passes through training in advance and obtains.
Further, the step S3 includes:
Step S3-1, training one is used to from the sequencing nerve for having been subjected to calculating feature representation in pretreated facial image
Network model.The neural network model includes sequencing neural network unit.Sequencing neural network unit is neural network model
In a kind of activation primitive.It is different from the activation primitive of single-input single-outputs such as sigmoi, ReLU common in deep learning, it is fixed
Sequence neural network unit takes the form of multi input, can obtain the sequencing expression between multiple inputs.
In one embodiment, a canonical form of the sequencing neural network unit are as follows:
Wherein I1, I2Respectively two inputs of sequencing neural network unit, Y are the output of sequencing neural network unit.I,
J is the index of input, output image on two x, y direction respectively.It can be seen that from above formula, the maximum of sequencing neural network unit
Value Operations are that step-by-step carries out, that is, exporting image, each is each maximum value for inputting corresponding position.That is the sequencing neural network list
The output of member is to express a sequencing of input.It is worth noting that, the operation that is maximized in formula is not that sequencing nerve is single
The sole operation form of member, maxima operation also can be replaced the common mathematical operations such as minimum value, average value and poor, product quotient.It is fixed
The input of sequence neural network unit is also not necessarily limited to two, can be the combination of multiple inputs.I.e. sequencing neural network unit may be used also
It is extended to following form:
Sequencing neural network uses sequencing neural network unit as activation primitive, can learn to export out automatically and appoint to target
Sequencing expression between effective feature of being engaged in.
Specifically, step S3-1 further comprises:
Use softmax loss, comparison loss and triple loss as objective function one facial image feature of training
Model is extracted, the input of the network is the facial image x of normal size, exports the facial image feature representation f for regular length
(x)。
Step S3-1-1, using the training data of collection, training convolutional neural networks disaggregated model is used to training sample
In facial image classification.The output layer of convolutional neural networks disaggregated model is a classifier, receives facial image spy
Sign expression f (x) can be used for the classification of calculating input image as input, output valve.For the disaggregated model with N class, network
Output have N number of node.
In the step, using a series of Classification Loss functions that softmax loss function is representative as optimization aim, instruction
Facial image disaggregated model is got, wherein softmax loss function is as follows:
Wherein, N is class number；X is input facial image；y∈RN×1It is the categorization vector for indicating facial image classification,
If training sample belongs to the i-th class, only i-th dimension is 1 in categorization vector y, other dimensions are equal to 0；Represent Neural Network Science
The classifier arrived,Neural network is represented in the output of i-th of node of output layer.
S3-1-2: using in step S3-1-1 trained facial image Classification Neural model as the mould of pre-training
Type is continued using comparison loss (contrastive loss) and triple loss (triplet loss) to neural network model
Optimize training.
Training obtains the output layer of model in removal S3-1-1, and the output f (x) for obtaining model rest layers is facial image
Feature representation.
Compare the optimization aim of loss are as follows:
Wherein f (x) is the input of disaggregated model output layer in step S3-1-1, i.e., the output of secondary last layer is facial image
Feature representation.D () can be for using L2 distance, COS distance as a series of distance functions of representative.It, can be random in training process
The sample that combined training is concentrated constructs sample pair between sample pair and class in class.θ is a ginseng for controlling between class distance difference in class
Number, is set based on experience value.
The optimization aim of triple loss are as follows:
L=max (d (f (xa), f (xp))-d(f(xa)+α, f (xn)), 0)
Wherein f (x) is the input of disaggregated model output layer in step S3-1-1, i.e., the output of secondary last layer is facial image
Feature representation.D () can be for using L2 distance, COS distance as a series of distance functions of representative.xaReferred to as central sample, xpFor
With central sample xaBelong to of a sort positive sample, xnFor with central sample xaBelong to inhomogeneous negative sample.A is in control class
One parameter of between class distance difference, is set based on experience value.In training process, sample can be selected at random in training set
Then sample centered on this chooses sample similar with its/inhomogeneous in remaining sample and constructs triple.
In one embodiment, in the training process of step S3-1-2, filtering out difficult sample group cooperation is training data.Difficult sample
The specific steps of this screening are as follows:
It is random to generate binary combination and triple for large capacity training set, calculate separately its corresponding comparison loss with
Triple loss, skimming loss value are lower than the combination of preset threshold, only retain the biggish a collection of training sample combination of penalty values and send
Enter model to be trained.
For low capacity training set, the similarity between all samples is calculated first.Then minimum similar of similarity is chosen
Sample is used as difficult negative sample as difficult positive sample and the highest inhomogeneity sample of similarity, finally the positive sample of hardly possible to filter out
This is sent into model as training sample with difficult negative sample and is trained.
With the progress of network training, continuous regularized learning algorithm rate simultaneously screens difficult sample feeding training, until training loss is not
It reduces again, to obtain final model.
Step S3-2, it using the neural network model that training obtains in step S3-1, will be pre-processed obtained in step S2
Input of the facial image afterwards as neural network, executes the forward calculation of neural network model, obtained neural network model
Output be input facial image feature representation.
Step S4, the similarity for calculating facial image feature in the facial image feature and database that step S3 is obtained, sentences
User identity in disconnected input facial image.
Case study on implementation:
For the specific embodiment and verifying effectiveness of the invention that the present invention will be described in detail, we propose the present invention
Method be applied to a disclosed face database --- LFW face database.The database includes 5749 people, altogether
13233 width images.
In our embodiment, we prove effectiveness of the invention using the standard test protocols of LFW data set.
The standard test protocols of LFW data set are made of 6000 pairs of facial images, wherein comprising 3000 pairs of same persons facial image with
And the facial image of 3000 pairs of different peoples.
Specific step is as follows:
Training process:
Step S3-1 collects a large amount of facial images as training data, designs neural network model.Particularly, Wo Mensuo
The use of neural network model includes 4 convolutional layers and 4 pond layers, output is divided into two groups after each pond layer, connection is most
The sequencing neural unit of big Value Operations.According to the sequence of step S3-1, damaged using softmax loss, comparison loss with triple
It loses and model is trained as optimization object function.With the progress of network training, continuous regularized learning algorithm rate simultaneously screens difficult sample
This feeding training, until training loss no longer reduces, to obtain final model.
Test process:
Step S1, we carry out Face datection and critical point detection to all input pictures first, obtain all input figures
The face location information and key point location information of picture.
Step S2, the face location information and key point information obtained according to previous step carry out posture school to facial image
Just, the pretreatment operations such as illumination balance.Specifically, for LFW data set, we will input face using rotation and scaling
Image rectification is to front face.
Input of the pretreated facial image that step S3, step S2 are obtained as neural network executes neural network
The forward calculation of model obtains the feature representation of 6000 pairs of facial images.
Step S4 calculates the COS distance of 6000 pairs of facial images as its similarity.Similarity threshold is adjusted, can be obtained
To correct percent of pass, misclassification rate and the discrimination under each threshold value.
| TPR@FPR=1% | 92.73% |
| TPR@FPR=0.1% | 84.83% |
| TPR@FPR=0 | 74.63% |
| Accuracy rate | 98.18% |
| Calculate the time | 78ms |
| Parameter model size | 26.3MB |
Table 1 is recognition accuracy of the present invention on LFW database
Table 1 illustrates our correct percent of pass of the method when misclassification rate is 0,0.1%, 1%, and under optimal threshold
Discrimination.The storage size and single picture that the deep learning model parameter file that table 1 also illustrates us needs are special
Sign extracts the time used.The comparable human face recognition model of performance is compared in the world, our method calculating speed faster, storage
Expense is smaller.
Particular embodiments described above has carried out further in detail the purpose of the present invention, technical scheme and beneficial effects
It describes in detail bright, it should be understood that the above is only a specific embodiment of the present invention, is not intended to restrict the invention, it is all
Within the spirit and principles in the present invention, any modification, equivalent substitution, improvement and etc. done should be included in guarantor of the invention
Within the scope of shield.
Claims (8)
1. a kind of face identification method based on sequencing neural network model characterized by comprising
Step S0, using the training sample in training set, using Classification Loss function as optimization aim, training convolutional nerve net
Network disaggregated model, to classify to the face in training sample, using trained convolutional neural networks disaggregated model as
The model of pre-training, using comparison loss function and triple loss function continue to the convolutional neural networks disaggregated model into
Row training, obtains sequencing neural network model；
Step S1, the images to be recognized of input is read in, face location and key point location information in images to be recognized are detected；
Step S2, pretreatment operation is carried out to images to be recognized according to the face location information and key point information；
Step S3, pretreated images to be recognized is input in sequencing neural network model, obtains the spy of images to be recognized
Sign expression；
Step S4, the similarity of known facial image feature in the feature representation and database of images to be recognized is calculated, with identification
Images to be recognized；
Wherein, the comparison loss function indicates as follows:
Wherein, θ is to compare the parameter that between class distance difference in class is controlled in loss function, the triple loss function such as following table
Show:
L=max (d (f (xa), f (xp))-d(f(xa)+α, f (xn)), 0)
Wherein f (x) is the input of the model output layer of the pre-training, i.e. the output of the model of pre-training time last layer；D () is
Distance function, xaReferred to as central sample, xpFor with central sample xaBelong to of a sort positive sample, xnFor with central sample xaBelong to
Inhomogeneous negative sample；In training process, central sample xaFor the sample selected at random in training set, a is triple damage
Lose the parameter that between class distance difference in class is controlled in function.
2. the face identification method according to claim 1 based on sequencing neural network model, which is characterized in that step S1
Further include:
For images to be recognized, application image recognizer detects face location information, and according to resulting face location information,
The key point location information of application image recognizer acquisition face；Wherein, the key point of face is pre-defined.
3. the face identification method according to claim 1 based on sequencing neural network model, which is characterized in that the step
Pretreatment operation described in rapid S2 includes attitude updating and light correction；The attitude updating comprises determining that the key of standard face
Point position and illumination condition, then according to the face location information and key point information by the face key point of input picture
It sets and is aligned to standard face key point position, to achieve the purpose that correct human face posture；Wherein, the key of standard face can be pre-defined
Point position and illumination condition, or be directly used on training set and average face is calculated as standard face, then determine its key
Dot position information and illumination condition standard face；
The light correction includes passing through image processing algorithm, and the light change of facial image is extremely consistent with standard face.
4. the face identification method according to claim 1 based on sequencing neural network model, which is characterized in that the step
The sequencing neural network model in rapid S3 includes sequencing neural network unit, for obtaining the sequencing table between multiple inputs
It reaches.
5. the face identification method according to claim 4 based on sequencing neural network model, which is characterized in that described fixed
Sequence expression includes maximum value, minimum value, average value and difference or product quotient.
6. the face identification method as described in claim 1 based on sequencing neural network model, which is characterized in that the classification
Loss function is softmax loss function, following to indicate:
Wherein, N is class number；X is input；F (x) is input facial image feature representation；y∈RN×1It is the class for indicating output
Other vector；The classifier that training obtains is represented,Represent the output of convolutional neural networks disaggregated model
The output of i-th of node of layer.
7. the face identification method as described in claim 1 based on sequencing neural network model, which is characterized in that the use
Comparison loss function and/or triple loss function filter out in continuing to be trained the convolutional neural networks disaggregated model
Difficult positive sample and difficult negative sample are as training data, the specific steps of the hardly possible positive sample and difficult negative sample screening are as follows:
It is random to generate binary combination and triple for large capacity training set, calculate separately its corresponding comparison loss and/or three
Tuple loss, skimming loss value are lower than the combination of preset threshold, are trained using the training sample combination remained；
For low capacity training set, the similarity between all training samples is calculated first, then chooses minimum similar of similarity
Training sample is used as difficult negative sample as difficult positive sample and the highest inhomogeneity sample of similarity, finally the difficulty to filter out
Positive sample is trained with difficult negative sample as training sample.
8. a kind of face identification device based on sequencing neural network model characterized by comprising
Training module, for utilizing the training sample in training set, using Classification Loss function as optimization aim, training convolutional
Neural network classification model, to classify to the face in training sample, by trained convolutional neural networks classification mould
Model of the type as pre-training continues to classify to the convolutional neural networks using comparison loss function and triple loss function
Model is trained, and obtains sequencing neural network model；
Input module detects the face location in images to be recognized and key point position for reading in the images to be recognized of input
Information；
Preprocessing module, for carrying out pretreatment behaviour to images to be recognized according to the face location information and key point information
Make；
Feature obtains module and obtains for pretreated images to be recognized to be input in sequencing neural network model wait know
The feature representation of other image；
Identification module, the similarity of known facial image feature in the feature representation and database for calculating images to be recognized,
To identify images to be recognized；
Wherein, the comparison loss function indicates as follows:
Wherein, θ is to compare the parameter that between class distance difference in class is controlled in loss function, the triple loss function such as following table
Show:
L=max (d (f (xa), f (xp))-d(f(xa)+α, f (xn)), 0)
Wherein f (x) is the input of the model output layer of the pre-training, i.e. the output of the model of pre-training time last layer；D () is
Distance function, xaReferred to as central sample, xpFor with central sample xaBelong to of a sort positive sample, xnFor with central sample xaBelong to
Inhomogeneous negative sample；In training process, central sample xaFor the sample selected at random in training set, α is triple damage
Lose the parameter that between class distance difference in class is controlled in function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610403028.9A CN106096538B (en) | 2016-06-08 | 2016-06-08 | Face identification method and device based on sequencing neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610403028.9A CN106096538B (en) | 2016-06-08 | 2016-06-08 | Face identification method and device based on sequencing neural network model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106096538A CN106096538A (en) | 2016-11-09 |
| CN106096538B true CN106096538B (en) | 2019-08-23 |
Family
ID=57229125
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610403028.9A Active CN106096538B (en) | 2016-06-08 | 2016-06-08 | Face identification method and device based on sequencing neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106096538B (en) |
Families Citing this family (48)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108229269A (en) * | 2016-12-31 | 2018-06-29 | 深圳市商汤科技有限公司 | Method for detecting human face, device and electronic equipment |
| CN106709528A (en) * | 2017-01-10 | 2017-05-24 | 深圳大学 | Method and device of vehicle reidentification based on multiple objective function deep learning |
| CN108460464A (en) * | 2017-02-22 | 2018-08-28 | 中兴通讯股份有限公司 | Deep learning training method and device |
| CN107016708B (en) * | 2017-03-24 | 2020-06-05 | 杭州电子科技大学 | Image hash coding method based on deep learning |
| CN107480785B (en) * | 2017-07-04 | 2021-01-29 | 北京小米移动软件有限公司 | Convolutional neural network training method and device |
| CN107609493B (en) * | 2017-08-25 | 2021-04-13 | 广州视源电子科技股份有限公司 | Method and device for optimizing human face image quality evaluation model |
| CN107678845B (en) * | 2017-09-30 | 2020-03-10 | Oppo广东移动通信有限公司 | Application program control method and device, storage medium and electronic equipment |
| CN107679513B (en) * | 2017-10-20 | 2021-07-13 | 北京达佳互联信息技术有限公司 | Image processing method and device and server |
| CN107729873A (en) * | 2017-11-03 | 2018-02-23 | 北京奇虎科技有限公司 | Special object detection method, device and computing device in video file |
| CN107797831B (en) * | 2017-11-14 | 2021-06-01 | Oppo广东移动通信有限公司 | Background application cleaning method and device, storage medium and electronic equipment |
| CN107944363B (en) * | 2017-11-15 | 2019-04-26 | 北京达佳互联信息技术有限公司 | Face image processing process, system and server |
| CN107818314B (en) * | 2017-11-22 | 2019-03-19 | 北京达佳互联信息技术有限公司 | Face image processing method, device and server |
| CN108182394B (en) * | 2017-12-22 | 2021-02-02 | 浙江大华技术股份有限公司 | Convolutional neural network training method, face recognition method and face recognition device |
| CN108205661A (en) * | 2017-12-27 | 2018-06-26 | 浩云科技股份有限公司 | A kind of ATM abnormal human face detection based on deep learning |
| CN108108807B (en) * | 2017-12-29 | 2020-06-02 | 北京达佳互联信息技术有限公司 | Learning type image processing method, system and server |
| CN108171209B (en) * | 2018-01-18 | 2021-07-16 | 中科视拓（北京）科技有限公司 | Face age estimation method for metric learning based on convolutional neural network |
| CN108921019B (en) * | 2018-05-27 | 2022-03-08 | 北京工业大学 | Gait recognition method based on GEI and TripletLoss-DenseNet |
| CN108830201B (en) * | 2018-06-01 | 2020-06-23 | 平安科技（深圳）有限公司 | Method and device for acquiring sample triple, computer equipment and storage medium |
| CN108921791A (en) * | 2018-07-03 | 2018-11-30 | 苏州中科启慧软件技术有限公司 | Lightweight image super-resolution improved method based on adaptive important inquiry learning |
| CN109190446A (en) * | 2018-07-06 | 2019-01-11 | 西北工业大学 | Pedestrian's recognition methods again based on triple focused lost function |
| CN109190654A (en) * | 2018-07-09 | 2019-01-11 | 上海斐讯数据通信技术有限公司 | The training method and device of human face recognition model |
| CN110766152B (en) * | 2018-07-27 | 2023-08-04 | 富士通株式会社 | Method and apparatus for training deep neural networks |
| CN109359541A (en) * | 2018-09-17 | 2019-02-19 | 南京邮电大学 | A kind of sketch face identification method based on depth migration study |
| CN109977639B (en) * | 2018-10-26 | 2021-05-04 | 招商银行股份有限公司 | Identity authentication method and device and computer readable storage medium |
| CN111402027B (en) * | 2018-12-14 | 2021-05-07 | 马上消费金融股份有限公司 | Identity recognition method, commodity loan auditing method, device and terminal equipment |
| CN109657612B (en) * | 2018-12-19 | 2023-12-12 | 苏州纳智天地智能科技有限公司 | Quality sorting system based on facial image features and application method thereof |
| CN111723613A (en) * | 2019-03-20 | 2020-09-29 | 广州慧睿思通信息科技有限公司 | Face image data processing method, device, equipment and storage medium |
| CN112001211B (en) * | 2019-05-27 | 2024-04-19 | 商汤集团有限公司 | Object detection method, device, equipment and computer readable storage medium |
| CN110569737A (en) * | 2019-08-15 | 2019-12-13 | 深圳华北工控软件技术有限公司 | Face recognition deep learning method and face recognition acceleration camera |
| CN110674688B (en) * | 2019-08-19 | 2023-10-31 | 深圳力维智联技术有限公司 | Face recognition model acquisition method, system and medium for video monitoring scene |
| CN110674941B (en) * | 2019-09-25 | 2023-04-18 | 南开大学 | Data encryption transmission method and system based on neural network |
| CN110674881B (en) * | 2019-09-27 | 2022-02-11 | 长城计算机软件与系统有限公司 | Trademark image retrieval model training method, system, storage medium and computer equipment |
| CN110837869A (en) * | 2019-11-11 | 2020-02-25 | 深圳市商汤科技有限公司 | Image classification model training method, image processing method and device |
| CN110889348A (en) * | 2019-11-15 | 2020-03-17 | 亚信科技（中国）有限公司 | Method and device for improving success rate of face recognition under complex light |
| CN110866962B (en) * | 2019-11-20 | 2023-06-16 | 成都威爱新经济技术研究院有限公司 | Virtual portrait and expression synchronization method based on convolutional neural network |
| CN111598124B (en) * | 2020-04-07 | 2022-11-11 | 深圳市商汤科技有限公司 | Image processing device, image processing apparatus, processor, electronic apparatus, and storage medium |
| CN111582141B (en) * | 2020-04-30 | 2023-05-09 | 京东方科技集团股份有限公司 | Face recognition model training method, face recognition method and device |
| CN111582199B (en) * | 2020-05-12 | 2023-05-23 | 佛山市玖章智能科技有限公司 | Face recognition model training method and face recognition method |
| CN112232117A (en) * | 2020-09-08 | 2021-01-15 | 深圳微步信息股份有限公司 | Face recognition method, face recognition device and storage medium |
| CN112115860B (en) * | 2020-09-18 | 2024-07-26 | 深圳市威富视界有限公司 | Face key point positioning method, device, computer equipment and storage medium |
| WO2022126464A1 (en) * | 2020-12-17 | 2022-06-23 | 深圳市优必选科技股份有限公司 | Face recognition method, robot, and storage medium |
| CN112508130A (en) * | 2020-12-25 | 2021-03-16 | 商汤集团有限公司 | Clustering method and device, electronic equipment and storage medium |
| CN113515999A (en) * | 2021-01-13 | 2021-10-19 | 腾讯科技（深圳）有限公司 | Training method, device and equipment of image processing model and readable storage medium |
| CN113743176A (en) * | 2021-01-29 | 2021-12-03 | 北京沃东天骏信息技术有限公司 | Image recognition method, device and computer readable storage medium |
| CN112818887B (en) * | 2021-02-08 | 2023-04-07 | 中国科学院自动化研究所 | Human skeleton sequence behavior identification method based on unsupervised learning |
| CN113282718B (en) * | 2021-07-26 | 2021-12-10 | 北京快鱼电子股份公司 | Language identification method and system based on self-adaptive center anchor |
| CN114266046A (en) * | 2021-12-13 | 2022-04-01 | 安天科技集团股份有限公司 | Network virus identification method and device, computer equipment and storage medium |
| CN115424335B (en) * | 2022-11-03 | 2023-08-04 | 智慧眼科技股份有限公司 | Living body recognition model training method, living body recognition method and related equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102629320A (en) * | 2012-03-27 | 2012-08-08 | 中国科学院自动化研究所 | Ordinal measurement statistical description face recognition method based on feature level |
| CN105608450A (en) * | 2016-03-01 | 2016-05-25 | 天津中科智能识别产业技术研究院有限公司 | Heterogeneous face identification method based on deep convolutional neural network |
-
2016
- 2016-06-08 CN CN201610403028.9A patent/CN106096538B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102629320A (en) * | 2012-03-27 | 2012-08-08 | 中国科学院自动化研究所 | Ordinal measurement statistical description face recognition method based on feature level |
| CN105608450A (en) * | 2016-03-01 | 2016-05-25 | 天津中科智能识别产业技术研究院有限公司 | Heterogeneous face identification method based on deep convolutional neural network |
Non-Patent Citations (2)
| Title |
|---|
| A Lightened CNN for Deep Face Representation;Xiang Wu, et al.;《Computer Science》;20151109;正文第1节-第4.5节，附图1 * |
| 基于卷积神经网络的三维人脸识别研究;赵亚龙;《中国优秀硕士学位论文全文数据库信息科技辑》;20160315;第2016年卷(第03期);I138-6611 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106096538A (en) | 2016-11-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106096538B (en) | Face identification method and device based on sequencing neural network model | |
| CN107273845B (en) | Facial expression recognition method based on confidence region and multi-feature weighted fusion | |
| CN106815566B (en) | Face retrieval method based on multitask convolutional neural network | |
| Guo et al. | Face recognition based on convolutional neural network and support vector machine | |
| CN107194341B (en) | Face recognition method and system based on fusion of Maxout multi-convolution neural network | |
| CN104143079B (en) | The method and system of face character identification | |
| CN109409297B (en) | Identity recognition method based on dual-channel convolutional neural network | |
| CN109492529A (en) | A kind of Multi resolution feature extraction and the facial expression recognizing method of global characteristics fusion | |
| CN106407958B (en) | Face feature detection method based on double-layer cascade | |
| CN108268859A (en) | A kind of facial expression recognizing method based on deep learning | |
| CN109858466A (en) | A kind of face critical point detection method and device based on convolutional neural networks | |
| CN108090830B (en) | Credit risk rating method and device based on facial portrait | |
| CN104834941A (en) | Offline handwriting recognition method of sparse autoencoder based on computer input | |
| CN108563999A (en) | A kind of piece identity's recognition methods and device towards low quality video image | |
| CN106372581A (en) | Method for constructing and training human face identification feature extraction network | |
| CN102270308B (en) | Facial feature location method based on five sense organs related AAM (Active Appearance Model) | |
| CN111401145B (en) | Visible light iris recognition method based on deep learning and DS evidence theory | |
| CN104700076A (en) | Face image virtual sample generating method | |
| Yang et al. | An ensemble classification algorithm for convolutional neural network based on AdaBoost | |
| CN110781897A (en) | Semantic edge detection method based on deep learning | |
| CN103065158A (en) | Action identification method of independent subspace analysis (ISA) model based on relative gradient | |
| CN107220598A (en) | Iris Texture Classification based on deep learning feature and Fisher Vector encoding models | |
| CN110472495A (en) | A kind of deep learning face identification method based on graphical inference global characteristics | |
| CN103544478A (en) | All-dimensional face detection method and system | |
| Wang et al. | Research on face recognition technology based on PCA and SVM |
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 |