CN108009528A - Face authentication method, device, computer equipment and storage medium based on Triplet Loss - Google Patents
Face authentication method, device, computer equipment and storage medium based on Triplet Loss Download PDFInfo
- Publication number
- CN108009528A CN108009528A CN201711436879.4A CN201711436879A CN108009528A CN 108009528 A CN108009528 A CN 108009528A CN 201711436879 A CN201711436879 A CN 201711436879A CN 108009528 A CN108009528 A CN 108009528A
- Authority
- CN
- China
- Prior art keywords
- sample
- facial image
- scene
- training
- certificate
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Collating Specific Patterns (AREA)
Abstract
The present invention relates to a kind of face authentication method based on Triplet Loss, device, computer equipment and storage medium, this method includes：Asked based on face authentication, obtain certificate photograph and the scene photo of personage；Carry out Face datection, crucial point location and image preprocessing respectively to scene photo and certificate photograph, obtain the corresponding scene facial image of scene photo, and the corresponding certificate facial image of certificate photograph；Scene facial image and certificate facial image are input to the advance trained convolutional neural networks model for face authentication, and the corresponding first eigenvector of scene facial image of convolutional neural networks model output is obtained, and the corresponding second feature vector of certificate facial image；Calculate the COS distance of first eigenvector and second feature vector；Compare COS distance and predetermined threshold value, and face authentication result is determined according to comparative result.It the method increase the reliability of face authentication.
Description
Technical field
The present invention relates to technical field of image processing, more particularly to a kind of face authentication side based on Triplet Loss
Method, device, computer equipment and storage medium.
Background technology
Face authentication, refers to contrast the certificate photograph in the personage's scene photo and identity information of collection in worksite, judges
Whether it is same person.The key technology of face authentication is recognition of face.
With the rise of depth learning technology, the relevant issues of recognition of face constantly break through traditional technical bottleneck, performance
Level is highly improved.In the research work of recognition of face is solved the problems, such as with deep learning, mainly there are two to send mainstream
Method：Method based on classification learning and the method based on metric learning.Wherein, the method based on classification learning mainly exists
Depth convolutional network extraction feature after calculate sample Classification Loss function (such as softmax loss, center loss and
Related variants) network is optimized, last layer of network is the full articulamentum for classification, the quantity of its output node is past
Toward to be consistent with total classification number of training dataset, such method is more suitable for training sample, especially same category
Training sample than more rich situation, network can obtain preferable training effect and generalization ability.But when classification number reaches
During hundreds thousand of or higher amount level, the last classification layer of network (full articulamentum) parameter amount can linearly increase and quite huge,
Network is caused to be difficult to train.
Another kind of method is the method based on metric learning, this method tissue training's sample (such as two in a manner of tuple
Tuple pair or triple triplet), need not be by layer of classifying after depth convolutional network, and it is based on directly on convolution
The measurement loss (such as contrastive loss, triplet loss etc.) that feature vector is calculated between sample to carry out network
Optimization, this method need not train classification layer, therefore the influence that network parameter amount increases from classification number, to training dataset
Classification number is not limited, it is only necessary to chooses the similar or suitable tuple of foreign peoples's sample architecture according to corresponding strategy.Compared to classification
Learning method, metric learning method is more suitable for that training data range is larger but depth is insufficient, and (sample class number is more, but similar sample
Originally situation less), by the various combination between sample, can construct quite abundant tuple data and be used to train, spend at the same time
Amount mode of learning focuses more on tuple internal relations, for 1:This kind of judgement of 1 face verification is that problem has that its is congenital with no
Advantage.
In practical applications, many mechanisms require that system of real name is registered, for example, open a bank account, phone number registration, gold
Melt account to open an account etc..System of real name register request user carry identity card arrive the place specified, by staff verification in person with
After the photo of identity card corresponds to, can open an account success.And with Internet technology develop, more and more mechanisms are proposed just
The people service, and are no longer strictly required client to specified site.The geographical location of user is unrestricted, uploads identity card, and utilize shifting
Personage's scene photo at the image acquisition device scene of dynamic terminal, carries out face authentication, and lead in face authentication by system
Later, you can success of opening an account.And the learning method of measurement is traditionally based on, measured using Euclidean distance similar between sample
Degree, and Euclidean distance weigh be spatial points absolute distance, directly related with the position coordinates where each point, this is not
Meet the properties of distributions in face characteristic space, cause the reliability of recognition of face relatively low.
The content of the invention
Based on this, it is necessary to for traditional face authentication method reliability it is low the problem of, there is provided one kind is based on Triplet
Face authentication method, device, computer equipment and the storage medium of Loss.
A kind of face authentication method based on Triplet Loss, including：
Asked based on face authentication, obtain certificate photograph and the scene photo of personage；
Carry out Face datection, crucial point location and image preprocessing respectively to the scene photo and the certificate photograph,
Obtain the corresponding scene facial image of the scene photo, and the corresponding certificate facial image of the certificate photograph；
The scene facial image and certificate facial image are input to the advance trained convolution for face authentication
Neural network model, and obtain the corresponding fisrt feature of the scene facial image of convolutional neural networks model output to
Amount, and the corresponding second feature vector of the certificate facial image；Wherein, the convolutional neural networks model is based on triple
The supervised training of loss function obtains；
Calculate the COS distance of the first eigenvector and second feature vector；
Compare the COS distance and predetermined threshold value, and face authentication result is determined according to comparative result.
In one embodiment, the method further includes：
The training sample of tape label is obtained, the training sample includes marked a certificate for belonging to each tagged object
Facial image and at least a scene facial image；
According to the training sample training convolutional neural networks module, the corresponding ternary of each training sample is produced by OHEM
Constituent element element；The triple element includes reference sample, positive sample and negative sample；
According to the triple element of each training sample, based on the supervision of triple loss function, the training convolutional Neural
Network model；The triple loss function, using COS distance as metric form, optimizes mould by stochastic gradient descent algorithm
Shape parameter；
Verification collection data are inputted into the convolutional neural networks model, when reaching trained termination condition, are obtained trained
Convolutional neural networks model for face authentication.
In another embodiment, according to the training sample training convolutional neural networks model, produced by OHEM each
The step of training sample corresponding triple element, including：
One image of random selection selects to belong to same label object, different from reference sample classification as sample refer to
Image as positive sample；
It is right using the COS distance between currently trained convolutional neural networks model extraction feature according to OHEM strategies
In each reference sample, it is not belonging to from other in the image of the label object, chosen distance minimum and the reference sample
The image to belong to a different category, the negative sample as the reference sample.
In another embodiment, the triple loss function includes the restriction to the COS distance of similar sample, with
And the restriction of the COS distance to foreign peoples's sample.
In another embodiment, the triple loss function is：
Wherein, cos () represents COS distance, its calculation isN is triple quantity,Represent the feature vector of reference sample,Represent the feature vector of similar positive sample,Represent foreign peoples's negative sample
Feature vector, []+Implication it is as follows：α1The spacing parameter between class, α2To be spaced ginseng in class
Number.
In another embodiment, the method further includes：Increase income the trained basis of human face data using based on magnanimity
Model parameter is initialized, and addition normalization layer and triple loss function layer, obtain to be trained after feature output layer
Convolutional neural networks model.
A kind of face authentication device based on Triplet Loss, including：Image collection module, image pre-processing module,
Feature acquisition module, computing module and authentication module；
Described image acquisition module, for being asked based on face authentication, obtains certificate photograph and the scene photo of personage；
Described image pretreatment module, for the scene photo and the certificate photograph are carried out respectively Face datection,
Crucial point location and image preprocessing, obtain the corresponding scene facial image of the scene photo, and the certificate photograph pair
The certificate facial image answered；
The feature acquisition module, trains in advance for the scene facial image and certificate facial image to be input to
The convolutional neural networks model for face authentication, and obtain the scene face of convolutional neural networks model output
The corresponding first eigenvector of image, and the corresponding second feature vector of the certificate facial image；Wherein, the convolution god
Through network model, the supervised training based on triple loss function obtains；
The computing module, for calculating the COS distance of the first eigenvector and second feature vector；
The authentication module, determines that face is recognized for the COS distance and predetermined threshold value, and according to comparative result
Demonstrate,prove result.
In another embodiment, described device further includes：Sample acquisition module, triple acquisition module, training module
And authentication module；
The sample acquisition module, for obtaining the training sample of tape label, the training sample includes marked belonging to
A certificate facial image and an at least scene facial image for each tagged object；
The triple acquisition module, for according to the training sample training convolutional neural networks model, passing through OHEM
Produce the corresponding triple element of each training sample；The triple element includes reference sample, positive sample and negative sample；
The training module, for the triple element according to each training sample, the supervision of base triple loss function, instruction
Practice the convolutional neural networks model；The triple loss function, using COS distance as metric form, by under stochastic gradient
Drop algorithm carrys out Optimized model parameter；
The authentication module, for verification collection data to be inputted the convolutional neural networks model, reaches training and terminates bar
During part, the trained convolutional neural networks model for face authentication is obtained.
A kind of computer equipment, including memory, processor and storage can be run on a memory and on a processor
Computer program, the processor realize the above-mentioned face authentication based on Triplet Loss when performing the computer program
The step of method.
A kind of storage medium, is stored thereon with computer program, it is characterised in that the computer program is executed by processor
When, the step of realizing above-mentioned face authentication method based on Triplet Loss.
Face authentication method based on Triplet Loss, device, computer equipment and storage medium of the present invention,
Face authentication is carried out using convolutional neural networks trained in advance, since convolutional neural networks model is based on triple loss function
Supervised training obtain, and the similarity of scene facial image and certificate facial image is according to scene facial image corresponding first
The COS distance of feature vector and the corresponding second feature vector of certificate facial image is calculated, and what COS distance was weighed is empty
Between vectorial angle, the difference being more embodied on direction, so as to more meet the properties of distributions in face characteristic space, improves people
The reliability of face certification.
Brief description of the drawings
Fig. 1 is the structure diagram of the face authentication system based on Triplet Loss of one embodiment；
Fig. 2 is the flow chart of the face authentication method based on Triplet Loss in one embodiment；
Fig. 3 is the flow for the step of training obtains the convolutional neural networks model for face authentication in one embodiment
Figure；
Fig. 4 be spaced between class consistent, variance within clusters it is larger in the case of, the probability schematic diagram that wrong point of sample；
Fig. 5 be spaced between class consistent, variance within clusters it is smaller in the case of, the probability schematic diagram that wrong point of sample；
Fig. 6 is the schematic diagram of the transfer learning process of the face authentication based on Triplet Loss in one embodiment；
Fig. 7 is the structure diagram for the convolutional neural networks model for being used for face authentication in one embodiment；
Fig. 8 is the flow diagram of the face authentication method based on Triplet Loss in one embodiment；
Fig. 9 is the structure diagram of the face authentication device based on Triplet Loss in one embodiment；
Figure 10 is the structure diagram of the face authentication device based on Triplet Loss in another embodiment.
Embodiment
Fig. 1 is the structure diagram of the face authentication system based on Triplet Loss of one embodiment.Such as Fig. 1 institutes
Show, face authentication system includes server 101 and image collecting device 102.Wherein, server 101 and image collecting device 102
Network connection.Image collecting device 102 gathers the real-time scene photo of user to be certified, and certificate photograph, and by collection
Real-time scene photo and certificate photograph are sent to server 101.Server 101 is judged in the personage and certificate photo of scene photo
Whether personage is same people, and the identity of user to be certified is authenticated.Based on specific application scenarios, image collecting device
102 can be camera, or the user terminal with camera function.Exemplified by the scene of opening an account, image collecting device 102 can
Think camera；Exemplified by financial account is carried out by internet and is opened an account, image collecting device 102 can be with camera function
Mobile terminal.
In other embodiments, face authentication system can also include card reader, for reading certificate (such as identity card
Deng) certificate photo in chip.
Fig. 2 is the flow chart of the face authentication method based on Triplet Loss in one embodiment.As shown in Fig. 2, should
Method includes：
S202, is asked based on face authentication, obtains certificate photograph and the scene photo of personage.
Wherein, certificate photograph refers to be able to demonstrate that the photo corresponding to the certificate of piece identity, such as is printed on identity card
Certificate photo in the certificate photo or chip of system.The acquisition modes of certificate photograph can use and carry out acquisition of taking pictures to certificate, also may be used
With the certificate photograph stored by card reader reading certificate chip.Certificate in the present embodiment can be identity card, driver's license
Or social security card etc..
The scene photo of personage refers to that user to be certified is gathered in certification, the photograph of the user to be certified environment at the scene
Piece.Site environment refers to local environment of the user when taking pictures, and site environment is unrestricted.The acquisition modes of scene photo can be with
Using the mobile terminal collection scene photo with camera function and to send to server.
Face authentication, refers to contrast the certificate photograph in the personage's scene photo and identity information of collection in worksite, judges
Whether it is same person.Face authentication request is triggered based on actual application operating, for example, the account opening request based on user, is touched
Send out face authentication request.Application program carries out the acquisition operations of photo in the display interface prompting user of user terminal, and is shining
After the completion of piece collection, the photo of collection is sent to server, carries out face authentication.
S204, carries out scene photo and certificate photograph Face datection, crucial point location and image preprocessing, obtains respectively
The corresponding scene facial image of scene photo, and the corresponding certificate facial image of certificate photograph.
Face datection refers to identify photo and obtains the human face region in photo.
Crucial point location, refers to the human face region to being detected in photo, obtains position of the face key point in every photos
Put.Face key point includes eyes, nose, corners of the mouth point, eyebrow and each component outline point of face.
In the present embodiment, the concatenated convolutional neutral net MTCNN methods based on multitask combination learning can be used complete at the same time
Into Face datection and face critical point detection, can also be returned using the method for detecting human face based on LBP features and based on shape
Face critical point detection method.
Image preprocessing refers to the position according to the face key point of detection in every pictures, carry out portrait alignment and
Shear treatment, so as to obtain the normalized scene facial image of size and certificate facial image.Wherein, scene facial image refers to
The facial image obtained after Face datection, crucial point location and image preprocessing is carried out to scene photo, certificate facial image is
Refer to the facial image for carrying out being obtained after Face datection, crucial point location and image preprocessing to certificate photograph.
S206, the advance trained convolution for face authentication is input to by scene facial image and certificate facial image
Neural network model, and the corresponding first eigenvector of scene facial image of convolutional neural networks model output is obtained, and
The corresponding second feature vector of certificate facial image.
Wherein, supervision of the convolutional neural networks model based on triple loss function is trained in advance previously according to training sample
Alright.Convolutional neural networks include convolutional layer, pond layer, activation primitive layer and full articulamentum, every layer of each neuron parameter
Determined by training.Using trained convolutional neural networks, by network propagated forward, convolutional neural networks model is obtained
The first eigenvector of the scene facial image of full articulamentum output, and the corresponding second feature vector of certificate facial image.
Triple (triplet) refers to concentrate from training data selects a sample at random, which is known as reference sample, so
A sample for belonging to same people with reference sample is randomly selected again afterwards as positive sample, chooses the sample work for being not belonging to same people
For negative sample, (reference sample, positive sample, negative sample) triple is thus formed.Card is mainly based upon since the testimony of a witness compares
Part is according to the comparison shone with scene, rather than certificate photo and certificate photo or scene be according to the comparison shone with scene, therefore triple
Pattern mainly has two kinds of combinations：During using certificate photo image as reference sample, positive sample and negative sample are that scene is shone；Shone with scene
When image is reference sample, positive sample and negative sample are certificate photo.
For each sample in triple, the network of one parameter sharing of training, obtains the feature representation of three elements.
The purpose for improving triple loss (triplet loss) be exactly by learning, allow reference sample and the feature representation of positive sample it
Between distance it is as small as possible, and the distance between feature representation of reference sample and negative sample is as big as possible, and to allow reference
Have between the distance between feature representation of the distance between sample and the feature representation of positive sample and reference sample and negative sample
One minimum interval.
S208, calculates the COS distance of first eigenvector and second feature vector.
COS distance, also referred to as cosine similarity, are to be used as measurement by the use of two vectorial angle cosine values in vector space
The measurement of the size of two inter-individual differences.The COS distance of first eigenvector and second feature vector is bigger, represents scene
The similarity of facial image and certificate facial image is bigger, and the COS distance of first eigenvector and second feature vector is smaller,
Represent that the similarity of scene facial image and certificate facial image is smaller.When scene facial image and the cosine of certificate facial image
When distance is more received in 1, the probability that two images belong to same people is bigger, more than scene facial image and certificate facial image
Chordal distance is smaller, and the probability that two images belong to same people is smaller.
In traditional triple loss (triplet loss) method, measured using Euclidean distance similar between sample
Degree.And Euclidean distance weigh be spatial points absolute distance, directly related with the position coordinates where each point, this is not
Meet the properties of distributions in face characteristic space.In the present embodiment, properties of distributions and the practical application field in face characteristic space are considered
Scape, the similarity between sample is measured using COS distance.What COS distance was weighed is the angle of space vector, is more embodied
Difference on direction, rather than position, so as to more meet the properties of distributions in face characteristic space.
Specifically, the calculation formula of COS distance is：
Wherein, x represents first eigenvector, and y represents second feature vector.
S210, compares COS distance and predetermined threshold value, and determines face authentication result according to comparative result.
Authentication result includes certification by the way that i.e. certificate photograph and scene photo belongs to same people.Authentication result, which further includes, recognizes
Card failure, i.e. certificate photograph and scene photo are not belonging to same people.
Specifically, by COS distance compared with predetermined threshold value, when COS distance is more than predetermined threshold value, represent to demonstrate,prove
The similarity of part photo and scene photo is more than predetermined threshold value, and certification success, when COS distance is less than predetermined threshold value, expression is
The similarity of certificate photograph and scene photo is less than predetermined threshold value, authentification failure.
The above-mentioned face authentication method based on Triplet Loss, using convolutional neural networks trained in advance into pedestrian
Face certification, since supervised training of the convolutional neural networks model based on triple loss function obtains, and scene facial image and
The similarity of certificate facial image is according to the corresponding first eigenvector of scene facial image and certificate facial image corresponding
The COS distance of two feature vectors is calculated, and what COS distance was weighed is the angle of space vector, is more embodied on direction
Difference, so as to more meet the properties of distributions in face characteristic space, improve the reliability of face authentication.
In another embodiment, face authentication method further includes training and obtains the convolutional neural networks for face authentication
The step of model.Fig. 3 is the stream for the step of training obtains the convolutional neural networks model for face authentication in one embodiment
Cheng Tu.As shown in figure 3, the step includes：
S302, obtains the training sample of tape label, and training sample includes marked one that belongs to each tagged object card
Part facial image and at least a scene facial image.
In the present embodiment, tagged object, that is, people, training sample marked the scene people for belonging to a people taking human as unit
Face image and certificate facial image.Specifically, scene facial image and certificate facial image can be by shining the scene of tape label
Piece and certificate photograph carry out Face datection, crucial point location and image preprocessing and obtain.
Face datection refers to identify photo and obtains the human face region in photo.
Crucial point location, refers to the human face region to being detected in photo, obtains position of the face key point in every photos
Put.Face key point includes eyes, nose, corners of the mouth point, eyebrow and each component outline point of face.
In the present embodiment, the concatenated convolutional neutral net MTCNN methods based on multitask combination learning can be used complete at the same time
It is crucial into Face datection and face, it can also use the method for detecting human face based on LBP features and the face returned based on shape to close
Key point detecting method.
Image preprocessing refers to the position according to the face key point of detection in every pictures, carry out portrait alignment and
Shear treatment, so as to obtain size normalization scene facial image and certificate facial image.Wherein, scene facial image refers to pair
Scene photo carries out the facial image obtained after Face datection, crucial point location and image preprocessing, and certificate facial image refers to
The facial image obtained after Face datection, crucial point location and image preprocessing is carried out to certificate photograph.
S304, according to training sample training convolutional neural networks model, each training sample corresponding three is produced by OHEM
Tuple elements；Triple element includes reference sample, positive sample and negative sample.
Triple has two kinds of combinations：During using certificate photo image as reference sample, positive sample and negative sample are scene
According to image；When shining image as reference sample using scene, positive sample and negative sample are certificate photo image.
Specifically, by taking certificate photo is reference picture as an example, the random certificate photo sample for selecting a people is concentrated from training data,
The sample is known as reference sample, then randomly select one again and reference sample belong to the scene of same people in the same old way this as positive sample
This, choose be not belonging to the scene of same people in the same old way this be used as negative sample, thus forming one, (reference sample, positive sample, bear sample
This) triple.
I.e. positive sample and reference sample are similar sample, that is, belong to same people's image.Negative sample is the foreign peoples of reference sample
Sample, that is, be not belonging to the image of same people.Wherein, the reference sample in triple element and positive sample have been marked in training sample
Note, negative sample is in the training process of convolutional neural networks, using OHEM (Online Hard Example Mining) plan
Triple is slightly constructed online, i.e., during each iteration optimization of network, before being carried out using current network to candidate's triple
To calculating, select to be not belonging to same user in training sample with reference sample, and the nearest image of COS distance is as negative sample,
So as to obtain the corresponding triple element of each training sample.
In one embodiment, according to training sample training convolutional neural networks, and the corresponding ternary of each training sample is produced
The step of constituent element element, comprises the following steps S1 and S2：
S1：One image of random selection belongs to same label object and reference sample classification not as sample, selection is referred to
Same image is as positive sample.
Classification refers to affiliated image type, and in the present embodiment, the classification of training sample includes scene facial image and card
Part facial image.Because face authentication is mainly the contrast between certificate photo and scene photograph, therefore, reference sample and positive sample should
When belonging to different classifications, if reference sample is scene facial image, positive sample is certificate facial image；If reference sample is
Certificate facial image, then positive sample is scene facial image.
S2：According to OHEM strategies, using the COS distance between currently trained convolutional neural networks model extraction feature,
For each reference sample, it is not belonging to from other in the image of same label object, chosen distance minimum and reference sample category
In different classes of image, the negative sample as the reference sample.
Negative sample is selected from the facial image of label that same people is not belonging to reference sample, and specifically, negative sample exists
In the training process of convolutional neural networks, using the online construction triple of OHEM strategies, the i.e. mistake in each iteration optimization of network
Cheng Zhong, carries out forward calculation to candidate's triple using current network, selects to be not belonging to reference sample in training sample same
User, and COS distance is not belonging to same category of image as negative sample recently, with reference sample.That is, negative sample and reference
The classification of sample is different.If it is believed that using certificate photo as reference sample in triple, positive sample and negative sample are scenes
According to；If on the contrary shone for reference sample with scene, in addition positive sample and negative sample are certificate photos.
S306, according to the triple element of each training sample, based on the supervision of triple loss function, training convolutional nerve
Network model, the triple loss function, using COS distance as metric form, optimizes mould by stochastic gradient descent algorithm
Shape parameter.
Whether testimony of a witness verification terminal is shone with scene according to unanimously testing user identity by comparing user certificate chip
Card, background acquisition to data be often that the sample of single people only has two figures, i.e., certificate photo is with comparing the field captured constantly
Jing Zhao, and the quantity of Different Individual can be thousands of.If the few data of this larger and similar sample of categorical measure with
Be trained based on the method for classification, classification layer parameter can be excessively huge and cause network to be very difficult to learn, therefore consider
Solved with the method for metric learning.Wherein the typical of metric learning is usually to lose (triplet loss) side with triple
Method, learns a kind of effective Feature Mapping, the characteristic distance of similar sample is small under the mapping by structural map as triple
In the characteristic distance of foreign peoples's sample, so as to achieve the purpose that correctly to compare.
The purpose of triple loss (triplet loss) is exactly by learning, allowing reference sample and the mark sheet of positive sample
Up to the distance between it is as small as possible, and the distance between feature representation of reference sample and negative sample is as big as possible, and to allow
The distance between feature representation of the distance between reference sample and the feature representation of positive sample and reference sample and negative sample it
Between have a minimum interval.
In another embodiment, triple loss function includes the restriction to the COS distance of similar sample, and right
The restriction of the COS distance of foreign peoples's sample.
Wherein, similar sample refers to reference sample and positive sample, and foreign peoples's sample refers to reference sample and negative sample.Similar sample
This COS distance refers to reference sample and the COS distance of positive sample, and the COS distance of foreign peoples's sample refers to reference sample and bears
The COS distance of sample.
On the one hand, gap is without considering gap in class between original triplet loss methods simply consider class, such as
Fruit distribution within class is not enough amassed wealth by heavy taxation, and the generalization ability of network will weaken, and scene adaptability can also be decreased.On the other hand,
Original triplet loss methods measure the similarity between sample, actually faceform portion using Euclidean distance
Link can more be measured using COS distance in aspect ratio after administration.What Euclidean distance was weighed is spatial points
Absolute distance, it is directly related with the position coordinates where each point；And COS distance weigh be space vector angle, more
The difference being embodied on direction, rather than position, so as to more meet the properties of distributions in face characteristic space.
Using triplet loss methods, network is inputted by constructing triple data online, then backpropagation ternary
The measurement of group is lost to be iterated optimization.Each triple includes three images, is a reference sample respectively, one with
The similar positive sample of reference sample, and one with the negative sample of reference sample foreign peoples, labeled as (anchor, positive,
negative).The basic thought of original triplet loss is, by between metric learning reference sample and positive sample
Distance is less than the distance between reference sample and negative sample, and is more than a minimum interval parameter alpha apart from its difference.Therefore it is original
Triplet loss loss functions it is as follows：
Wherein, N is triple quantity,Represent the feature vector of reference sample (anchor),Represent similar
The feature vector of positive sample (positive),Represent the feature vector of foreign peoples's negative sample (negative).Represent L2
Normal form, i.e. Euclidean distance.[·]+Implication it is as follows：
It can be seen that from above formula, original triplet loss functions merely define similar sample (anchor, positive)
The distance between foreign peoples's sample (anchor, negative), i.e., increase between class distance as far as possible by spacing parameter α, and right
Inter- object distance is not limited in any way, i.e., does not make any constraint to the distance between similar sample.If inter- object distance is more dispersed,
Variance is excessive, and the generalization ability of network will weaken, and sample will bigger by the probability of mistake point.Fig. 4 be spaced between class it is consistent,
In the case of variance within clusters are larger, the wrong probability schematic diagram divided of sample, Fig. 5 is that consistent, the smaller situation of variance within clusters is spaced between class
Under, the wrong probability schematic diagram divided of sample, as shown in Figure 4 and Figure 5, dash area represent the probability of wrong point of sample, are spaced between class
Unanimously, in the case of variance within clusters are larger, the wrong probability divided of sample is spaced consistent, the smaller situation of variance within clusters between being significantly greater than class
Under, the wrong probability divided of sample.
In view of the above-mentioned problems, the present invention proposes improved triplet loss methods, on the one hand remain in original method
Restriction between class distance, while add the bound term to inter- object distance so that inter- object distance is amassed wealth by heavy taxation as far as possible.Its loss letter
Counting expression formula is：
Wherein, cos () represents COS distance, its calculation isN is triple quantity,Represent the feature vector of reference sample,Represent the feature vector of similar positive sample,Represent foreign peoples's negative sample
Feature vector, []+Implication it is as follows：α1The spacing parameter between class, α2To be spaced ginseng in class
Number.
Compared to original triplet loss functions, the metric forms of improved triplet loss functions by Euclidean away from
From COS distance is changed to, the uniformity of training stage and deployment phase metric form can be so kept, improves feature learning
Continuity.It is consistent with original triplet loss effects with stylish triplet loss functions Section 1, for increasing class
Between gap, Section 2 with the addition of the distance restraint to (positive tuple) to similar sample, for reducing gap in class.α1Between between class
Every parameter, value range is 0~0.2, α2For spacing parameter in class, value range is 0.8~1.0.Significantly, since
It is to be measured with cosine manner, the similarity between obtained metric two samples of correspondence, thereforeIt is only negative in expression formula
Tuple cosine similarity is in α1In the range of be more than positive tuple cosine similarity sample, just can really participate in training.
Based on improved triple loss function come training pattern, constraint is combined by Inter-class loss and Intra-class loss
Come to model carry out backpropagation optimization train so that similar sample feature space as close possible to and foreign peoples's sample in spy
Sign space is far as possible from improving the sense of model, so as to improve the reliability of face authentication.
S308, by verification collection data input convolutional neural networks, when reaching trained termination condition, obtains trained be used for
The convolutional neural networks of face authentication.
Specifically, 90% data are taken from testimony of a witness view data pond, and as training set, residue 10% is as verification collection.It is based on
Above formula calculates improved triplet loss values, feeds back in convolutional neural networks and is iterated optimization.Observe mould at the same time
The performance that type is concentrated in verification, when verifying that performance no longer raises, model reaches convergence state, and the training stage terminates.
Above-mentioned face authentication method, on the one hand adds to sample in class in the loss function of original triplet loss
The constraint of this distance, so as to reduce gap in class, the generalization ability of lift scheme while gap between increasing class；The opposing party
Face, is changed to COS distance by Euclidean distance by the metric form of original triplet loss, keeps the measurement one of training and deployment
Cause property, improves the continuity of feature learning.
In another embodiment, the step of training convolutional neural networks further include：Increase income face number using based on magnanimity
Initialized according to trained basic model parameter, addition normalization layer and improved triple damage after feature output layer
Function layer is lost, obtains convolutional neural networks to be trained.
Specifically, it is conventional to be instructed based on internet mass human face data when solving testimony of a witness unification problem with deep learning
The testimony of a witness of the depth human face recognition model got under special scenes compares and can decline to a great extent using upper performance, and application-specific
Testimony of a witness data source under scene is again than relatively limited, and directly study is often due to sample deficiency causes training result undesirable,
Therefore pole need to research and develop it is a kind of be effectively extended trained method for the contextual data of small data set, to lift face knowledge
Accuracy rate of the other model under application-specific scene, meets market application demand.
Deep learning algorithm tends to rely on the training of mass data, and in testimony of a witness unification application, certificate photo shines with scene
Comparison is belonged to heterogeneous sample and compares problem, the conventional depth recognition of face mould trained based on magnanimity internet human face data
Type is compared in the testimony of a witness and can declined to a great extent using upper performance.But testimony of a witness data source is limited (needs to be provided simultaneously with same person
ID Card Image and corresponding scene image), less available for trained data volume, direct training can be caused due to sample deficiency
Training effect is undesirable, therefore when carrying out the model training of testimony of a witness unification with deep learning, often utilizes transfer learning
Thought, first the internet human face data based on magnanimity train one in the basic model of dependable performance on test set of increasing income, so
Secondary spread training is carried out in limited testimony of a witness data again afterwards, model is learnt the character representation of modality-specific automatically, carries
Rise model performance.This process is as shown in Figure 6.
During second training, whole network is initialized with the good basic model parameter of pre-training, Ran Hou
A L2 normalization layer and triplet loss layers improved, convolution to be trained are added after the feature output layer of network
Neural network structure figure is as shown in Figure 7.
In one embodiment, a kind of flow diagram of face authentication method is as shown in figure 8, including three phases, difference
For data acquisition and pretreatment stage, training stage and deployment phase.
Data acquisition and pretreatment stage, read certificate chip by the card reader module of testimony of a witness verification terminal equipment and shine, with
And front camera crawl scene photograph, by human-face detector, Keypoint detector, face alignment with being obtained after shear module
To the normalized certificate facial image of size and scene facial image.
Training stage, 90% data are taken from testimony of a witness view data pond, and as training set, residue 10% is as verification collection.By
Comparison between the testimony of a witness compares mainly certificate photo and scene is shone, if because using certificate photo as reference chart in triple
(anchor), then other two figures are that scene is shone；If on the contrary shone for reference chart with scene, in addition two figures are certificates
According to.Construct the strategy of triple online using OHEM, i.e., during each iteration optimization of network, using current network to waiting
Triple is selected to carry out forward calculation, screening meets effective triple of condition, improved triplet is calculated according to above formula
Loss values, feed back in network and are iterated optimization.At the same time observation model verification concentrate performance, when verification performance not
When raising again, model reaches convergence state, and the training stage terminates.
Deployment phase, is deployed to testimony of a witness verification terminal by trained model and carries out in use, the image that equipment collects
By the preprocessor identical with the training stage, then obtained by network forward calculation the feature of every facial image to
Amount, obtains the similarity of two images by calculating COS distance, is then made decisions according to predetermined threshold value, more than predetermined threshold value
For same people, otherwise is different people.
Above-mentioned face authentication method, original triplet loss functions merely define the study relation of between class distance, on
The face authentication method stated, the bound term of inter- object distance is added by improving original triplet loss loss functions, can be with
So that network reduces gap in class as far as possible while increasing gap between class in the training process, so as to improve the extensive energy of network
Power, and then the scene adaptability of lift scheme.In addition, it instead of the Euclidean distance in original triplet loss with COS distance
Metric form, more meets the properties of distributions in face characteristic space, and it is consistent with deployment phase metric form to maintain the training stage
Property so that comparison result is relatively reliable.
In one embodiment, there is provided a kind of face authentication device, as shown in figure 9, including：Image collection module 902, figure
As pretreatment module 904, feature acquisition module 906, computing module 908 and authentication module 910.
Image collection module 902, for being asked based on face authentication, obtains certificate photograph and the scene photo of personage.
Image pre-processing module 904, for carrying out Face datection, crucial point location respectively to scene photo and certificate photograph
And image preprocessing, obtain the corresponding scene facial image of scene photo, and the corresponding certificate facial image of certificate photograph.
Feature acquisition module 906, for scene facial image and certificate facial image to be input to advance trained use
In the convolutional neural networks model of face authentication, and obtain the scene facial image corresponding the of convolutional neural networks model output
One feature vector, and the corresponding second feature vector of certificate facial image；Wherein, convolutional neural networks model is based on triple
The supervised training of loss function obtains.
Computing module 908, for calculating the COS distance of first eigenvector and second feature vector.
Authentication module 910, face authentication knot is determined for comparing COS distance and predetermined threshold value, and according to comparative result
Fruit.
Above-mentioned face authentication device, face authentication is carried out using convolutional neural networks trained in advance, due to convolution god
Through network model, the supervised training based on improved triple loss function obtains, and scene facial image and certificate face figure
The similarity of picture is according to the corresponding first eigenvector of scene facial image and the corresponding second feature vector of certificate facial image
COS distance be calculated, COS distance weigh be space vector angle, the difference being more embodied on direction, without
It is position, so as to more meet the properties of distributions in face characteristic space, improves the reliability of face authentication.
As shown in figure 9, in another embodiment, face authentication device further includes：Sample acquisition module 912, triple
Acquisition module 914, training module 916 and authentication module 918.
Sample acquisition module 912, for obtaining the training sample of tape label, the training sample includes marked belonging to every
One certificate facial image of a tagged object and at least a scene facial image.
Triple acquisition module 914, for according to training sample training convolutional neural networks model, being produced by OHEM each
The corresponding triple element of training sample；Triple element includes reference sample, positive sample and negative sample.
Specifically, triple acquisition module 914, belongs to same for randomly choosing an image as sample, selection is referred to
One label object, the image different from reference sample classification are additionally operable to, according to OHEM strategies, utilize current training as positive sample
Convolutional neural networks model extraction feature between COS distance, for each reference sample, from it is other have be not belonging to
The image that in the facial image of same label object, chosen distance is minimum, belongs to a different category with reference sample, as the reference
The negative sample of sample.
Specifically, using certificate photo as when referring to sample, positive sample and negative sample are that scene is shone；Shone using scene and be used as ginseng
When examining sample, positive sample and negative sample are certificate photo.
Training module 916, for the triple element according to each training sample, based on the supervision of triple loss function,
Training convolutional neural networks model, the triple loss function, using COS distance as metric form, passes through stochastic gradient descent
Algorithm carrys out Optimized model parameter.
Specifically, modified triple loss function includes the restriction to the COS distance of similar sample, and to foreign peoples
The restriction of the COS distance of sample.
Modified triple loss function is：
Wherein, cos () represents COS distance, its calculation isN is triple quantity,Represent the feature vector of reference sample,Represent the feature vector of similar positive sample,Represent foreign peoples's negative sample
Feature vector, []+Implication it is as follows：α1The spacing parameter between class, α2To be spaced ginseng in class
Number.
Authentication module 918, for verification collection data to be inputted convolutional neural networks model, when reaching trained termination condition,
Obtain the trained convolutional neural networks model for face authentication.
In another embodiment, face authentication device further includes model initialization module 920, and magnanimity is based on for utilizing
The trained basic model parameter of human face data of increasing income is initialized, addition normalization layer and triple after feature output layer
Loss function layer, obtains convolutional neural networks to be trained.Above-mentioned face authentication device, on the one hand in original triplet
The constraint to sample distance in class is added in the loss function of loss, so as to reduce class internal difference while gap between increasing class
Away from the generalization ability of lift scheme；On the other hand, the metric form of original triplet loss is changed to cosine by Euclidean distance
Distance, keeps the measurement uniformity of training and deployment, improves the continuity of feature learning.
A kind of computer equipment, including memory, processor and storage can be run on a memory and on a processor
The step of computer program, processor realizes the face authentication method of the various embodiments described above when performing computer program.
A kind of storage medium, is stored thereon with computer program, it is characterised in that the computer program is executed by processor
When, the step of realizing the face authentication method of the various embodiments described above.
Each technical characteristic of embodiment described above can be combined arbitrarily, to make description succinct, not to above-mentioned reality
Apply all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, the scope that this specification is recorded all is considered to be.
Embodiment described above only expresses the several embodiments of the present invention, its description is more specific and detailed, but simultaneously
Cannot therefore it be construed as limiting the scope of the patent.It should be pointed out that come for those of ordinary skill in the art
Say, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to the protection of the present invention
Scope.Therefore, the protection domain of patent of the present invention should be determined by the appended claims.
Claims (10)
1. a kind of face authentication method based on Triplet Loss, including：
Asked based on face authentication, obtain certificate photograph and the scene photo of personage；
Carry out Face datection, crucial point location and image preprocessing respectively to the scene photo and the certificate photograph, obtain
The corresponding scene facial image of the scene photo, and the corresponding certificate facial image of the certificate photograph；
The scene facial image and certificate facial image are input to the advance trained convolutional Neural for face authentication
Network model, and the corresponding first eigenvector of the scene facial image of the convolutional neural networks model output is obtained,
And the corresponding second feature vector of the certificate facial image；Wherein, the convolutional neural networks model is damaged based on triple
The supervised training for losing function obtains；
Calculate the COS distance of the first eigenvector and second feature vector；
Compare the COS distance and predetermined threshold value, and face authentication result is determined according to comparative result.
2. according to the method described in claim 1, it is characterized in that, the method further includes：
The training sample of tape label is obtained, the training sample includes marked a certificate face for belonging to each tagged object
Image and at least a scene facial image；
According to the training sample training convolutional neural networks model, the corresponding ternary constituent element of each training sample is produced by OHEM
Element；The triple element includes reference sample, positive sample and negative sample；
According to the triple element of each training sample, based on the supervision of triple loss function, the training convolutional neural networks
Model；The triple loss function, using COS distance as metric form, is joined by stochastic gradient descent algorithm come Optimized model
Number；
Verification collection data are inputted into the convolutional neural networks model, when reaching trained termination condition, obtain trained be used for
The convolutional neural networks model of face authentication.
3. according to the method described in claim 2, it is characterized in that, according to the training sample training convolutional neural networks mould
Type, the step of corresponding triple element of each training sample is produced by OHEM, including：
One image of random selection belongs to same label object, the figure different from reference sample classification as sample, selection is referred to
As being used as positive sample；
According to OHEM strategies, using the COS distance between currently trained convolutional neural networks model extraction feature, for every
One reference sample, is not belonging in the image of same label object from other, and chosen distance is minimum, belongs to the reference sample
Different classes of image, the negative sample as the reference sample.
4. according to the method described in claim 2, it is characterized in that, the triple loss function is included to more than similar sample
The restriction of chordal distance, and the restriction of the COS distance to foreign peoples's sample.
5. according to the method described in claim 4, it is characterized in that, the triple loss function is：
<mrow>
<mover>
<mi>L</mi>
<mo>~</mo>
</mover>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mi>i</mi>
<mi>N</mi>
</munderover>
<msub>
<mrow>
<mo>&lsqb;</mo>
<mi>c</mi>
<mi>o</mi>
<mi>s</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>a</mi>
</msubsup>
<mo>)</mo>
<mo>,</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>n</mi>
</msubsup>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>c</mi>
<mi>o</mi>
<mi>s</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>a</mi>
</msubsup>
<mo>)</mo>
<mo>,</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>p</mi>
</msubsup>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msub>
<mi>&alpha;</mi>
<mn>1</mn>
</msub>
<mo>&rsqb;</mo>
</mrow>
<mo>+</mo>
</msub>
<mo>+</mo>
<munderover>
<mo>&Sigma;</mo>
<mi>i</mi>
<mi>N</mi>
</munderover>
<msub>
<mrow>
<mo>&lsqb;</mo>
<msub>
<mi>&alpha;</mi>
<mn>2</mn>
</msub>
<mo>-</mo>
<mi>c</mi>
<mi>o</mi>
<mi>s</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>a</mi>
</msubsup>
<mo>)</mo>
<mo>,</mo>
<mi>f</mi>
<mo>(</mo>
<msubsup>
<mi>x</mi>
<mi>i</mi>
<mi>p</mi>
</msubsup>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
</mrow>
<mo>+</mo>
</msub>
</mrow>
Wherein, cos () represents COS distance, its calculation isN is triple quantity,Table
Show the feature vector of reference sample,Represent the feature vector of similar positive sample,Represent the feature of foreign peoples's negative sample
Vector, []+Implication it is as follows：α1The spacing parameter between class, α2For spacing parameter in class.
6. according to the method described in claim 2, it is characterized in that, the method further includes：Increase income face using based on magnanimity
The trained basic model parameter of data is initialized, addition normalization layer and triple loss function after feature output layer
Layer, obtains convolutional neural networks model to be trained.
7. a kind of face authentication device based on Triplet Loss, including：Image collection module, image pre-processing module, spy
Levy acquisition module, computing module and authentication module；
Described image acquisition module, for being asked based on face authentication, obtains certificate photograph and the scene photo of personage；
Described image pretreatment module, for carrying out Face datection, key respectively to the scene photo and the certificate photograph
Point location and image preprocessing, obtain the corresponding scene facial image of the scene photo, and the certificate photograph is corresponding
Certificate facial image；
The feature acquisition module, for the scene facial image and certificate facial image to be input to advance trained use
In the convolutional neural networks model of face authentication, and obtain the scene facial image of the convolutional neural networks model output
Corresponding first eigenvector, and the corresponding second feature vector of the certificate facial image；Wherein, the convolutional Neural net
Supervised training of the network model based on triple loss function obtains；
The computing module, for calculating the COS distance of the first eigenvector and second feature vector；
The authentication module, face authentication knot is determined for the COS distance and predetermined threshold value, and according to comparative result
Fruit.
8. device according to claim 7, it is characterised in that described device further includes：Sample acquisition module, triple obtain
Modulus block, training module and authentication module；
The sample acquisition module, for obtaining the training sample of tape label, the training sample includes marked belonging to each
One certificate facial image of tagged object and at least a scene facial image；
The triple acquisition module, for according to the training sample training convolutional neural networks model, being produced by OHEM
The corresponding triple element of each training sample；The triple element includes reference sample, positive sample and negative sample；
The training module, for the triple element according to each training sample, based on the supervision of triple loss function, training
The convolutional neural networks model；The triple loss function, using COS distance as metric form, passes through stochastic gradient descent
Algorithm carrys out Optimized model parameter；
The authentication module, for verification collection data to be inputted the convolutional neural networks model, when reaching trained termination condition,
Obtain the trained convolutional neural networks model for face authentication.
9. a kind of computer equipment, including memory, processor and storage are on a memory and the meter that can run on a processor
Calculation machine program, it is characterised in that the processor is realized described in any one of claim 1 to 6 when performing the computer program
The face authentication method based on Triplet Loss the step of.
10. a kind of storage medium, is stored thereon with computer program, it is characterised in that the computer program is executed by processor
When, the step of realizing face authentication method of claim 1 to 6 any one of them based on Triplet Loss.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711436879.4A CN108009528B (en) | 2017-12-26 | 2017-12-26 | Triple Loss-based face authentication method and device, computer equipment and storage medium |
| PCT/CN2018/109169 WO2019128367A1 (en) | 2017-12-26 | 2018-09-30 | Face verification method and apparatus based on triplet loss, and computer device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711436879.4A CN108009528B (en) | 2017-12-26 | 2017-12-26 | Triple Loss-based face authentication method and device, computer equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108009528A true CN108009528A (en) | 2018-05-08 |
| CN108009528B CN108009528B (en) | 2020-04-07 |
Family
ID=62061566
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711436879.4A Active CN108009528B (en) | 2017-12-26 | 2017-12-26 | Triple Loss-based face authentication method and device, computer equipment and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN108009528B (en) |
| WO (1) | WO2019128367A1 (en) |
Cited By (67)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921033A (en) * | 2018-06-04 | 2018-11-30 | 北京京东金融科技控股有限公司 | Face picture comparison method, device, medium and electronic equipment |
| CN108921952A (en) * | 2018-06-15 | 2018-11-30 | 深圳大学 | Object functionality prediction technique, device, computer equipment and storage medium |
| CN108922542A (en) * | 2018-06-01 | 2018-11-30 | 平安科技（深圳）有限公司 | Acquisition methods, device, computer equipment and the storage medium of sample triple |
| CN108960342A (en) * | 2018-08-01 | 2018-12-07 | 中国计量大学 | Based on the image similarity calculation method for improving SoftMax loss function |
| CN108960209A (en) * | 2018-08-09 | 2018-12-07 | 腾讯科技（深圳）有限公司 | Personal identification method, device and computer readable storage medium |
| CN108985198A (en) * | 2018-07-02 | 2018-12-11 | 四川斐讯信息技术有限公司 | A kind of COS distance calculation method based on big data feature vector |
| CN109145991A (en) * | 2018-08-24 | 2019-01-04 | 北京地平线机器人技术研发有限公司 | Image group generation method, image group generating means and electronic equipment |
| CN109145956A (en) * | 2018-07-26 | 2019-01-04 | 上海慧子视听科技有限公司 | Methods of marking, device, computer equipment and storage medium |
| CN109145704A (en) * | 2018-06-14 | 2019-01-04 | 西安电子科技大学 | A kind of human face portrait recognition methods based on face character |
| CN109165589A (en) * | 2018-08-14 | 2019-01-08 | 北京颂泽科技有限公司 | Vehicle based on deep learning recognition methods and device again |
| CN109214361A (en) * | 2018-10-18 | 2019-01-15 | 康明飞（北京）科技有限公司 | A kind of face identification method and device and ticket verification method and device |
| CN109271877A (en) * | 2018-08-24 | 2019-01-25 | 北京智芯原动科技有限公司 | A kind of human figure identification method and device |
| CN109344740A (en) * | 2018-09-12 | 2019-02-15 | 上海了物网络科技有限公司 | Face identification system, method and computer readable storage medium |
| 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 |
| CN109492583A (en) * | 2018-11-09 | 2019-03-19 | 安徽大学 | A kind of recognition methods again of the vehicle based on deep learning |
| CN109522850A (en) * | 2018-11-22 | 2019-03-26 | 中山大学 | A kind of movement similarity estimating method based on small-sample learning |
| CN109543524A (en) * | 2018-10-18 | 2019-03-29 | 同盾控股有限公司 | A kind of image-recognizing method, device |
| CN109685121A (en) * | 2018-12-11 | 2019-04-26 | 中国科学院苏州纳米技术与纳米仿生研究所 | Training method, image search method, the computer equipment of image encrypting algorithm |
| CN109685106A (en) * | 2018-11-19 | 2019-04-26 | 深圳博为教育科技有限公司 | A kind of image-recognizing method, face Work attendance method, device and system |
| CN109711358A (en) * | 2018-12-28 | 2019-05-03 | 四川远鉴科技有限公司 | Neural network training method, face identification method and system and storage medium |
| CN109711443A (en) * | 2018-12-14 | 2019-05-03 | 平安城市建设科技(深圳)有限公司 | Floor plan recognition methods, device, equipment and storage medium neural network based |
| CN109815801A (en) * | 2018-12-18 | 2019-05-28 | 北京英索科技发展有限公司 | Face identification method and device based on deep learning |
| CN109886186A (en) * | 2019-02-18 | 2019-06-14 | 上海骏聿数码科技有限公司 | A kind of face identification method and device |
| CN109948568A (en) * | 2019-03-26 | 2019-06-28 | 东华大学 | Embedded human face identifying system based on ARM microprocessor and deep learning |
| WO2019128367A1 (en) * | 2017-12-26 | 2019-07-04 | 广州广电运通金融电子股份有限公司 | Face verification method and apparatus based on triplet loss, and computer device and storage medium |
| CN110147732A (en) * | 2019-04-16 | 2019-08-20 | 平安科技（深圳）有限公司 | Refer to vein identification method, device, computer equipment and storage medium |
| CN110147833A (en) * | 2019-05-09 | 2019-08-20 | 北京迈格威科技有限公司 | Facial image processing method, apparatus, system and readable storage medium storing program for executing |
| CN110213660A (en) * | 2019-05-27 | 2019-09-06 | 广州荔支网络技术有限公司 | Distribution method, system, computer equipment and the storage medium of program |
| CN110598840A (en) * | 2018-06-13 | 2019-12-20 | 富士通株式会社 | Knowledge migration method, information processing apparatus, and storage medium |
| CN110674688A (en) * | 2019-08-19 | 2020-01-10 | 深圳力维智联技术有限公司 | Face recognition model acquisition method, system and medium for video monitoring scene |
| CN110738071A (en) * | 2018-07-18 | 2020-01-31 | 浙江中正智能科技有限公司 | face algorithm model training method based on deep learning and transfer learning |
| CN110765933A (en) * | 2019-10-22 | 2020-02-07 | 山西省信息产业技术研究院有限公司 | Dynamic portrait sensing comparison method applied to driver identity authentication system |
| CN110874602A (en) * | 2018-08-30 | 2020-03-10 | 北京嘀嘀无限科技发展有限公司 | Image identification method and device |
| CN110929099A (en) * | 2019-11-28 | 2020-03-27 | 杭州趣维科技有限公司 | Short video frame semantic extraction method and system based on multitask learning |
| CN111062338A (en) * | 2019-12-19 | 2020-04-24 | 厦门商集网络科技有限责任公司 | Certificate portrait consistency comparison method and system |
| CN111178249A (en) * | 2019-12-27 | 2020-05-19 | 杭州艾芯智能科技有限公司 | Face comparison method and device, computer equipment and storage medium |
| CN111325223A (en) * | 2018-12-13 | 2020-06-23 | 中国电信股份有限公司 | Deep learning model training method and device and computer readable storage medium |
| WO2020125404A1 (en) * | 2018-12-19 | 2020-06-25 | 北京世纪好未来教育科技有限公司 | Method and apparatus for constructing neural network and computer-readable medium |
| CN111368644A (en) * | 2020-02-14 | 2020-07-03 | 深圳市商汤科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN111401277A (en) * | 2020-03-20 | 2020-07-10 | 深圳前海微众银行股份有限公司 | Face recognition model updating method, device, equipment and medium |
| CN111401257A (en) * | 2020-03-17 | 2020-07-10 | 天津理工大学 | Non-constraint condition face recognition method based on cosine loss |
| WO2020147408A1 (en) * | 2019-01-16 | 2020-07-23 | 平安科技（深圳）有限公司 | Facial recognition model evaluation method and apparatus, and storage medium and computer device |
| CN111461152A (en) * | 2019-01-21 | 2020-07-28 | 同方威视技术股份有限公司 | Cargo detection method and device, electronic equipment and computer readable medium |
| CN111709313A (en) * | 2020-05-27 | 2020-09-25 | 杭州电子科技大学 | Pedestrian re-identification method based on local and channel combination characteristics |
| CN111723222A (en) * | 2019-03-19 | 2020-09-29 | Sap欧洲公司 | Image search and training system |
| CN111832364A (en) * | 2019-04-22 | 2020-10-27 | 普天信息技术有限公司 | Face recognition method and device |
| CN112052821A (en) * | 2020-09-15 | 2020-12-08 | 浙江智慧视频安防创新中心有限公司 | Fire fighting channel safety detection method, device, equipment and storage medium |
| CN112116021A (en) * | 2020-09-27 | 2020-12-22 | 广州华多网络科技有限公司 | Gem similarity measurement data processing method and related equipment |
| CN112492383A (en) * | 2020-12-03 | 2021-03-12 | 珠海格力电器股份有限公司 | Video frame generation method and device, storage medium and electronic equipment |
| CN112580406A (en) * | 2019-09-30 | 2021-03-30 | 北京中关村科金技术有限公司 | Face comparison method and device and storage medium |
| WO2021073150A1 (en) * | 2019-10-16 | 2021-04-22 | 平安科技（深圳）有限公司 | Data detection method and apparatus, and computer device and storage medium |
| CN112949780A (en) * | 2020-04-21 | 2021-06-11 | 佳都科技集团股份有限公司 | Feature model training method, device, equipment and storage medium |
| CN113065495A (en) * | 2021-04-13 | 2021-07-02 | 深圳技术大学 | Image similarity calculation method, target object re-identification method and system |
| CN113269155A (en) * | 2021-06-28 | 2021-08-17 | 苏州市科远软件技术开发有限公司 | End-to-end face recognition method, device, equipment and storage medium |
| CN113283359A (en) * | 2021-06-02 | 2021-08-20 | 万达信息股份有限公司 | Authentication method and system for handheld certificate photo and electronic equipment |
| CN113344875A (en) * | 2021-06-07 | 2021-09-03 | 武汉象点科技有限公司 | Abnormal image detection method based on self-supervision learning |
| CN113486804A (en) * | 2021-07-07 | 2021-10-08 | 科大讯飞股份有限公司 | Object identification method, device, equipment and storage medium |
| CN113538075A (en) * | 2020-04-14 | 2021-10-22 | 阿里巴巴集团控股有限公司 | Data processing method, model training method, device and equipment |
| CN113642468A (en) * | 2021-08-16 | 2021-11-12 | 中国银行股份有限公司 | Identity authentication method and device |
| CN113688793A (en) * | 2021-09-22 | 2021-11-23 | 万章敏 | Training method of face model and face recognition system |
| CN114283483A (en) * | 2021-12-31 | 2022-04-05 | 中国联合网络通信集团有限公司 | Face recognition method and device |
| CN114387457A (en) * | 2021-12-27 | 2022-04-22 | 腾晖科技建筑智能(深圳)有限公司 | Face intra-class interval optimization method based on parameter adjustment |
| CN116127298A (en) * | 2023-02-22 | 2023-05-16 | 北京邮电大学 | Small sample radio frequency fingerprint identification method based on triplet loss |
| WO2023093838A1 (en) * | 2021-11-25 | 2023-06-01 | 北京字跳网络技术有限公司 | Super-resolution image processing method and apparatus, and device and medium |
| US20230169792A1 (en) * | 2021-11-26 | 2023-06-01 | GenkiTek Technology Co., Ltd. | System and method of mode selection face recognition with parallel cnns |
| EP4383212A1 (en) * | 2022-12-06 | 2024-06-12 | Bundesdruckerei GmbH | Authentication device and method for authenticating a person by means of an identification document associated with the person, and identification document and method for producing |
| CN118337532A (en) * | 2024-06-13 | 2024-07-12 | 浙江鹏信信息科技股份有限公司 | Zero trust-based traffic safety audit protection method and system and readable medium |
Families Citing this family (77)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110516533B (en) * | 2019-07-11 | 2023-06-02 | 同济大学 | Pedestrian re-identification method based on depth measurement |
| CN110414431B (en) * | 2019-07-29 | 2022-12-27 | 广州像素数据技术股份有限公司 | Face recognition method and system based on elastic context relation loss function |
| CN110647880A (en) * | 2019-08-12 | 2020-01-03 | 深圳市华付信息技术有限公司 | Mobile terminal identity card image shielding judgment method |
| CN110458233B (en) * | 2019-08-13 | 2024-02-13 | 腾讯云计算（北京）有限责任公司 | Mixed granularity object recognition model training and recognition method, device and storage medium |
| CN110705357A (en) * | 2019-09-02 | 2020-01-17 | 深圳中兴网信科技有限公司 | Face recognition method and face recognition device |
| CN110555478B (en) * | 2019-09-05 | 2023-02-03 | 东北大学 | Fan multi-fault diagnosis method based on depth measurement network of difficult sample mining |
| CN111008550A (en) * | 2019-09-06 | 2020-04-14 | 上海芯灵科技有限公司 | Identification method for finger vein authentication identity based on Multiple loss function |
| CN110674637B (en) * | 2019-09-06 | 2023-07-11 | 腾讯科技（深圳）有限公司 | Character relationship recognition model training method, device, equipment and medium |
| CN110705393B (en) * | 2019-09-17 | 2023-02-03 | 中国计量大学 | Method for improving face recognition performance of community population |
| CN110647938B (en) * | 2019-09-24 | 2022-07-15 | 北京市商汤科技开发有限公司 | Image processing method and related device |
| CN112560880A (en) * | 2019-09-25 | 2021-03-26 | 中国电信股份有限公司 | Object classification method, object classification apparatus, and computer-readable storage medium |
| CN112733574B (en) * | 2019-10-14 | 2023-04-07 | 中移(苏州)软件技术有限公司 | Face recognition method and device and computer readable storage medium |
| CN110796057A (en) * | 2019-10-22 | 2020-02-14 | 上海交通大学 | Pedestrian re-identification method and device and computer equipment |
| CN110852367B (en) * | 2019-11-05 | 2023-10-31 | 上海联影智能医疗科技有限公司 | Image classification method, computer device, and storage medium |
| CN110956098B (en) * | 2019-11-13 | 2023-05-12 | 深圳数联天下智能科技有限公司 | Image processing method and related equipment |
| CN111126360B (en) * | 2019-11-15 | 2023-03-24 | 西安电子科技大学 | Cross-domain pedestrian re-identification method based on unsupervised combined multi-loss model |
| CN111222411B (en) * | 2019-11-28 | 2023-09-01 | 中国船舶重工集团公司第七一三研究所 | Laser emission safety rapid alarm method and device |
| CN111079566B (en) * | 2019-11-28 | 2023-05-02 | 深圳市信义科技有限公司 | Large-scale face recognition model optimization system |
| CN111144240B (en) * | 2019-12-12 | 2023-02-07 | 深圳数联天下智能科技有限公司 | Image processing method and related equipment |
| CN111062430B (en) * | 2019-12-12 | 2023-05-09 | 易诚高科（大连）科技有限公司 | Pedestrian re-identification evaluation method based on probability density function |
| CN111091089B (en) * | 2019-12-12 | 2022-07-29 | 新华三大数据技术有限公司 | Face image processing method and device, electronic equipment and storage medium |
| CN111126240B (en) * | 2019-12-19 | 2023-04-07 | 西安工程大学 | Three-channel feature fusion face recognition method |
| CN111191563A (en) * | 2019-12-26 | 2020-05-22 | 三盟科技股份有限公司 | Face recognition method and system based on data sample and test data set training |
| CN111241925B (en) * | 2019-12-30 | 2023-08-18 | 新大陆数字技术股份有限公司 | Face quality assessment method, system, electronic equipment and readable storage medium |
| CN111209839B (en) * | 2019-12-31 | 2023-05-23 | 上海涛润医疗科技有限公司 | Face recognition method |
| CN111198964B (en) * | 2020-01-10 | 2023-04-25 | 中国科学院自动化研究所 | Image retrieval method and system |
| CN111274946B (en) * | 2020-01-19 | 2023-05-05 | 杭州涂鸦信息技术有限公司 | Face recognition method, system and equipment |
| CN113362096A (en) * | 2020-03-04 | 2021-09-07 | 驰众信息技术（上海）有限公司 | Frame advertisement image matching method based on deep learning |
| CN111368766B (en) * | 2020-03-09 | 2023-08-18 | 云南安华防灾减灾科技有限责任公司 | Deep learning-based cow face detection and recognition method |
| CN111539247B (en) * | 2020-03-10 | 2023-02-10 | 西安电子科技大学 | Hyper-spectrum face recognition method and device, electronic equipment and storage medium thereof |
| CN111429414B (en) * | 2020-03-18 | 2023-04-07 | 腾讯科技（深圳）有限公司 | Artificial intelligence-based focus image sample determination method and related device |
| CN111414862B (en) * | 2020-03-22 | 2023-03-24 | 西安电子科技大学 | Expression recognition method based on neural network fusion key point angle change |
| CN111507289A (en) * | 2020-04-22 | 2020-08-07 | 上海眼控科技股份有限公司 | Video matching method, computer device and storage medium |
| CN111582107B (en) * | 2020-04-28 | 2023-09-29 | 浙江大华技术股份有限公司 | Training method and recognition method of target re-recognition model, electronic equipment and device |
| CN111639535B (en) * | 2020-04-29 | 2023-08-22 | 深圳英飞拓智能技术有限公司 | Face recognition method and device based on deep learning |
| CN111626212B (en) * | 2020-05-27 | 2023-09-26 | 腾讯科技（深圳）有限公司 | Method and device for identifying object in picture, storage medium and electronic device |
| CN113807122A (en) * | 2020-06-11 | 2021-12-17 | 阿里巴巴集团控股有限公司 | Model training method, object recognition method and device, and storage medium |
| CN111738157B (en) * | 2020-06-23 | 2023-07-21 | 平安科技（深圳）有限公司 | Face action unit data set construction method and device and computer equipment |
| CN112257738A (en) * | 2020-07-31 | 2021-01-22 | 北京京东尚科信息技术有限公司 | Training method and device of machine learning model and classification method and device of image |
| CN111988614B (en) * | 2020-08-14 | 2022-09-13 | 深圳前海微众银行股份有限公司 | Hash coding optimization method and device and readable storage medium |
| CN112069993B (en) * | 2020-09-04 | 2024-02-13 | 西安西图之光智能科技有限公司 | Dense face detection method and system based on five-sense organ mask constraint and storage medium |
| CN112084956A (en) * | 2020-09-11 | 2020-12-15 | 上海交通大学烟台信息技术研究院 | Special face crowd screening system based on small sample learning prototype network |
| CN112287765B (en) * | 2020-09-30 | 2024-06-04 | 新大陆数字技术股份有限公司 | Face living body detection method, device, equipment and readable storage medium |
| CN112329826B (en) * | 2020-10-24 | 2024-10-18 | 中国人民解放军空军军医大学 | Training method of image recognition model, image recognition method and device |
| CN112307968B (en) * | 2020-10-30 | 2022-11-08 | 天地伟业技术有限公司 | Face recognition feature compression method |
| CN112328786A (en) * | 2020-11-03 | 2021-02-05 | 平安科技（深圳）有限公司 | Text classification method and device based on BERT, computer equipment and storage medium |
| GB2600922B (en) * | 2020-11-05 | 2024-04-10 | Thales Holdings Uk Plc | One shot learning for identifying data items similar to a query data item |
| CN112347957A (en) * | 2020-11-12 | 2021-02-09 | 广联达科技股份有限公司 | Pedestrian re-identification method and device, computer equipment and storage medium |
| CN112200154A (en) * | 2020-11-17 | 2021-01-08 | 苏州方正璞华信息技术有限公司 | Face recognition method and device for mask, electronic equipment and storage medium |
| CN112836566A (en) * | 2020-12-01 | 2021-05-25 | 北京智云视图科技有限公司 | Multitask neural network face key point detection method for edge equipment |
| CN112836719B (en) * | 2020-12-11 | 2024-01-05 | 南京富岛信息工程有限公司 | Indicator diagram similarity detection method integrating two classifications and triplets |
| CN112580011B (en) * | 2020-12-25 | 2022-05-24 | 华南理工大学 | Portrait encryption and decryption system facing biological feature privacy protection |
| CN112861626B (en) * | 2021-01-04 | 2024-03-08 | 西北工业大学 | Fine granularity expression classification method based on small sample learning |
| CN113762019B (en) * | 2021-01-22 | 2024-04-09 | 北京沃东天骏信息技术有限公司 | Training method of feature extraction network, face recognition method and device |
| CN112836629B (en) * | 2021-02-01 | 2024-03-08 | 清华大学深圳国际研究生院 | Image classification method |
| CN112966724B (en) * | 2021-02-07 | 2024-04-09 | 惠州市博实结科技有限公司 | Method and device for classifying image single categories |
| CN112766237A (en) * | 2021-03-12 | 2021-05-07 | 东北林业大学 | Unsupervised pedestrian re-identification method based on cluster feature point clustering |
| CN113157956B (en) * | 2021-04-23 | 2022-08-05 | 雅马哈发动机（厦门）信息系统有限公司 | Picture searching method, system, mobile terminal and storage medium |
| CN113344031B (en) * | 2021-05-13 | 2022-12-27 | 清华大学 | Text classification method |
| CN113469253B (en) * | 2021-07-02 | 2024-05-14 | 河海大学 | Electric larceny detection method based on triple twinning network |
| CN113705383A (en) * | 2021-08-12 | 2021-11-26 | 南京英诺森软件科技有限公司 | Cross-age face recognition method and system based on ternary constraint |
| CN113435545A (en) * | 2021-08-14 | 2021-09-24 | 北京达佳互联信息技术有限公司 | Training method and device of image processing model |
| CN113642481A (en) * | 2021-08-17 | 2021-11-12 | 百度在线网络技术（北京）有限公司 | Recognition method, training method, device, electronic equipment and storage medium |
| CN113569991B (en) * | 2021-08-26 | 2024-05-28 | 深圳市捷顺科技实业股份有限公司 | Person evidence comparison model training method, computer equipment and computer storage medium |
| CN113780461B (en) * | 2021-09-23 | 2022-08-05 | 中国人民解放军国防科技大学 | Robust neural network training method based on feature matching |
| CN113887653B (en) * | 2021-10-20 | 2024-02-06 | 西安交通大学 | Positioning method and system for tight coupling weak supervision learning based on ternary network |
| CN114049479A (en) * | 2021-11-10 | 2022-02-15 | 苏州魔视智能科技有限公司 | Self-supervision fisheye camera image feature point extraction method and device and storage medium |
| CN114445871A (en) * | 2021-12-22 | 2022-05-06 | 天翼云科技有限公司 | Intelligent video monitoring method and system for face recognition |
| CN114863425B (en) * | 2022-03-31 | 2024-10-29 | 哈尔滨工程大学 | Urine erythrocyte classification method based on supervised contrast learning |
| CN114882558B (en) * | 2022-04-29 | 2024-02-23 | 陕西师范大学 | Learning scene real-time identity authentication method based on face recognition technology |
| CN114663965B (en) * | 2022-05-24 | 2022-10-21 | 之江实验室 | Testimony comparison method and device based on two-stage alternative learning |
| CN114926445B (en) * | 2022-05-31 | 2024-03-26 | 哈尔滨工业大学 | Small sample crop disease image identification method and system based on twin network |
| CN114817888A (en) * | 2022-06-27 | 2022-07-29 | 中国信息通信研究院 | Certificate registering and issuing method, device and storage medium |
| CN116206355A (en) * | 2023-04-25 | 2023-06-02 | 鹏城实验室 | Face recognition model training, image registration and face recognition method and device |
| CN116959064B (en) * | 2023-06-25 | 2024-04-26 | 上海腾桥信息技术有限公司 | Certificate verification method and device, computer equipment and storage medium |
| CN116977461B (en) * | 2023-06-30 | 2024-03-08 | 北京开普云信息科技有限公司 | Portrait generation method, device, storage medium and equipment for specific scene |
| CN118410192B (en) * | 2024-06-24 | 2024-10-11 | 广州思德医疗科技有限公司 | Endoscopic image deduplication method, endoscopic image deduplication apparatus, endoscopic image deduplication computer device, endoscopic image deduplication program product, and endoscopic image deduplication program product |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106599827A (en) * | 2016-12-09 | 2017-04-26 | 浙江工商大学 | Small target rapid detection method based on deep convolution neural network |
| CN107194341A (en) * | 2017-05-16 | 2017-09-22 | 西安电子科技大学 | The many convolution neural network fusion face identification methods of Maxout and system |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9129216B1 (en) * | 2013-07-15 | 2015-09-08 | Xdroid Kft. | System, method and apparatus for computer aided association of relevant images with text |
| CN107423690B (en) * | 2017-06-26 | 2020-11-13 | 广东工业大学 | Face recognition method and device |
| CN108009528B (en) * | 2017-12-26 | 2020-04-07 | 广州广电运通金融电子股份有限公司 | Triple Loss-based face authentication method and device, computer equipment and storage medium |
-
2017
- 2017-12-26 CN CN201711436879.4A patent/CN108009528B/en active Active
-
2018
- 2018-09-30 WO PCT/CN2018/109169 patent/WO2019128367A1/en active Application Filing
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106599827A (en) * | 2016-12-09 | 2017-04-26 | 浙江工商大学 | Small target rapid detection method based on deep convolution neural network |
| CN107194341A (en) * | 2017-05-16 | 2017-09-22 | 西安电子科技大学 | The many convolution neural network fusion face identification methods of Maxout and system |
Non-Patent Citations (1)
| Title |
|---|
| BRANDON，AMOS等: "OpenFace:A general-purpose face recognition library with mobile applications", 《CMU SCHOOL OF COMPUTER SCIENCE，TECH. REP.》 * |
Cited By (91)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019128367A1 (en) * | 2017-12-26 | 2019-07-04 | 广州广电运通金融电子股份有限公司 | Face verification method and apparatus based on triplet loss, and computer device and storage medium |
| CN108922542A (en) * | 2018-06-01 | 2018-11-30 | 平安科技（深圳）有限公司 | Acquisition methods, device, computer equipment and the storage medium of sample triple |
| WO2019227613A1 (en) * | 2018-06-01 | 2019-12-05 | 平安科技（深圳）有限公司 | Method and device for obtaining triple of samples, computer device and storage medium |
| CN108921033A (en) * | 2018-06-04 | 2018-11-30 | 北京京东金融科技控股有限公司 | Face picture comparison method, device, medium and electronic equipment |
| CN110598840B (en) * | 2018-06-13 | 2023-04-18 | 富士通株式会社 | Knowledge migration method, information processing apparatus, and storage medium |
| CN110598840A (en) * | 2018-06-13 | 2019-12-20 | 富士通株式会社 | Knowledge migration method, information processing apparatus, and storage medium |
| CN109145704A (en) * | 2018-06-14 | 2019-01-04 | 西安电子科技大学 | A kind of human face portrait recognition methods based on face character |
| CN108921952A (en) * | 2018-06-15 | 2018-11-30 | 深圳大学 | Object functionality prediction technique, device, computer equipment and storage medium |
| WO2019237450A1 (en) * | 2018-06-15 | 2019-12-19 | 深圳大学 | Object functionality prediction method, computer device, and storage medium |
| US11288538B2 (en) | 2018-06-15 | 2022-03-29 | Shenzhen University | Object functionality predication methods, computer device, and storage medium |
| CN108985198A (en) * | 2018-07-02 | 2018-12-11 | 四川斐讯信息技术有限公司 | A kind of COS distance calculation method based on big data feature vector |
| CN110738071A (en) * | 2018-07-18 | 2020-01-31 | 浙江中正智能科技有限公司 | face algorithm model training method based on deep learning and transfer learning |
| CN109145956A (en) * | 2018-07-26 | 2019-01-04 | 上海慧子视听科技有限公司 | Methods of marking, device, computer equipment and storage medium |
| CN109145956B (en) * | 2018-07-26 | 2021-12-14 | 上海慧子视听科技有限公司 | Scoring method, scoring device, computer equipment and storage medium |
| CN108960342A (en) * | 2018-08-01 | 2018-12-07 | 中国计量大学 | Based on the image similarity calculation method for improving SoftMax loss function |
| CN108960209A (en) * | 2018-08-09 | 2018-12-07 | 腾讯科技（深圳）有限公司 | Personal identification method, device and computer readable storage medium |
| CN108960209B (en) * | 2018-08-09 | 2023-07-21 | 腾讯科技（深圳）有限公司 | Identity recognition method, identity recognition device and computer readable storage medium |
| CN109165589B (en) * | 2018-08-14 | 2021-02-23 | 北京颂泽科技有限公司 | Vehicle weight recognition method and device based on deep learning |
| CN109165589A (en) * | 2018-08-14 | 2019-01-08 | 北京颂泽科技有限公司 | Vehicle based on deep learning recognition methods and device again |
| CN109145991A (en) * | 2018-08-24 | 2019-01-04 | 北京地平线机器人技术研发有限公司 | Image group generation method, image group generating means and electronic equipment |
| CN109145991B (en) * | 2018-08-24 | 2020-07-31 | 北京地平线机器人技术研发有限公司 | Image group generation method, image group generation device and electronic equipment |
| CN109271877A (en) * | 2018-08-24 | 2019-01-25 | 北京智芯原动科技有限公司 | A kind of human figure identification method and device |
| CN110874602A (en) * | 2018-08-30 | 2020-03-10 | 北京嘀嘀无限科技发展有限公司 | Image identification method and device |
| CN109344740A (en) * | 2018-09-12 | 2019-02-15 | 上海了物网络科技有限公司 | Face identification system, method and computer readable storage medium |
| 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 |
| CN109543524A (en) * | 2018-10-18 | 2019-03-29 | 同盾控股有限公司 | A kind of image-recognizing method, device |
| CN109214361A (en) * | 2018-10-18 | 2019-01-15 | 康明飞（北京）科技有限公司 | A kind of face identification method and device and ticket verification method and device |
| CN109492583A (en) * | 2018-11-09 | 2019-03-19 | 安徽大学 | A kind of recognition methods again of the vehicle based on deep learning |
| CN109685106A (en) * | 2018-11-19 | 2019-04-26 | 深圳博为教育科技有限公司 | A kind of image-recognizing method, face Work attendance method, device and system |
| CN109522850B (en) * | 2018-11-22 | 2023-03-10 | 中山大学 | Action similarity evaluation method based on small sample learning |
| CN109522850A (en) * | 2018-11-22 | 2019-03-26 | 中山大学 | A kind of movement similarity estimating method based on small-sample learning |
| CN109685121B (en) * | 2018-12-11 | 2023-07-18 | 中国科学院苏州纳米技术与纳米仿生研究所 | Training method of image retrieval model, image retrieval method and computer equipment |
| CN109685121A (en) * | 2018-12-11 | 2019-04-26 | 中国科学院苏州纳米技术与纳米仿生研究所 | Training method, image search method, the computer equipment of image encrypting algorithm |
| CN111325223B (en) * | 2018-12-13 | 2023-10-24 | 中国电信股份有限公司 | Training method and device for deep learning model and computer readable storage medium |
| CN111325223A (en) * | 2018-12-13 | 2020-06-23 | 中国电信股份有限公司 | Deep learning model training method and device and computer readable storage medium |
| CN109711443A (en) * | 2018-12-14 | 2019-05-03 | 平安城市建设科技(深圳)有限公司 | Floor plan recognition methods, device, equipment and storage medium neural network based |
| CN109815801A (en) * | 2018-12-18 | 2019-05-28 | 北京英索科技发展有限公司 | Face identification method and device based on deep learning |
| WO2020125404A1 (en) * | 2018-12-19 | 2020-06-25 | 北京世纪好未来教育科技有限公司 | Method and apparatus for constructing neural network and computer-readable medium |
| CN109711358A (en) * | 2018-12-28 | 2019-05-03 | 四川远鉴科技有限公司 | Neural network training method, face identification method and system and storage medium |
| WO2020147408A1 (en) * | 2019-01-16 | 2020-07-23 | 平安科技（深圳）有限公司 | Facial recognition model evaluation method and apparatus, and storage medium and computer device |
| CN111461152A (en) * | 2019-01-21 | 2020-07-28 | 同方威视技术股份有限公司 | Cargo detection method and device, electronic equipment and computer readable medium |
| CN111461152B (en) * | 2019-01-21 | 2024-04-05 | 同方威视技术股份有限公司 | Cargo detection method and device, electronic equipment and computer readable medium |
| CN109886186A (en) * | 2019-02-18 | 2019-06-14 | 上海骏聿数码科技有限公司 | A kind of face identification method and device |
| CN111723222A (en) * | 2019-03-19 | 2020-09-29 | Sap欧洲公司 | Image search and training system |
| CN109948568A (en) * | 2019-03-26 | 2019-06-28 | 东华大学 | Embedded human face identifying system based on ARM microprocessor and deep learning |
| CN110147732A (en) * | 2019-04-16 | 2019-08-20 | 平安科技（深圳）有限公司 | Refer to vein identification method, device, computer equipment and storage medium |
| CN111832364A (en) * | 2019-04-22 | 2020-10-27 | 普天信息技术有限公司 | Face recognition method and device |
| CN111832364B (en) * | 2019-04-22 | 2024-04-23 | 普天信息技术有限公司 | Face recognition method and device |
| CN110147833A (en) * | 2019-05-09 | 2019-08-20 | 北京迈格威科技有限公司 | Facial image processing method, apparatus, system and readable storage medium storing program for executing |
| CN110213660A (en) * | 2019-05-27 | 2019-09-06 | 广州荔支网络技术有限公司 | Distribution method, system, computer equipment and the storage medium of program |
| CN110213660B (en) * | 2019-05-27 | 2021-08-20 | 广州荔支网络技术有限公司 | Program distribution method, system, computer device and storage medium |
| CN110674688A (en) * | 2019-08-19 | 2020-01-10 | 深圳力维智联技术有限公司 | Face recognition model acquisition method, system and medium for video monitoring scene |
| 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 |
| CN112580406A (en) * | 2019-09-30 | 2021-03-30 | 北京中关村科金技术有限公司 | Face comparison method and device and storage medium |
| WO2021073150A1 (en) * | 2019-10-16 | 2021-04-22 | 平安科技（深圳）有限公司 | Data detection method and apparatus, and computer device and storage medium |
| CN110765933A (en) * | 2019-10-22 | 2020-02-07 | 山西省信息产业技术研究院有限公司 | Dynamic portrait sensing comparison method applied to driver identity authentication system |
| CN110929099A (en) * | 2019-11-28 | 2020-03-27 | 杭州趣维科技有限公司 | Short video frame semantic extraction method and system based on multitask learning |
| CN111062338B (en) * | 2019-12-19 | 2023-11-17 | 厦门商集网络科技有限责任公司 | License and portrait consistency comparison method and system |
| CN111062338A (en) * | 2019-12-19 | 2020-04-24 | 厦门商集网络科技有限责任公司 | Certificate portrait consistency comparison method and system |
| CN111178249A (en) * | 2019-12-27 | 2020-05-19 | 杭州艾芯智能科技有限公司 | Face comparison method and device, computer equipment and storage medium |
| CN111368644A (en) * | 2020-02-14 | 2020-07-03 | 深圳市商汤科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN111368644B (en) * | 2020-02-14 | 2024-01-05 | 深圳市商汤科技有限公司 | Image processing method, device, electronic equipment and storage medium |
| CN111401257A (en) * | 2020-03-17 | 2020-07-10 | 天津理工大学 | Non-constraint condition face recognition method based on cosine loss |
| CN111401257B (en) * | 2020-03-17 | 2022-10-04 | 天津理工大学 | Face recognition method based on cosine loss under non-constraint condition |
| CN111401277A (en) * | 2020-03-20 | 2020-07-10 | 深圳前海微众银行股份有限公司 | Face recognition model updating method, device, equipment and medium |
| CN113538075A (en) * | 2020-04-14 | 2021-10-22 | 阿里巴巴集团控股有限公司 | Data processing method, model training method, device and equipment |
| CN112949780A (en) * | 2020-04-21 | 2021-06-11 | 佳都科技集团股份有限公司 | Feature model training method, device, equipment and storage medium |
| CN111709313A (en) * | 2020-05-27 | 2020-09-25 | 杭州电子科技大学 | Pedestrian re-identification method based on local and channel combination characteristics |
| CN111709313B (en) * | 2020-05-27 | 2022-07-29 | 杭州电子科技大学 | Pedestrian re-identification method based on local and channel combination characteristics |
| CN112052821A (en) * | 2020-09-15 | 2020-12-08 | 浙江智慧视频安防创新中心有限公司 | Fire fighting channel safety detection method, device, equipment and storage medium |
| CN112052821B (en) * | 2020-09-15 | 2023-07-07 | 浙江智慧视频安防创新中心有限公司 | Fire-fighting channel safety detection method, device, equipment and storage medium |
| CN112116021A (en) * | 2020-09-27 | 2020-12-22 | 广州华多网络科技有限公司 | Gem similarity measurement data processing method and related equipment |
| CN112492383A (en) * | 2020-12-03 | 2021-03-12 | 珠海格力电器股份有限公司 | Video frame generation method and device, storage medium and electronic equipment |
| CN113065495B (en) * | 2021-04-13 | 2023-07-14 | 深圳技术大学 | Image similarity calculation method, target object re-recognition method and system |
| CN113065495A (en) * | 2021-04-13 | 2021-07-02 | 深圳技术大学 | Image similarity calculation method, target object re-identification method and system |
| CN113283359A (en) * | 2021-06-02 | 2021-08-20 | 万达信息股份有限公司 | Authentication method and system for handheld certificate photo and electronic equipment |
| CN113344875A (en) * | 2021-06-07 | 2021-09-03 | 武汉象点科技有限公司 | Abnormal image detection method based on self-supervision learning |
| CN113269155A (en) * | 2021-06-28 | 2021-08-17 | 苏州市科远软件技术开发有限公司 | End-to-end face recognition method, device, equipment and storage medium |
| CN113486804B (en) * | 2021-07-07 | 2024-02-20 | 科大讯飞股份有限公司 | Object identification method, device, equipment and storage medium |
| CN113486804A (en) * | 2021-07-07 | 2021-10-08 | 科大讯飞股份有限公司 | Object identification method, device, equipment and storage medium |
| CN113642468A (en) * | 2021-08-16 | 2021-11-12 | 中国银行股份有限公司 | Identity authentication method and device |
| CN113688793A (en) * | 2021-09-22 | 2021-11-23 | 万章敏 | Training method of face model and face recognition system |
| WO2023093838A1 (en) * | 2021-11-25 | 2023-06-01 | 北京字跳网络技术有限公司 | Super-resolution image processing method and apparatus, and device and medium |
| US20230169792A1 (en) * | 2021-11-26 | 2023-06-01 | GenkiTek Technology Co., Ltd. | System and method of mode selection face recognition with parallel cnns |
| US12080101B2 (en) * | 2021-11-26 | 2024-09-03 | GenkiTek Technology Co., Ltd. | System and method of mode selection face recognition with parallel CNNS |
| CN114387457A (en) * | 2021-12-27 | 2022-04-22 | 腾晖科技建筑智能(深圳)有限公司 | Face intra-class interval optimization method based on parameter adjustment |
| CN114283483A (en) * | 2021-12-31 | 2022-04-05 | 中国联合网络通信集团有限公司 | Face recognition method and device |
| EP4383212A1 (en) * | 2022-12-06 | 2024-06-12 | Bundesdruckerei GmbH | Authentication device and method for authenticating a person by means of an identification document associated with the person, and identification document and method for producing |
| CN116127298A (en) * | 2023-02-22 | 2023-05-16 | 北京邮电大学 | Small sample radio frequency fingerprint identification method based on triplet loss |
| CN116127298B (en) * | 2023-02-22 | 2024-03-19 | 北京邮电大学 | Small sample radio frequency fingerprint identification method based on triplet loss |
| CN118337532A (en) * | 2024-06-13 | 2024-07-12 | 浙江鹏信信息科技股份有限公司 | Zero trust-based traffic safety audit protection method and system and readable medium |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2019128367A1 (en) | 2019-07-04 |
| CN108009528B (en) | 2020-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108009528A (en) | Face authentication method, device, computer equipment and storage medium based on Triplet Loss | |
| KR102125379B1 (en) | Apparatus and method for verifying identification based on deep learning | |
| WO2020155939A1 (en) | Image recognition method and device, storage medium and processor | |
| CN109784186A (en) | A kind of pedestrian recognition methods, device, electronic equipment and computer readable storage medium again | |
| CN108182409A (en) | Biopsy method, device, equipment and storage medium | |
| CN106022317A (en) | Face identification method and apparatus | |
| CN104820924B (en) | A kind of online safety payment system based on handwriting verification | |
| CN108304858A (en) | Fight specimen discerning model generating method, verification method and its system | |
| CN105844206A (en) | Identity authentication method and identity authentication device | |
| CN109284733A (en) | A kind of shopping guide's act of omission monitoring method based on yolo and multitask convolutional neural networks | |
| CN104504362A (en) | Face detection method based on convolutional neural network | |
| CN109902202A (en) | A kind of video classification methods and device | |
| CN109886222A (en) | Face identification method, neural network training method, device and electronic equipment | |
| Longjam et al. | Writer independent handwritten signature verification on multi-scripted signatures using hybrid CNN-BiLSTM: A novel approach | |
| US20190147218A1 (en) | User specific classifiers for biometric liveness detection | |
| CN106709418A (en) | Face identification method based on scene photo and identification photo and identification apparatus thereof | |
| CN109635643A (en) | A kind of fast human face recognition based on deep learning | |
| CN110008664A (en) | Authentication information acquisition, account-opening method, device and electronic equipment | |
| CN109344709A (en) | A kind of face generates the detection method of forgery image | |
| CN110008861A (en) | A kind of recognition methods again of the pedestrian based on global and local feature learning | |
| Alfarsi et al. | Techniques for face verification: Literature review | |
| Xie et al. | Writer-independent online signature verification based on 2D representation of time series data using triplet supervised network | |
| Kumar et al. | Enhancing Fingerprint Security Using CNN for Robust Biometric Authentication and Spoof Detection | |
| Goud et al. | Smart attendance notification system using SMTP with face recognition | |
| CN115457622A (en) | Method, system and equipment for detecting deeply forged faces based on identity invariant features |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: Face authentication method, device, computer device and storage medium based on triplet loss Effective date of registration: 20210621 Granted publication date: 20200407 Pledgee: Bank of China Limited by Share Ltd. Guangzhou Tianhe branch Pledgor: GRG Banking Equipment Co.,Ltd. Registration number: Y2021980004993 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Date of cancellation: 20230302 Granted publication date: 20200407 Pledgee: Bank of China Limited by Share Ltd. Guangzhou Tianhe branch Pledgor: GRG BANKING EQUIPMENT Co.,Ltd. Registration number: Y2021980004993 |