CN109558827A - A kind of finger vein identification method and system based on personalized convolutional neural networks - Google Patents
A kind of finger vein identification method and system based on personalized convolutional neural networks Download PDFInfo
- Publication number
- CN109558827A CN109558827A CN201811416590.0A CN201811416590A CN109558827A CN 109558827 A CN109558827 A CN 109558827A CN 201811416590 A CN201811416590 A CN 201811416590A CN 109558827 A CN109558827 A CN 109558827A
- Authority
- CN
- China
- Prior art keywords
- sample
- personalized
- convolutional neural
- neural networks
- user
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06V40/14—Vascular patterns
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
- Collating Specific Patterns (AREA)
Abstract
The present invention discloses a kind of finger vein identification method based on personalized convolutional neural networks, it is related to finger vein identification technology field, this method includes including training and identification two parts, personalized weighting is carried out in the importance that training part is primarily based on positive class sample, sample deposit customized information after personalized weighting is generated into center, sample is constituted to being trained simultaneously and by the positive class sample after weighting and the finger venous image of collection, to construct personalized convolutional neural networks, central configuration then is generated by customized information in identification division, identification is completed by personalized convolutional neural networks, and export recognition result, as a result it is divided into and is verified and verifying not by two kinds.Invention additionally discloses a kind of finger vein recognition systems, combine with recognition methods, and the identity information of Very Important Person is obtained for different scene requirements, improve discrimination and user satisfaction.
Description
Technical field
It is specifically a kind of based on personalized convolutional neural networks the present invention relates to finger vein identification technology field
Finger vein identification method and system.
Background technique
The main mode of information age authentication is had become based on living things feature recognition.In recent years, finger vena
Identification has become a kind of authentication mode of great potential due to its exclusive advantage.Finger vein identification technology utilizes finger
Interior vein distributed image carries out identification.Its working principle is according to the absorbable spy of blood flowed in human finger
The long light of standing wave, and referred to using specific wavelength light opponent and be irradiated, the clear image of finger vena can be obtained.Utilize this
One intrinsic scientific feature will be realized and be analyzed the image of acquisition, be handled, so that the biological characteristic of finger vena is obtained,
Obtained finger vein features information is compared with the finger vein features of registration in advance again, to confirm the body of registrant
Part.
Compared with other biological identification technology, fin- ger vein authentication technology has following main advantage: biological identification technology, no
It can lose, will not be stolen, memoryless password burden.Original finger vena image is captured and digitized processing, image compare by
The proprietary finger vena extraction algorithm of Hitachi is completed, and whole process was less than 1 second；Inside of human body information, not by rough coat, outside
The influence of environment (temperature, humidity).User's psychological reactance is low, and the factor influenced by physiology and environment is also low, comprising: dry
Skin, greasy dirt, the pollution such as dust, skin surface exception etc.；Wide application of the crowd, accuracy rate are high, not reproducible, can not forge, and pacify
It is complete convenient.Vein is hidden in body interior, the probability very little for being replicated or usurping.Accuracy of system identification is 0.0001%, refuse sincere be
0.01%, registration failure rate is less than 0.03%；Vivo identification, when carrying out identification with finger vena, acquisition is that finger is quiet
The characteristics of image of arteries and veins, feature existing for ability when being finger living body.
However, conventional finger vein identification method only makes discrimination highest, special screne (low rejection or low mistake are had ignored
Knowledge rate) under specified conditions.For example, usually only specifying several people to possess financial process power for a financial system
Limit.For this scene, need the authentication to this several people that there is high accuracy of identification, otherwise, if allowing wrong to other
Personnel, which assign financial process permission, will bring huge loss.For another example for runaway convict's identifying system, if wrong by runaway convict
Good citizen are mistakenly identified as, then the best chance for arresting runaway convict will be lost, to threaten to social safety.In order to solve this
Specific application class problem under class special scenes, the invention proposes a kind of finger venas based on personalized convolutional neural networks
Recognition methods and system.
Summary of the invention
The present invention is directed to the demand and shortcoming of current technology development, provides a kind of based on personalized convolutional neural networks
Finger vein identification method and system.
A kind of finger vein identification method based on personalized convolutional neural networks of the invention solves above-mentioned technical problem
The technical solution adopted is as follows:
A kind of finger vein identification method based on personalized convolutional neural networks, includes the following steps:
1) training part:
1a) firstly, obtaining positive class sample, positive class sample refers to the finger vena that responsible consumer is concerned in special scenes
Image；
1b) then, according to application scenarios, the customized information for introducing user generates center, to assign to each positive class sample
Different weights is given, the significance level of user is distinguished；
1c) again, the sample for collecting finger venous image takes one of sample and customized information to generate the sample at center
This progress manually compares, constitute a sample to and be trained so that the sample of collected finger venous image and personalized
Information generates when the sample at center is similar and exports 1, output -1 when inhomogeneity；The sample of finger venous image can be positive class sample
Originally, negative class sample, negative class sample refer to the finger venous image for not being concerned user in special scenes；
1d) finally, introducing the personalized weight of user in convolutional neural networks, input sample is constructed to being trained
Personalized convolutional neural networks；
2) identification division:
The positive class sample that finger venous image and customized information to be verified generate center 2a) is constituted into sample pair；
2b) the convolutional neural networks that sample is personalized to input, output 1 cannot be passed through when inputting -1 by verifying
Verifying.
Center is generated in customized information, each positive class sample has different weights, introduces formula (1):
In formula (1), N is the number of positive class sample, RiIt is the weight of i-th of user, measures i-th of user identity
Importance, UiThe data of i-th of user, T indicate the summation after N number of positive class sample personalization weighting.
When introducing the personalized weight of user in convolutional neural networks, need to establish loss function, such as formula (2) institute
Show:
s.t Li=wzi+b
s.t zi=xi-T
In formula (2), C1It is the loss weight for responsible consumer mistake being divided into general user, C2It is general user's mistake point
For the loss weight of responsible consumer；QiIt is the classification indicator function of i-th of sample；If i-th of sample is responsible consumer, Qi=
1；Conversely, Qi=-1；LiIndicate the prediction category result of i-th of sample；W is the weight of trained part, and b is the inclined of trained part
It sets；ziIt is the sample pair of input, passes through sample xiJudged with the positive class center T of generation, if the two samples are similar
, then it is labeled as 1.
Loss function, that is, formula (2) is optimized, in optimization process, acquires loss weight C according to following algorithm1And C2,
And parameter w and b:
Step 1, fixed C1And C2Two variables.Initial value is assigned to two variables, so that C1>>C2；
Step 2, by C1And C2It substitutes into formula (2), the parameter w and b of model is acquired using stochastic gradient descent method；
Then step 3 brings the w acquired and b into formula (2), by enabling C again1And C2Local derviation be equal to 0, acquire C1With
C2Value；
Step 4 repeats step step 2 and step 3, until reaching the condition of convergence.
Based on above-mentioned finger vein identification method, the present invention also provides a kind of fingers based on personalized convolutional neural networks
Vein recognition system, the system include training part and identification division；
Training department divides
Acquisition module, for acquiring positive class sample, positive class sample refers to the hand that responsible consumer is concerned in special scenes
Refer to vein image；
Generation module is set, for assigning different weights to each positive class sample according to the significance level of user, is distinguished
The significance level of user, and then generate customized information and generate center module；
Output module is constructed, for the finger venous image manually compared and the positive class sample of weighting to be constituted sample pair, root
Sample is indicated to for similar according to artificial comparison result output 1 or -1,1, and to for inhomogeneity, finger venous image is -1 expression sample
Any of all samples, all samples include positive class sample and negative class sample, and negative class sample refers in special scenes not
It is concerned the finger venous image of user；
Training building module, based on the different weights that each sample has, convolutional neural networks training sample pair, building
Property convolutional neural networks module；
Identification division includes:
Customized information generate center module, for store weight after positive class sample, and with finger vena to be verified
Image construction sample pair, wherein positive class sample refers to the finger venous image for being concerned responsible consumer in special scenes；
Personalized convolutional neural networks module exports recognition result 1 or -1,1 indicates verifying for analyzing contrast sample couple
Pass through, -1 indicates that verifying does not pass through.
When generating customized information generation center module, each positive class sample has different weights, introduces formula
(1):
In formula (1), N is the number of positive class sample, RiIt is the weight of i-th of user, measures i-th of user identity
Importance, UiThe data of i-th of user, T indicate the summation after N number of positive class sample personalization weighting.
When training building module to construct personalized convolutional neural networks module, weighed based on the difference that each sample has
Weight, establishes loss function, as shown in formula (2):
s.t Li=wzi+b
s.t zi=xi-T
In formula (2), C1It is the loss weight for responsible consumer mistake being divided into general user, C2It is general user's mistake point
For the loss weight of responsible consumer；QiIt is the classification indicator function of i-th of sample；If i-th of sample is responsible consumer, Qi=
1；Conversely, Qi=-1；LiIndicate the prediction category result of i-th of sample；W is the weight of trained part, and b is the inclined of trained part
It sets；ziIt is the sample pair of input, passes through sample xiJudged with the positive class center T of generation, if the two samples are similar
, then it is labeled as 1.
Loss function, that is, formula (2) is optimized, in optimization process, acquires loss weight C according to following algorithm1And C2,
And parameter w and b:
Step 1, fixed C1And C2Two variables.Initial value is assigned to two variables, so that C1>>C2；
Step 2, by C1And C2It substitutes into formula (2), the parameter w and b of model is acquired using stochastic gradient descent method；
Then step 3 brings the w acquired and b into formula (2), by enabling C again1And C2Local derviation be equal to 0, acquire C1With
C2Value；
Step 4 repeats step step 2 and step 3, until reaching the condition of convergence.
A kind of finger vein identification method and system based on personalized convolutional neural networks of the invention, with the prior art
Compared to having the beneficial effect that
1) finger vein identification method of the invention includes training and identification two parts, is primarily based on positive class in training part
The importance of sample carries out personalized weighting, the sample deposit customized information after personalized weighting is generated center, while simultaneously
Positive class sample after weighting and the finger venous image of collection are constituted into sample to being trained, to construct personalized convolution mind
Through network, central configuration then is generated by customized information in identification division, completes to know by personalized convolutional neural networks
Not, and recognition result is exported, is as a result divided into and is verified and verifies not by two kinds, this recognition methods is especially for different fields
Scape requires to obtain the identity information of Very Important Person, improves discrimination and user satisfaction；
2) finger vein recognition system of the invention is combined with recognition methods, obtains weight for different scene requirements
The identity information of client is wanted, discrimination and user satisfaction are improved；
3) finger vein identification method of the invention and system are particularly suitable for the needs such as financial system, runaway convict's identifying system
The special scenes of low reject rate or low misclassification rate, it is enough that efficiently the finger vena of specific user is correctly identified, it improves
Discrimination and user satisfaction.
Detailed description of the invention
Attached drawing 1 is the connection block diagram of the embodiment of the present invention two.
Each label information indicates in attached drawing:
10, acquisition module, 20, setting generation module, 30, construction output module, 40, training building module, 50, personalization
Information generates center module, 60, personalized convolutional neural networks module.
Specific embodiment
The technical issues of to make technical solution of the present invention, solving and technical effect are more clearly understood, below in conjunction with tool
Body embodiment is checked technical solution of the present invention, is completely described, it is clear that described embodiment is only this hair
Bright a part of the embodiment, instead of all the embodiments.Based on the embodiment of the present invention, those skilled in the art are not doing
All embodiments obtained under the premise of creative work out, all within protection scope of the present invention.
Embodiment one:
The present embodiment proposes a kind of finger vein identification method based on personalized convolutional neural networks, including walks as follows
It is rapid:
1) training part:
1a) firstly, obtaining positive class sample, positive class sample refers to the finger vena that responsible consumer is concerned in special scenes
Image；
1b) then, according to application scenarios, the customized information for introducing user generates center, to assign to each positive class sample
Different weights is given, the significance level of user is distinguished；
1c) again, the sample for collecting finger venous image takes one of sample and customized information to generate the sample at center
This progress manually compares, constitute a sample to and be trained so that the sample of collected finger venous image and personalized
Information generates when the sample at center is similar and exports 1, output -1 when inhomogeneity；The sample of finger venous image can be positive class sample
Originally, negative class sample, negative class sample refer to the finger venous image for not being concerned user in special scenes；
1d) finally, introducing the personalized weight of user in convolutional neural networks, input sample is constructed to being trained
Personalized convolutional neural networks；
2) identification division:
The positive class sample that finger venous image and customized information to be verified generate center 2a) is constituted into sample pair；
2b) the convolutional neural networks that sample is personalized to input, output 1 cannot be passed through when inputting -1 by verifying
Verifying.
Center is generated in customized information, each positive class sample has different weights, introduces formula (1):
In formula (1), N is the number of positive class sample, RiIt is the weight of i-th of user, measures i-th of user identity
Importance, UiThe data of i-th of user, T indicate the summation after N number of positive class sample personalization weighting.
When introducing the personalized weight of user in convolutional neural networks, need to establish loss function, such as formula (2) institute
Show:
s.t Li=wzi+b
S.t zi=xi-T
In formula (2), C1It is the loss weight for responsible consumer mistake being divided into general user, C2It is general user's mistake point
For the loss weight of responsible consumer；QiIt is the classification indicator function of i-th of sample；If i-th of sample is responsible consumer, Qi=
1；Conversely, Qi=-1；LiIndicate the prediction category result of i-th of sample；W is the weight of trained part, and b is the inclined of trained part
It sets；ziIt is the sample pair of input, passes through sample xiJudged with the positive class center T of generation, if the two samples are similar
, then it is labeled as 1.
Loss function, that is, formula (2) is optimized, in optimization process, acquires loss weight C according to following algorithm1And C2,
And parameter w and b:
Step 1, fixed C1And C2Two variables.Initial value is assigned to two variables, so that C1>>C2；
Step 2, by C1And C2It substitutes into formula (2), the parameter w and b of model is acquired using stochastic gradient descent method；
Then step 3 brings the w acquired and b into formula (2), by enabling C again1And C2Local derviation be equal to 0, acquire C1With
C2Value；
Step 4 repeats step step 2 and step 3, until reaching the condition of convergence.
The finger vein identification method of the present embodiment includes training and identification two parts, is primarily based on positive class in training part
The importance of sample carries out personalized weighting, the sample deposit customized information after personalized weighting is generated center, while simultaneously
Positive class sample after weighting and the finger venous image of collection are constituted into sample to being trained, to construct personalized convolution mind
Through network, central configuration then is generated by customized information in identification division, completes to know by personalized convolutional neural networks
Not, and recognition result is exported, is as a result divided into and is verified and verifies not by two kinds, this recognition methods is especially for different fields
Scape requires to obtain the identity information of Very Important Person, improves discrimination and user satisfaction；
Embodiment two:
In conjunction with attached drawing 1, the present embodiment also provides a kind of finger vein recognition system based on personalized convolutional neural networks,
The system includes training part and identification division；
Training department divides
Acquisition module 10, for acquiring positive class sample, positive class sample refers to being concerned responsible consumer in special scenes
Finger venous image；
Generation module 20 is set, for assigning different weights, area to each positive class sample according to the significance level of user
Divide the significance level of user, and then generates customized information and generate center module；
Output module 30 is constructed, for the finger venous image manually compared and the positive class sample of weighting to be constituted sample pair,
Sample is indicated to for similar according to artificial comparison result output 1 or -1,1, and -1 expression sample is to for inhomogeneity, finger venous image
It is any of all samples, all samples include positive class sample and negative class sample, and negative class sample refers in special scenes
It is not concerned the finger venous image of user；
Training building module 40, based on the different weights that each sample has, convolutional neural networks training sample pair, building
Personalized convolutional neural networks module；
Identification division includes:
Customized information generates center module 50, for storing the positive class sample after weighting, and it is quiet with finger to be verified
Arteries and veins image construction sample pair, wherein positive class sample refers to the finger venous image for being concerned responsible consumer in special scenes；
Personalized convolutional neural networks module 60 exports recognition result 1 or -1,1 indicates to test for analyzing contrast sample couple
Card passes through, and -1 indicates that verifying does not pass through.
When generating customized information generation center module 50, each positive class sample has different weights, introduces formula
(1):
In formula (1), N is the number of positive class sample, RiIt is the weight of i-th of user, measures i-th of user identity
Importance, UiThe data of i-th of user, T indicate the summation after N number of positive class sample personalization weighting.
When training building module 40 constructs personalized convolutional neural networks module 60, the difference that is had based on each sample
Weight establishes loss function, as shown in formula (2):
s.t Li=wzi+b
s.t zi=xi-T
In formula (2), C1It is the loss weight for responsible consumer mistake being divided into general user, C2It is general user's mistake point
For the loss weight of responsible consumer；QiIt is the classification indicator function of i-th of sample；If i-th of sample is responsible consumer, Qi=
1；Conversely, Qi=-1；LiIndicate the prediction category result of i-th of sample；W is the weight of trained part, and b is the inclined of trained part
It sets；ziIt is the sample pair of input, passes through sample xiJudged with the positive class center T of generation, if the two samples are similar
, then it is labeled as 1.
Loss function, that is, formula (2) is optimized, in optimization process, acquires loss weight C according to following algorithm1And C2,
And parameter w and b:
Step 1, fixed C1And C2Two variables.Initial value is assigned to two variables, so that C1>>C2；
Step 2, by C1And C2It substitutes into formula (2), the parameter w and b of model is acquired using stochastic gradient descent method；
Then step 3 brings the w acquired and b into formula (2), by enabling C again1And C2Local derviation be equal to 0, acquire C1With
C2Value；
Step 4 repeats step step 2 and step 3, until reaching the condition of convergence.
The finger vein recognition system of the present embodiment includes training and identification two parts.In training part, by acquiring mould
Block 10 acquires positive class sample, according to the significance level of positive class sample, is assigned by setting generation module 20 to positive class sample different
Weight, and generate customized information generate center module 50, for store weight after positive class sample, then, by collection
A wherein finger venous image, with the positive class sample of weighting constitute sample to and manually compared, be to be similar in the two
Output 1, input -1 when inhomogeneity, finally, based on the positive class sample after personalized weighting, training simultaneously constructs personalized convolutional Neural
Network module 60, that is to say, that complete customized information in training part and generate center module 50 and personalized convolutional Neural
The building of network module 60.In identification division, we can will be to be verified finger venous image and during customized information generates
The personalization of core module 50 weights positive class sample and constitutes sample pair, and inputs the personalized convolutional neural networks module that building is completed
60 are identified, output result, which is 1, indicates that, by verifying, output result is that -1 expression does not pass through verifying.The present embodiment
Finger vein recognition system is combined with the recognition methods of embodiment one, obtains Very Important Person for different scene requirements
Discrimination and user satisfaction can be improved in identity information.
For above-mentioned two embodiment, it can be appreciated that finger vein identification method and system of the invention is especially suitable
The special scenes of low reject rate or low misclassification rate are needed for financial system, runaway convict's identifying system etc., it is enough efficiently to specific
The finger vena of user is correctly identified, discrimination and user satisfaction are improved.
Use above specific case elaborates the principle of the present invention and embodiment, these embodiments are
It is used to help understand core of the invention technology contents, the protection scope being not intended to restrict the invention, technical side of the invention
Case is not limited in above-mentioned specific embodiment.Based on above-mentioned specific embodiment of the invention, those skilled in the art
Without departing from the principle of the present invention, any improvement and modification to made by the present invention should all be fallen into of the invention special
Sharp protection scope.
Claims (8)
1. a kind of finger vein identification method based on personalized convolutional neural networks, which comprises the steps of:
1) training part:
1a) firstly, obtaining positive class sample, positive class sample refers to the finger vena figure that responsible consumer is concerned in special scenes
Picture；
1b) then, according to application scenarios, the customized information for introducing user generates center, to assign not to each positive class sample
Same weight, distinguishes the significance level of user；
1c) again, the sample for collecting finger venous image, take one of sample and customized information generate the sample at center into
Pedestrian's work compare, constitute a sample to and be trained so that the sample and customized information of collected finger venous image
The sample at generation center exports 1 when being similar, output -1 when inhomogeneity；The sample of finger venous image can be positive class sample,
Negative class sample, negative class sample refer to the finger venous image for not being concerned user in special scenes；
1d) finally, introducing the personalized weight of user in convolutional neural networks, input sample constructs individual character to being trained
The convolutional neural networks of change；
2) identification division:
The positive class sample that finger venous image and customized information to be verified generate center 2a) is constituted into sample pair；
2b) the convolutional neural networks that sample is personalized to input, output 1 are cannot to pass through verifying when inputting -1 by verifying.
2. a kind of finger vein identification method based on personalized convolutional neural networks according to claim 1, feature
It is, generates center in customized information, each positive class sample has different weights, introduces formula (1):
In formula (1), N is the number of positive class sample, RiIt is the weight of i-th of user, measures the important of i-th of user identity
Property, UiThe data of i-th of user, T indicate the summation after N number of positive class sample personalization weighting.
3. a kind of finger vena identification based on personalized convolutional neural networks according to claim 2, which is characterized in that
When introducing the personalized weight of user in convolutional neural networks, need to establish loss function, as shown in formula (2):
s.t Li=wzi+b
s.t zi=xi-T
In formula (2), C1It is the loss weight for responsible consumer mistake being divided into general user, C2It is general user's mistake to be divided into weight
Want the loss weight of user；QiIt is the classification indicator function of i-th of sample；If i-th of sample is responsible consumer, Qi=1；Instead
It, Qi=-1；LiIndicate the prediction category result of i-th of sample；W is the weight of trained part, and b is the biasing of trained part；zi
It is the sample pair of input, passes through sample xiJudged with the positive class center T of generation, if the two samples be it is similar,
Labeled as 1.
4. a kind of finger vena identification based on personalized convolutional neural networks according to claim 3, which is characterized in that
Loss function, that is, formula (2) is optimized, in optimization process, acquires loss weight C according to following algorithm1And C2And parameter
W and b:
Step 1, fixed C1And C2Two variables.Initial value is assigned to two variables, so that C1>>C2；
Step 2, by C1And C2It substitutes into formula (2), the parameter w and b of model is acquired using stochastic gradient descent method；
Then step 3 brings the w acquired and b into formula (2), by enabling C again1And C2Local derviation be equal to 0, acquire C1And C2's
Value；
Step 4 repeats step step 2 and step 3, until reaching the condition of convergence.
5. a kind of finger vein recognition system based on personalized convolutional neural networks, which is characterized in that the system includes training
Part and identification division；
The training department divides
Acquisition module, for acquiring positive class sample, positive class sample refers to that the finger that responsible consumer is concerned in special scenes is quiet
Arteries and veins image；
Generation module is set, for assigning different weights to each positive class sample according to the significance level of user, distinguishes user
Significance level, and then generate customized information generate center module；
Output module is constructed, for the finger venous image manually compared and the positive class sample of weighting to be constituted sample pair, according to people
Work comparison result output 1 or -1,1 indicates sample to for similar, and for -1 expression sample to for inhomogeneity, finger venous image is all
Any of sample, all samples include positive class sample and negative class sample, and negative class sample refers to not closed in special scenes
Infuse the finger venous image of user；
Training building module, based on the different weights that each sample has, convolutional neural networks training sample pair, building is personalized
Convolutional neural networks module；
The identification division includes:
Customized information generate center module, for store weight after positive class sample, and with finger venous image to be verified
Sample pair is constituted, wherein positive class sample refers to the finger venous image for being concerned responsible consumer in special scenes；
Personalized convolutional neural networks module exports recognition result 1 or -1,1 indicates that verifying is logical for analyzing contrast sample couple
It crosses, -1 indicates that verifying does not pass through.
6. a kind of finger vein recognition system based on personalized convolutional neural networks according to claim 5, feature
It is, when generating customized information generation center module, each positive class sample has different weights, it introduces formula (1):
In formula (1), N is the number of positive class sample, RiIt is the weight of i-th of user, measures the important of i-th of user identity
Property, UiThe data of i-th of user, T indicate the summation after N number of positive class sample personalization weighting.
7. a kind of finger vein recognition system based on personalized convolutional neural networks according to claim 6, feature
It is, when training building module constructs personalized convolutional neural networks module, based on the different weights that each sample has, builds
Vertical loss function, as shown in formula (2):
s.t Li=wzi+b
s.t zi=xi-T
In formula (2), C1It is the loss weight for responsible consumer mistake being divided into general user, C2It is general user's mistake to be divided into weight
Want the loss weight of user；QiIt is the classification indicator function of i-th of sample；If i-th of sample is responsible consumer, Qi=1；Instead
It, Qi=-1；LiIndicate the prediction category result of i-th of sample；W is the weight of trained part, and b is the biasing of trained part；zi
It is the sample pair of input, passes through sample xiJudged with the positive class center T of generation, if the two samples be it is similar,
Labeled as 1.
8. a kind of finger vein recognition system based on personalized convolutional neural networks according to claim 7, feature
It is, loss function, that is, formula (2) is optimized, in optimization process, acquires loss weight C according to following algorithm1And C2, with
And parameter w and b:
Step 1, fixed C1And C2Two variables.Initial value is assigned to two variables, so that C1>>C2；
Step 2, by C1And C2It substitutes into formula (2), the parameter w and b of model is acquired using stochastic gradient descent method；
Then step 3 brings the w acquired and b into formula (2), by enabling C again1And C2Local derviation be equal to 0, acquire C1And C2's
Value；
Step 4 repeats step step 2 and step 3, until reaching the condition of convergence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811416590.0A CN109558827A (en) | 2018-11-26 | 2018-11-26 | A kind of finger vein identification method and system based on personalized convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811416590.0A CN109558827A (en) | 2018-11-26 | 2018-11-26 | A kind of finger vein identification method and system based on personalized convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109558827A true CN109558827A (en) | 2019-04-02 |
Family
ID=65867597
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811416590.0A Pending CN109558827A (en) | 2018-11-26 | 2018-11-26 | A kind of finger vein identification method and system based on personalized convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109558827A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414373A (en) * | 2019-07-08 | 2019-11-05 | 武汉大学 | A deep learning palm vein recognition system and method based on cloud-edge-device collaborative computing |
| CN113269080A (en) * | 2021-05-20 | 2021-08-17 | 南京邮电大学 | Palm vein identification method based on multi-channel convolutional neural network |
| WO2023243796A1 (en) * | 2022-06-17 | 2023-12-21 | Samsung Electronics Co., Ltd. | Method and system for personalising machine learning models |
| US12488801B2 (en) | 2022-06-17 | 2025-12-02 | Samsung Electronics Co., Ltd. | Method and system for personalising machine learning models |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108537206A (en) * | 2018-04-23 | 2018-09-14 | 济南浪潮高新科技投资发展有限公司 | A kind of face verification method based on convolutional neural networks |
-
2018
- 2018-11-26 CN CN201811416590.0A patent/CN109558827A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108537206A (en) * | 2018-04-23 | 2018-09-14 | 济南浪潮高新科技投资发展有限公司 | A kind of face verification method based on convolutional neural networks |
Non-Patent Citations (1)
| Title |
|---|
| 袭肖明等: "基于手指静脉和手指轮廓的个性化融合方法", 《计算机研究与发展》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414373A (en) * | 2019-07-08 | 2019-11-05 | 武汉大学 | A deep learning palm vein recognition system and method based on cloud-edge-device collaborative computing |
| CN110414373B (en) * | 2019-07-08 | 2021-09-24 | 武汉大学 | A deep learning palm vein recognition system and method based on cloud-side-terminal collaborative computing |
| CN113269080A (en) * | 2021-05-20 | 2021-08-17 | 南京邮电大学 | Palm vein identification method based on multi-channel convolutional neural network |
| CN113269080B (en) * | 2021-05-20 | 2022-07-12 | 南京邮电大学 | Palm vein identification method based on multi-channel convolutional neural network |
| WO2023243796A1 (en) * | 2022-06-17 | 2023-12-21 | Samsung Electronics Co., Ltd. | Method and system for personalising machine learning models |
| US12488801B2 (en) | 2022-06-17 | 2025-12-02 | Samsung Electronics Co., Ltd. | Method and system for personalising machine learning models |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jyotishi et al. | An LSTM-based model for person identification using ECG signal | |
| CN107220600B (en) | A kind of Picture Generation Method and generation confrontation network based on deep learning | |
| CN108776788A (en) | A kind of recognition methods based on brain wave | |
| CN105989266B (en) | Authentication method, device and system based on ECG signal | |
| CN109558827A (en) | A kind of finger vein identification method and system based on personalized convolutional neural networks | |
| CN111062021B (en) | Method and device for identity authentication based on wearable equipment | |
| US20200401846A1 (en) | System and method for detecting potential matches between a candidate biometric and a dataset of biometrics | |
| Pinto et al. | An end-to-end convolutional neural network for ECG-based biometric authentication | |
| Patro et al. | Machine learning classification approaches for biometric recognition system using ECG signals. | |
| CN113208593A (en) | Multi-modal physiological signal emotion classification method based on correlation dynamic fusion | |
| CN108256452A (en) | A kind of method of the ECG signal classification of feature based fusion | |
| CN108323201A (en) | A kind of identity authentication method and device | |
| CN116861217B (en) | Identity recognition method and system for mobile terminal | |
| KR20210023170A (en) | Deep learning based emotional recognition system and methods using PPG signals | |
| CN115439892A (en) | Palm print palm vein multi-mode fusion-based identity recognition method | |
| Karthiga et al. | Feature selection using multi-objective modified genetic algorithm in multimodal biometric system | |
| Xie et al. | WT feature based emotion recognition from multi-channel physiological signals with decision fusion | |
| CN109620260A (en) | Psychological condition recognition methods, equipment and storage medium | |
| Mohamed et al. | Recurrent Neural Networks (RNNs) to improve EEG-based person identification | |
| Guelta et al. | An Improved Behavioral Biometric System based on Gait and ECG signals. | |
| CN119293772B (en) | Method and system for carrying out joint verification on digital identity by using combination model | |
| CN120196873A (en) | A lightweight mobile PPG signal identification method and system | |
| Jien et al. | Age-based facial recognition using convoluted neural network deep learning algorithm | |
| Rehkha et al. | Differentiating monozygotic twins by facial features | |
| Wanis et al. | A Face Recognition using Convolutional Neural Network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190402 |
|
| RJ01 | Rejection of invention patent application after publication |