CN107704822A - The extraction of finger vein further feature and matching process and system based on incomplete completion - Google Patents
The extraction of finger vein further feature and matching process and system based on incomplete completion Download PDFInfo
- Publication number
- CN107704822A CN107704822A CN201710914184.6A CN201710914184A CN107704822A CN 107704822 A CN107704822 A CN 107704822A CN 201710914184 A CN201710914184 A CN 201710914184A CN 107704822 A CN107704822 A CN 107704822A
- Authority
- CN
- China
- Prior art keywords
- finger vein
- vein image
- further feature
- registered
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- 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
- G06V40/14—Vascular patterns
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Collating Specific Patterns (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the extraction of the finger vein further feature based on incomplete completion and matching process and system, it is characterised in that refer to the extraction of vein further feature includes with matching process：Finger vein image to be registered is pre-processed, obtains pretreated finger vein image to be registered；Further feature extraction training managing is carried out to pretreated finger vein image to be registered using depth convolutional neural networks, obtains finger vein image further feature to be registered；Registered using finger vein image further feature to be registered and build ATL processing, obtain the retention further feature ATL after registration；Finger vein image to be identified is gathered, finger vein image to be identified is subjected to feature extraction, obtains finger vein image further feature to be identified；Match cognization processing is carried out using finger vein image further feature to be identified and the retention further feature ATL after registration, obtains match cognization result.In embodiments of the present invention, the high discrimination feature for referring to vein image is extracted, improves matching precision.
Description
Technical field
The present invention relates to biometrics identification technology field, more particularly to finger vein further feature based on incomplete completion carries
Take and matching process and system.
Background technology
As a member of bio-identification family, refer to vein identification technology and be widely studied, be often divided into pretreatment, feature carries
Take, three parts of characteristic matching；Existing technical scheme is mostly based on knowledge in specific area, mainly covers physiological characteristic research
Field, digital image processing field so that refer to the characteristic extraction part in vein identification technology and be often difficult to extract or realize, and
And lack the redundancy from acquired original image zooming-out feature.
It is difficult to extract in addition, referring to vein identification technology and desirable features currently still be present, characteristic dimension is too high, and matching is difficult
The problems such as, certain obstacle is caused to the development for referring to vein.
Existing technical scheme one is the projection change detection feature based on prevalence study；In the training stage, by a large amount of
Training obtains best projection matrix, and a finger training just needs several samples of same finger；In cognitive phase, training is utilized
The eigenmatrix of obtained best projection matrix computations sample to be matched；Try to achieve after eigenmatrix with all samples in registry
This calculating Euclidean distance, classified using nearest neighbour classification device；Existing technical scheme two is to be entered based on deep neural network
The fin- ger vein authentication method of row feature extraction；Author refers to venosomes using other existing finger vein Enhancement Method marks,
Vein is divided into circle of good definition and confusion region and background area, is then trained using convolutional neural networks, obtains extracting feature
Neural network model.
Existing one each finger of technical scheme need at least 45 sample trainings be only possible to training obtain it is relatively good most
Good projection matrix, and must be the more complete image after pretreatment；In practical problem, user can not possibly be just like
Sample more than this is trained, and will also result in the damage of precision to a certain extent when classification in addition using nearest neighbour classification device
Lose；Existing technical scheme two needs to use other finger vena characteristic extracting methods to extract vein in the training process, due to original
Larger difference be present in the feature for having feature and extracting；Again by study, the feature learnt and actual characteristic gap meeting
It is bigger, reduce accuracy of identification.
The content of the invention
It is deep the invention provides the finger vein based on incomplete completion it is an object of the invention to overcome the deficiencies in the prior art
Layer feature extracting and matching method and system, extract the high discrimination feature for referring to vein image, improve matching precision.
In order to solve the above-mentioned technical problem, the embodiments of the invention provide the finger vein further feature based on incomplete completion to carry
Take and matching process, the finger vein further feature extraction include with matching process：
S11：Finger vein image to be registered is pre-processed, obtains pretreated finger vein image to be registered；
It is described that finger vein image to be registered is pre-processed, include progress region of interesting extraction processing successively, image
Normalized, histogram remap processing；
S12：Further feature is carried out to the pretreated finger vein image to be registered using depth convolutional neural networks
Training managing is extracted, obtains finger vein image further feature to be registered；
It is described that further feature is carried out to the pretreated finger vein image to be registered using depth convolutional neural networks
Training managing is extracted, detailed process is as follows：
1) the pretreated finger vein image to be registered is carried out into incomplete feature completion to handle, after obtaining feature completion
Finger vein image to be registered；
2) it is quiet to the finger to be registered after the feature completion using the first convolution pond layer is formed by foundation characteristic convolution kernel
Arteries and veins image carries out the processing of first time convolution pondization, obtains the foundation characteristic to be registered for referring to vein image；
3) using the foundation characteristic by the second convolution pond layer that high-order spy's convolution kernel forms to finger vein image to be registered
Second of convolution pondization processing is carried out, finger vein image depth to be registered is obtained and hides feature；
4) the full connection processing processing of feature progress is hidden to the finger vein image depth to be registered, it is quiet obtains finger to be registered
Arteries and veins image further feature；
S13：Registered using the finger vein image further feature to be registered and build ATL processing, obtain registration
Retention further feature ATL afterwards；
S14：Finger vein image to be identified is gathered, the finger vein image to be identified is subjected to S1, S2 step process successively,
Obtain finger vein image further feature to be identified；
S15：Using the finger vein image further feature to be identified and the retention further feature ATL after the registration
Match cognization processing is carried out, obtains match cognization result.
Preferably, the hierarchical structure of the depth convolutional neural networks is 6 layers, and wherein first layer is input layer, the second layer
For incomplete completion layer, third layer be the first convolution pond layer, the 4th layer be the second convolution pond layer, layer 5 be full articulamentum,
Layer 6 is output layer.
Preferably, it is described that the pretreated finger vein image to be registered is subjected to incomplete feature completion processing, including：
Line data set amplification processing is entered to the pretreated finger vein image to be registered, obtains and waits to train after amplification
Refer to vein image；
It will treat that training refers to vein image in the incomplete completion layer and is iterated training managing after the amplification, pass through
Repetitive exercise obtains the finger vein image to be registered after feature completion.
Preferably, it is described using the finger vein image further feature to be identified and the retention further feature after the registration
ATL carries out match cognization processing, including：
By the finger vein image further feature to be identified and staying in the retention further feature ATL after the registration
Deposit depth characteristic and carry out Euclidean distance calculating one by one, obtain Euclidean distance set；
Euclidean distance in the Euclidean distance set is ranked up, Euclidean distance minimum in ranking results is chosen and makees
For matching result；
Judge whether the matching result is less than given threshold, if so, then the matching result is to identify successfully, if it is not,
Then the matching result is recognition failures.
Preferably, it is described using the finger vein image further feature to be identified and the retention further feature after the registration
ATL, which carries out match cognization processing, also to be included：
By the identification retention further feature that successfully finger vein image further feature to be identified is added after the registration
In ATL, using the successful finger vein image further feature to be identified of the identification to the retention further feature after the registration
ATL is updated, and identity ID number corresponding to mark.
In addition, the embodiment of the present invention additionally provides the extraction of finger vein further feature and matching system based on incomplete completion,
The finger vein further feature extraction includes with matching system：
Pretreatment module：For being pre-processed to finger vein image to be registered, it is quiet to obtain pretreated finger to be registered
Arteries and veins image；
It is described that finger vein image to be registered is pre-processed, include progress region of interesting extraction processing successively, image
Normalized, histogram remap processing；
Characteristic extracting module：For using depth convolutional neural networks to the pretreated finger vein image to be registered
Further feature extraction training managing is carried out, obtains finger vein image further feature to be registered；
The characteristic extracting module includes：
Incomplete completion unit：For the pretreated finger vein image to be registered to be carried out at incomplete feature completion
Reason, obtain the finger vein image to be registered after feature completion；
First convolution pond unit：For forming the first convolution pond layer to the feature using by foundation characteristic convolution kernel
Finger vein image to be registered after completion carries out the processing of first time convolution pondization, and it is special to obtain the basis to be registered for referring to vein image
Sign；
Second convolution pond unit：For using the second convolution pond layer being made up of high-order spy's convolution kernel to finger to be registered
The foundation characteristic of vein image carries out second of convolution pondization processing, obtains finger vein image depth to be registered and hides feature；
Full connection unit：Full connection processing processing is carried out for hiding feature to the finger vein image depth to be registered,
Obtain finger vein image further feature to be registered；
ATL builds module：For being registered using the finger vein image further feature to be registered and building template
Storehouse is handled, and obtains the retention further feature ATL after registration；
Second feature extraction module：For gathering finger vein image to be identified, by the finger vein image to be identified successively
Pretreatment and feature extraction processing are carried out, obtains finger vein image further feature to be identified；
Matching module：For special using the finger vein image further feature to be identified and the retention deep layer after the registration
Levy ATL and carry out match cognization processing, obtain match cognization result.
Preferably, the hierarchical structure of the depth convolutional neural networks is 6 layers, and wherein first layer is input layer, the second layer
For incomplete completion layer, third layer be the first convolution pond layer, the 4th layer be the second convolution pond layer, layer 5 be full articulamentum,
Layer 6 is output layer.
Preferably, the incomplete completion unit includes：
Data set expands subelement：For entering to the pretreated finger vein image to be registered at line data set amplification
Reason, obtain and treat that training refers to vein image after amplification；
Repetitive exercise subelement：For will treat that training refers to vein image in the incomplete completion layer and entered after the amplification
Row iteration training managing, the finger vein image to be registered after feature completion is obtained by repetitive exercise.
Preferably, the matching module includes：
Computing unit：For by the retention further feature to be identified referred to after vein image further feature and the registration
Retention depth characteristic in ATL carries out Euclidean distance calculating one by one, obtains Euclidean distance set；
Sequencing unit：For being ranked up to the Euclidean distance in the Euclidean distance set, choose in ranking results most
Small Euclidean distance is as matching result；
Judging unit：For judging whether the matching result is less than given threshold, if so, then the matching result is knowledge
Not Cheng Gong, if it is not, then the matching result is recognition failures.
Preferably, the matching module also includes：
Updating block：After by the identification, successfully finger vein image further feature to be identified adds the registration
Retain further feature ATL in, using it is described identification successfully it is to be identified finger vein image further feature to the registration after
Further feature ATL is retained to be updated, and identity ID number corresponding to mark.
In embodiments of the present invention, the finger vein pattern of high discrimination is extracted by the embodiment of the present invention；Due to depth
The particularity of model structure, the function of referring to vein incompleteness completion can be not only played, enhancing refers to vein image, quiet by refer to
Arteries and veins incompleteness completion, it can require relatively low to referring to vein image quality, algorithm redundancy is higher；And due to proposing different convolution
Convolution kernel corresponding to the layer of pond so that the vein pattern that refers to extracted has very strong feature descriptive, improves matching precision.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the required accompanying drawing used in technology description to be briefly described, it is clear that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is to refer to the extraction of vein further feature and the method for matching process based on incomplete completion in the embodiment of the present invention
Schematic flow sheet；
Fig. 2 is the schematic flow sheet of the progress match cognization processing step in the embodiment of the present invention；
Fig. 3 is to refer to the extraction of vein further feature and matching process based on incomplete completion in another embodiment of the present invention
Method flow schematic diagram；
Fig. 4 is to refer to the extraction of vein further feature and the system of matching system based on incomplete completion in the embodiment of the present invention
Structure composition schematic diagram.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, rather than whole embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained all other under the premise of creative work is not made
Embodiment, belong to the scope of protection of the invention.
Embodiment one：
Fig. 1 is to refer to the extraction of vein further feature and the method for matching process based on incomplete completion in the embodiment of the present invention
Schematic flow sheet, as shown in figure 1, the finger vein further feature extraction includes with matching process：
S11：Finger vein image to be registered is pre-processed, obtains pretreated finger vein image to be registered；
S12：Further feature is carried out to the pretreated finger vein image to be registered using depth convolutional neural networks
Training managing is extracted, obtains finger vein image further feature to be registered；
S13：Registered using the finger vein image further feature to be registered and build ATL processing, obtain registration
Retention further feature ATL afterwards；
S14：Finger vein image to be identified is gathered, the finger vein image to be identified is carried out at S11, S12 step successively
Reason, obtain finger vein image further feature to be identified；
S15：Using the finger vein image further feature to be identified and the retention further feature ATL after the registration
Match cognization processing is carried out, obtains match cognization result.
S11 is described further：
Finger vein image to be registered is pre-processed, wherein pretreatment includes ROI extractions, (interesting image regions carry
Take), image normalization, histogram the processing such as remap, it is specific as follows：
The pretreatment includes：
Interesting image regions extraction process is carried out to the finger vein image to be registered, interesting image regions is obtained and carries
Take the finger vein image to be registered of processing；The finger vein image to be registered progress image of interesting image regions extraction process is returned
One change is handled, and obtains the finger vein image after normalized；Column hisgram weight is entered to the finger vein image after normalized
Mapping is handled, and obtains pretreated finger vein image to be registered；By above-mentioned pretreatment, finger vein to be registered can be obtained
The enhancing image of image, wherein ROI extractions realize that extraction obtains finger areas, goes unless hand using Canny edge detection algorithms
Refer to region；Image normalization is realized by bilinear interpolation method so that all finger vein images have identical size, gray value
Scope is 0-1.
S12 is described further：
A depth convolutional neural networks are built first, the hierarchical structure of the depth convolutional neural networks of structure is 6 layers, its
Middle first layer is input layer, the second layer is incomplete completion layer, third layer be the first convolution pond layer, the 4th layer be the second convolution pond
Change layer, layer 5 is full articulamentum, layer 6 is output layer.
Large area training, percentage regulation are carried out to depth convolutional neural networks by referring to the image in vein image database
The parameter of 6 layer models in study module, get the depth convolutional neural networks trained.
The input layer that pretreated finger vein image to be registered is entered to the depth convolutional neural networks by training is defeated
Enter into the depth convolutional neural networks trained, be successively input layer by first layer, the second layer be incomplete completion layer, the 3rd
Layer for the first convolution pond layer, the 4th layer be the second convolution pond layer, layer 5 is full articulamentum, layer 6 is at output layer
Reason, finger vein image further feature to be registered is got when output.
Referring to vein image has a fixed attribute, i.e., statistical property a part of in image with other parts is the same,
Another part can be also used in by referring to the feature that a portion learns in vein image, and all positions on image can be used
Identical learning characteristic；And convolution kernel can be used for extracting the features such as the line referred in vein image, angle, network, edge, therefore this
Selection is applicable different convolution kernels to extract the finger vein pattern of different levels in invention.
Need to do classification with these features after by image convolution, generally go to be matched with the feature extracted
Or training grader, but the number of the feature quantity due to acquiring in the present invention is different according to the hidden layer number of plies, it is each
The convolution property of a very higher-dimension can be all obtained after individual feature and image convolution, so causes very high-dimensional grader, and
Easily there is over-fitting；The present invention is determined to solve this problem from pond neutralizing；Why decision uses the feature after convolution
It is that an image-region useful feature is also suitable in another region because image has the attribute of a kind of " nature static " i.e..
Therefore in order to describe big image, aggregate statistics are carried out to the feature of diverse location, i.e. pondization operates.
S13 is described further：
The finger vein image further feature to be registered got is registered in the identification matching database, and identified
Subscriber identity information ID corresponding to device, after registration is good, corresponding further feature and subscriber identity information ID renewals are entered
Into the ATL built, the retention further feature ATL for obtaining the renewal after having new user's registration is finally completed.
S14 is described further：
Finger vein image to be identified is subjected to above-mentioned S11 and S12 step process successively, obtains finger vein figure to be identified
As further feature；On specific implementation process, just do not repeat herein, specifically may be referred to the detailed implementation of S11 and S12 steps
Process.
S15 is described further：
Carried out using the vein image further feature to be identified that refers to the retention further feature ATL after the registration
Match cognization processing, get finger vein further feature to be identified and retain the match cognization knot between further feature ATL
Fruit.
Specifically, Fig. 2 is the schematic flow sheet of the progress match cognization processing step in the embodiment of the present invention, such as Fig. 2 institutes
Show, it is as follows to carry out match cognization processing step：
S151：By in the retention further feature ATL after the finger vein image further feature to be identified and the registration
Retention depth characteristic carry out Euclidean distance calculating one by one, obtain Euclidean distance set；
By the finger vein image further feature to be identified and staying in the retention further feature ATL after the registration
Deposit depth characteristic and carry out Euclidean distance calculating, specific Euclidean distance calculation formula one by one：
Wherein, a represents finger vein image feature to be identified, and b represents one in the retention further feature ATL after registration
Individual retention refers to vein image feature, and i is represented in the retention further feature ATL after finger vein image feature to be identified and registration
Retention refer to the small feature of vein image further feature, i=1,2,3 ..., n；D represents to calculate the Euclidean distance obtained；According to
Above-mentioned calculating obtains Euclidean distance and builds an Euclidean distance set.
S152：Euclidean distance in the Euclidean distance set is ranked up, chooses Euclidean minimum in ranking results
Distance is used as matching result；
After above-mentioned calculating, Euclidean distance set is obtained, by carrying out one to the distance in this Euclidean distance set
Individual sequence, sort method can be depending on user's actual conditions, and the minimum Eustachian distance chosen in ranking results is tied as matching
Fruit.
S153：Judge whether the matching result is less than given threshold, if so, then the matching result is to identify successfully,
If it is not, then the matching result is recognition failures.
Pass through above-mentioned steps, it is determined that matching result in the matching process, by the Euclidean distance in matching result with setting
Fixed threshold value μ is compared, and threshold value μ is the result chosen by test of many times, chooses μ=1 in embodiments of the present invention, is ensured
Exact rate is identified, its smaller matching precision of threshold value μ is higher, and the match is successful, and rate is lower, and security is higher, and user can be according to reality
Situation sets threshold value μ, and the embodiment of the present invention is not limited；When the Euclidean distance in matching result is less than or equal to threshold value μ, then
Think to identify successfully, if the Euclidean distance in matching result is more than threshold value μ, then it is assumed that recognition failures.
After above-mentioned identification success, in addition to the retention further feature ATL after registration is updated, specifically such as
Under：
By the identification retention further feature that successfully finger vein image further feature to be identified is added after the registration
In ATL, using the successful finger vein image further feature to be identified of the identification to the retention further feature after the registration
ATL is updated, and identity ID number corresponding to mark.
In embodiments of the present invention, the finger vein pattern of high discrimination is extracted by the embodiment of the present invention；Due to depth
The particularity of model structure, the function of referring to vein incompleteness completion can be not only played, enhancing refers to vein image, quiet by refer to
Arteries and veins incompleteness completion, it can require relatively low to referring to vein image quality, algorithm redundancy is higher；And due to proposing different convolution
Convolution kernel corresponding to the layer of pond so that the vein pattern that refers to extracted has very strong feature descriptive, improves matching precision.
Embodiment two：
Fig. 3 is to refer to the extraction of vein further feature and matching process based on incomplete completion in another embodiment of the present invention
Method flow schematic diagram, as shown in figure 3, the finger vein further feature extraction includes with matching process：
S21：Finger vein image to be registered is pre-processed, obtains pretreated finger vein image to be registered；
Specific implementation process refers to embodiment one, will not be repeated here.
S22：The pretreated finger vein image to be registered is subjected to incomplete feature completion processing, obtains feature completion
Finger vein image to be registered afterwards；
Line data set amplification processing is entered to the pretreated finger vein image to be registered, obtains and waits to train after amplification
Refer to vein image；It will treat that training refers to vein image in the incomplete completion layer and is iterated training managing after the amplification,
Finger vein image to be registered after feature completion is obtained by repetitive exercise.
Specifically, enter line data set amplification；On the basis of pretreated data set is passed through, it is quiet to randomly select finger
Pixel corresponding to any 1/6 part of arteries and veins image is labeled as non-vein area, plays the increased effect of sample set, and every width refers to vein
Image amplification is 10 width images.Image mends layer in incompleteness and passes through a large amount of repetitive exercises, the model for training to obtain can realize refer to it is quiet
Arteries and veins incompleteness polishing.
S23：The first convolution pond layer is formed to the finger to be registered after the feature completion using by foundation characteristic convolution kernel
Vein image carries out the processing of first time convolution pondization, obtains the foundation characteristic to be registered for referring to vein image；
Possess 15 convolution kernels in first convolution pond layer, for detecting the primary features such as lines, angle, in the first convolution pond
Change layer and mainly realize the step of convolution kernel pondization two, refer to the foundation characteristic (foundation characteristic of vein image by being extracted after this layer
For features such as lines, side, angles).
Convolution：
In this layer, image size is 40x80, and the area size of selection is 10x10, and convolution pattern uses valid, i.e.,
Remember input matrix：W × H, convolution kernel F × F, step-length stride, high, a width of new_height, new_width are exported, then：
New_width=(W-F+1)/S
New_height=(H-F+1)/S
Rounded up after result is calculated；Herein, input picture 40x80, step-length 1, it is ensured that input and output
It is consistent big.After each image has carried out convolution, it is saved in matrix and carries out subsequent arithmetic.
Chi Hua：
Often translated due to referring to venous collection image, and pondization operation shows well in terms of image translation, i.e.,
When image have passed through translation, image tagged, which is appointed, so keeps constant, therefore the dimensionality reduction with pond.
Processing procedure in the present invention is：After getting convolution feature, pond area size m*n is determined, then convolution
Feature is divided on the disjoint range that several sizes are m*n, after then obtaining pond with the average characteristics in these regions
Feature after convolution feature, most terminal cistern is used for being classified；First layer selection pond area size is 4x4, after the second layer
It is 2x2 to select size.
S24：Using special to the basis of finger vein image to be registered by the second convolution pond layer that high-order spy's convolution kernel forms
Sign carries out second of convolution pondization processing, obtains finger vein image depth to be registered and hides feature；
Convolution kernel in second convolution pond layer is the high-order spy's convolution kernel group obtained by own coding neural network learning
Into；After this layer, the advanced features for treating that training refers to vein image are extracted, the method for obtaining convolution kernel is as follows：
An own coding neural network is defined, whole own coding neutral net is divided into three layers, input layer, hidden layer, exported
Layer；Make output layer equal with input layer, be by pretreated finger vein image, by limiting the number of concealed nodes,
Can be so that from the feature for acquiring fine discrimination, the high-order spy convolution kernel finally learnt forms the volume in the second convolution pond layer
Product core.
S25：Feature is hidden to the finger vein image depth to be registered and carries out full connection processing processing, obtains finger to be registered
Vein image further feature；
Connected entirely by full articulamentum, the size of full articulamentum is 128*1.
S26：Registered using the finger vein image further feature to be registered and build ATL processing, obtain registration
Retention further feature ATL afterwards；
Specific implementation process refers to embodiment one, will not be repeated here.
S27：Finger vein image to be identified is gathered, the finger vein image to be identified is carried out at S21-S25 steps successively
Reason, obtain finger vein image further feature to be identified；
Specific implementation process refers to embodiment one, will not be repeated here.
S28：Using the finger vein image further feature to be identified and the retention further feature ATL after the registration
Match cognization processing is carried out, obtains match cognization result.
Specific implementation process refers to embodiment one, will not be repeated here.
Embodiment three：
Fig. 4 is to refer to the extraction of vein further feature and the system of matching system based on incomplete completion in the embodiment of the present invention
Structure composition schematic diagram, as shown in figure 4, the finger vein further feature extraction includes with matching system：
Pretreatment module 11：For being pre-processed to finger vein image to be registered, pretreated finger to be registered is obtained
Vein image；
It is described that finger vein image to be registered is pre-processed, include progress region of interesting extraction processing successively, image
Normalized, histogram remap processing；
Characteristic extracting module 12：For using depth convolutional neural networks to the pretreated finger vein figure to be registered
As carrying out further feature extraction training managing, finger vein image further feature to be registered is obtained；
The characteristic extracting module 12 includes：
Incomplete completion unit 121：For the pretreated finger vein image to be registered to be carried out into incomplete feature completion
Processing, obtain the finger vein image to be registered after feature completion；
First convolution pond unit 122：For forming the first convolution pond layer to described using by foundation characteristic convolution kernel
Finger vein image to be registered after feature completion carries out the processing of first time convolution pondization, obtains the basis to be registered for referring to vein image
Feature；
Second convolution pond unit 123：For treating note using the second convolution pond layer being made up of high-order spy's convolution kernel
The foundation characteristic that volume refers to vein image carries out second of convolution pondization processing, obtains finger vein image depth to be registered and hides spy
Sign；
Full connection unit 124：Carried out for hiding feature to the finger vein image depth to be registered at full connection processing
Reason, obtain finger vein image further feature to be registered；
ATL builds module 13：For being registered using the finger vein image further feature to be registered and building mould
Plate storehouse is handled, and obtains the retention further feature ATL after registration；
Second feature extraction module 14：For gathering finger vein image to be identified, by it is described it is to be identified finger vein image according to
It is secondary to carry out pretreatment and feature extraction processing, obtain finger vein image further feature to be identified；
Matching module 15：For using the finger vein image further feature to be identified and the retention deep layer after the registration
Feature templates storehouse carries out match cognization processing, obtains match cognization result.
Preferably, the hierarchical structure of the depth convolutional neural networks is 6 layers, and wherein first layer is input layer, the second layer
For incomplete completion layer, third layer be the first convolution pond layer, the 4th layer be the second convolution pond layer, layer 5 be full articulamentum,
Layer 6 is output layer.
Preferably, the incomplete completion unit 121 includes：
Data set expands subelement：For entering to the pretreated finger vein image to be registered at line data set amplification
Reason, obtain and treat that training refers to vein image after amplification；
Repetitive exercise subelement：For will treat that training refers to vein image in the incomplete completion layer and entered after the amplification
Row iteration training managing, the finger vein image to be registered after feature completion is obtained by repetitive exercise.
Preferably, the matching module 15 includes：
Computing unit：For by the retention further feature to be identified referred to after vein image further feature and the registration
Retention depth characteristic in ATL carries out Euclidean distance calculating one by one, obtains Euclidean distance set；
Sequencing unit：For being ranked up to the Euclidean distance in the Euclidean distance set, choose in ranking results most
Small Euclidean distance is as matching result；
Judging unit：For judging whether the matching result is less than given threshold, if so, then the matching result is knowledge
Not Cheng Gong, if it is not, then the matching result is recognition failures.
Preferably, the matching module 15 also includes：
Updating block：After by the identification, successfully finger vein image further feature to be identified adds the registration
Retain further feature ATL in, using it is described identification successfully it is to be identified finger vein image further feature to the registration after
Further feature ATL is retained to be updated, and identity ID number corresponding to mark.
Specifically, the operation principle of the system related functions module of the embodiment of the present invention can be found in the correlation of embodiment of the method
Description, is repeated no more here.
In embodiments of the present invention, the finger vein pattern of high discrimination is extracted by the embodiment of the present invention；Due to depth
The particularity of model structure, the function of referring to vein incompleteness completion can be not only played, enhancing refers to vein image, quiet by refer to
Arteries and veins incompleteness completion, it can require relatively low to referring to vein image quality, algorithm redundancy is higher；And due to proposing different convolution
Convolution kernel corresponding to the layer of pond so that the vein pattern that refers to extracted has very strong feature descriptive, improves matching precision.
One of ordinary skill in the art will appreciate that all or part of step in the various methods of above-described embodiment is can
To instruct the hardware of correlation to complete by program, the program can be stored in a computer-readable recording medium, storage
Medium can include：Read-only storage (ROM, Read Only Memory), random access memory (RAM, Random
Access Memory), disk or CD etc..
In addition, the finger vein further feature based on incomplete completion provided above the embodiment of the present invention is extracted with matching
Method and system are described in detail, and should employ specific case herein and the principle and embodiment of the present invention are carried out
Illustrate, the explanation of above example is only intended to help the method and its core concept for understanding the present invention；Meanwhile for this area
Those skilled in the art, according to the thought of the present invention, there will be changes in specific embodiments and applications, to sum up
Described, this specification content should not be construed as limiting the invention.
Claims (10)
1. the extraction of finger vein further feature and matching process based on incomplete completion, it is characterised in that the finger vein deep layer is special
Sign extraction includes with matching process：
S11：Finger vein image to be registered is pre-processed, obtains pretreated finger vein image to be registered；
It is described that finger vein image to be registered is pre-processed, include progress region of interesting extraction processing successively, image normalizing
Change is handled, and histogram remaps processing；
S12：Further feature extraction is carried out to the pretreated finger vein image to be registered using depth convolutional neural networks
Training managing, obtain finger vein image further feature to be registered；
It is described that further feature extraction is carried out to the pretreated finger vein image to be registered using depth convolutional neural networks
Training managing, detailed process are as follows：
1) the pretreated finger vein image to be registered is carried out into incomplete feature completion to handle, obtains and treated after feature completion
Registration refers to vein image；
2) the first convolution pond layer is formed to the finger vein figure to be registered after the feature completion using by foundation characteristic convolution kernel
As carrying out the processing of first time convolution pondization, the foundation characteristic to be registered for referring to vein image is obtained；
3) foundation characteristic of finger vein image to be registered is carried out using by the second convolution pond layer that high-order spy's convolution kernel forms
Second of convolution pondization processing, obtain finger vein image depth to be registered and hide feature；
4) feature is hidden to the finger vein image depth to be registered and carries out full connection processing processing, obtain finger vein figure to be registered
As further feature；
S13：Registered using the finger vein image further feature to be registered and build ATL processing, after obtaining registration
Retain further feature ATL；
S14：Finger vein image to be identified is gathered, the finger vein image to be identified is subjected to S11, S12 step process successively, obtained
Take finger vein image further feature to be identified；
S15：Carried out using the vein image further feature to be identified that refers to the retention further feature ATL after the registration
Match cognization processing, obtains match cognization result.
2. the extraction of finger vein further feature and matching process, its feature according to claim 1 based on incomplete completion exist
In the hierarchical structure of, depth convolutional neural networks be 6 layers, wherein first layer be input layer, the second layer be incomplete completion layer,
Third layer be the first convolution pond layer, the 4th layer be the second convolution pond layer, layer 5 is full articulamentum, layer 6 is output
Layer.
3. the extraction of finger vein further feature and matching process, its feature according to claim 1 based on incomplete completion exist
In, it is described that the pretreated finger vein image to be registered is subjected to incomplete feature completion processing, including：
Line data set amplification processing is entered to the pretreated finger vein image to be registered, obtain after amplification to treat that training refers to quiet
Arteries and veins image；
It will treat that training refers to vein image in the incomplete completion layer and is iterated training managing after the amplification, pass through iteration
Training obtains the finger vein image to be registered after feature completion.
4. the extraction of finger vein further feature and matching process, its feature according to claim 1 based on incomplete completion exist
In described using the finger vein image further feature to be identified and the retention further feature ATL progress after the registration
With identifying processing, including：
The finger vein image further feature to be identified and the retention in the retention further feature ATL after the registration is deep
Degree feature carries out Euclidean distance calculating one by one, obtains Euclidean distance set；
Euclidean distance in the Euclidean distance set is ranked up, chooses Euclidean distance conduct minimum in ranking results
With result；
Judge whether the matching result is less than given threshold, if so, then the matching result is identifies successfully, if it is not, then institute
It is recognition failures to state matching result.
5. the extraction of finger vein further feature and matching process, its feature according to claim 4 based on incomplete completion exist
In described using the finger vein image further feature to be identified and the retention further feature ATL progress after the registration
Also include with identifying processing：
By the identification retention further feature template that successfully finger vein image further feature to be identified is added after the registration
In storehouse, using the successful finger vein image further feature to be identified of the identification to the retention further feature template after the registration
Storehouse is updated, and identity ID number corresponding to mark.
6. the extraction of finger vein further feature and matching system based on incomplete completion, it is characterised in that the finger vein deep layer is special
Sign extraction includes with matching system：
Pretreatment module：For being pre-processed to finger vein image to be registered, pretreated finger vein figure to be registered is obtained
Picture；
It is described that finger vein image to be registered is pre-processed, include progress region of interesting extraction processing successively, image normalizing
Change is handled, and histogram remaps processing；
Characteristic extracting module：For being carried out using depth convolutional neural networks to the pretreated finger vein image to be registered
Further feature extracts training managing, obtains finger vein image further feature to be registered；
The characteristic extracting module includes：
Incomplete completion unit：Handle, obtain for the pretreated finger vein image to be registered to be carried out into incomplete feature completion
Take the finger vein image to be registered after feature completion；
First convolution pond unit：For forming the first convolution pond layer to the feature completion using by foundation characteristic convolution kernel
Finger vein image to be registered afterwards carries out the processing of first time convolution pondization, obtains the foundation characteristic to be registered for referring to vein image；
Second convolution pond unit：For using the second convolution pond layer being made up of high-order spy's convolution kernel to finger vein to be registered
The foundation characteristic of image carries out second of convolution pondization processing, obtains finger vein image depth to be registered and hides feature；
Full connection unit：Full connection processing processing is carried out for hiding feature to the finger vein image depth to be registered, is obtained
Finger vein image further feature to be registered；
ATL builds module：For being registered and being built at ATL using the finger vein image further feature to be registered
Reason, obtain the retention further feature ATL after registration；
Second feature extraction module：For gathering finger vein image to be identified, the finger vein image to be identified is carried out successively
Pretreatment and feature extraction processing, obtain finger vein image further feature to be identified；
Matching module：For using the finger vein image further feature to be identified and the retention further feature mould after the registration
Plate storehouse carries out match cognization processing, obtains match cognization result.
7. the extraction of finger vein further feature and matching system, its feature according to claim 6 based on incomplete completion exist
In the hierarchical structure of, depth convolutional neural networks be 6 layers, wherein first layer be input layer, the second layer be incomplete completion layer,
Third layer be the first convolution pond layer, the 4th layer be the second convolution pond layer, layer 5 is full articulamentum, layer 6 is output
Layer.
8. the extraction of finger vein further feature and matching system, its feature according to claim 6 based on incomplete completion exist
In the incomplete completion unit includes：
Data set expands subelement：For entering line data set amplification processing to the pretreated finger vein image to be registered,
Obtain and treat that training refers to vein image after amplification；
Repetitive exercise subelement：For will treat that training refers to vein image in the incomplete completion layer and changed after the amplification
For training managing, the finger vein image to be registered after feature completion is obtained by repetitive exercise.
9. the extraction of finger vein further feature and matching system, its feature according to claim 6 based on incomplete completion exist
In the matching module includes：
Computing unit：For by the retention further feature template to be identified referred to after vein image further feature and the registration
Retention depth characteristic in storehouse carries out Euclidean distance calculating one by one, obtains Euclidean distance set；
Sequencing unit：For being ranked up to the Euclidean distance in the Euclidean distance set, choose minimum in ranking results
Euclidean distance is as matching result；
Judging unit：For judging whether the matching result is less than given threshold, if so, then the matching result is to be identified as
Work(, if it is not, then the matching result is recognition failures.
10. the extraction of finger vein further feature and matching system, its feature according to claim 9 based on incomplete completion exist
In the matching module also includes：
Updating block：For the retention for adding the identification successful finger vein image further feature to be identified after the registration
In further feature ATL, using the successful finger vein image further feature to be identified of the identification to the retention after the registration
Further feature ATL is updated, and identity ID number corresponding to mark.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710914184.6A CN107704822A (en) | 2017-09-30 | 2017-09-30 | The extraction of finger vein further feature and matching process and system based on incomplete completion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710914184.6A CN107704822A (en) | 2017-09-30 | 2017-09-30 | The extraction of finger vein further feature and matching process and system based on incomplete completion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107704822A true CN107704822A (en) | 2018-02-16 |
Family
ID=61174717
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710914184.6A Pending CN107704822A (en) | 2017-09-30 | 2017-09-30 | The extraction of finger vein further feature and matching process and system based on incomplete completion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107704822A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109190566A (en) * | 2018-09-10 | 2019-01-11 | 中国民航大学 | A kind of fusion local code and CNN model finger vein identification method |
| CN109740569A (en) * | 2019-01-21 | 2019-05-10 | 中国民航大学 | Garbled information restored method when finger vena based on coder-decoder is imaged |
| CN110555382A (en) * | 2019-07-31 | 2019-12-10 | 浙江理工大学 | Finger vein identification method based on deep learning and Wasserstein distance measurement |
| CN113139999A (en) * | 2021-05-14 | 2021-07-20 | 广东工业大学 | Transparent object single-view multi-scale depth estimation method and system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020028004A1 (en) * | 2000-09-06 | 2002-03-07 | Naoto Miura | Personal identification device and method |
| CN103729640A (en) * | 2013-12-24 | 2014-04-16 | 小米科技有限责任公司 | Finger vein feature extraction method and device and terminal |
| US20150186634A1 (en) * | 2013-12-31 | 2015-07-02 | Digit Security, LLC | Biometric access system |
| CN105975974A (en) * | 2016-05-10 | 2016-09-28 | 深圳市金脉智能识别科技有限公司 | ROI image extraction method in finger vein identification |
| CN106127151A (en) * | 2016-06-23 | 2016-11-16 | 北京海鑫科金高科技股份有限公司 | Based on the finger vein identification method and the device that improve local binary model |
| CN106529468A (en) * | 2016-11-07 | 2017-03-22 | 重庆工商大学 | Finger vein identification method and system based on convolutional neural network |
-
2017
- 2017-09-30 CN CN201710914184.6A patent/CN107704822A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020028004A1 (en) * | 2000-09-06 | 2002-03-07 | Naoto Miura | Personal identification device and method |
| CN103729640A (en) * | 2013-12-24 | 2014-04-16 | 小米科技有限责任公司 | Finger vein feature extraction method and device and terminal |
| US20150186634A1 (en) * | 2013-12-31 | 2015-07-02 | Digit Security, LLC | Biometric access system |
| CN105975974A (en) * | 2016-05-10 | 2016-09-28 | 深圳市金脉智能识别科技有限公司 | ROI image extraction method in finger vein identification |
| CN106127151A (en) * | 2016-06-23 | 2016-11-16 | 北京海鑫科金高科技股份有限公司 | Based on the finger vein identification method and the device that improve local binary model |
| CN106529468A (en) * | 2016-11-07 | 2017-03-22 | 重庆工商大学 | Finger vein identification method and system based on convolutional neural network |
Non-Patent Citations (3)
| Title |
|---|
| HUAFENG QIN 等: "Deep Representation-Based Feature Extraction and Recovering for Finger-Vein Verification", 《IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY》 * |
| YIHUA SHI 等: "Image restoration and enhancement for finger-vein recognition", 《2012 IEEE 11TH INTERNATIONAL CONFERENCE ON SIGNAL PROCESSING》 * |
| 王科俊 等: "使用改进的方向滤波与修正的Hausdorff距离的指静脉识别方法", 《计算机辅助设计与图形学学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109190566A (en) * | 2018-09-10 | 2019-01-11 | 中国民航大学 | A kind of fusion local code and CNN model finger vein identification method |
| CN109190566B (en) * | 2018-09-10 | 2021-09-14 | 中国民航大学 | Finger vein recognition method integrating local coding and CNN model |
| CN109740569A (en) * | 2019-01-21 | 2019-05-10 | 中国民航大学 | Garbled information restored method when finger vena based on coder-decoder is imaged |
| CN109740569B (en) * | 2019-01-21 | 2022-12-27 | 中国民航大学 | Incomplete information restoration method during finger vein imaging based on encoder-decoder |
| CN110555382A (en) * | 2019-07-31 | 2019-12-10 | 浙江理工大学 | Finger vein identification method based on deep learning and Wasserstein distance measurement |
| CN113139999A (en) * | 2021-05-14 | 2021-07-20 | 广东工业大学 | Transparent object single-view multi-scale depth estimation method and system |
| CN113139999B (en) * | 2021-05-14 | 2023-04-07 | 广东工业大学 | Transparent object single-view multi-scale depth estimation method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wu et al. | Image copy-move forgery detection via an end-to-end deep neural network | |
| You et al. | A Siamese CNN for image steganalysis | |
| Jeon et al. | Plant leaf recognition using a convolution neural network | |
| Elaskily et al. | A novel deep learning framework for copy-moveforgery detection in images | |
| Zahisham et al. | Food recognition with resnet-50 | |
| CN107704822A (en) | The extraction of finger vein further feature and matching process and system based on incomplete completion | |
| Yu et al. | Leaf spot attention network for apple leaf disease identification | |
| Salman et al. | Classification of real and fake human faces using deep learning | |
| Zhao et al. | High‐quality retinal vessel segmentation using generative adversarial network with a large receptive field | |
| Ahmed et al. | Automated weed classification with local pattern-based texture descriptors. | |
| CN107967442A (en) | A kind of finger vein identification method and system based on unsupervised learning and deep layer network | |
| Bouzouina et al. | Multimodal biometric: Iris and face recognition based on feature selection of iris with GA and scores level fusion with SVM | |
| Bhatt et al. | Evolutionary granular approach for recognizing faces altered due to plastic surgery | |
| CN109871905A (en) | A kind of plant leaf identification method based on attention mechanism depth model | |
| Manoharan | Flawless detection of herbal plant leaf by machine learning classifier through two stage authentication procedure | |
| Gragnaniello et al. | Biometric spoofing detection by a domain-aware convolutional neural network | |
| Kanda et al. | A deep learning-based recognition technique for plant leaf classification | |
| Zhang et al. | HCNN: A neural network model for combining local and global features towards human-like classification | |
| da Silva et al. | Improved texture image classification through the use of a corrosion-inspired cellular automaton | |
| Uma et al. | Copy-move forgery detection of digital images using football game optimization | |
| Narvaez et al. | Painting authorship and forgery detection challenges with AI image generation algorithms: Rembrandt and 17th century Dutch painters as a case study | |
| CN111127407A (en) | Fourier transform-based style migration counterfeit image detection device and method | |
| Diqi et al. | Implementation of CNN for plant leaf classification | |
| Zulkifli | Plant leaf identification using moment invariants & general regression neural network | |
| De Vries | Object recognition: a shape-based approach using artificial neural networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180216 |