CN107967442A - A kind of finger vein identification method and system based on unsupervised learning and deep layer network - Google Patents

A kind of finger vein identification method and system based on unsupervised learning and deep layer network Download PDF

Info

Publication number
CN107967442A
CN107967442A CN201710915095.3A CN201710915095A CN107967442A CN 107967442 A CN107967442 A CN 107967442A CN 201710915095 A CN201710915095 A CN 201710915095A CN 107967442 A CN107967442 A CN 107967442A
Authority
CN
China
Prior art keywords
image
vein image
registered
finger
finger vein
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
Application number
CN201710915095.3A
Other languages
Chinese (zh)
Inventor
胡建国
王金鹏
王德明
丁颜玉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangzhou Smart City Development Research Institute
Original Assignee
Guangzhou Smart City Development Research Institute
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangzhou Smart City Development Research Institute filed Critical Guangzhou Smart City Development Research Institute
Priority to CN201710915095.3A priority Critical patent/CN107967442A/en
Publication of CN107967442A publication Critical patent/CN107967442A/en
Pending legal-status Critical Current

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
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • 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

Abstract

The invention discloses a kind of finger vein identification method and system based on unsupervised learning and deep layer network, its middle finger vein identification method includes：Finger vein image to be registered is pre-processed successively, obtains pretreated finger vein image to be registered；Feature extraction processing to be registered is carried out to pretreated finger vein image to be registered using trained deep neural network model, obtains the further feature to be registered for referring to vein image；And build the retention further feature template library after registration；Gather finger vein image to be identified and finger vein image to be identified is subjected to pretreatment and further feature processing successively, obtain finger vein image further feature to be identified；Classification and matching identifying processing is carried out to finger vein image further feature to be identified and the retention further feature template library after registration using Softmax graders, obtains match cognization result.In embodiments of the present invention, the finger vein further feature of deeper can be extracted, and match cognization efficiency is faster, more accurately.

Description

A kind of finger vein identification method and system based on unsupervised learning and deep layer network
Technical field
The present invention relates to biometrics identification technology field, more particularly to one kind to be based on unsupervised learning and deep layer network Finger vein identification method and system.
Background technology
The research and application of automatic identity authentication technology based on physiological characteristic and behavioural characteristic are increasingly extensive；Current society It can cause the requirement that higher is proposed to biological identification technology for the demand of high safety and more friendly authentication；And refer to Vein has live body and uniqueness, will not produce feature repetition situation and permission is non-contact, therefore become field of biological recognition Inside more concerned one kind.
Current finger vein recognition system or method are mostly based on knowledge in field, are related to image enhancement, filtering etc.； Some scholars propose to refer to vein authentication based on traditional neural fusion also have new researched and proposed based on self-study Habit realizes the schemes such as authentication.But there are a large amount of critical problems to need to solve referring to hand vein recognition；It is urgently to be resolved hurrily at present The problem of have：Have descriptive feature to be difficult to extract, recognition success rate is dependent on finger vein image image quality etc..
Existing technical solution one is proposed using the neural network classifier based on probability statistics and Radon features come real Now identify certification；Author proposes new finger vein pattern i.e. Radon features, completes classification and matching using neutral net, passes through Experiment achieves good effect on self-built database；In technical solution one in model training database sample number Very little, the feature extracted is difficult that determine whether can be in the especially data set containing much noise on other data set to amount Upper performance is good；In addition the operational data in scheme one is not disclosed, and is simply tested in self-built database, on Radon The validity of feature lacks persuasion, it is likely that the simply over-fitting under single model.
Existing technical solution two proposes based on the wide line tracker and refers to the normalized fin- ger vein authentication method of vein； Author is constantly detected using line detector to referring to vein image, refers to the change of vein cross section curvature by analysis to determine Direction is followed the trail of, finger areas is determined using least square method, better effects is achieved on disclosed data set, and apply Among actual, data set sample is abundant；Method based on existing curve tracing in technical solution two is improved, Finger is projected, although matching process precision is higher, author is to carry out chasing along curve based on the hypothesis of oneself so that The migration of model is not good enough, and author simply carries out finger areas extraction with simple least square method in addition, extracts Finger-image effect is very undesirable.
The content of the invention
It is an object of the invention to overcome the deficiencies in the prior art, the present invention provides one kind based on unsupervised learning and The finger vein identification method and system of deep layer network, can extract the finger vein further feature of deeper, and match cognization Efficiency faster, more accurately.
In order to solve the above-mentioned technical problem, an embodiment of the present invention provides one kind to be based on unsupervised learning and deep layer network Finger vein identification method, it is described finger vein identification method include：
S11：Finger areas extraction, data normalization and image enhancement are carried out successively to finger vein image to be registered in advance to locate Reason, obtains pretreated finger vein image to be registered；
S12：The pretreated finger vein image to be registered is carried out using trained deep neural network model Feature extraction processing to be registered, obtains the further feature to be registered for referring to vein image；
S13：Registered according to the further feature to be registered for referring to vein image and build template library processing, obtained Retention further feature template library after registration；
S14：Finger vein image to be identified is gathered to carry out the finger vein image to be identified at S11, S12 step successively Reason, obtains finger vein image further feature to be identified；
S15：Using Softmax graders to the finger vein image further feature to be identified and staying after the registration Deposit further feature template library and carry out classification and matching identifying processing, obtain match cognization result.
Preferably, it is described to finger vein image progress finger areas extraction pretreatment to be registered, including：
To finger vein image to be registered first using Sobel edge detection operators extraction edge, then carry out profile and carry Take, extract the profile to be registered for referring to vein image, then the ductility and flatness of the profile according to finger vein image to be registered, The branch for not meeting finger extension direction is removed, and the profile of incomplete finger vein image to be registered is extended, acquisition is prolonged Long image is denoted as x1；
Edge in Prewitt operator detection images is used first to finger vein figure to be registered, is changed into bianry image, then By the finger beginning and end position on image both sides, it is baseline to take its both ends midpoint line, selects least square method to centre Line is fitted, and then takes the point set of both sides same distance to form finger contours, is obtained the image among profile and is denoted as x2；
The image being denoted as using extension image among x1 and profile is denoted as x2 and is overlapped, and obtains complete finger to be registered The area image of vein image is denoted as x3.
Preferably, it is described that data normalization and image enhancement pretreatment are carried out to finger vein image to be registered, including：
The complete area image to be registered for referring to vein image is subjected to size normalization and standardization, obtains ruler The area image of very little normalization and standardization；
The size is normalized using Histogram Mapping and the area image of standardization is carried out at contrast enhancing Reason, obtains the area image after image enhancement；
Wherein, the size normalization is realized using cubic interpolation, is changed into 0 to all area image averages, variance becomes For 1；The mapping coefficient of the Histogram Mapping is obtained by genetic algorithm.
Preferably, the deep neural network model training process, including：
Build initial depth neural network model；
Finger vein image to be trained at least is divided into three set, the first subclass, yield in the second subset is closed and the 3rd subset Close；
Respectively to treating that training refers to vein image in first subclass, yield in the second subset conjunction and the 3rd subclass Finger areas extraction, standardization and image enhancement are carried out successively to pre-process, and obtain pretreated first subclass, the second son Set and the 3rd subclass；
Self study feature extraction processing is carried out to treating that training refers to vein image in pretreated first subclass, Obtain the alignment features for treating that training refers to vein image in first subclass；
Using treated in first subclass training refer to vein image alignment features and it is described it is pretreated second son Treat that training refers to vein image and processing is trained to initial depth neural network model in set, obtain the depth god after training Through network model；
Refer to vein image to the depth nerve after the training after training using in pretreated 3rd subclass Network model is tested, and obtains test result；
Judge that the exact discrimination identification of the test result is less than predetermined threshold value, if so, the then depth after the training Neural network model is trained deep neural network model, continues to train if it is not, then returning to training step.
Preferably, it is described to carry out self study spy to treating that training refers to vein image in pretreated first subclass Extraction process is levied, including：
Three layers of self-editing neural network model, including input layer, hidden layer and output layer are built, and limits the hidden layer Hidden unit number be 50；
By in first subclass treat training refer to vein image by the input layer input in the hidden layer into After row iteration training, treating that training refers to the different position and direction of vein image and carry out edge detection using different hidden units, Obtain in first subclass and treat that training refers to vein image alignment features；
It will treat that training refers to vein image alignment features and exported by the output layer in first subclass.
Preferably, the initial depth neural network model includes input layer, the first convolution pond layer, volume Two successively Product pond layer, the 3rd convolution pond layer, full articulamentum and output layer；
The convolution kernel size of the first convolution pond layer is 3x3, and pond core size is 2x2, step-length 1；
The convolution kernel size of the second convolution pond layer is 3x3, and pond core size is 2x2, step-length 1；
The convolution kernel size of 3rd convolution pond layer is 3x3, and pond core size is 2x2, step-length 1；
The full connection size of the full articulamentum is 210。
Preferably, the finger vein identification method further includes：
Judge whether the match cognization result is less than predetermined threshold value, if then identifying successfully, if otherwise recognition failures；
If identify successfully, after the identification successfully the finger vein image further feature addition to be identified registration Retain further feature template library in, using it is described identification successfully it is to be identified finger vein image further feature to the registration after Retention further feature template library be updated, and identify corresponding identity ID number.
In addition, the embodiment of the present invention additionally provides a kind of finger hand vein recognition system based on unsupervised learning and deep layer network System, the finger vein recognition system include：
Pretreatment module：For carrying out finger areas extraction, data normalization and figure successively to finger vein image to be registered Image intensifying pre-processes, and obtains pretreated finger vein image to be registered；
Fisrt feature extraction module：For pretreated being treated to described using trained deep neural network model Registration refers to vein image and carries out feature extraction processing to be registered, obtains the further feature to be registered for referring to vein image；
Template library builds module：For being registered and being built according to the further feature to be registered for referring to vein image Template library processing, obtains the retention further feature template library after registration；
Second feature extraction module：For gather it is to be identified finger vein image will it is described it is to be identified refer to vein image successively Pretreatment and feature extraction processing step are carried out, obtains finger vein image further feature to be identified；
Classification and Identification module：For using Softmax graders to the finger vein image further feature to be identified and institute State the retention further feature template library after registration and carry out classification and matching identifying processing, obtain match cognization result.
Preferably, the pretreatment module includes area extracting unit, data normalization and image enhancing unit；
The area extracting unit includes：
The first profile extracts subelement：For being carried first using Sobel edge detection operators to finger vein image to be registered Edge is taken, then carries out contours extract, extracts the profile to be registered for referring to vein image, then according to finger vein image to be registered The ductility and flatness of profile, remove the branch for not meeting finger extension direction, and to incomplete finger vein image to be registered Profile extended, obtain extend image be denoted as x1；
Second contours extract subelement：For being used first in Prewitt operator detection images to finger vein figure to be registered Edge, is changed into bianry image, and then by the finger beginning and end position on image both sides, it is base to take its both ends midpoint line Line, selects least square method to be fitted medium line, then takes the point set of both sides same distance to form finger contours, obtains Image among profile is denoted as x2
It is superimposed subelement：Image for being denoted as using extension image among x1 and profile is denoted as x2 and is overlapped, and is obtained The complete area image to be registered for referring to vein image is denoted as x3；
Data normalization includes with image enhancing unit：
Data normalization subelement：For the complete area image to be registered for referring to vein image to be carried out size normalizing Change and standardization, acquisition size normalize the area image with standardization；
Image enhancement subelement：For being normalized using Histogram Mapping to the size and the region of standardization Image carries out contrast enhancement processing, obtains the area image after image enhancement；
Wherein, the size normalization is realized using cubic interpolation, is changed into 0 to all area image averages, variance becomes For 1；The mapping coefficient of the Histogram Mapping is obtained by genetic algorithm.
Preferably, the finger vein recognition system further includes：
Judgment module：For judging whether the match cognization result is less than predetermined threshold value, if then identifying successfully, if Otherwise recognition failures；
Update module：If for identifying that successfully, successfully finger vein image further feature to be identified adds by the identification In retention further feature template library after the registration, refer to vein image further feature using the identification is successfully to be identified Retention further feature template library after the registration is updated, and identifies corresponding identity ID number.
In embodiments of the present invention, can only be extracted when referring to vein image feature extraction using the embodiment of the present invention Include crucial finger vein image further feature；Preprocessing part propose a kind of new finger areas extracting method and Histogram Mapping is carried out by genetic algorithm, can be preferably to the carry out extracted region of finger vein image；For with description Property finger vein pattern be difficult to extract and find, the embodiment of the present invention employs self study feature extraction mode, using training number Part sample in carries out self study to sparse autocoder, effectively improves subsequently to deep neural network model Training effect so that deep neural network model can fast and accurately extract the further feature for referring to vein image；Follow-up Returned using Softmax and carry out quick and precisely identification matching.
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 to embodiment or Attached drawing needed to be used in the description of the prior art is briefly described, it is clear that, drawings in the following description are only Some embodiments of the present invention, for those of ordinary skill in the art, without creative efforts, also Other attached drawings can be obtained according to these attached drawings.
Fig. 1 is the flow of the finger vein identification method based on unsupervised learning and deep layer network in the embodiment of the present invention Schematic diagram；
Fig. 2 is the finger vein identification method based on unsupervised learning and deep layer network in another embodiment of the present invention Flow diagram；
Fig. 3 is the system of the finger vein recognition system based on unsupervised learning and deep layer network in the embodiment of the present invention Structure composition schematic diagram；
Fig. 4 is the finger vein recognition system based on unsupervised learning and deep layer network in another embodiment of the present invention System structure composition schematic diagram.
Embodiment
Below in conjunction with the attached drawing in the embodiment of the present invention, the technical solution 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, instead of all the embodiments.It is based on Embodiment in the present invention, those of ordinary skill in the art obtained without creative efforts it is all its Its embodiment, belongs to the scope of protection of the invention.
Embodiment one：
Fig. 1 is the flow of the finger vein identification method based on unsupervised learning and deep layer network in the embodiment of the present invention Schematic diagram, the finger vein identification method include：
S11：Finger areas extraction, data normalization and image enhancement are carried out successively to finger vein image to be registered in advance to locate Reason, obtains pretreated finger vein image to be registered；
S12：The pretreated finger vein image to be registered is carried out using trained deep neural network model Feature extraction processing to be registered, obtains the further feature to be registered for referring to vein image；
S13：Registered according to the further feature to be registered for referring to vein image and build template library processing, obtained Retention further feature template library after registration；
S14：Finger vein image to be identified is gathered to carry out the finger vein image to be identified at S11, S12 step successively Reason, obtains finger vein image further feature to be identified；
S15：Using Softmax graders to the finger vein image further feature to be identified and staying after the registration Deposit further feature template library and carry out classification and matching identifying processing, obtain match cognization result.
S11 is described further：
Finger areas extraction pretreatment and data normalization are carried out successively to finger vein image to be registered and image enhancement is pre- Processing.
It is as follows that finger areas extraction preprocessing process is carried out to finger vein image to be registered：
To finger vein image to be registered first using Sobel edge detection operators extraction edge, then carry out profile and carry Take, extract the profile to be registered for referring to vein image, then the ductility and flatness of the profile according to finger vein image to be registered, The branch for not meeting finger extension direction is removed, and the profile of incomplete finger vein image to be registered is extended, acquisition is prolonged Long image is denoted as x1；Edge in Prewitt operator detection images is used first to finger vein figure to be registered, is changed into bianry image, Then by the finger beginning and end position on image both sides, it is baseline to take its both ends midpoint line, selects least square method pair Medium line is fitted, and then takes the point set of both sides same distance to form finger contours, is obtained the image among profile and is denoted as x2；The image being denoted as using extension image among x1 and profile is denoted as x2 and is overlapped, and obtains complete finger vein figure to be registered The area image of picture is denoted as x3.
Data normalization and image enhancement preprocessing process are carried out to finger vein image to be registered：
The complete area image to be registered for referring to vein image is subjected to size normalization and standardization, obtains ruler The area image of very little normalization and standardization；Using Histogram Mapping the size is normalized and standardization Area image carries out contrast enhancement processing, obtains the area image after image enhancement；Wherein, the size normalization uses Cubic interpolation is realized, is changed into 0 to all area image averages, variance is changed into 1；The mapping coefficient of the Histogram Mapping is Obtained by genetic algorithm.
Further, finger areas is chosen：
Since the particularity of harvester and the finger of acquisition target put action uncertainty, SDUMLA-HMT data The image collected in storehouse always contains incoherent information and background；In addition the finger venous collection device of entity is base In infrared light projection, CCD camera shooting is imaged, and in gained 2D images in addition to finger, also has device background, light causes Shade etc.；Therefore preprocessing part carries out finger extraction first in the present invention, only retains the finger for including and referring to vein segment Parts of images, but since the single finger contours determined by least square method can often produce incompleteness, the present invention Middle to propose the new finger areas extraction scheme based on Sobel operators, finger ductility and least square method, detailed process is such as Under：
(1) extracts edge with Sobel edge detection operators first to original image, then using contours extract；Carry The principal outline in image, then ductility and flatness according to finger contours are taken out, removal does not meet finger extension direction Branch, specific method are to judge branch direction and position；Due to the image that collects, illumination condition is unstable sometimes, and together The same finger of one people may be different in different collection moment placement positions, angle etc., it is difficult to pass through from image Edge detection and contours extract extract complete finger contours, thus also need to incomplete finger edge according to finger its The angular characteristics of its part carry out finger lengthening, and the image obtained after extension is denoted as x1.
(2) first with edge in Prewitt operator detection images, is changed into bianry image to same original image, Then by the finger beginning and end position on image both sides, it is baseline to take its both ends midpoint line, selects least square method pair Medium line is fitted, and then takes the point set of both sides same distance to form finger contours, the image among contouring is denoted as x2.
(3), which is overlapped image x1 and x2, can obtain more complete finger areas image x3, avoid finger Problem is lost in region, is carried out in next step as a data set sample.
Data normalization and image enhancement：
For the original image in database, need processing consistent and good for specification after finger areas extraction The input picture of processing is to autocoder；Size normalization and standardization are carried out first, in order to which protrusion refers to venous information, and The contrast based on Histogram Mapping is used to strengthen afterwards, in order to select most appropriate mapping coefficient and fitness function, using something lost Propagation algorithm is realized；Size normalization is realized using cubic interpolation method, causes average to be changed into 0 in all samples, variance is changed into 1.
Histogram equalization (Histogram Equalization (HE)), which is often used in living things feature recognition field, schemes Image intensifying part, but there may be unnatural image and inapplicable and light and shade for HE and most of contrast enhancement process Degree needs the occasion retained；Therefore the optimum mapping of image is found in selection with genetic algorithm in the present invention, then is mapped.
The process of whole genetic algorithm is as follows:
1) first generation individual is randomly generated, first generation individual amount is identical with sample total in data set.The dye of the first generation Colour solid element is made of random number.Individual can be all evaluated in per a generation, fitness numerical value is obtained by calculating fitness function, According to ranking fitness individual.
2) present age is made choice and bred, according to fitness by wheel disc method assortative mating individual, breeding includes intersecting With mutation, produce individual of future generation and form colony, wherein what is do not intersected directly arrives the next generation.
3) second step is constantly repeated, until meeting end condition i.e. iterations.
After mapping coefficient is obtained by genetic algorithm, original image is mapped；And the image dimension obtained at this time It is higher, it is unfavorable for being trained；And some features extracted may be useless, and dimensionality reduction is carried out using PCA.
S12 is described further：
Depth characteristic is carried out to pretreated finger vein image to be registered using trained deep neural network model Extraction, pretreated finger vein image to be registered is input in trained deep neural network model, is passed through in a model After crossing a series of processing, depth characteristic is exported.
Deep neural network model training process is as follows：
1) initial depth neural network model is built；
The initial depth neural network model includes input layer, the first convolution pond layer, the second convolution pond successively Layer, the 3rd convolution pond layer, full articulamentum and output layer；The convolution kernel size of the first convolution pond layer is 3x3, Chi Hua Core size is 2x2, step-length 1；The convolution kernel size of the second convolution pond layer is 3x3, and pond core size is 2x2, step-length For 1；The convolution kernel size of 3rd convolution pond layer is 3x3, and pond core size is 2x2, step-length 1；The full articulamentum Full connection size be 210。
It is r × c to define and treat that training refers to vein image size, is denoted as xlarge, first middle self-study through this embodiment The size dimension for practising a × b that characteristic extraction step is extracted from large-size images treats that training refers to vein image sample xsamll instructions Practice sparse own coding, calculate
F=σ (W(1)xsmal+b(1))
Wherein, σ is sigmoid activation primitives, after obtaining k feature, wherein, W(1)With b(1)It is visualization and hidden Containing the weight and deviation between unit, for the small image x of each a × b sizes, calculated accordingly by above-mentioned formula Value be denoted as fconvolved；Convolution is done to these fconvolved values, it is possible to it is a to obtain k × (r-a+1) × (c-b+1) Eigenmatrix after convolution.
In embodiments of the present invention, image size is W × H, and the area size of selection is 15*15, and convolution pattern uses Valid, that is, remember output matrix：W × H, convolution kernel F × F, step-length stride, output is high, width new_height, new_ Width, then：
New_width=(W-F+1)/S
New_height=(H-F+1)/S
Round 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.
Accordingly calculated by above-mentioned method, the first convolution pond layer, the second convolution pond layer, the 3rd convolution pond It is 3x3 to change the convolution kernel size in layer, and pond core size is 2x2, and step-length 1, the full connection size of full articulamentum is 210For most Properly.
2) finger vein image to be trained at least is divided into three set, the first subclass, yield in the second subset are closed and the 3rd son Set；
Finger vein image to be trained at least is divided into three set, the first subclass, yield in the second subset is closed and the 3rd subset Close, treat that training refers to vein image quantity and for the first subclass is more than yield in the second subset and closes and be more than the 3rd subclass in subclass, three A subclass treats that training refers to vein image quantity and all to treat that training refers to vein quantity.
3) respectively to treating that training refers to vein figure in first subclass, yield in the second subset conjunction and the 3rd subclass Pre-processed as carrying out finger areas extraction, standardization and image enhancement successively, obtain pretreated first subclass, second Subclass and the 3rd subclass；
Training, which refers to vein image and carries out finger areas extraction, standardization and image successively, to be treated to above-mentioned subclass respectively Enhancing pretreatment, specific implementation process refer to S11 steps, and details are not described herein.
4) carried out to treating that training refers to vein image in pretreated first subclass at self study feature extraction Reason, obtains the alignment features for treating that training refers to vein image in first subclass；
First, three layers of self-editing neural network model, including input layer, hidden layer and output layer are built, and is limited described hidden The hidden unit number for hiding layer is 50；It will treat that training refers to vein image and passes through the input layer in first subclass Input in the hidden layer be iterated training after, using different hidden units treat training refer to the different position of vein image Edge detection is carried out with direction, obtains in first subclass and treats that training refers to vein image alignment features；By described first Treat that training refers to vein image alignment features and exported by the output layer in subclass.
Further, supervised learning is needed using a large amount of marker samples, it is necessary to put into substantial amounts of human and material resources； And own coding neutral net is a kind of unsupervised learning algorithm, it has used back-propagation algorithm, and it is defeated to allow desired value to be equal to Enter value, i.e. y(i)=x(i)；Three layers of own coding neural network model are used in the present invention, comprising input layer, output layer and are hidden Layer.
In vein image is referred to, it is to be mutually related between pixel, is not completely independent；And own coding neutral net Trial learning approaches identity function hW,b(x) ≈ x, by limiting the quantity of hidden neuron in the present invention so that own coding god Carry out feature self study through network, study obtains the efficient expression of feature, limited in the present invention hidden layer hidden unit number as 50。
It is trained using the first subclass, is to hide after 1000 iteration, retaining concealed nodes weights Feature, the different hidden units that final training obtains carry out edge detection in the different position and direction of image, i.e. study is arrived Extracted in specific direction and treat that training refers to vein image alignment features.
5) alignment features and described pretreated second for treating that training refers to vein image in first subclass are used Treat that training refers to vein image and processing is trained to initial depth neural network model in subclass, obtain the depth after training Neural network model；
It will wait to train in the yield in the second subset conjunction in the first subclass after the alignment features for referring to vein image and processing is trained Refer to vein image to input in initial neural network model by the input layer in initial depth neural network model, be trained Processing, by training, adjusts the parameter in initial depth neural network model, and the preferable scope of parameter adjustment best friend obtains Deep neural network model after to training.
6) vein image is referred to the depth god after the training after training using in pretreated 3rd subclass Tested through network model, obtain test result；
The deep neural network model after training is carried out accordingly using vein image is referred to after training in the 3rd subclass Depth characteristic extraction test, is matched using the depth characteristic, whether judges the depth characteristic extracted by matching result In perfect condition.
7) judge that the exact discrimination of the test result identifies whether to be less than predetermined threshold value, if so, then after the training Deep neural network model be trained deep neural network model, if it is not, then return training step continue to train.
By the matching result obtained in above-mentioned steps, judge whether the true discrimination of matching result is less than threshold value, if so, The depth characteristic for then proving to extract is more complete depth characteristic, and the deep neural network model after the training is training Good deep neural network model, continues to train if it is not, then returning to training step.
S13 is described further：
The finger vein image further feature to be registered got is registered in the identification matching database, and is marked Know the corresponding subscriber identity information ID of device, after registration is good, by corresponding further feature and subscriber identity information ID update into Enter into the template library built, finally complete the retention further feature template library for obtaining the renewal after having new user's registration.
S14 is described further：
Specific implementation process may be referred to S11 and S12 steps, and details are not described herein.
S15 is described further：
It is deep to the finger vein image further feature to be identified and the retention after the registration using Softmax graders Layer feature templates storehouse carries out classification and matching identifying processing.
In matching stage, classification is realized by Softmax graders, Softmax is that logistic recurrence is being classified more Popularization in problem, in the present invention, is identified N number of people, and class label y takes 1 to arrive N.
Note number of types is k, i.e. y can take the different values of k, then to training set { (x(1),y(1)),…,(x(m),y(m)), Wherein input feature vector x(i)∈Rn+1, wherein assuming that function (hypothesis function) is as follows in logistic recurrence:
Pass through training pattern parameter θ so that it can minimize cost function；When having multiple classification, for Fixed input x, wants that therefore, the output of function needs with assuming that each type of function pair j estimates probable value p (y=j | x) If the vector (vector element and be 1) of k dimension represents the k probable value estimated, and form is as follows：
Wherein θ1,θ2,…,θk∈Rn+1It is the parameter of model, equation the right face is that operation is normalized so that probability With for 1；Define Softmax regression algorithms cost function be：
Wherein, 1 { y(i)=j } represent y(i)=j durations are 1；Wherein, x is categorized as y probability is：
In order to enable J (θ) is minimized, therefore using the optimization algorithm (L-BFGS) of iteration, obtaining gradient after derivation is：
Wherein,It is vector, first element is J (θ) to θjOne-component partial derivative；Try to achieve local derviation After number, cost function can be minimized with gradient descent algorithm, i.e.,：
Wherein, α is speed, but is worked as from vectorial θjIn subtract vectorial ψ, λ is sparse coefficient, it is assumed that function is still constant, I.e. the prediction result of influence function, i.e. Softmax graders are not over parameterized completely, it is therefore desirable to weight weight decay To solve redundancy issue, final cost function is as follows：
Constantly trained by above-mentioned steps Softmax graders, reach certain number or Softmax classification After device convergence, preservation refers to hand vein recognition model, for above-mentioned carry out Classification and Identification and obtains match cognization result.
In embodiments of the present invention, can only be extracted when referring to vein image feature extraction using the embodiment of the present invention Include crucial finger vein image further feature；Preprocessing part propose a kind of new finger areas extracting method and Histogram Mapping is carried out by genetic algorithm, can be preferably to the carry out extracted region of finger vein image；For with description Property finger vein pattern be difficult to extract and find, the embodiment of the present invention employs self study feature extraction mode, using training number Part sample in carries out self study to sparse autocoder, effectively improves subsequently to deep neural network model Training effect so that deep neural network model can fast and accurately extract the further feature for referring to vein image；Follow-up Returned using Softmax and carry out quick and precisely identification matching.
Embodiment two：
Fig. 2 is the finger vein identification method based on unsupervised learning and deep layer network in another embodiment of the present invention Flow diagram, as shown in Fig. 2, the finger vein identification method includes：
S21：Finger areas extraction, data normalization and image enhancement are carried out successively to finger vein image to be registered in advance to locate Reason, obtains pretreated finger vein image to be registered；
S22：The pretreated finger vein image to be registered is carried out using trained deep neural network model Feature extraction processing to be registered, obtains the further feature to be registered for referring to vein image；
S23：Registered according to the further feature to be registered for referring to vein image and build template library processing, obtained Retention further feature template library after registration；
S24：Finger vein image to be identified is gathered to carry out the finger vein image to be identified at S21, S22 step successively Reason, obtains finger vein image further feature to be identified；
S25：Using Softmax graders to the finger vein image further feature to be identified and staying after the registration Deposit further feature template library and carry out classification and matching identifying processing, obtain match cognization result
S26：Judge whether the match cognization result is less than predetermined threshold value, if then identifying successfully, if otherwise identification is lost Lose；
S27：If identifying, successfully, by the identification, successfully finger vein image further feature to be identified adds the registration In retention further feature template library afterwards, using the successful finger vein image further feature to be identified of the identification to the note Retention further feature template library after volume is updated, and identifies corresponding identity ID number.
As soon as it may be referred to the S11-S15 steps in embodiment for the detailed implementation process in S21-S25, herein not Repeat again.
S26 is described further：
Judge whether match cognization result is less than predetermined threshold value, wherein predetermined threshold value can be set according to user's self-demand Put, such as it is 1 to set threshold value；Compared with being carried out accordingly with match cognization result using predetermined threshold value, match cognization result is less than pre- If threshold value, then it is judged as identifying successfully, is otherwise determined as recognition failures.
S27 is described further：
By the identification, successfully finger vein image further feature to be identified adds the retention further feature after the registration In template library, using the identification, successfully the vein image further feature to be identified that refers to is special to the retention deep layer after the registration Sign template library is updated, and identifies corresponding identity ID number.
In embodiments of the present invention, can only be extracted when referring to vein image feature extraction using the embodiment of the present invention Include crucial finger vein image further feature；Preprocessing part propose a kind of new finger areas extracting method and Histogram Mapping is carried out by genetic algorithm, can be preferably to the carry out extracted region of finger vein image；For with description Property finger vein pattern be difficult to extract and find, the embodiment of the present invention employs self study feature extraction mode, using training number Part sample in carries out self study to sparse autocoder, effectively improves subsequently to deep neural network model Training effect so that deep neural network model can fast and accurately extract the further feature for referring to vein image；Follow-up Returned using Softmax and carry out quick and precisely identification matching.
Embodiment three：
Fig. 3 is the system of the finger vein recognition system based on unsupervised learning and deep layer network in the embodiment of the present invention Structure composition schematic diagram, as shown in figure 3, the finger vein recognition system includes：
Pretreatment module 11：For to it is to be registered finger vein image carry out successively finger areas extraction, data normalization with Image enhancement pre-processes, and obtains pretreated finger vein image to be registered；
Fisrt feature extraction module 12：For using trained deep neural network model to described pretreated Finger vein image to be registered carries out feature extraction processing to be registered, obtains the further feature to be registered for referring to vein image；
Template library builds module 13：For register simultaneously structure according to the further feature to be registered for referring to vein image Template library processing is built, obtains the retention further feature template library after registration；
Second feature extraction module 14：For gather it is to be identified finger vein image will it is described it is to be identified refer to vein image according to It is secondary to carry out pretreatment and feature extraction processing step, obtain finger vein image further feature to be identified；
Classification and Identification module 15：For using Softmax graders to it is described it is to be identified finger vein image further feature with Retention further feature template library after the registration carries out classification and matching identifying processing, obtains match cognization result.
Preferably：The pretreatment module includes area extracting unit, data normalization and image enhancing unit；
The area extracting unit includes：
The first profile extracts subelement：For being carried first using Sobel edge detection operators to finger vein image to be registered Edge is taken, then carries out contours extract, extracts the profile to be registered for referring to vein image, then according to finger vein image to be registered The ductility and flatness of profile, remove the branch for not meeting finger extension direction, and to incomplete finger vein image to be registered Profile extended, obtain extend image be denoted as x1；
Second contours extract subelement：For being used first in Prewitt operator detection images to finger vein figure to be registered Edge, is changed into bianry image, and then by the finger beginning and end position on image both sides, it is base to take its both ends midpoint line Line, selects least square method to be fitted medium line, then takes the point set of both sides same distance to form finger contours, obtains Image among profile is denoted as x2；
It is superimposed subelement：Image for being denoted as using extension image among x1 and profile is denoted as x2 and is overlapped, and is obtained The complete area image to be registered for referring to vein image is denoted as x3；
Data normalization includes with image enhancing unit：
Data normalization subelement：For the complete area image to be registered for referring to vein image to be carried out size normalizing Change and standardization, acquisition size normalize the area image with standardization；
Image enhancement subelement：For being normalized using Histogram Mapping to the size and the region of standardization Image carries out contrast enhancement processing, obtains the area image after image enhancement；
Wherein, the size normalization is realized using cubic interpolation, is changed into 0 to all area image averages, variance becomes For 1；The mapping coefficient of the Histogram Mapping is obtained by genetic algorithm.
Preferably, the deep neural network model training process, including：
Build initial depth neural network model；
Finger vein image to be trained at least is divided into three set, the first subclass, yield in the second subset is closed and the 3rd subset Close；
Respectively to treating that training refers to vein image in first subclass, yield in the second subset conjunction and the 3rd subclass Finger areas extraction, standardization and image enhancement are carried out successively to pre-process, and obtain pretreated first subclass, the second son Set and the 3rd subclass；
Self study feature extraction processing is carried out to treating that training refers to vein image in pretreated first subclass, Obtain the alignment features for treating that training refers to vein image in first subclass；
Using treated in first subclass training refer to vein image alignment features and it is described it is pretreated second son Treat that training refers to vein image and processing is trained to initial depth neural network model in set, obtain the depth god after training Through network model；
Refer to vein image to the depth nerve after the training after training using in pretreated 3rd subclass Network model is tested, and obtains test result；
Judge that the exact discrimination of the test result identifies whether to be less than predetermined threshold value, if so, then after the training Deep neural network model is trained deep neural network model, continues to train if it is not, then returning to training step.
Preferably, it is described to carry out self study spy to treating that training refers to vein image in pretreated first subclass Extraction process is levied, including：
Three layers of self-editing neural network model, including input layer, hidden layer and output layer are built, and limits the hidden layer Hidden unit number be 50；
By in first subclass treat training refer to vein image by the input layer input in the hidden layer into After row iteration training, treating that training refers to the different position and direction of vein image and carry out edge detection using different hidden units, Obtain in first subclass and treat that training refers to vein image alignment features；
It will treat that training refers to vein image alignment features and exported by the output layer in first subclass.
Preferably, the initial depth neural network model includes input layer, the first convolution pond layer, volume Two successively Product pond layer, the 3rd convolution pond layer, full articulamentum and output layer；
The convolution kernel size of the first convolution pond layer is 3x3, and pond core size is 2x2, step-length 1；
The convolution kernel size of the second convolution pond layer is 3x3, and pond core size is 2x2, step-length 1；
The convolution kernel size of 3rd convolution pond layer is 3x3, and pond core size is 2x2, step-length 1；
The full connection size of the full articulamentum is 210。
Fig. 4 is the finger vein recognition system based on unsupervised learning and deep layer network in another embodiment of the present invention System structure composition schematic diagram, as shown in figure 4, the finger vein recognition system further includes：
Judgment module 16：For judging whether the match cognization result is less than predetermined threshold value, if then identifying successfully, If otherwise recognition failures；
Update module 17：If for identifying that successfully, successfully finger vein image further feature to be identified adds by the identification Enter in the retention further feature template library after the registration, using the identification, successfully finger vein image deep layer to be identified is special Sign is updated the retention further feature template library after the registration, and identifies corresponding identity ID number.
Specifically, the operation principle of the system related functions module of the embodiment of the present invention can be found in the phase of embodiment of the method Description is closed, which is not described herein again.
In embodiments of the present invention, can only be extracted when referring to vein image feature extraction using the embodiment of the present invention Include crucial finger vein image further feature；Preprocessing part propose a kind of new finger areas extracting method and Histogram Mapping is carried out by genetic algorithm, can be preferably to the carry out extracted region of finger vein image；For with description Property finger vein pattern be difficult to extract and find, the embodiment of the present invention employs self study feature extraction mode, using training number Part sample in carries out self study to sparse autocoder, effectively improves subsequently to deep neural network model Training effect so that deep neural network model can fast and accurately extract the further feature for referring to vein image；Follow-up Returned using Softmax and carry out quick and precisely identification matching.
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 relevant hardware to complete by program, which 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..
A kind of in addition, finger vein based on unsupervised learning and deep layer network provided above the embodiment of the present invention Recognition methods and system are described in detail, and should employ specific case herein to the principle of the present invention and embodiment It is set forth, the explanation of above example is only intended to help to understand method and its core concept of the invention；It is meanwhile right In those of ordinary skill in the art, according to the thought of the present invention, change is had in specific embodiments and applications Part, in conclusion this specification content should not be construed as limiting the invention.

Claims (10)

  1. A kind of 1. finger vein identification method based on unsupervised learning and deep layer network, it is characterised in that the finger hand vein recognition Method includes：
    S11：Finger areas extraction, data normalization and image enhancement are carried out successively to finger vein image to be registered to pre-process, and are obtained Take pretreated finger vein image to be registered；
    S12：The pretreated finger vein image to be registered is carried out waiting to note using trained deep neural network model Volume feature extraction processing, obtains the further feature to be registered for referring to vein image；
    S13：Registered according to the further feature to be registered for referring to vein image and build template library processing, after obtaining registration Retention further feature template library；
    S14：Gather finger vein image to be identified and the finger vein image to be identified is subjected to S11, S12 step process successively, obtain Take finger vein image further feature to be identified；
    S15：Using Softmax graders to the finger vein image further feature to be identified and the retention deep layer after the registration Feature templates storehouse carries out classification and matching identifying processing, obtains match cognization result.
  2. 2. the finger vein identification method according to claim 1 based on unsupervised learning and deep layer network, it is characterised in that It is described that finger vein image progress finger areas extraction to be registered is pre-processed, including：
    To finger vein image to be registered first using Sobel edge detection operators extraction edge, contours extract is then carried out, is extracted The profile to be registered for referring to vein image, then the ductility and flatness of the profile according to finger vein image to be registered, removal are not inconsistent The branch in finger extension direction is closed, and the profile of incomplete finger vein image to be registered is extended, obtains and extends image note For x1；
    Edge in Prewitt operator detection images is used first to finger vein figure to be registered, is changed into bianry image, then passes through figure As the finger beginning and end position on both sides, it is baseline to take its both ends midpoint line, selects least square method to carry out medium line Fitting, then takes the point set of both sides same distance to form finger contours, obtains the image among profile and is denoted as x2；
    The image being denoted as using extension image among x1 and profile is denoted as x2 and is overlapped, and obtains complete finger vein figure to be registered The area image of picture is denoted as x3.
  3. 3. the finger vein identification method according to claim 1 based on unsupervised learning and deep layer network, it is characterised in that It is described that data normalization and image enhancement pretreatment are carried out to finger vein image to be registered, including：
    The complete area image to be registered for referring to vein image is subjected to size normalization and standardization, obtains size normalizing Change the area image with standardization；
    The size is normalized using Histogram Mapping and the area image of standardization carries out contrast enhancement processing, is obtained Take the area image after image enhancement；
    Wherein, the size normalization is realized using cubic interpolation, is changed into 0 to all area image averages, variance is changed into 1； The mapping coefficient of the Histogram Mapping is obtained by genetic algorithm.
  4. 4. the finger vein identification method according to claim 1 based on unsupervised learning and deep layer network, it is characterised in that The deep neural network model training process, including：
    Build initial depth neural network model；
    Finger vein image to be trained at least is divided into three set, the first subclass, yield in the second subset is closed and the 3rd subclass；
    Respectively to first subclass, the yield in the second subset close and the 3rd subclass in treat training refer to vein image successively into The extraction of row finger areas, standardization and image enhancement pre-process, and obtain pretreated first subclass, yield in the second subset is closed and the Three subclass；
    Self study feature extraction processing is carried out to treating that training refers to vein image in pretreated first subclass, obtains institute State the alignment features for treating that training refers to vein image in the first subclass；
    Using treating that training refers to the alignment features of vein image and the pretreated yield in the second subset is closed in first subclass Inside treat that training refers to vein image and processing is trained to initial depth neural network model, obtain the deep neural network after training Model；
    Refer to vein image to the deep neural network after the training after training using in pretreated 3rd subclass Model is tested, and obtains test result；
    Judge that the exact discrimination of the test result identifies whether to be less than predetermined threshold value, if so, the then depth after the training Neural network model is trained deep neural network model, continues to train if it is not, then returning to training step.
  5. 5. the finger vein identification method according to claim 4 based on unsupervised learning and deep layer network, it is characterised in that It is described to carry out self study feature extraction processing to treating that training refers to vein image in pretreated first subclass, including：
    Three layers of self-editing neural network model, including input layer, hidden layer and output layer are built, and limits hiding for the hidden layer Unit number is 50；
    It will treat that training refers to vein image by changing in the input layer input hidden layer in first subclass After generation training, treating that training refers to the different position and direction of vein image and carry out edge detection using different hidden units, obtaining Treat that training refers to vein image alignment features in first subclass；
    It will treat that training refers to vein image alignment features and exported by the output layer in first subclass.
  6. 6. the finger vein identification method according to claim 4 based on unsupervised learning and deep layer network, it is characterised in that The initial depth neural network model includes input layer, the first convolution pond layer, the second convolution pond layer, the 3rd convolution successively Pond layer, full articulamentum and output layer；
    The convolution kernel size of the first convolution pond layer is 3x3, and pond core size is 2x2, step-length 1；
    The convolution kernel size of the second convolution pond layer is 3x3, and pond core size is 2x2, step-length 1；
    The convolution kernel size of 3rd convolution pond layer is 3x3, and pond core size is 2x2, step-length 1；
    The full connection size of the full articulamentum is 210。
  7. 7. the finger vein identification method according to claim 1 based on unsupervised learning and deep layer network, it is characterised in that The finger vein identification method further includes：
    Judge whether the match cognization result is less than predetermined threshold value, if then identifying successfully, if otherwise recognition failures；
    If identifying, successfully, by the identification, successfully the vein image further feature to be identified that refers to adds the retention depth after the registration In layer feature templates storehouse, using the identification, successfully the vein image further feature to be identified that refers to is deep to the retention after the registration Layer feature templates storehouse is updated, and identifies corresponding identity ID number.
  8. A kind of 8. finger vein recognition system based on unsupervised learning and deep layer network, it is characterised in that the finger hand vein recognition System includes：
    Pretreatment module：Increase for carrying out finger areas extraction, data normalization and image successively to finger vein image to be registered Strong pretreatment, obtains pretreated finger vein image to be registered；
    Fisrt feature extraction module：For using trained deep neural network model to the pretreated finger to be registered Vein image carries out feature extraction processing to be registered, obtains the further feature to be registered for referring to vein image；
    Template library builds module：For being registered according to the further feature to be registered for referring to vein image and building template library Processing, obtains the retention further feature template library after registration；
    Second feature extraction module：For gather it is to be identified finger vein image by it is described it is to be identified finger vein image carry out successively it is pre- Processing and feature extraction processing step, obtain finger vein image further feature to be identified；
    Classification and Identification module：For using Softmax graders to the finger vein image further feature to be identified and the note Retention further feature template library after volume carries out classification and matching identifying processing, obtains match cognization result.
  9. 9. the finger vein recognition system according to claim 8 based on unsupervised learning and deep layer network, it is characterised in that The pretreatment module includes area extracting unit, data normalization and image enhancing unit；
    The area extracting unit includes：
    The first profile extracts subelement：For extracting side using Sobel edge detection operators first to finger vein image to be registered Edge, then carries out contours extract, extracts the profile to be registered for referring to vein image, then according to the profile to be registered for referring to vein image Ductility and flatness, remove the branch for not meeting finger extension direction, and to the profile of incomplete finger vein image to be registered Extended, obtain extension image and be denoted as x1；
    Second contours extract subelement：For using edge in Prewitt operator detection images first to finger vein figure to be registered, It is changed into bianry image, then by the finger beginning and end position on image both sides, it is baseline to take its both ends midpoint line, is selected Least square method is fitted medium line, then takes the point set of both sides same distance to form finger contours, obtains among profile Image be denoted as x2；
    It is superimposed subelement：Image for being denoted as using extension image among x1 and profile is denoted as x2 and is overlapped, and is obtained complete The area image of finger vein image to be registered be denoted as x3；
    Data normalization includes with image enhancing unit：
    Data normalization subelement：For the complete area image to be registered for referring to vein image to be carried out size normalization and mark Quasi-ization processing, obtains the area image of size normalization and standardization；
    Image enhancement subelement：For using Histogram Mapping the size is normalized and the area image of standardization into Row contrast enhancement processing, obtains the area image after image enhancement；
    Wherein, the size normalization is realized using cubic interpolation, is changed into 0 to all area image averages, variance is changed into 1； The mapping coefficient of the Histogram Mapping is obtained by genetic algorithm.
  10. 10. the finger vein recognition system according to claim 8 based on unsupervised learning and deep layer network, its feature exist In the finger vein recognition system further includes：
    Judgment module：For judging whether the match cognization result is less than predetermined threshold value, if then identifying successfully, if otherwise knowing Do not fail；
    Update module：If for identifying successfully, by described in the identification successfully finger vein image further feature addition to be identified In retention further feature template library after registration, using the successful finger vein image further feature to be identified of the identification to described Retention further feature template library after registration is updated, and identifies corresponding identity ID number.
CN201710915095.3A 2017-09-30 2017-09-30 A kind of finger vein identification method and system based on unsupervised learning and deep layer network Pending CN107967442A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710915095.3A CN107967442A (en) 2017-09-30 2017-09-30 A kind of finger vein identification method and system based on unsupervised learning and deep layer network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710915095.3A CN107967442A (en) 2017-09-30 2017-09-30 A kind of finger vein identification method and system based on unsupervised learning and deep layer network

Publications (1)

Publication Number Publication Date
CN107967442A true CN107967442A (en) 2018-04-27

Family

ID=61997541

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710915095.3A Pending CN107967442A (en) 2017-09-30 2017-09-30 A kind of finger vein identification method and system based on unsupervised learning and deep layer network

Country Status (1)

Country Link
CN (1) CN107967442A (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110084238A (en) * 2019-04-09 2019-08-02 五邑大学 Finger vena image segmentation method, device and storage medium based on LadderNet network
CN110135253A (en) * 2019-04-11 2019-08-16 重庆工商大学 A kind of finger vena identification method based on long-term recursive convolution neural network
CN110147732A (en) * 2019-04-16 2019-08-20 平安科技（深圳）有限公司 Refer to vein identification method, device, computer equipment and storage medium
CN110321871A (en) * 2019-07-11 2019-10-11 电子科技大学成都学院 A kind of palm vein identification system and method based on LSTM
CN110390309A (en) * 2019-05-22 2019-10-29 西南科技大学 A kind of finger vena illegal user's recognition methods based on residual distribution
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
CN110674681A (en) * 2019-08-13 2020-01-10 平安科技（深圳）有限公司 Identity verification method and device based on attention mechanism
CN111915541A (en) * 2020-07-31 2020-11-10 平安科技（深圳）有限公司 Image enhancement processing method, device, equipment and medium based on artificial intelligence
CN113076927A (en) * 2021-04-25 2021-07-06 华南理工大学 Finger vein identification method and system based on multi-source domain migration
WO2023036322A1 (en) * 2021-09-13 2023-03-16 广州广电运通金融电子股份有限公司 Rapid authentication method based on finger vein biometric recognition technology, and device and medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106326886A (en) * 2016-11-07 2017-01-11 重庆工商大学 Finger-vein image quality evaluation method and system based on convolutional neural network
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
CN106971174A (en) * 2017-04-24 2017-07-21 华南理工大学 A kind of CNN models, CNN training methods and the vein identification method based on CNN

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106326886A (en) * 2016-11-07 2017-01-11 重庆工商大学 Finger-vein image quality evaluation method and system based on convolutional neural network
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
CN106971174A (en) * 2017-04-24 2017-07-21 华南理工大学 A kind of CNN models, CNN training methods and the vein identification method based on CNN

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ALEXEY DOSOVITSKIY ET AL.: "Discriminative Unsupervised Feature Learning with Convolutional Neural Networks", 《ARXIV》 *
贾旭 等: "基于稀疏编码的手背静脉识别算法", 《计算机应用》 *
高世杰: "手掌浅层静脉血管图像获取技术的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110084238A (en) * 2019-04-09 2019-08-02 五邑大学 Finger vena image segmentation method, device and storage medium based on LadderNet network
CN110084238B (en) * 2019-04-09 2023-01-03 五邑大学 Finger vein image segmentation method and device based on LadderNet network and storage medium
CN110135253B (en) * 2019-04-11 2022-05-20 重庆工商大学 Finger vein authentication method based on long-term recursive convolutional neural network
CN110135253A (en) * 2019-04-11 2019-08-16 重庆工商大学 A kind of finger vena identification method based on long-term recursive convolution neural network
CN110147732A (en) * 2019-04-16 2019-08-20 平安科技（深圳）有限公司 Refer to vein identification method, device, computer equipment and storage medium
CN110390309A (en) * 2019-05-22 2019-10-29 西南科技大学 A kind of finger vena illegal user's recognition methods based on residual distribution
CN110321871A (en) * 2019-07-11 2019-10-11 电子科技大学成都学院 A kind of palm vein identification system and method based on LSTM
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
CN110674681A (en) * 2019-08-13 2020-01-10 平安科技（深圳）有限公司 Identity verification method and device based on attention mechanism
WO2021189770A1 (en) * 2020-07-31 2021-09-30 平安科技（深圳）有限公司 Image enhancement processing method and apparatus based on artificial intelligence, and device and medium
US20220245803A1 (en) * 2020-07-31 2022-08-04 Ping An Technology (Shenzhen) Co., Ltd. Image enhancement processing method, device, equipment, and medium based on artificial intelligence
CN111915541A (en) * 2020-07-31 2020-11-10 平安科技（深圳）有限公司 Image enhancement processing method, device, equipment and medium based on artificial intelligence
CN113076927A (en) * 2021-04-25 2021-07-06 华南理工大学 Finger vein identification method and system based on multi-source domain migration
CN113076927B (en) * 2021-04-25 2023-02-14 华南理工大学 Finger vein identification method and system based on multi-source domain migration
WO2023036322A1 (en) * 2021-09-13 2023-03-16 广州广电运通金融电子股份有限公司 Rapid authentication method based on finger vein biometric recognition technology, and device and medium

Similar Documents

Publication Publication Date Title
CN107967442A (en) A kind of finger vein identification method and system based on unsupervised learning and deep layer network
Jin et al. Weed identification using deep learning and image processing in vegetable plantation
CN110321813B (en) Cross-domain pedestrian re-identification method based on pedestrian segmentation
Acharya et al. TallyQA: Answering complex counting questions
Siagian et al. Biologically inspired mobile robot vision localization
JP2020038658A (en) Learning method and learning device for detecting object by using edge image, and test method and test device using the same
CN108830188A (en) Vehicle checking method based on deep learning
CN108304788A (en) Face identification method based on deep neural network
Salman et al. Classification of real and fake human faces using deep learning
Yu et al. An object-based visual attention model for robotic applications
Shin et al. Effective feature extraction by trace transform for insect footprint recognition
Hohberg Wildfire smoke detection using convolutional neural networks
Foroozandeh et al. Offline handwritten signature verification and recognition based on deep transfer learning
CN111539320B (en) Multi-view gait recognition method and system based on mutual learning network strategy
CN113269103A (en) Abnormal behavior detection method and system based on space map convolutional network
CN107704822A (en) The extraction of finger vein further feature and matching process and system based on incomplete completion
CN110738166A (en) Fishing administration monitoring system infrared target identification method based on PCNN and PCANet and storage medium
Wu et al. Boosted convolutional neural networks (BCNN) for pedestrian detection
Magistri et al. From one field to another—Unsupervised domain adaptation for semantic segmentation in agricultural robotics
Ghazali et al. RGBT tracking based on prior least absolute shrinkage and selection operator and quality aware fusion of deep and handcrafted features
Damarla et al. An approach for optimization of features using gorilla troop optimizer for classification of melanoma
Alraddady et al. Fusion Strategy for Improving Medical Image Segmentation.
Filus et al. Nam: What does a neural network see?
Gera et al. Smart farming: Industry 4.0 in agriculture using artificial intelligence
Ivanovska et al. Y-gan: Learning dual data representations for efficient anomaly detection

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

Application publication date: 20180427

WD01 Invention patent application deemed withdrawn after publication