CN104166842B - It is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods - Google Patents
It is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods Download PDFInfo
- Publication number
- CN104166842B CN104166842B CN201410359335.2A CN201410359335A CN104166842B CN 104166842 B CN104166842 B CN 104166842B CN 201410359335 A CN201410359335 A CN 201410359335A CN 104166842 B CN104166842 B CN 104166842B
- Authority
- CN
- China
- Prior art keywords
- palm print
- point
- type
- dimensional
- dimensional palm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Abstract
The invention discloses it is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods, the sampled images of three-dimensional palm print are divided into some sub-regions by it, for every sub-regions, the surface type of palmmprint is calculated first, then palmmprint surface type in the every sub-regions of statistics with histogram is used, finally the surface type histogram of every sub-regions is stitched together, operator is described as the feature of palmmprint depth image, and classified using joint representational framework, so as to improve recognition efficiency and accuracy, can be used for the occasion for having strict demand to identification.Efficiently solve the alignment offset issue between multiple three-dimensional palm print samplings.
Description
Technical field
The invention belongs to area of pattern recognition, it is related to a kind of method of identity information checking, especially a kind of personal recognition
Method.
Background technology
In recent years, industrial quarters, academia be constantly devoted to improve identity information verification the verifying results, with meet access control,
In the multiple different field such as aviation safety, e-bank, the harsh demand of the identity for recognizing people.Based on living things feature recognition
Method just attract increasing concern, personal recognition is the biological feather recognition method of the great representative of one of which.
Palm grain identification method have distinction high, strong robustness, it is user friendly many advantages, such as.The skin line on palmmprint fingers and palms heart surface
Reason, mainly includes two category features:Friction ridge and flexion crease.Both features be for human individual it is constant, permanent,
It is unique.
Two-dimentional Palm Print Recognition System is limited to imaging factors, is influenceed larger by illumination condition.Additionally, two-dimentional palmmprint is also subject to
Replacement is replicated by other people, it is difficult to meet the application field for having harsh demand to identification.At present, with the development of science and technology I
, while obtaining depth image and the Two-dimensional Color Image data of three-dimensional palm print, two dimension can be solved by using structured light technique
Defect present in Palm Print Recognition System, to meet the requirement in different field for authentication.
The content of the invention
It is an object of the invention to provide it is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods.
It needs to match palmmprint to be verified one by one with sample data set in being directed to conventional three-dimensional palm print matching process, efficiency with
The problem that the capacity of sample data set increases and is greatly reduced, the framework represented using joint is carried out a pair to three-dimensional palm print to be measured
Many identification;The minute alignment error existed after being matched for three-dimensional palm print, using the description operator based on block statistics feature.
And then, establish a kind of three-dimensional palm print recognition methods accurately and fast.
To reach above-mentioned purpose, solution of the invention is:
It is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods,
(1) the three-dimensional data feature of palmmprint is determined:
(1) defineIt is i-th sampled images of the jth of palm time three-dimensional palm print, its length and width is respectively M, N, includes altogether
M × N number of point, arbitrfary point P positions in three dimensions are described with (x, y, f (x, y));
(2) willMake m deciles and n deciles respectively on length and width direction, obtain the m × n son of three-dimensional palm print sampled images
Region, then often the length and width of sub-regions is respectivelyNoteForP-th on length direction, on cross direction
Q sub-regions, wherein, 1≤p≤m, 1≤q≤n;
In the step (2), if the M long of three-dimensional palm print image cannot by m divide exactly and/or three-dimensional palm print image N wide without
Method is divided exactly by n, and during piecemeal, retaining the point positioned at central area of three-dimensional palm print image carries out piecemeal as subregion, deletes
Three-dimensional palm print image surrounding cannot divided evenly region point, not as the subregion being identified.
(3) the description operator of subregion, the information for describing three-dimensional palm print surface in subregion are created:
(3-1) determines the mean curvature H of each point in subregion;
In the step (3-1), according to formulaDetermine each point in subregion
Mean curvature H, wherein fx,fySingle order local derviation of the depth of the point on x, y direction, f are represented respectivelyxx,fyy,fxyRepresent respectively
The second order local derviation of the depth of the point.
(3-2) determines the Gaussian curvature K of each point in subregion;
In the step (3-2), according to formulaThe Gaussian curvature K of each point in subregion is calculated, its
Middle fx,fySingle order local derviation of the depth of the point on x, y direction, f are represented respectivelyxx,fyy,fxyThe two of the depth of the point are represented respectively
Rank local derviation.
Mean curvature H and the symbol of Gaussian curvature K that (3-3) puts according to each, determine each type put:If H<0 and K
>0, the point is Class1;If H<0 and K=0, the point is type 2;If H<0 and K<0, the point is type 3;If H=0 and K>0, should
Point is type 4;If H=0 and K=0, the point is type 5;If H=0 and K<0, the point is type 6;If H>0 and K>0, the point is
Type 7;If H>0 and K=0, the point is type 8;If H>0 and K<0, the point is type 9;
In the step (3-3), Class1 is in crest, and type 2 is in ridge shape, and type 3 is in saddle ridge shape, and type 5 is in plane
Shape, type 6 is minimum curve surface, and type 7 is recessed, and type 8 is in mountain valley shape, and type 9 is in saddle paddy shape, and type 4 is specific type,
Including the other shapes in addition to above-mentioned 8 type.
(3-4) obtains vectorial h=[h using the number of times of all types of appearance of statistics with histogram1,h2,…hi…,h8,h9] with
In the description subregion, wherein hiRepresent the number of the point in the subregion for type i, 1≤i≤9;
(4) the description operator of all subregion is stitched together, is constitutedDescription operator
Wherein hp,qRepresentP-th on length direction, the description operator of q-th subregion on cross direction;
(2) according to the three-dimensional data feature construction three-dimensional palm print database dictionary of the palmmprint to be measured determined in step (1) N represents the palm sum in database, CjRepresent j-th sampling sum of palm;
(3) relatively three-dimensional palm print to be measured and data message in database is determining the identity of palmmprint to be measured:
(I) joint representational framework is used, coefficient vector of the palmmprint to be identified for S is determined:By three-dimensional palm print to be identified
Description operator FprobeFrame representation is represented using joint, energy equation is minimizedObtain
Coefficient vector x0, wherein, g (x) is the regularization term of energy equation, and λ is the coefficient of predefined regularization term to control energy
The Relative Contribution of two in equation;
In the step (I), the dictionary S is preferably two dimension of the line number much smaller than columns in joint representational framework
Matrix.
Further, in the step (I), if S ∈ Rr×c, when if the line number r of the three-dimensional palm print dictionary for constituting is larger, or S
Line number r when being more than columns c, use formula S '=Φ S are to S dimensionality reductions, wherein Φ=[Φ1Φ2…Φr]∈Rk×r(k<R) it is height
This white noise accidental projection matrix, any one row Φ in ΦiMeet ‖ Φi‖2=1, and in step (3) and afterwards in the step of,
Use S ' to replace S, carry out three-dimensional palm print identification.
In the step (I), λ should be greater than 0 and less than 1;Preferably, the λ values are 0.01.
In the step (I), the energy equation selects least square regularization, i.e.,Then coefficient vector x0
By closing equation x0=(STS+λI)-1STFprobeDraw, wherein I represents unit matrix, its row (OK) counts the columns phase with S
Together;
Or, the energy equation selects a norm regularization, i.e. g (x)=| x |1, | x |1Each unit in coefficient vector x is asked in expression
The absolute value sum of element, coefficient vector x is determined using DALM, Homotopy method0。
Further, in the step (I), when the energy equation selects least square regularization, (STS+λI)-1ST
Determine after S construction completes, before palm print identity checking, to reduce the overall used time of three-dimensional palm print identification.
(II) residual error is calculated, palm print identity to be identified is determined:Calculate respectively the description operator of three-dimensional palm print to be identified with
In database dictionary it is all kinds of between correlation degree, choose identity of the maximum object of the degree of association as face to be measured, the degree of association
Obtained by calculating residual error, residual error is smaller, the degree of association is higher.
In the step (II), the computing formula of residual error isWherein,
δj(x0) mean selection x0In the only coefficient related to classification j.
Due to using such scheme, the beneficial effects of the invention are as follows:
Step (1) describes to carry out three-dimensional palm print data piecemeal, and uses the palmmprint table in each piecemeal of statistics with histogram
Noodles type, the final process for obtaining three-dimensional palm print data feature description operator.In the present invention, due to using each point of block statistics
Palmmprint surface type in block, significantly reduce the alignment error that is still suffered from after aliging between the multiple repairing weld of same palmmprint for
The influence that three-dimensional palm print identity identification is caused.By this step, for the three-dimensional palm print data being input into, each piecemeal can be distinguished
Statistics with histogram surface type is used, so that the feature for constituting the three-dimensional palm print describes operator.
Three-dimensional data feature construction three-dimensional palm print database of the step (2) according to the palmmprint to be measured determined in step (1).
In order to ensure that the database has the property for supporting the multiple registration of many people, step (2) is for the three-dimensional palm print number registered each time
According to the upper affiliated Customs Assigned Number of mark and times of registration, and then the bases such as registration, identification, the checking of three-dimensional palm print data can be met
This demand.
Step (3) describes operator with the feature that step (1) extraction obtains three-dimensional palm print to be measured, the institute in step (2)
The three-dimensional palm print database of determination compares the method to determine palm print identity to be measured:
First, the framework for being represented using joint minimizes energy equation, obtains the description operator of three-dimensional palm print to be identified
Coefficient vector.According to joint represent theory, be required to solve the energy equation of coefficient vector The energy equation includes twoWith g (x), the former for the classification of measured signal discrimination capabilities compared with
The latter seem stronger, therefore in arrange parameter λ, λ should be greater than 0 and less than 1.Herein, g (x) can beOr g
(x)=| x |1.The former can be by closing equation x0=(STS+λI)-1STFprobeDraw, wherein I represents unit matrix, its row is (OK)
Number is identical with the columns of S;The latter can be solved using methods such as DALM, Homotopy.The single order that g (x) can be obtained using the latter
Discrimination slightly better than the single order discrimination that can be obtained using the former, is about higher by 0.4%.However, in energy process is minimized
The time of consumption, according to the difference of method for solving, using the time required for the latter than the former is used, will increase sharply tens times very
To hundred times.
Secondly, according to coefficient vector, all kinds of reconstructed errors are calculated, the minimum class of selection reconstructed error is chosen and treated
The maximum class of the degree of association of three-dimensional palm print is surveyed, as the classification of three-dimensional palm print to be measured.
The effect of step (3) is discrimination high, and elapsed time is few, and recognition time is not by number in three-dimensional palm print database
According to increase and it is unprecedented soaring.
By performing three above step and slightly adjusting according to the actual requirements, you can realize the note of three-dimensional palm print data
The basic functions such as volume, identification and checking, meet recognition time and are not substantially improved by the increase of data in three-dimensional palm print database,
Meanwhile, error that three-dimensional palm print data produce in alignment procedure is also solved for influence that three-dimensional palm print identification is caused.
Brief description of the drawings
Fig. 1 is the present invention based on block statistics feature and the workflow diagram for combining the three-dimensional palm print recognition methods for representing.
Specific embodiment
The present invention will be further described with reference to the accompanying drawings.
For needing to match palmmprint to be verified one by one with sample data set in conventional three-dimensional palm print matching process,
The problem that efficiency increasing and being greatly reduced with the capacity of sample data set, the present invention is using the framework of joint expression to three-dimensional to be measured
Palmmprint carries out one-to-many identification;The minute alignment error existed after being matched for three-dimensional palm print, the present invention is using based on piecemeal
The description operator of statistical nature is setting up a kind of three-dimensional palm print recognition methods accurately and fast, its specific workflow such as Fig. 1 institutes
Show:
(1) the three-dimensional data feature of palmmprint to be measured is determined:
(1) defineIt is i-th sampled images of the jth of palm time three-dimensional palm print, its length and width is respectively M, N, includes altogether
M × N number of point, arbitrfary point P therein positions in three dimensions are described with (x, y, z), and wherein z=f (x, y), f represents z-axis
Function of the value on value on x-axis, y-axis direction on direction;
(2) willMake m deciles and n deciles respectively on length and width direction, obtain the m × n son of three-dimensional palm print sampled images
Region.NoteForP-th on length direction, q-th subregion on cross direction, meet 1≤p≤m, 1≤q≤n, its
Length and width is respectively
In step (2), if the M long of three-dimensional palm print image cannot be divided exactly by m and/or the N wide of three-dimensional palm print image cannot be by n
Divide exactly, during piecemeal, retaining the point positioned at central area of three-dimensional palm print image carries out piecemeal as subregion, deletes the three-dimensional palm
Print image surrounding cannot divided evenly region point, not as the subregion being identified.
For example, when the image of three-dimensional palm print is 128*128 sizes, by row by each 10 decile of row, then 100 phases can be obtained
With the region of size.For each region, its length and width should be 128/10=12.8, i.e., 128 cannot divide exactly 10.Due to three-dimensional
The image midpoint of palmmprint is discrete, it is impossible to get the such as the 12.8th point, and in this case, the length and width in each region are respectively
12.8 round downwards, are 12.Consequently, it is possible to length and width have respectively taken 120 points, because original image length and width are respectively on 128, therefore length and width
8 points are all had more.In the present embodiment, the point for choosing the picture centre region of three-dimensional palm print carries out piecemeal as subregion,
Delete three-dimensional palm print image surrounding cannot divided evenly region point, that is, the left side retains 4 points and do not select, and the right retains 4 points
Do not select;Retain 4 points above not select, 4 points are retained below and is not selected, therefore it is to be located at center that length and width respectively select at the 5th~the 124th point
The point in region, then every 12 o'clock used as a region, and statistics surface type is carried out respectively.
(3) the description operator of subregion, the information for describing three-dimensional palm print surface in subregion are created:
It is according to formula in (3-1) the present embodimentDetermine each in subregion
The mean curvature H of point, wherein fx,fySingle order local derviation of the depth of the point on x, y direction, f are represented respectivelyxx,fyy,fxyDifference table
Show the second order local derviation of the depth of the point;
(3-2) is according to formulaDetermine the Gaussian curvature K of each point in subregion;
Mean curvature H and the symbol of Gaussian curvature K that (3-3) puts according to each, determine each type put:If H<0 and K
>0, the point is Class1;If H<0 and K=0, the point is type 2;If H<0 and K<0, the point is type 3;If H=0 and K>0, should
Point is type 4;If H=0 and K=0, the point is type 5;If H=0 and K<0, the point is type 6;If H>0 and K>0, the point is
Type 7;If H>0 and K=0, the point is type 8;If H>0 and K<0, the point is type 9.
In step (3-3), Class1 is in crest, and type 2 is in ridge shape, and type 3 is in saddle ridge shape, and type 5 is in plane, class
Type 6 is minimum curve surface, and type 7 is recessed, and type 8 is in mountain valley shape, and type 9 is in saddle paddy shape, and type 4 is specific type.
(3-4) obtains vectorial h=[h using the number of times of all types of appearance of statistics with histogram1,h2,…hi…,h8,h9] with
In the description subregion, wherein hiRepresent the number of the point in the subregion for type i, 1≤i≤9;
(4) the description operator of all subregion is stitched together, is constitutedDescription operator
Wherein hp,qRepresentP-th on length direction, the description operator of q-th subregion on cross direction;
(2) according to the three-dimensional data feature construction three-dimensional palm print database dictionary of the palmmprint to be measured determined in step (1) N represents the palm sum in database, CjRepresent j-th sampling sum of palm.
(3) relatively three-dimensional palm print to be measured and data message in database is determining the identity of palmmprint to be measured:
(I) joint representational framework is used, coefficient vector of the palmmprint to be identified for S is determined:By three-dimensional palm print to be identified
Description operator Fprobe, using joint representational framework, minimize energy equationObtain
Coefficient vector x0, wherein, g (x) is the regularization term of energy equation, and λ is the coefficient of predefined regularization term, controls energy side
The Relative Contribution of two in journey.
In step (I), dictionary S is preferably two-dimensional matrix of the line number less than columns, and row in joint representational framework
What number was tried one's best is set less than columns, so can obtain more preferable recognition effect.When to three-dimensional palm print image divided block, due to drawing
Block number after point is different, such as can be by row divided by column into 25 pieces, it is also possible to be divided into 100 pieces, therefore each three-dimensional palm print
Feature descriptor dimension it is different, with above-mentioned data instance, the dimension for dividing 25 pieces of descriptor is 25*9=225 dimensions, is drawn
It is divided into 100 pieces of the dimension of descriptor for 100*9=900 is tieed up.Because dictionary S is two-dimensional matrix of the line number much smaller than columns
When, preferable recognition effect can be obtained, therefore in the present embodiment, if S ∈ Rr×cIf the line number r of the three-dimensional palm print dictionary for constituting is larger
When, or the line number r of S is when being more than columns c, uses formula S '=Φ S are to S dimensionality reductions, wherein Φ=[Φ1Φ2…Φr]∈Rk×r(k<
R) it is white Gaussian noise accidental projection matrix, any one row Φ in ΦiMeet ‖ Φi‖2=1, and step in step (3) and afterwards
In rapid, use S ' to replace S, carry out three-dimensional palm print identification.
In step (I), in arrange parameter λ, in order to prevent working as energy equationIt is acquired when minimum
As a result x0Over-fitting, therefore g (x) items of non-negative are increased in energy equation, therefore λ should also be the item more than 0, it is ensured that it is non-
Negative property, λ should be greater than 0 and less than 1, be verified through test of many times, and taking for preferable recognition result can be obtained when λ values 0.01
Value.
In step (1), the theory that (a) is represented according to joint, energy equation can use least square regularization, i.e.,
The energy equation includes twoWithThe former discrimination capabilities for the classification of measured signal
Seem stronger than the latter, λ should be greater than 0 and less than 1, coefficient vector x0By closing equation x0=(STS+λI)-1STFprobe
Go out, wherein I represents unit matrix, (OK) number is identical with the columns of S for its row.
Energy equation selects least square regularization, i.e.,Now, (STS+λI)-1STCan be in S construction completes
Afterwards, determine before palm print identity checking, can so reduce the overall used time of three-dimensional palm print identification.Because (STS+λI)-1ST
In, what one side S was represented is database palmmprint dictionary, i.e., one two-dimensional matrix, each row from a feature for three-dimensional palm print to
Amount is constituted, therefore S is known;On the other hand, I is unit matrix, also known.In view of calculating (STS+λI)- 1STConsumption time it is relatively more, it is impossible to meet real-time (if i.e. every time calculate consumption time use needs if common PC
Be more than 40 milliseconds, the reason for cause the sampling included in database dictionary S quantity it is more, number of samples is more, and calculating needs
The time wanted is more long, and in accompanying drawing is tested, the listed calculating time is and precalculates ((STS+λI)-1STResult), therefore
Design factor vector x0Before, it may be predetermined that ((STS+λI)-1STResult, solve x0When directly using being previously obtained
Result is calculated, and can significantly accelerate recognition time.
B () energy equation also can select a norm regularization, that is, minimize energy equation when solving coefficient vector |x|1The absolute value sum of each element in coefficient vector x, a norm regularization are asked in expression
Energy equation can be solved using the method such as DALM, Homotopy.
Meanwhile, energy equation selects a norm regularization, i.e. g (x)=| x |1, verified through test of many times, when λ values 0.01
When can obtain the value of preferable recognition result.
Using the single order discrimination of a norm regularization slightly better than the single order discrimination of least square regularization, about it is higher by
0.4%, but the time consumed in energy process is minimized is according to the difference of method for solving, will increase sharply tens times even hundreds of
Times, the selection of the two can be determined as the case may be.
(II) residual error is calculated, palm print identity to be identified is determined:Calculate respectively the description operator of three-dimensional palm print to be identified with
In database dictionary it is all kinds of between correlation degree, choose identity of the maximum object of the degree of association as face to be measured, the degree of association
Obtained by calculating residual error, residual error is smaller, corresponding class is higher with the degree of association of three-dimensional palm print to be measured.
In step (II), can be by formulaObtain the class of three-dimensional palm print
Not, wherein δj(x0) mean reservation x0In the only coefficient related to classification j, by x0Element in middle other positions is all set to 0.
Beneficial effects of the present invention are illustrated below in conjunction with specific experiment:
Experiment one:PolyU three-dimensional palm prints database includes 8000 palmmprint samplings, belongs to 200 the 400 of volunteer hands
The palm.In 200 volunteers, 136 entitled males, remaining is women.The each palm of each volunteer is gathered at twice,
10 samples of collection every time, every palm totally 20 sampled datas after gathering twice.
In an experiment, the three-dimensional palm data composition known collection for collecting for the first time is chosen, second is chosen and is gathered
The three-dimensional palm data for obtaining constitute set to be measured.Designed according to experiment condition, it is known that 400 classes are included in set, each
Class has 10 sampled datas.Experiment runs on Hewlett-Packard's Z620 work stations, and it is furnished with 3.2GHZ Intel to strong E5-1650 centers
Processor and 8G internal memories, are run using MatlabR2013b.
Table 1 is the schematic diagram of the single order discrimination obtained using different characteristic on PolyU 3D palm print data collection;
Table 1
From table 1 it follows that on the premise of the same sorting algorithm represented using least square regularization joint, fortune
With block statistics surface type as feature, acquired single order discrimination is far beyond based on main element analysis and different parameters
Under the conditions of LBP features obtain result.There it can be seen that the character description method used in the present invention is for three-dimensional data
Local shape structure to portray ability stronger.
Experiment two:Setting in the same experiment one of setting of the known collection of three-dimensional palm print and set to be measured, using the present invention
The block statistics surface type of middle proposition compares different classifications method and obtains single order discrimination and complete one time three as feature
Tie up the time that the ID inquiring of palmmprint needs to expend.
In an experiment, withBe reference as the sorting technique of the regularization term of energy equation, choose with g (x)=
|x|1Compared as the different solutions of the sorting technique of the regularization term of energy equation, including Homotopy, FISTA, l1_
Ls, SpaRSA, DALM, table 2 are the signal of the single order discrimination obtained using different classifications method on PolyU3D palm print data collection
Figure.CR_L is respectively labeled as in table 21_ Homotopy, CR_L1_ FISTA, CR_L1_l1_ ls, CR_L1_ SpaRSA, CR_L1_
DALM;D.Zhang, G.Lu, W.Li, L.Zhang and N.Luo are in " Palmprint recognition using3-D
The tri- kinds of methods of MCI, GCI, ST proposed in information ";D.Zhang, V.Kanhangad, N.Luo and A.Kumar exist
The LC methods proposed in " Robust palmprint verification using 2D and 3D features ".
Table 2
Following three points conclusion as can be drawn from Table 2:First, method proposed by the present invention from for single order discrimination, than
The effect of other method is more preferable, and especially the method such as MCI, GCI, LC cannot obtain the recognition effect of same level completely;Second,
WithAs the sorting technique of the regularization term of energy equation, and with g (x)=| x |1As the rule of energy equation
The sorting technique of item, acquired recognition effect is all sufficiently close to.However, the former needs in a three-dimensional palm print identification is completed
The time of consumption, be the latter different solutions required for the time 1/tens, or even more than one percent, therefore combine piecemeal
Statistical form noodles not and use the former as the regularization term of energy equation, and recognition effect is good, it is often more important that, the used time is less;The
Three, different solutions are with g (x)=| x |1As the time required for the sorting technique of the regularization term of energy equation because of the side for realizing
Formula difference difference great disparity.Homotopy and DALM methods are than FISTA, l1_ ls and SpaRSA are rapider.
The above-mentioned description to embodiment is to be understood that and use this hair for ease of those skilled in the art
It is bright.Person skilled in the art obviously easily can make various modifications to these embodiments, and described herein
General Principle is applied in other embodiment without by performing creative labour.Therefore, the invention is not restricted to above-described embodiment,
Those skilled in the art's announcement of the invention, does not depart from improvement that scope made and modification all should be in this hair
Within bright protection domain.
Claims (10)
1. it is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods, it is characterised in that:
(1) the three-dimensional data feature of palmmprint is determined:
(1) defineIt is i-th sampled images of the jth of palm time three-dimensional palm print, its length and width is respectively M, N, altogether comprising M × N
It is individual, P positions in three dimensions in arbitrfary point are described with (x, y, f (x, y));
(2) willMake m deciles and n deciles respectively on length and width direction, obtain the m × n sub-district of three-dimensional palm print sampled images
Domain, then often the length and width of sub-regions is respectivelyNoteForP-th on length direction, the q on cross direction
Sub-regions, wherein, 1≤p≤m, 1≤q≤n;
(3) the description operator of subregion, the information for describing three-dimensional palm print surface in subregion are created:
(3-1) determines the mean curvature H of each point in subregion;
(3-2) determines the Gaussian curvature K of each point in subregion;
Mean curvature H and the symbol of Gaussian curvature K that (3-3) puts according to each, determine each type put:If H<0 and K>0,
The point is Class1;If H<0 and K=0, the point is type 2;If H<0 and K<0, the point is type 3;If H=0 and K>0, the point is
Type 4;If H=0 and K=0, the point is type 5;If H=0 and K<0, the point is type 6;If H>0 and K>0, the point is type
7;If H>0 and K=0, the point is type 8;If H>0 and K<0, the point is type 9;
(3-4) obtains vectorial h=[h using the number of times of all types of appearance of statistics with histogram1,h2,…hi…,h8,h9] for retouching
State the subregion, wherein hiRepresent the number of the point in the subregion for type i, 1≤i≤9;
(4) the description operator of all subregion is stitched together, is constitutedDescription operatorIts
Middle hp,qRepresentP-th on length direction, the description operator of q-th subregion on cross direction;
(2) according to the three-dimensional data feature construction three-dimensional palm print database dictionary of the palmmprint to be measured determined in step (1) N represents the palm sum in database, CjRepresent j-th sampling sum of palm;
(3) relatively three-dimensional palm print to be measured and data message in database is determining the identity of palmmprint to be measured:
(I) joint representational framework is used, coefficient vector of the palmmprint to be identified for S is determined:By retouching for three-dimensional palm print to be identified
State operator FprobeFrame representation is represented using joint, energy equation is minimized
To coefficient vector x0, wherein, g (x) is the regularization term of energy equation, and λ is the coefficient of predefined regularization term to control energy
The Relative Contribution of two in amount equation;
(II) residual error is calculated, palm print identity to be identified is determined:The description operator and data of three-dimensional palm print to be identified are calculated respectively
In the dictionary of storehouse it is all kinds of between correlation degree, choose identity of the maximum object of the degree of association as face to be measured, the degree of association passes through
Calculate residual error and obtain, residual error is smaller, and the degree of association is higher.
2. three-dimensional palm print recognition methods according to claim 1, it is characterised in that:In the step (2), if three-dimensional palm print
The M long of image cannot be divided exactly by m and/or the N wide of three-dimensional palm print image cannot be divided exactly by n, during piecemeal, retains three-dimensional palm print image
The point positioned at central area carry out piecemeal as subregion, deleting three-dimensional palm print image surrounding cannot divided evenly region
Point, not as the subregion being identified.
3. three-dimensional palm print recognition methods according to claim 1, it is characterised in that:In the step (3-1), according to formulaDetermine the mean curvature H of each point in subregion, wherein fx,fyThe point is represented respectively
Single order local derviation of the depth on x, y direction, fxx,fyy,fxyThe second order local derviation of the depth of the point is represented respectively.
4. three-dimensional palm print recognition methods according to claim 1, it is characterised in that:In the step (3-2), according to formulaCalculate the Gaussian curvature K, wherein f of each point in subregionx,fyRepresent the depth of the point in x, y side respectively
Upward single order local derviation, fxx,fyy,fxyThe second order local derviation of the depth of the point is represented respectively.
5. three-dimensional palm print recognition methods according to claim 1, it is characterised in that:In the step (3-3), Class1 is in
Crest, type 2 is in ridge shape, and type 3 is in saddle ridge shape, and type 5 is in plane, and type 6 is minimum curve surface, and type 7 is recessed,
Type 8 is in mountain valley shape, and type 9 is in saddle paddy shape, and type 4 is specific type.
6. three-dimensional palm print recognition methods according to claim 1, it is characterised in that:In the step (I), the energy side
Cheng Xuanyong least square regularization, i.e.,Then coefficient vector x0By closing equation x0=(STS+λI)-1STFprobe
Draw, wherein I represents unit matrix, the columns of its columns and its line number all with S is identical;
Or the energy equation selects a norm regularization, i.e. g (x)=| x |1, | x |1Each element in coefficient vector x is sought in expression
Absolute value sum, coefficient vector x is determined using DALM, Homotopy method0。
7. three-dimensional palm print recognition methods according to claim 6, it is characterised in that:In the step (I), the energy side
During Cheng Xuanyong least square regularization, (STS+λI)-1STDetermine after S construction completes, before palm print identity checking, to reduce
The overall used time of three-dimensional palm print identification.
8. three-dimensional palm print recognition methods according to claim 1, it is characterised in that:In the step (I), λ should be greater than 0 and
Less than 1.
9. three-dimensional palm print recognition methods according to claim 1, it is characterised in that:In the step (I), joint represents frame
The dictionary S is two-dimensional matrix of the line number less than columns in frame.
10. three-dimensional palm print recognition methods according to claim 1, it is characterised in that:In the step (II), the meter of residual error
Calculating formula isWherein, δj(x0) mean selection x0In only it is related to classification j
Coefficient.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410359335.2A CN104166842B (en) | 2014-07-25 | 2014-07-25 | It is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410359335.2A CN104166842B (en) | 2014-07-25 | 2014-07-25 | It is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104166842A CN104166842A (en) | 2014-11-26 |
CN104166842B true CN104166842B (en) | 2017-06-13 |
Family
ID=51910646
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410359335.2A Active CN104166842B (en) | 2014-07-25 | 2014-07-25 | It is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104166842B (en) |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104636760B (en) * | 2015-03-11 | 2017-09-05 | 西安科技大学 | A kind of localization method of weld seam |
JP6604008B2 (en) * | 2015-03-13 | 2019-11-13 | 富士通株式会社 | Image processing apparatus, image processing program, and image processing method |
CN105069403B (en) * | 2015-07-20 | 2018-10-26 | 同济大学 | A kind of three-dimensional human ear identification based on block statistics feature and the classification of dictionary learning rarefaction representation |
CN106250823A (en) * | 2016-07-21 | 2016-12-21 | 同济大学 | A kind of palm print and palm vein recognition methods and palm print and palm vein image capture instrument |
CN108154066B (en) * | 2016-12-02 | 2021-04-27 | 中国科学院沈阳自动化研究所 | Three-dimensional target identification method based on curvature characteristic recurrent neural network |
CN107169445B (en) * | 2017-05-11 | 2020-07-03 | 北京东方金指科技有限公司 | Large-scale palm print coding and comparing method |
CN107330382A (en) * | 2017-06-16 | 2017-11-07 | 深圳大学 | The single sample face recognition method and device represented based on local convolution characteristic binding |
CN107704846A (en) * | 2017-10-27 | 2018-02-16 | 济南大学 | Palm grain identification method based on two-value direction commensal vector and bloom wave filters |
CN108573212B (en) * | 2018-03-08 | 2022-03-25 | 广东工业大学 | Palm feature identity authentication method and device |
CN108960125B (en) * | 2018-06-29 | 2020-03-13 | 河北工业大学 | Three-dimensional palm print recognition method |
CN110390268B (en) * | 2019-06-26 | 2023-06-30 | 东南大学 | Three-dimensional palmprint recognition method based on geometric characteristics and direction characteristics |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101196988A (en) * | 2007-12-25 | 2008-06-11 | 哈尔滨工业大学 | Palm locating and center area extraction method of three-dimensional palm print identity identification system |
CN103793692A (en) * | 2014-01-29 | 2014-05-14 | 五邑大学 | Low-resolution multi-spectral palm print and palm vein real-time identity recognition method and system |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7609865B2 (en) * | 2004-11-08 | 2009-10-27 | Biomagnetics | 3D fingerprint and palm print data model and capture devices using multi structured lights and cameras |
US8265347B2 (en) * | 2008-04-24 | 2012-09-11 | The Hong Kong Polytechnic University | Method and system for personal identification using 3D palmprint imaging |
-
2014
- 2014-07-25 CN CN201410359335.2A patent/CN104166842B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101196988A (en) * | 2007-12-25 | 2008-06-11 | 哈尔滨工业大学 | Palm locating and center area extraction method of three-dimensional palm print identity identification system |
CN103793692A (en) * | 2014-01-29 | 2014-05-14 | 五邑大学 | Low-resolution multi-spectral palm print and palm vein real-time identity recognition method and system |
Non-Patent Citations (2)
Title |
---|
A novel classification method for palmprint recognition based on reconstrunction error and normalized distance;Zhonghua LIU等;《Applied Intelligence》;20130223;全文 * |
三维掌纹身份鉴别系统;肖琼林;《万方数据》;20130305;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN104166842A (en) | 2014-11-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104166842B (en) | It is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods | |
Herzlinger et al. | A new method for 3D geometric morphometric shape analysis: The case study of handaxe knapping skill | |
Chen et al. | Schelling points on 3D surface meshes | |
Vittayakorn et al. | Runway to realway: Visual analysis of fashion | |
CN107977671A (en) | A kind of tongue picture sorting technique based on multitask convolutional neural networks | |
CN106462771A (en) | 3D image significance detection method | |
CN104008375B (en) | The integrated face identification method of feature based fusion | |
CN105893925A (en) | Human hand detection method based on complexion and device | |
CN106296638A (en) | Significance information acquisition device and significance information acquisition method | |
CN104636721B (en) | A kind of palm grain identification method based on profile Yu Edge texture Fusion Features | |
Sun et al. | Monte Carlo convex hull model for classification of traditional Chinese paintings | |
CN105678235B (en) | Three-dimensional face expression recognition methods based on representative region various dimensions feature | |
CN103413145B (en) | Intra-articular irrigation method based on depth image | |
CN103488415B (en) | Personal letter word based on paper medium and pattern recognition system and recognition methods | |
CN105009171A (en) | Method and apparatus for learning-enhanced altas-based auto-segmentation | |
Blomley et al. | Classification of airborne laser scanning data using geometric multi-scale features and different neighbourhood types | |
CN105005765A (en) | Facial expression identification method based on Gabor wavelet and gray-level co-occurrence matrix | |
CN104573707B (en) | A kind of license plate Chinese character recognition method based on multi-feature fusion | |
Ozdarici-Ok | Automatic detection and delineation of citrus trees from VHR satellite imagery | |
Casanova et al. | Texture analysis using fractal descriptors estimated by the mutual interference of color channels | |
CN110119688A (en) | A kind of Image emotional semantic classification method using visual attention contract network | |
CN106897669A (en) | A kind of pedestrian based on consistent iteration various visual angles transfer learning discrimination method again | |
CN110135364A (en) | A kind of Objects recognition method and device | |
Li et al. | Dating ancient paintings of Mogao Grottoes using deeply learnt visual codes | |
Zhang et al. | Automatic detection of invasive ductal carcinoma based on the fusion of multi-scale residual convolutional neural network and SVM |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |