CN106407921B - Vein identification method based on Riesz small echo and SSLM model - Google Patents
Vein identification method based on Riesz small echo and SSLM model Download PDFInfo
- Publication number
- CN106407921B CN106407921B CN201610808628.3A CN201610808628A CN106407921B CN 106407921 B CN106407921 B CN 106407921B CN 201610808628 A CN201610808628 A CN 201610808628A CN 106407921 B CN106407921 B CN 106407921B
- Authority
- CN
- China
- Prior art keywords
- sslm
- model
- riesz
- sample
- image
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V10/443—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 by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/14—Vascular patterns
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Human Computer Interaction (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Biodiversity & Conservation Biology (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
- Collating Specific Patterns (AREA)
Abstract
A kind of vein identification method based on Riesz small echo and SSLM model.Feature extraction is based on Riesz small echo in this method.Riesz transformation is the Multi-Dimensional Extension of Hilbert transformation, has good frequency spectrum directionality, and the change of amplitude will not occur before and after transformation.The characteristics of Riesz small echo for combining and generating with small echo not only keeps original directionality, also adds scale invariability.Vein image is handled using Riesz small echo can efficiently extract its different directions, the textural characteristics on different scale.In addition, positive sample point can be surrounded using hypersphere small as far as possible using SSLM model, while negative sample point is separated using larger space, so often using less negative sample with regard to achievable model construction;And its hypersphere centre of sphere can preferably represent positive sample, and the texture signature of multiple scales can be generated, therefore the method for the present invention has the characteristics that speed is fast, effect is good, physical significance is strong.
Description
Technical field
The invention belongs to image identification technical field, more particularly to a kind of quiet based on Riesz small echo and SSLM model
Arteries and veins recognition methods.
Background technique
Modern biotechnology feature identification technique develops towards accurate, safe, quick direction, and traditional biological characteristic is known
Not, such as fingerprint, sound are unable to satisfy people for the demand of secure context already.Finger vena is special as a kind of new bio
Identification technology is levied, there are many advantages compared to more traditional method.First, refer to that vein detection has living body requirement, is only examining
It can obtain referring to vein image when surveying living body, there is good safety;Second, refer to that vein is internal feature, extraneous factor is not
The obstacle that will lead to identification aspect, there is preferable anti-interference;Third, everyone finger venous image are different, tools
There is unique and irreplaceability.
Summary of the invention
To solve the above-mentioned problems, the purpose of the present invention is to provide a kind of quiet based on Riesz small echo and SSLM model
Arteries and veins recognition methods.
In order to achieve the above object, the vein identification method packet provided by the invention based on Riesz small echo and SSLM model
Include the following steps carried out in order:
1) every of acquisition original finger venous image is normalized into 2m×2mThe rectangular image of size;
2) above-mentioned every normalized finger venous image is subjected to N rank Riesz transformation and obtains N+1 Riesz transformation
Image;
3) above-mentioned N+1 Riesz changing images are subjected to J layers of small wavelength-division respectively using high-pass filter and low-pass filter
It solves and obtains J (N+1) Zhang Xiaobo and decompose image;
4) it calculates above-mentioned every Zhang Xiaobo and decomposes the energy of image, and obtain the energy feature of every original finger venous image
Vector;
5) database is established, using all finger venous images of individual each in database as a classification, from data
Belong in library according to classification to choose the positive negative sample of a certain classification as training sample, and assigns respective labels;It then will just
Negative sample is respectively by above-mentioned steps 1) method of-step 4) handled and obtains respective J (N+1) Riesz wavelet decomposition
Image simultaneously acquires the feature parameter vectors;The corresponding the feature parameter vectors of each positive negative sample are known as a sample in feature space
This point;
6) a SSLM model is established using the sample point of above-mentioned each classification;
7) important parameter in above-mentioned each SSLM model is calculated;
8) using the important parameter in above-mentioned SSLM model, order is converted in conjunction with Riesz and Decomposition order generates certain classification
The different scale texture signature map of training sample;
9) differentiated respectively using the feature parameter vectors of the decision function to all positive samples of a certain classification, then will
Differentiate that result is encoded, constitutes coding schedule by all positive sample encoded radios of all categories;
10) finger venous image to be detected is identified using above-mentioned SSLM model and coding schedule.
In step 2), shown in transformation for mula such as formula (1) used by the Riesz is converted:
Wherein RNIndicate N rank Riesz transformation, which is by N+1 monokaryon R(n, N-n)It constitutes;The calculating of Riesz monokaryon is public
Shown in formula such as formula (2):
W is frequency in formula, by horizontal direction frequency w1With vertical direction frequency w2Composition;F (x) is original finger vena figure
Picture, f (w) are the two-dimensional Fourier transform image of original finger venous image f (x).
In step 3), the formula of the high-pass filter H (w) and low-pass filter G (w) are respectively formula (3) and formula
(4):
Ω is passband or stopband cutoff frequency in formula.
In step 4), the every Zhang Xiaobo of calculating decomposes the energy of image, and obtains every original finger vena figure
The method of the feature parameter vectors of picture are as follows:
The energy balane formula of the wavelet decomposition image are as follows:
I, j indicate pixel coordinate in formula,Indicate the decomposition of jth layer, n-th of wavelet decomposition image
In gray value at the position (i, j),It is effective for gray scale in the wavelet decomposition image, i.e. the picture of gray value non-zero
Vegetarian refreshments number;
After calculating the energy that J (N+1) Zhang Xiaobo decomposes image, by a energy value of above-mentioned J (N+1) first according to Decomposition order
Further according to transformation order and the ascending sequence of numerical value is arranged, thus to obtain the energy of every original finger venous image
Feature vector E.
In step 5), described establishes database, using all finger venous images of individual each in database as
One classification belongs to choose the positive negative sample of a certain classification as training sample according to classification from database, and assigns phase
Answer label;Then by positive negative sample respectively by above-mentioned steps 1) method of-step 4) handled and obtains respective J (N+1)
It opens Riesz wavelet decomposition image and acquires the feature parameter vectors;It is in feature space that the corresponding energy of each positive negative sample is special
Sign vector, which is known as the method for a sample point, is;
Acquire multiple finger venous images of multiple and different same fingers of individual and composition data library, it will be each in database
All finger venous images of individual extract m from each classification as a classification1Finger venous image is opened as m1It is a just
Sample, and the label for assigning positive sample is+1, extracts m from the remaining non-category2Finger venous image is opened as m2A negative sample
This, and the label for assigning negative sample is -1;Extract the training sample that the positive negative sample obtained is the category;Then by above-mentioned m1
A positive sample and m2A negative sample is respectively by above-mentioned steps 1) method of-step 4) handled and obtains respective J (N+1)
Zhang Xiaobo decomposes image and the feature parameter vectors E.By each positive sample or the corresponding energy feature of negative sample in feature space
Vector E is referred to as positive sample point or negative sample point.
In step 6), the method that the sample point using above-mentioned each classification establishes a SSLM model is:
The hypersphere that a centre of sphere is C, radius is R is established in feature space to include above-mentioned a certain classification training sample
In all positive sample points, and by negative sample points all in category training sample be placed in away from centre of sphere C be R+ ρ interval spherical surface it
Outside, ρ is the spacing distance outside hypersphere, thus sets up SSLM model;
The objective function of SSLM model are as follows:
Constraint condition are as follows:
||Φ(xi)-C||2≤R2+ζi 1≤i≤m1
||Φ(xi)-C||2≥R2+ρ+ζj m1≤j≤n
ζk≥0 1≤k≤n
In formula (6), R indicates that hyperspherical radius, ρ indicate the spacing distance outside hypersphere, m1And m2Respectively indicate positive and negative sample
This number, ξiAnd ξjExpression slack variable, and v, v1And v2For the Optimal Parameters of positive value;In the present invention, SSLM model is defeated
Enter amount xiThe as the feature parameter vectors E that acquires of step 4).
In step 7), the method for the important parameter in the above-mentioned each SSLM model of calculating is:
Important parameter is the interval outside hyperspherical centre of sphere C, hyperspherical radius of a ball R and hypersphere in SSLM model
Distance ρ, each calculation formula are as follows:
α indicates the Lagrange coefficient during optimization, y in above formulaiIndicate the label of i-th of sample;n1And n2
Respectively indicate the number that optimization forms effective positive and negative sample point to hypersphere in the process;P1And P2Then by following public
Formula, which calculates, to be obtained:
S in formula (10), (11)1And S2Respectively indicate positive and negative samples point set;K (x, y) indicates used kernel function,
Gaussian kernel function is referred here to, as shown in formula (12);
In step 8), the important parameter using in above-mentioned SSLM model converts order and decomposition in conjunction with Riesz
The method that the number of plies generates the different scale texture signature map of certain classification training sample is:
The hyperspherical centre of sphere C of the corresponding SSLM model of a certain classification is grouped according to scale, i.e., by hyperspherical ball
Every N+1 number is divided into one group in heart C, as the signature coefficient under a scale, by the hyperspherical centre of sphere for being grouped the category
The signature coefficient w of J scale can be obtainedj, j=1,2,3 ... J can obtain the category under jth scale using formula (13)
Texture signature map:
W in formulajIndicate that the signature coefficient under jth scale, G (x) indicate Gaussian function.
It is described to be distinguished using the feature parameter vectors of the decision function to all positive samples of a certain classification in step 9)
Differentiated, then will differentiate that result encodes, the method for constituting coding schedule by all positive sample encoded radios of all categories
It is:
(1) any l finger vein images in the training sample of a certain classification are chosen first, and extract this l images respectively
The feature parameter vectors Eq, q=1,2,3 ... l;
(2) then utilize decision function shown in formula (14) successively to the l of the category using the SSLM model of all categories
A the feature parameter vectors EqDifferentiated respectively, differentiate that result is as follows:
Differentiate that input quantity x is the feature parameter vectors E that previous step is extracted in formula (14)q;BCiIndicate training sample Eq
As a result ,+1 indicates that differentiation belongs to the i-th class, -1 indicates that differentiation is not belonging to the i-th class for differentiation relative to the i-th class SSLM model;RiTable
Show the radius of a ball of the i-th class SSLM model, Ci、αiRespectively indicate the corresponding centre of sphere of SSLM model, Lagrange coefficient, yiAnd xi
Indicate the supporting vector of category SSLM model;
(3) by the differentiation result BC of above-mentioned acquisition afteriIt is encoded, coding formula is as follows:
ECR in formulac(i) indicate that the category in i-th of bits of coded, that is, corresponds to the coding situation of i-th of SSLM classifier,
And length be classification sum, be made of 0 and 1 ECR coding be exactly the category coding result;Due to m1The coding of a positive sample
As a result it is possible that inconsistent situation, so by m1A coding result step-by-step progress and operation, using obtained result as
The encoded radio of the category;
(4) finally the encoded radio of all categories is put into a table and generates coding schedule.
It is described that finger venous image to be detected is known using above-mentioned SSLM model and coding schedule in step 10)
Method for distinguishing is:
A finger venous image to be detected is chosen from database, then according to above-mentioned steps 2) method of-step 8)
And its encoded radio is obtained, encoded radio of all categories in the encoded radio and above-mentioned coding schedule is finally subjected to Euclidean distance solution, choosing
Take the recognition result apart from the smallest classification as the finger venous image to be detected.
Feature extraction in vein identification method provided by the invention based on Riesz small echo and SSLM model is to be based on
Riesz small echo.Riesz transformation is the Multi-Dimensional Extension of Hilbert transformation, has good frequency spectrum directionality, and before transformation
The change of amplitude will not occur afterwards.The Riesz small echo for combining and generating with small echo not only keeps original directionality, also increases
The characteristics of scale invariability.Vein image is handled using Riesz small echo can efficiently extract its different directions, different rulers
Textural characteristics on degree.In addition, positive sample point can be surrounded using hypersphere small as far as possible using SSLM model, use simultaneously
Biggish interval often can use less negative sample with regard to achievable model construction to separate negative sample point in this way;And it is super
The spherical surface centre of sphere can preferably represent positive sample, the texture signature of multiple scales can be generated, therefore the method for the present invention has speed
The features such as degree is fast, effect is good, physical significance is strong.
Detailed description of the invention
Fig. 1 is monokaryon spectral image of Riesz transformation order N when being 8.
Fig. 2 is that Riesz transformation order N one opens normalized finger venous image and the image by Riesz change when being 8
The Riesz changing image obtained after changing.
Fig. 3 is wavelet decomposition process schematic.
Fig. 4 is that Ω isWhen high-pass filter H (w) and low-pass filter G (w) spectrogram.
Fig. 5 be Riesz transformation order N be 8, J (N+1) Zhang Xiaobo that obtains when Decomposition order J is 3 decomposes image.
Fig. 6 is SSLM model schematic.
Fig. 7 is the texture signature map under different scale.
Specific embodiment
Known in the following with reference to the drawings and specific embodiments to provided by the invention based on the vein of Riesz small echo and SSLM model
Other method is described in detail.
Vein identification method provided by the invention based on Riesz small echo and SSLM model include carry out in order it is following
Step:
1) every of acquisition original finger venous image is normalized into 2m×2mThe rectangular image of size;
2) above-mentioned every normalized finger venous image is subjected to N rank Riesz transformation and obtains N+1 Riesz transformation
Image, shown in transformation for mula such as formula (1) used by Riesz is converted:
Wherein RNIndicate N rank Riesz transformation, which is by N+1 monokaryon R(n, N-n)It constitutes;The calculating of Riesz monokaryon is public
Shown in formula such as formula (2):
W is frequency in formula, by horizontal direction frequency w1With vertical direction frequency w2Composition;F (x) is original finger vena figure
Picture, f (w) are the two-dimensional Fourier transform image of original finger venous image f (x).
It is converted by Riesz, a normalized finger venous image will generate N+1 Riesz changing images.Riesz
The list of the monokaryon spectral image of transformation as shown in Figure 1, wherein from top to bottom, be from left to right followed successively by order n=0, when 1 ... 8
Core spectral image.Riesz changing image such as Fig. 2 institute that one normalized finger venous image obtains after Riesz is converted
Show, wherein topmost an image is finger venous image after normalizing, below two rows from top to bottom, be from left to right followed successively by rank
Riesz changing image when 8 n=0,1 ... of number.
3) above-mentioned N+1 Riesz changing images are subjected to J layers of small wavelength-division respectively using high-pass filter and low-pass filter
It solves and obtains J (N+1) Zhang Xiaobo and decompose image;
Since Riesz changing image does not have scale invariability, it can make Riesz changing image that there is scale not in conjunction with small echo
The characteristic of change, wavelet decomposition process schematic are as shown in Figure 3.
The formula of high-pass filter H (w) and low-pass filter G (w) is respectively formula (3) and formula (4).It should be noted that
There is a down-sampled process after low-pass filtering, picture size can be made to become smaller in this way.
Ω is passband or stopband cutoff frequency, frequency spectrum such as Fig. 4 of high-pass filter H (w) and low-pass filter G (w) in formula
It is shown:
Fig. 5 be Riesz transformation order N be 8, J (N+1) Zhang Xiaobo that obtains when Decomposition order J is 3 decomposes image.In Fig. 5
N+1 Zhang Xiaobo when each behavior different layers decomposes image, is respectively classified as wavelet decomposition image when different rank n.
N+1 Riesz changing images are known after J layers of wavelet decomposition, J (N+1) Zhang Xiaobo is obtained altogether and decomposes image
4) it calculates above-mentioned every Zhang Xiaobo and decomposes the energy of image, and obtain the energy feature of every original finger venous image
Vector.
Image is decomposed for J (N+1) Zhang XiaoboThe energy that every Zhang Xiaobo decomposes image is solved using formula (5)
Amount:
I, j indicate pixel coordinate in formula,Indicate the decomposition of jth layer, n-th of wavelet decomposition image
In gray value at the position (i, j),For the picture of gray scale in the wavelet decomposition image effectively (i.e. gray value non-zero)
Vegetarian refreshments number.
After calculating the energy that J (N+1) Zhang Xiaobo decomposes image, by a energy value of above-mentioned J (N+1) first according to Decomposition order
Further according to transformation order and the ascending sequence of numerical value is arranged, thus to obtain the energy of every original finger venous image
Feature vector E.
5) database is established, using all finger venous images of individual each in database as a classification, from data
Belong in library according to classification to choose the positive negative sample of a certain classification as training sample, and assigns respective labels;It then will just
Negative sample is respectively by above-mentioned steps 1) method of-step 4) handled and obtains respective J (N+1) Riesz wavelet decomposition
Image simultaneously acquires the feature parameter vectors;The corresponding the feature parameter vectors of each positive negative sample are known as a sample in feature space
This point;
Acquire multiple finger venous images of multiple and different same fingers of individual and composition data library, it will be each in database
All finger venous images of individual extract m from each classification as a classification1Finger venous image is opened as m1It is a just
Sample, and the label for assigning positive sample is+1, extracts m from the remaining non-category2Finger venous image is opened as m2A negative sample
This, and the label for assigning negative sample is -1;Extract the training sample that the positive negative sample obtained is the category;Then by above-mentioned m1
A positive sample and m2A negative sample is respectively by above-mentioned steps 1) method of-step 4) handled and obtains respective J (N+1)
Zhang Xiaobo decomposes image and the feature parameter vectors E.By each positive sample or the corresponding energy feature of negative sample in feature space
Vector E is referred to as positive sample point or negative sample point.
6) a SSLM model is established using the sample point of above-mentioned each classification;
SSLM model (Small Sphere and Large Margin) is two classifiers, is mainly used for singular point
Detection, positive sample point is surrounded using hypersphere small as far as possible, while separating negative sample using biggish spacing distance
Point often can use less negative sample point with regard to achievable model construction;And its hyperspherical centre of sphere can be represented preferably
Positive sample point.
SSLM model is as shown in fig. 6, establish the hypersphere that a centre of sphere is C, radius is R in feature space to include upper
All positive sample points in a certain classification training sample are stated, and negative sample points all in category training sample are placed in away from the centre of sphere
C is except the interval spherical surface of R+ ρ, and ρ is the spacing distance outside hypersphere, thus sets up SSLM model, and wherein square is positive sample
This point, cross are negative sample point, and solid line circle is the hypersphere that radius is R, and circle of dotted line is then the interval spherical surface of R+ ρ.It should use up
Hypersphere radius surface R may be reduced simultaneously and maximize spacing distance ρ to obtain model optimal solution.
The objective function of SSLM model are as follows:
Constraint condition are as follows:
||Φ(xi)-C||2≤R2+ζi 1≤i≤m1
||Φ(xi)-C||2≥R2+ρ+ζj m1≤j≤n
ζk≥0 1≤k≤n
In formula (6), R indicates that hyperspherical radius, ρ indicate the spacing distance outside hypersphere, m1And m2Respectively indicate positive and negative sample
This number, ξiAnd ξjExpression slack variable, and v, v1And v2For the Optimal Parameters of positive value;In the present invention, SSLM model is defeated
Enter amount xiThe as the feature parameter vectors E that acquires of step 4).
7) important parameter in above-mentioned each SSLM model is calculated;
Important parameter is the interval outside hyperspherical centre of sphere C, hyperspherical radius of a ball R and hypersphere in SSLM model
Distance ρ, each calculation formula are as follows:
α indicates the Lagrange coefficient during optimization, y in above formulaiIndicate the label of i-th of sample;n1And n2
Respectively indicate the number that optimization forms effective positive and negative sample point to hypersphere in the process;P1And P2Then by following public
Formula, which calculates, to be obtained:
S in formula (10), (11)1And S2Respectively indicate positive and negative samples point set;K (x, y) indicates used kernel function,
Gaussian kernel function is referred here to, as shown in formula (12);
8) using the important parameter in above-mentioned SSLM model, order is converted in conjunction with Riesz and Decomposition order generates certain classification
The different scale texture signature map of training sample;
Texture signature is the exclusive expression of a texture, has irreversibility, can safely and effectively indicate a kind of texture.
After the hyperspherical centre of sphere C for obtaining a certain classification training sample by SSLM model, order N is converted in conjunction with Riesz
The different scale texture signature map of category sample is produced with Decomposition order J.
Specific method is: the hyperspherical centre of sphere C of the corresponding SSLM model of a certain classification being grouped according to scale, i.e., will
Every N+1 number is divided into one group in hyperspherical centre of sphere C, as the signature coefficient under a scale, passes through the super of the grouping category
The centre of sphere of spherical surface can obtain the signature coefficient w of J scalej, j=1,2,3 ... J can obtain the category using formula (13) and exist
Texture signature map under jth scale:
W in formulajIndicate that the signature coefficient under jth scale, G (x) indicate Gaussian function.Texture under the different scale of generation
Signature map is as shown in Figure 7.
9) differentiated respectively using the feature parameter vectors of the decision function to all positive samples of a certain classification, then will
Differentiate that result is encoded, constitutes coding schedule by all positive sample encoded radios of all categories;
It is the model generated using a small amount of sample, therefore to a certain degree since SSLM model is two classifiers
On sacrifice precision, frequently can lead to a sample in this way and be mistaken for " this class " by the classifier of multiple classifications, that is, a figure occur
The case where multiclass, the present invention are re-defined in the way of coding, thus to solve the problems, such as this.
The specific method is as follows:
(1) any l finger vein images in the training sample of a certain classification are chosen first, and extract this l images respectively
The feature parameter vectors Eq, q=1,2,3 ... l;
(2) then utilize decision function shown in formula (14) successively to the l of the category using the SSLM model of all categories
A the feature parameter vectors EqDifferentiated respectively, differentiate that result is as follows:
Differentiate that input quantity x is the feature parameter vectors E that previous step is extracted in formula (14)q;BCiIndicate training sample Eq
As a result ,+1 indicates that differentiation belongs to the i-th class, -1 indicates that differentiation is not belonging to the i-th class for differentiation relative to the i-th class SSLM model;RiTable
Show the radius of a ball of the i-th class SSLM model, Ci、αiRespectively indicate the corresponding centre of sphere of SSLM model, Lagrange coefficient, yiAnd xi
Indicate the supporting vector of category SSLM model.
(3) by the differentiation result BC of above-mentioned acquisition afteriIt is encoded, coding formula is as follows:
ECR in formulac(i) indicate that the category in i-th of bits of coded, that is, corresponds to the coding situation of i-th of SSLM classifier,
And length be classification sum, be made of 0 and 1 ECR coding be exactly the category coding result.Due to m1The coding of a positive sample
As a result it is possible that inconsistent situation, so by m1A coding result step-by-step progress and operation, using obtained result as
The encoded radio of the category.
(4) finally the encoded radio of all categories is put into a table and generates coding schedule.
It should be noted that carrying out identification using SSLM model and coding schedule has preferable scalability, if had again new
Classification occurs, and only need to generate the SSLM model of the category and update coding schedule, without changing original model.
10) finger venous image to be detected is identified using above-mentioned SSLM model and coding schedule;
A finger venous image to be detected is chosen from database, then according to above-mentioned steps 2) method of-step 8)
And its encoded radio is obtained, encoded radio of all categories in the encoded radio and above-mentioned coding schedule is finally subjected to Euclidean distance solution, choosing
Take the recognition result apart from the smallest classification as the finger venous image to be detected.
Experimental result
Database in the present invention includes 500 Different Individuals altogether, each individual 10 original finger venous images, in total
The original finger venous image of 5000 width, experimental situation are PC machine, Matlab R2014a.
In an experiment, original finger venous image is having a size of 91*210, when normalizing is found through experiments that having a size of 128*128
Effect is best;It is 8 that Riesz, which converts the default order N selected, and the Decomposition order J for carrying out default when wavelet decomposition is 5;In experiment
Due to there is 500 classifications, also correspondence produces 500 SSLM models, when forming training sample, positive sample quantity m1For
5, and negative sample quantity m2It is 11, i.e., extracts 11 from 4990 original finger venous images of remaining non-category sample
?.
1) it is compared based on finger venous image with OCSVM, SVDD
OCSVM model and SVDD model are two classical single category classifiers, are also commonly used for the detection of abnormal point.Due to
SSLM model is obtained on the basis of the two, is also compared herein, and N=8, J=5 are selected in feature extraction
Riesz small echo is identified for the finger venous image of 500 people:
The discrimination of table 1, OCSVM model, SVDD model and SSLM model for finger venous image
Reduce the requirement for sample using single category classifier by comparing can be seen that, positive sample can be only used
This.But due to lacking negative sample, so that the resolving effect of classifier is reduced very much, so not only being subtracted using SSLM model
Lack the demand to training sample, improves recognizer for the scalability of new samples, while also having taken into account the accurate of algorithm
Rate.
2) the method for the present invention compares the accuracy rate for referring to that vein, fingerprint, phalangeal configurations identify
Fingerprint, phalangeal configurations are similarly handled using the method for the present invention, fingerprint, the phalangeal configurations database used is
500 classifications, each classification take 10 images, and each database totally 5000 images, the results are shown in Table 2:
Table 2 refers to vein, fingerprint, phalangeal configurations in the asynchronous accuracy rate of Decomposition order
It can be seen that using the method for the present invention, fingerprint, the recognition effect for referring to vein and phalangeal configurations are different.Due to
Refer to that vein is in-vivo image, reduces the interference of external factor, recognition effect is best;Relatively almost be phalangeal configurations;And refer to
Head abrasion as caused by daily routines is more, therefore the recognition effect of fingerprint image is worst, this benefit described before also having confirmed
The identification lower judgement of accuracy is carried out with traditional fingerprint characteristic.
In addition, also listing influence of the scale selection for recognition effect in table 2, it can be seen that with Decomposition order
Increase, discrimination is generally in rising trend, and it is unobvious that effect is promoted when Decomposition order is higher.This is because Decomposition order compared with
The majority of high frequency information of Gao Shi, image are filtered out, and the energy feature provided difference between class reduces, so Decomposition order is answered
Appropriate choosing is high, but also unsuitable excessively high, but only has the accuracy of very little to be promoted in order to avoid increasing significantly operand.
Claims (9)
1. a kind of vein identification method based on Riesz small echo and SSLM model, it is characterised in that: described small based on Riesz
The vein identification method of wave and SSLM model includes the following steps carried out in order:
1) every of acquisition original finger venous image is normalized into 2m×2mThe rectangular image of size;
2) above-mentioned every normalized finger venous image is subjected to N rank Riesz transformation and obtains N+1 Riesz changing images;
3) using high-pass filter and low-pass filter above-mentioned N+1 Riesz changing images are subjected to J layers of wavelet decomposition respectively and
It obtains J (N+1) Zhang Xiaobo and decomposes image;
4) calculate above-mentioned every Zhang Xiaobo and decompose the energy of image, and obtain the energy feature of every original finger venous image to
Amount;
5) database is established, using all finger venous images of individual each in database as a classification, from database
Belong to according to classification to choose the positive negative sample of a certain classification as training sample, and assigns respective labels;Then by positive and negative sample
This is respectively by above-mentioned steps 1) method of-step 4) handled and obtains respective J (N+1) Riesz wavelet decomposition image
And acquire the feature parameter vectors;The corresponding the feature parameter vectors of each positive negative sample are known as a sample in feature space
Point;
6) a SSLM model is established using the sample point of above-mentioned each classification;
7) important parameter in above-mentioned each SSLM model is calculated;
8) using the important parameter in above-mentioned SSLM model, order is converted in conjunction with Riesz and Decomposition order generates the training of certain classification
The different scale texture signature map of sample;
9) differentiated respectively using the feature parameter vectors of the decision function to all positive samples of a certain classification, then will be differentiated
As a result it is encoded, constitutes coding schedule by all positive sample encoded radios of all categories;
10) finger venous image to be detected is identified using above-mentioned SSLM model and coding schedule;
In step 6), the method that the sample point using above-mentioned each classification establishes a SSLM model is:
The hypersphere that a centre of sphere is C, radius is R is established in feature space to include in above-mentioned a certain classification training sample
All positive sample points, and negative sample points all in category training sample are placed in except the interval spherical surface for being R+ ρ away from centre of sphere C, ρ
Thus spacing distance outside for hypersphere sets up SSLM model;
The objective function of SSLM model are as follows:
Constraint condition are as follows: | | Φ (xi)-C||2≤R2+ζi 1≤i≤m1
||Φ(xi)-C||2≥R2+ρ+ζj m1≤j≤n
ζk≥0 1≤k≤n
In formula (6), R indicates that hyperspherical radius, ρ indicate the spacing distance outside hypersphere, m1And m2Respectively indicate positive negative sample
Number, ζiAnd ζjExpression slack variable, and v, v1And v2For the Optimal Parameters of positive value;In the present invention, the input quantity of SSLM model
xiThe as the feature parameter vectors E that acquires of step 4).
2. the vein identification method according to claim 1 based on Riesz small echo and SSLM model, it is characterised in that:
In step 2), shown in transformation for mula such as formula (1) used by the Riesz is converted:
Wherein RNIndicate N rank Riesz transformation, which is by N+1 monokaryon R(n, N-n)It constitutes;The calculation formula of Riesz monokaryon is such as
Shown in formula (2):
W is frequency in formula, by horizontal direction frequency w1With vertical direction frequency w2Composition;F (x) is original finger venous image, f
It (w) is the two-dimensional Fourier transform image of original finger venous image f (x).
3. the vein identification method according to claim 1 based on Riesz small echo and SSLM model, it is characterised in that:
In step 3), the formula of the high-pass filter H (w) and low-pass filter G (w) are respectively formula (3) and formula (4):
Ω is passband or stopband cutoff frequency in formula.
4. the vein identification method according to claim 1 based on Riesz small echo and SSLM model, it is characterised in that:
In step 4), the every Zhang Xiaobo of calculating decomposes the energy of image, and the energy for obtaining every original finger venous image is special
The method for levying vector are as follows:
The energy balane formula of the wavelet decomposition image are as follows:
I, j indicate pixel coordinate in formula,Indicate that jth layer decomposes, in n-th of wavelet decomposition image
Gray value at the position (i, j),It is effective for gray scale in the wavelet decomposition image, i.e. the pixel of gray value non-zero
Number;
After calculating the energy that J (N+1) Zhang Xiaobo decomposes image, by a energy value of above-mentioned J (N+1) first according to Decomposition order root again
It is arranged according to transformation order and the ascending sequence of numerical value, thus to obtain the energy feature of every original finger venous image
Vector E.
5. the vein identification method according to claim 1 based on Riesz small echo and SSLM model, it is characterised in that:
In step 5), described establishes database, using all finger venous images of individual each in database as a classification, from
Belong in database according to classification to choose the positive negative sample of a certain classification as training sample, and assigns respective labels;Then
By positive negative sample respectively by above-mentioned steps 1) method of-step 4) handled and obtains respective J (N+1) Riesz small echo
It decomposes image and acquires the feature parameter vectors;The corresponding the feature parameter vectors of each positive negative sample are known as one in feature space
The method of a sample point is;
Multiple finger venous images of multiple and different same fingers of individual are acquired and composition data library, by individual each in database
All finger venous images as a classification, extract m from each classification1Finger venous image is opened as m1A positive sample,
And the label for assigning positive sample is+1, extracts m from the remaining non-category2Finger venous image is opened as m2A negative sample, and
The label for assigning negative sample is -1;Extract the training sample that the positive negative sample obtained is the category;Then by above-mentioned m1It is a just
Sample and m2A negative sample is respectively by above-mentioned steps 1) method of-step 4) is handled and to obtain respective J (N+1) small
Wave Decomposition image and the feature parameter vectors E;By each positive sample or the corresponding the feature parameter vectors E of negative sample in feature space
It is referred to as positive sample point or negative sample point.
6. the vein identification method according to claim 1 based on Riesz small echo and SSLM model, it is characterised in that:
In step 7), the method for the important parameter in the above-mentioned each SSLM model of calculating is:
Important parameter is the spacing distance outside hyperspherical centre of sphere C, hyperspherical radius of a ball R and hypersphere in SSLM model
ρ, each calculation formula are as follows:
α in above formulaiIndicate the Lagrange coefficient during optimization, yiIndicate the label of i-th of sample;n1And n2Respectively
Indicate that optimization forms the number of effective positive and negative sample point to hypersphere in the process;P1And P2Then pass through following formula meter
It calculates and obtains:
S in formula (10), (11)1And S2Respectively indicate positive and negative samples point set;K (x, y) indicates used kernel function, here
Gaussian kernel function is referred to, as shown in formula (12);
7. the vein identification method according to claim 1 based on Riesz small echo and SSLM model, it is characterised in that:
In step 8), the important parameter using in above-mentioned SSLM model converts order in conjunction with Riesz and Decomposition order generates certain
The method of the different scale texture signature map of classification training sample is:
The hyperspherical centre of sphere C of the corresponding SSLM model of a certain classification is grouped according to scale, i.e., it will be in hyperspherical centre of sphere C
Every N+1 number is divided into one group, and as the signature coefficient under a scale, the hyperspherical centre of sphere by being grouped the category obtains J
The signature coefficient w of a scalej, j=1,2,3 ... J utilize formula (13) to obtain texture signature map of the category under jth scale:
W in formulajIndicate that the signature coefficient under jth scale, G (x) indicate Gaussian function.
8. the vein identification method according to claim 1 based on Riesz small echo and SSLM model, it is characterised in that:
It is described to be differentiated respectively using the feature parameter vectors of the decision function to all positive samples of a certain classification in step 9),
Then it will differentiate that result encodes, and is by the method that all positive sample encoded radios of all categories constitute coding schedule:
(1) any l finger vein images in the training sample of a certain classification are chosen first, and extract the energy of this l images respectively
Measure feature vector Eq, q=1,2,3 ... l;
(2) then utilize decision function shown in formula (14) successively to l energy of the category using the SSLM model of all categories
Measure feature vector EqDifferentiated respectively, differentiate that result is as follows:
Differentiate that input quantity x is the feature parameter vectors E that previous step is extracted in formula (14)q;BCiIndicate training sample EqRelatively
Belong to the i-th class as a result ,+1 expression differentiates in the differentiation of the i-th class SSLM model, -1 indicates that differentiation is not belonging to the i-th class;RiIndicate i-th
The radius of a ball of class SSLM model, Ci、αiRespectively indicate the corresponding centre of sphere of SSLM model, Lagrange coefficient, yiAnd xiIndicating should
The supporting vector of classification SSLM model;
(3) by the differentiation result BC of above-mentioned acquisition afteriIt is encoded, coding formula is as follows:
ECR in formulac(i) it indicates that the category in i-th of bits of coded, that is, corresponds to the coding situation of i-th of SSLM classifier, and grows
The ECR coding that degree is total for classification, is made of 0 and 1 is exactly the coding result of the category;Due to m1The coding result of a positive sample
It is possible that inconsistent situation, so by m1A coding result step-by-step progress and operation, using obtained result as such
Other encoded radio;
(4) finally the encoded radio of all categories is put into a table and generates coding schedule.
9. the vein identification method according to claim 1 based on Riesz small echo and SSLM model, it is characterised in that:
In step 10), it is described finger venous image to be detected is carried out knowing method for distinguishing using above-mentioned SSLM model and coding schedule be:
A finger venous image to be detected is chosen from database, then according to above-mentioned steps 2) method of-step 8) and obtain
Its encoded radio, encoded radio of all categories in the encoded radio and above-mentioned coding schedule is finally subjected to Euclidean distance solution, choose away from
Recognition result from the smallest classification as the finger venous image to be detected.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610808628.3A CN106407921B (en) | 2016-09-08 | 2016-09-08 | Vein identification method based on Riesz small echo and SSLM model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610808628.3A CN106407921B (en) | 2016-09-08 | 2016-09-08 | Vein identification method based on Riesz small echo and SSLM model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106407921A CN106407921A (en) | 2017-02-15 |
CN106407921B true CN106407921B (en) | 2019-05-03 |
Family
ID=57998967
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610808628.3A Expired - Fee Related CN106407921B (en) | 2016-09-08 | 2016-09-08 | Vein identification method based on Riesz small echo and SSLM model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106407921B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101093539A (en) * | 2007-07-27 | 2007-12-26 | 哈尔滨工程大学 | Matching identification method by extracting characters of vein from finger |
CN102890772A (en) * | 2011-07-21 | 2013-01-23 | 常熟安智生物识别技术有限公司 | Palm vein recognition technical scheme |
CN103870808A (en) * | 2014-02-27 | 2014-06-18 | 中国船舶重工集团公司第七一〇研究所 | Finger vein identification method |
CN104504370A (en) * | 2014-12-15 | 2015-04-08 | 广东智冠信息技术股份有限公司 | Finger vein recognition method combining bionic texture feature and linear texture feature |
CN104951774A (en) * | 2015-07-10 | 2015-09-30 | 浙江工业大学 | Palm vein feature extracting and matching method based on integration of two sub-spaces |
CN105184266A (en) * | 2015-09-14 | 2015-12-23 | 中国民航大学 | Finger vein image recognition method |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI536272B (en) * | 2012-09-27 | 2016-06-01 | 光環科技股份有限公司 | Bio-characteristic verification device and method |
US20150023572A1 (en) * | 2013-07-22 | 2015-01-22 | Rocky Williform | System and methods for providing finger vein authentication and signature for execution of electronic wallet transactions |
-
2016
- 2016-09-08 CN CN201610808628.3A patent/CN106407921B/en not_active Expired - Fee Related
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101093539A (en) * | 2007-07-27 | 2007-12-26 | 哈尔滨工程大学 | Matching identification method by extracting characters of vein from finger |
CN102890772A (en) * | 2011-07-21 | 2013-01-23 | 常熟安智生物识别技术有限公司 | Palm vein recognition technical scheme |
CN103870808A (en) * | 2014-02-27 | 2014-06-18 | 中国船舶重工集团公司第七一〇研究所 | Finger vein identification method |
CN104504370A (en) * | 2014-12-15 | 2015-04-08 | 广东智冠信息技术股份有限公司 | Finger vein recognition method combining bionic texture feature and linear texture feature |
CN104951774A (en) * | 2015-07-10 | 2015-09-30 | 浙江工业大学 | Palm vein feature extracting and matching method based on integration of two sub-spaces |
CN105184266A (en) * | 2015-09-14 | 2015-12-23 | 中国民航大学 | Finger vein image recognition method |
Non-Patent Citations (2)
Title |
---|
《手指静脉识别技术研究》;袁智;《中国优秀硕士学位论文全文数据库 信息科技辑》;20071015(第2007年第04期);全文 |
《静脉识别算法研究》;李铁钢;《中国博士学位论文全文数据库 信息科技辑》;20070915(第2007年第03期);全文 |
Also Published As
Publication number | Publication date |
---|---|
CN106407921A (en) | 2017-02-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106599854B (en) | Automatic facial expression recognition method based on multi-feature fusion | |
CN102938065B (en) | Face feature extraction method and face identification method based on large-scale image data | |
CN105718889B (en) | Based on GB (2D)2The face personal identification method of PCANet depth convolution model | |
Fallah et al. | A new online signature verification system based on combining Mellin transform, MFCC and neural network | |
CN107403084B (en) | Gait data-based identity recognition method | |
CN108681737B (en) | Method for extracting image features under complex illumination | |
Zalasiński et al. | New approach for the on-line signature verification based on method of horizontal partitioning | |
CN102156887A (en) | Human face recognition method based on local feature learning | |
CN105956570B (en) | Smiling face's recognition methods based on lip feature and deep learning | |
CN110516526A (en) | A kind of small sample target identification method based on Feature prototype metric learning | |
CN104504361B (en) | Palm vein principal direction feature extracting method based on direction character | |
CN101178767A (en) | Recognizing layer amalgamation for human face and iris mixed recognition | |
Ghafoor et al. | Fingerprint identification with shallow multifeature view classifier | |
Li et al. | A fingerprint indexing scheme with robustness against sample translation and rotation | |
Dubey et al. | Palmprint recognition using binary wavelet transform and LBP representation | |
CN103839047A (en) | Human motion track recognition method and device | |
CN106650685B (en) | Identity recognition method and device based on electrocardiogram signal | |
CN106407921B (en) | Vein identification method based on Riesz small echo and SSLM model | |
Saffar et al. | Online signature verification using deep representation: a new descriptor | |
CN110298159B (en) | Smart phone dynamic gesture identity authentication method | |
CN111950333B (en) | Electronic handwritten signature recognition method based on neural network | |
Sadeddine et al. | Sign language recognition using PCA, wavelet and neural network | |
Al-Rawi et al. | Feature Extraction of Human Facail Expressions Using Haar Wavelet and Neural network | |
Punyani et al. | Iris recognition system using morphology and sequential addition based grouping | |
Wang et al. | The research on footprint recognition method based on wavelet and fuzzy neural network |
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 | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190503 Termination date: 20190908 |