CN102708362A - Iris recognition method - Google Patents

Iris recognition method Download PDF

Info

Publication number
CN102708362A
CN102708362A CN201210146181XA CN201210146181A CN102708362A CN 102708362 A CN102708362 A CN 102708362A CN 201210146181X A CN201210146181X A CN 201210146181XA CN 201210146181 A CN201210146181 A CN 201210146181A CN 102708362 A CN102708362 A CN 102708362A
Authority
CN
China
Prior art keywords
iris
formula
svms
lineoid
function
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201210146181XA
Other languages
Chinese (zh)
Inventor
张旭
徐昊
陈再新
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
NANJING GUANGHUA TECHNOLOGY DEVELOPMENT CO LTD
Original Assignee
NANJING GUANGHUA TECHNOLOGY DEVELOPMENT CO LTD
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by NANJING GUANGHUA TECHNOLOGY DEVELOPMENT CO LTD filed Critical NANJING GUANGHUA TECHNOLOGY DEVELOPMENT CO LTD
Priority to CN201210146181XA priority Critical patent/CN102708362A/en
Publication of CN102708362A publication Critical patent/CN102708362A/en
Pending legal-status Critical Current

Links

Images

Abstract

The invention provides an improved iris recognition method and aims to improve the accuracy of iris recognition and shorten the time of iris recognition. The method includes decomposing an iris image through wavelet transform to obtain low frequency sub-band information of the iris image, extracting texture features of the iris by utilizing two-dimensional log-Gabor filter banks in different directions and dimensions, and matching generated iris feature codes by utilizing a support vector machine. The iris recognition method is more suitable for real-time iris recognition.

Description

A kind of iris recognition method
Technical field
The present invention relates to the field of biometrics identification technology, specifically, is a kind of improved iris recognition method based on two-dimentional log-Gabor filtering.
Background technology
In altitude figure society, identification has been penetrated in the daily life.Shortcomings such as the method for traditional identity identification has the personation of being prone to, be prone to lose, be prone to forget can not satisfy the demand of present informationization society.In recent years, cause people's extensive concern based on the identity recognizing technology of biological characteristic, with respect to the traditional identity recognition technology, it has advantages such as stable, reliable, convenient and difficult forgery, has become the focus in the identification research.And in various biological identification technologies, iris has characteristics such as the property of collection, non-infringement property, uniqueness and stability, and ten years has in the past obtained broad research.
Iris recognition mainly comprises Iris Texture Features Extraction and two committed steps of iris feature sign indicating number coupling; Both are to the Iris Texture Features Extraction method of main employing Gabor wave filter; There is DC component in common Gabor wave filter; Maximum bandwidth limit influences the extraction effect of iris texture characteristic in the defective of 1 frequency multiplication.There was the scholar to make improvements afterwards, proposed one dimension log-Gabor and two-dimentional log-Gabor texture characteristic extracting method, efficiently solved common Gabor wave filter defective, improved the iris recognition rate.
The main modes such as Euclidean distance, Hamming distance, neural network that adopt of current iris feature sign indicating number coupling judge that whether two irises are from same individual.Wherein, Euclidean distance, Hamming distance realize easily, speed is fast, but recognition correct rate is low; Neural network requires sample big, and the cost cost is high, has difficult problems such as the complicated network structure, recognition efficiency be low simultaneously, is not suitable for real-time and requires high application.
(support vector machine, SVM) be a kind of small sample to SVMs, non-linear classification can very excellent sorting algorithm.
Therefore, provide the high method of identification of a kind of improved iris recognition rate real for necessary.
Summary of the invention
In order to improve the accuracy of iris recognition, the present invention has developed a kind of iris recognition method, and this method is implemented according to the following steps:
The extraction of step 1. iris texture characteristic:
A. wavelet decomposition: iris image is decomposed into horizontal high frequency (LH1), vertical high frequency (HL1), diagonal angle high frequency (HH1) and low frequency approaches (LL1) four sub-band images;
B. the multi-channel filter group is extracted: construct a plurality of wave filters, through improved two-dimentional log-Gabor filtering algorithm, low frequency is approached (LL1) sub-band images from radially extract the information of iris texture characteristic with the angle both direction, function expression is following:
The log-Gabor function definition is:
Figure 201210146181X100002DEST_PATH_IMAGE001
In the formula;
Figure 193319DEST_PATH_IMAGE002
is the centre frequency of wave filter;
Figure 201210146181X100002DEST_PATH_IMAGE003
is the direction of wave filter, and is the Gaussian function standard deviation;
Two dimension log-Gabor function definition is:
Figure 201210146181X100002DEST_PATH_IMAGE005
In the formula, U, V are respectively the component of two axles of radial center frequency;
The formula that extracts iris texture characteristic is:
Figure 363717DEST_PATH_IMAGE006
In the formula;
Figure 201210146181X100002DEST_PATH_IMAGE007
is for handling the back iris image;
Figure 614701DEST_PATH_IMAGE008
is convolution algorithm;
Figure 201210146181X100002DEST_PATH_IMAGE009
representes
Figure 97635DEST_PATH_IMAGE010
individual yardstick;
Figure 201210146181X100002DEST_PATH_IMAGE011
representes individual direction, and comprises amplitude information and phase information.
The coding of step 2. iris texture characteristic:
Through Gray code the phase information of each characteristic
Figure 496704DEST_PATH_IMAGE013
is encoded:
Figure 484251DEST_PATH_IMAGE014
In the formula,
Figure 201210146181X100002DEST_PATH_IMAGE015
is phase information;
The coupling of step 3. iris feature sign indicating number:
A. adopt SVMs to carry out iris feature sign indicating number coupling: given data set:
Figure 708559DEST_PATH_IMAGE016
; In the formula;
Figure 201210146181X100002DEST_PATH_IMAGE017
;
Figure 583105DEST_PATH_IMAGE018
; When two iris texture characteristics belong to same iris;
Figure 201210146181X100002DEST_PATH_IMAGE019
, otherwise
Figure 929773DEST_PATH_IMAGE020
;
SVMs optimal classification lineoid is:
Figure DEST_PATH_IMAGE021
; In the formula;
Figure 342300DEST_PATH_IMAGE022
is the normal vector of lineoid, and b is the offset vector of lineoid;
Introduce the Lagrange multiplier and accelerate classification speed, the lineoid classification function of SVMs is:
Figure DEST_PATH_IMAGE023
In the formula; is sign function, and is the Lagrange multiplier;
For the inseparable classification of linearity; SVMs replaces dot product
Figure DEST_PATH_IMAGE027
with kernel function
Figure 427247DEST_PATH_IMAGE026
, and the final classification function of SVMs is:
Figure 616920DEST_PATH_IMAGE028
.
B. construct based on the iris adaptation of SVMs: input iris texture characteristic collection; Select optimum kernel function and nuclear parameter; With textural characteristics collection standardization processing; Structure nuclear matrix
Figure DEST_PATH_IMAGE029
; Find the solution Lagrangian coefficient
Figure 329793DEST_PATH_IMAGE030
, support vector
Figure DEST_PATH_IMAGE031
and premium class lineoid coefficient
Figure 958220DEST_PATH_IMAGE032
; Set up iris texture and hold the optimizing decision lineoid of levying; Calculate the corresponding decision output valve of iris to be identified; Whether obtain iris from same people, the output recognition result.
The invention has the beneficial effects as follows: improved the accuracy rate of iris recognition, accelerated the speed of iris recognition, be more suitable for real-time iris recognition.
Description of drawings
Fig. 1 is the structural drawing of iris image wavelet decomposition.
Fig. 2 is the optimum lineoid of SVMs.
Fig. 3 is characteristic matching method and the comparison diagram of other characteristic matching method recognition correct rates among the present invention.
Embodiment
Below in conjunction with accompanying drawing the present invention is carried out detailed explanation:
A kind of iris recognition method, this method is implemented according to the following steps:
The extraction of step 1. iris texture characteristic:
A. wavelet decomposition: wavelet decomposition is a kind of method of multiresolution analysis, iris image is decomposed into horizontal high frequency (LH1), vertical high frequency (HL1), diagonal angle high frequency (HH1) and low frequency approaches (LL1) four sub-band images, and is specifically as shown in Figure 1;
B. the multi-channel filter group is extracted: construct a plurality of wave filters, through improved two-dimentional log-Gabor filtering algorithm, low frequency is approached (LL1) sub-band images from radially extract the information of iris texture characteristic with the angle both direction, function expression is following:
The log-Gabor function definition is:
Figure 259888DEST_PATH_IMAGE001
In the formula;
Figure 620463DEST_PATH_IMAGE002
is the centre frequency of wave filter;
Figure 820631DEST_PATH_IMAGE003
is the direction of wave filter, and
Figure 190432DEST_PATH_IMAGE004
is the Gaussian function standard deviation;
Two dimension log-Gabor function definition is:
Figure 81028DEST_PATH_IMAGE005
In the formula, U, V are respectively the component of two axles of radial center frequency;
The formula that extracts iris texture characteristic is:
Figure 878083DEST_PATH_IMAGE006
In the formula;
Figure 549235DEST_PATH_IMAGE007
is for handling the back iris image;
Figure 722728DEST_PATH_IMAGE008
is convolution algorithm; representes
Figure 514414DEST_PATH_IMAGE010
individual yardstick;
Figure 672863DEST_PATH_IMAGE011
representes
Figure 384467DEST_PATH_IMAGE012
individual direction, and
Figure 249655DEST_PATH_IMAGE013
comprises amplitude information and phase information;
The coding of step 2. iris texture characteristic:
Through Gray code the phase information of each characteristic
Figure 201562DEST_PATH_IMAGE013
is encoded:
Figure 784990DEST_PATH_IMAGE014
In the formula,
Figure 97023DEST_PATH_IMAGE015
is phase information.
The coupling of step 3. iris feature sign indicating number:
SVMs (SVM) is a kind of machine learning method based on the structural risk minimization principle; Its classification basic thought is an optimum lineoid of seeking the restrictive condition of a foot classification; Separate needing institute in the grouped data set to have a few; And make a little to the greatest extent with the lineoid distance as far as possible farthest, its synoptic diagram of classifying is as shown in Figure 2.
A. adopt SVMs to carry out iris feature sign indicating number coupling: the problem that whether comes from these two classification of same iris for iris texture characteristic; Given data set: ; In the formula;
Figure 205104DEST_PATH_IMAGE017
;
Figure 338145DEST_PATH_IMAGE018
; When two iris texture characteristics belong to same iris;
Figure 391552DEST_PATH_IMAGE019
, otherwise
Figure 231332DEST_PATH_IMAGE020
;
SVMs optimal classification lineoid is:
Figure 525041DEST_PATH_IMAGE021
; In the formula;
Figure 817482DEST_PATH_IMAGE022
is the normal vector of lineoid, and b is the offset vector of lineoid;
In order to make all the data point vectors and the distance between the optimum lineoid of training set maximum; Convert the double optimization problem to regard to it, promptly have:
Figure DEST_PATH_IMAGE033
In the formula;
Figure 736897DEST_PATH_IMAGE034
is the punishment parameter, and expression divides sample punishment degree to mistake;
SVMs is for the classification problem of large sample, and its pace of learning is very slow, is translated into its dual problem through introducing the Lagrange multiplier, finds the solution through dual problem, accelerates classification speed, and the lineoid classification function changes:
Figure 431183DEST_PATH_IMAGE023
In the formula;
Figure 895794DEST_PATH_IMAGE024
is sign function, and
Figure 941110DEST_PATH_IMAGE025
is the Lagrange multiplier;
For the inseparable classification problem of linearity; SVMs replaces dot product through kernel function
Figure DEST_PATH_IMAGE035
, therefore supports to the final classification function of reason machine to be:
Figure 947429DEST_PATH_IMAGE028
；
B. construct based on the iris adaptation of SVM:
The steps include:
(1). the input iris texture is held collection;
(2). select optimum kernel function and nuclear parameter, texture is held collection carry out the Gui Fanhua processing, it is defined in the selected kernel function claimed range;
(3). structure nuclear matrix ; Find the solution Lagrangian coefficient , support vector
Figure 115553DEST_PATH_IMAGE031
and premium class lineoid coefficient
Figure 314454DEST_PATH_IMAGE038
;
(4). set up iris texture and hold the optimizing decision lineoid of levying, accomplish training;
(5). calculate the corresponding decision output valve of iris to be identified, whether obtain iris from same people, and the output recognition result.
Discrimination through emulation testing experimental verification the inventive method below:
A. contrast the accuracy of different characteristic extraction method: carry out the emulation testing experiment with the UBIRIS iris image database at CASIA 1.0; Adopting single wavelet decomposition, two-dimentional log-Gabor filtering and the two combined techniques of the present invention to carry out iris feature extracts; And adopt SVMs to carry out the characteristic matching algorithm; Evaluation index is correct recognition rata and error recognition rate, and their comparing results are shown in table 1 and table 2.
Table 1 is the CASIA library test result of various feature extracting methods:
Feature extracting method Recognition correct rate (%) Identification error rate (%)
Wavelet decomposition 85.85 14.15
log-Gabor 91.04 8.96
Wavelet decomposition and two-dimentional log-Gabor filtering combine 94.27 5.73
Table 2 is UBIRIS library test results of various feature extracting methods:
Feature extracting method Recognition correct rate (%) Identification error rate (%)
Wavelet decomposition 87.15 12.75
log-Gabor 90.77 9.23
Wavelet decomposition and two-dimentional log-Gabor filtering combine 95.08 4.92
Can know from the comparing result of table 1 and table 2; In three kinds of feature extracting methods, recognition correct rate of the present invention is the highest relatively, relative minimum of the error rate of identification; The feature extracting method that adopts wavelet decomposition and log-Gabor to combine is described, can be obtained better iris texture characteristic.
B. contrast the accuracy of different characteristic matching method: adopt Hamming distance, BP neural network and SVMs of the present invention to carry out the iris feature coupling, test result is as shown in Figure 3.
Can know that from figure under the same characteristic features method for distilling, SVMs matching method recognition correct rate of the present invention is the highest, shows that the performance of SVMs matching method of the present invention is more excellent, can effectively improve the effect of iris recognition, shorten match time, with the obvious advantage.

Claims (1)

1. an iris recognition method is characterized in that, this method is implemented according to the following steps:
The extraction of step 1, iris texture characteristic:
Wavelet decomposition: iris image is decomposed into horizontal high frequency (LH1), vertical high frequency (HL1), diagonal angle high frequency (HH1) and low frequency approaches (LL1) four sub-band images;
The multi-channel filter group is extracted: construct a plurality of wave filters, through improved two-dimentional log-Gabor filtering algorithm, low frequency is approached (LL1) sub-band images from radially extract the information of iris texture characteristic with the angle both direction, function expression is following:
The log-Gabor function definition is:
Figure 761015DEST_PATH_IMAGE001
；
In the formula;
Figure 89359DEST_PATH_IMAGE002
is the centre frequency of wave filter;
Figure 809053DEST_PATH_IMAGE003
is the direction of wave filter, and
Figure 118812DEST_PATH_IMAGE004
is the Gaussian function standard deviation;
Two dimension log-Gabor function definition is:
Figure 251853DEST_PATH_IMAGE005
；
In the formula, U, V are respectively the component of two axles of radial center frequency;
The formula that extracts iris texture characteristic is:
；
In the formula;
Figure 145039DEST_PATH_IMAGE007
is for handling the back iris image;
Figure 625699DEST_PATH_IMAGE008
is convolution algorithm; representes
Figure 588287DEST_PATH_IMAGE010
individual yardstick; representes
Figure 199714DEST_PATH_IMAGE012
individual direction, and
Figure 41768DEST_PATH_IMAGE013
comprises amplitude information and phase information;
The coding of step 2, iris texture characteristic:
Through Gray code the phase information of each characteristic
Figure 436978DEST_PATH_IMAGE013
is encoded:
；
In the formula,
Figure 73812DEST_PATH_IMAGE015
is phase information;
The coupling of step 3, iris feature sign indicating number:
A. adopt SVMs to carry out iris feature sign indicating number coupling: given data set:
Figure 153895DEST_PATH_IMAGE016
; In the formula; ;
Figure 21674DEST_PATH_IMAGE018
; When two iris texture characteristics belong to same iris;
Figure 77355DEST_PATH_IMAGE019
, otherwise
Figure 831684DEST_PATH_IMAGE020
;
SVMs optimal classification lineoid is:
Figure 568696DEST_PATH_IMAGE021
; In the formula;
Figure 92081DEST_PATH_IMAGE022
is the normal vector of lineoid, and b is the offset vector of lineoid;
Introduce the Lagrange multiplier and accelerate classification speed, the lineoid classification function of SVMs is:
Figure 334975DEST_PATH_IMAGE023
;
In the formula;
Figure 576600DEST_PATH_IMAGE024
is sign function, and
Figure 851724DEST_PATH_IMAGE025
is the Lagrange multiplier;
For the inseparable classification of linearity; SVMs replaces dot product
Figure 892678DEST_PATH_IMAGE027
with kernel function
Figure 495195DEST_PATH_IMAGE026
, and the final classification function of SVMs is: ;
B. construct based on the iris adaptation of SVMs: input iris texture characteristic collection, select optimum kernel function and nuclear parameter, with textural characteristics collection standardization processing, the structure nuclear matrix
Figure 700414DEST_PATH_IMAGE029
, find the solution Lagrangian coefficient
Figure 11441DEST_PATH_IMAGE030
, support vector
Figure 783088DEST_PATH_IMAGE031
And premium class lineoid coefficient
Figure 733726DEST_PATH_IMAGE032
, set up iris texture and hold the optimizing decision lineoid of levying, calculate the corresponding decision output valve of iris to be identified, whether obtain iris from same people, the output recognition result.
CN201210146181XA 2012-05-14 2012-05-14 Iris recognition method Pending CN102708362A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201210146181XA CN102708362A (en) 2012-05-14 2012-05-14 Iris recognition method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201210146181XA CN102708362A (en) 2012-05-14 2012-05-14 Iris recognition method

Publications (1)

Publication Number Publication Date
CN102708362A true CN102708362A (en) 2012-10-03

Family

ID=46901107

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201210146181XA Pending CN102708362A (en) 2012-05-14 2012-05-14 Iris recognition method

Country Status (1)

Country Link
CN (1) CN102708362A (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104537648A (en) * 2014-12-15 2015-04-22 深圳云派思科技有限公司 Fusing and encoding method for vein bionic texture feature and linear texture feature
CN104751150A (en) * 2015-04-21 2015-07-01 南京安穗智能科技有限公司 Method for recognizing iris on basis of odd-symmetry 2D (two-dimensional) Log-Gabor filter and Adaboost combinations
CN106326827A (en) * 2015-11-08 2017-01-11 北京巴塔科技有限公司 Palm vein recognition system
CN106778535A (en) * 2016-11-28 2017-05-31 北京无线电计量测试研究所 A kind of iris feature based on WAVELET PACKET DECOMPOSITION is extracted and matching process
CN107194260A (en) * 2017-04-20 2017-09-22 中国科学院软件研究所 A kind of Linux Kernel association CVE intelligent Forecastings based on machine learning
CN107256385A (en) * 2017-05-22 2017-10-17 西安交通大学 Infrared iris Verification System and method based on 2D Log Gabor Yu composite coding method
CN112560893A (en) * 2020-11-13 2021-03-26 贝壳技术有限公司 Picture texture matching method and device, electronic medium and storage medium

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
党宁娜: "虹膜识别技术的研究", 《中国优秀博硕士学位论文全文数据库（硕士） 信息科技辑》 *
卢珊，赵强，刘丽萍: "虹膜识别算法的应用研究", 《计算机仿真》 *
周治平，李雨淞，吴会军: "一种改进的Log-Gabor滤波和SVM的虹膜识别方法", 《中国图象图形学报》 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104537648A (en) * 2014-12-15 2015-04-22 深圳云派思科技有限公司 Fusing and encoding method for vein bionic texture feature and linear texture feature
CN104751150A (en) * 2015-04-21 2015-07-01 南京安穗智能科技有限公司 Method for recognizing iris on basis of odd-symmetry 2D (two-dimensional) Log-Gabor filter and Adaboost combinations
CN106326827A (en) * 2015-11-08 2017-01-11 北京巴塔科技有限公司 Palm vein recognition system
CN106326827B (en) * 2015-11-08 2019-05-24 北京巴塔科技有限公司 Palm vein identification system
CN106778535A (en) * 2016-11-28 2017-05-31 北京无线电计量测试研究所 A kind of iris feature based on WAVELET PACKET DECOMPOSITION is extracted and matching process
CN107194260A (en) * 2017-04-20 2017-09-22 中国科学院软件研究所 A kind of Linux Kernel association CVE intelligent Forecastings based on machine learning
CN107256385A (en) * 2017-05-22 2017-10-17 西安交通大学 Infrared iris Verification System and method based on 2D Log Gabor Yu composite coding method
CN112560893A (en) * 2020-11-13 2021-03-26 贝壳技术有限公司 Picture texture matching method and device, electronic medium and storage medium

Similar Documents

Publication Publication Date Title
CN102708362A (en) Iris recognition method
CN102521575B (en) Iris identification method based on multidirectional Gabor and Adaboost
CN107609497B (en) Real-time video face recognition method and system based on visual tracking technology
Nguyen et al. Leaf based plant identification system for android using surf features in combination with bag of words model and supervised learning
Cruz et al. Handwritten digit recognition using multiple feature extraction techniques and classifier ensemble
CN104036289A (en) Hyperspectral image classification method based on spatial and spectral features and sparse representation
CN102750526A (en) Identity verification and recognition method based on face image
Cruz et al. An ensemble classifier for offline cursive character recognition using multiple feature extraction techniques
CN102722699A (en) Face identification method based on multiscale weber local descriptor and kernel group sparse representation
CN1908960A (en) Feature classification based multiple classifiers combined people face recognition method
Dash et al. Handwritten numeral recognition using non‐redundant stockwell transform and bio‐inspired optimal zoning
CN102542243A (en) LBP (Local Binary Pattern) image and block encoding-based iris feature extracting method
CN101968850A (en) Method for extracting face feature by simulating biological vision mechanism
CN106778512A (en) Face identification method under the conditions of a kind of unrestricted based on LBP and depth school
CN113139512B (en) Depth network hyperspectral image classification method based on residual error and attention
CN104091181A (en) Injurious insect image automatic recognition method and system based on deep restricted Boltzmann machine
CN102708364A (en) Cascade-classifier-based fingerprint image classification method
CN108805027B (en) Face recognition method under low resolution condition
Zhong et al. Palmprint and dorsal hand vein dualmodal biometrics
Cruz et al. Feature representation selection based on classifier projection space and oracle analysis
CN102768732A (en) Face recognition method integrating sparse preserving mapping and multi-class property Bagging
CN104156628A (en) Ship radiation signal recognition method based on multi-kernel learning and discriminant analysis
Sethy et al. Off-line Odia handwritten character recognition: an axis constellation model based research
CN1209732C (en) Iris identifying method based on ripple analysis and zero passage description
CN103246877A (en) Image contour based novel human face recognition method

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C02 Deemed withdrawal of patent application after publication (patent law 2001)
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20121003