L Iride. Docente: Michele Nappi. biplab.unisa.it

Size: px

Start display at page:

Download "L Iride. Docente: Michele Nappi. biplab.unisa.it"

Joy Davis
5 years ago
Views:

1 L Iride Docente: Michele Nappi biplab.unisa.it

2 AGENDA 1. Biometric Overview 2. IRIS and Daugman 3. NICE I: Iris Segmentation (Detection) 4.NICE II: Iris Coding and Matching (Recognition) 5. Datasets and Evaluations 6.Multibiometric Involving Iris 18/05/ Conclusions

3 Some Desirable Properties 18/05/2016 3

4 Some Desirable Properties (cont) 18/05/2016 4

5 Biometric Overview

6 Comparison of biometric techniques 6

7 AGENDA 1. Biometric Overview 2. IRIS and Daugman 3. NICE I: Iris Segmentation (Detection) 4. NICE II: Iris Coding and Matching (Recognition) 5. Dataset Evaluations 6. MICHE: Mobile Iris Challenge Evaluation 7. Multibiometric Involving Iris 8. Conclusions 18/05/2016

Iris The iris is a muscle membrane of the eye, of variable color, with both shape and function of a diaphragm It is pigmented, located posterior to the cornea and in

It consists of a flat layer of muscle fibers which circularly surround the pupil, a thin layer of smooth muscle fibers by means of which the pupil is dilated (thereby

8 Iris The iris is a muscle membrane of the eye, of variable color, with both shape and function of a diaphragm It is pigmented, located posterior to the cornea and in front of the lens, and is perforated by pupil. It consists of a flat layer of muscle fibers which circularly surround the pupil, a thin layer of smooth muscle fibers by means of which the pupil is dilated (thereby regulating the amount of light that enters the eye) and posteriorly by two layers of epithelial pigmented cells Iris colour, regular texture (mostly by furrows) and irregular patterns (e.g., freckles and crypts) provide a very high level of discrimination, which is comparable to fingerprints

Iris Pros Iris is visible yet well protected It is a time invariant and

Acquisition: near infrared and visible wavelenghts Cons Iris surface is very

64 cm2 A good acquisition requires a distance of less than one meter to

9 Iris Pros Iris is visible yet well protected It is a time invariant and extremely distinguishing trait Its image can be acquired without direct contact Acquisition: near infrared and visible wavelenghts Cons Iris surface is very limited: only about 3.64 cm2 A good acquisition requires a distance of less than one meter to guarantee a sufficient resolution, depending on the input device. Near Infrared (NIR: nm) Visible Wavelenghts 18/05/2016 Visione delle Macchine 2012

10 Sharbat Gula: The Afghan Monna Lisa 18/05/2016

11 18/05/2016 The Afghan Monna Lisa

12 John Daugman and the Eyes (Iris) of Sharbat Gula Left eye: HD=0.24; Right eye: HD=0.31 If the HD (Hamming distance) is < 0.33 the chances of the two codes coming from different irises is 1 in 2.9 million

13 Processing Phases on the Periocular Region The presence of a number of noisy elements requires a good preprocessing/segmentation eyelashes

14 Segmentation Normalization Coding Processing phases Matching

15 The first and most famous: Daugman Template db Acquisition of iris image Iris location and unwrapping Iris feature extraction and coding Matching Result

16 Daugman: Iris Location The approach uses a kind of circular edge detector to localize both the pupil and the iris (integro-differential operator) The operator exploits the convolution of the image with a Gaussian smoothing function with center r 0 and standard deviation (1) The operator looks for a circular path along which pixel variation is maximized, by varying the center r and radius (x0, y0) of a candidate circular contour (2) * Is the convolution operator When the candidate circle has the same radius and center of the iris, the operator should provide a peak

17 Daugman: iris and eyelids location

18 General Iris Segmentation From

Daugman: iris unwrapping Polar coordinate make iris processing simpler (circular bands become horizontal stripes, the overall iris annulus becomes a rectangle) Determining the right centre for the

19 Daugman: iris unwrapping Polar coordinate make iris processing simpler (circular bands become horizontal stripes, the overall iris annulus becomes a rectangle) Determining the right centre for the polar coordinates is of paramount importante but pupil and iris are not perfectly concentric and size of the pupil can change due to illumination or pathological conditions (drunk or drugs) Gaze direction can change the relative positions of sclera, iris and pupil It is necessary to devise a normalization procedure: Rubber Sheet Model

20 Rubber Sheet Model

Daugman: feature extraction Feature are extracted by applying Gabor filters to the I(r, ) image in polar coordinates (r, ) is the position, and are the filter dimensions and its frequency Complex

21 Daugman: feature extraction Feature are extracted by applying Gabor filters to the I(r, ) image in polar coordinates (r, ) is the position, and are the filter dimensions and its frequency Complex pixel For each element with coordinates (r, ) in the image I(, ), the method computes a pair of bits as follows are discretized to obtain a 256 byte code, plus a mask of the same size to identify valid iris elements

22 Iris code Matching: Hamming distance Matching: Hamming distance with mask

23 AGENDA 1. Biometric Overview 2. IRIS and Daugman 3. NICE I: Iris Segmentation (Detection) 4. NICE II: Iris Coding and Matching (Recognition) 5. Dataset Evaluations 6. MICHE: Mobile Iris Challenge Evaluation 7. Multibiometric Involving Iris 8. Conclusions 18/05/2016

24 NICE: Noisy Iris Challenge Evaluation Iris biometric evaluation initiative that received worldwide participations Two phases: NICE.I evaluated iris segmentation and noise detection techniques NICE.II evaluated encoding and matching strategies for biometric signatures. 18/05/2016

25 NICE I Contest Chairs: NICE I Committees Hugo Proença, Department of Computer Science, SOCIA Lab., IT-Networks and Multimedia Group, University of Beira Interior. Luís A. Alexandre, Department of Computer Science, SOCIA Lab., IT-Networks and Multimedia Group, University of Beira Interior. Organizing Committee: David Carvalho, Department of Computer Science, University of Beira Interior. João Oliveira, Department of Computer Science, SOCIA Lab., University of Beira Interior. Ricardo Santos, Department of Computer Science, SOCIA Lab., University of Beira Interior. Sílvio Filipe, Department of Computer Science, SOCIA Lab., University of Beira Interior. 18/05/2016

26 NICE I (2) NICE I received a total of 97 participants from over 22 countries. September 30 th, 2008: The deadline for the final submission of the participations October 15 th, 2008: The classification of the best participants is available. 18/05/2016

27 NICE I (3) The UBIRIS databases were developed by the SOCIA Lab. (Soft Computing and Image Analysis Group) of the University of Beira Interior (Portugal). They contain visible wavelength iris images captured in heterogeneous lighting conditions, which leaded to the appearance of highly degraded images. The imaging framework used in the acquisition of the UBIRIS data set was installed in a lounge under both natural and artificial lighting sources. 18/05/2016

28 NICE I (4) A large majority of the volunteers were: Latin Caucasian (approximately 90%) black (8%) Asian (2%). Approximately 60% of the volunteers participated in both imaging sessions, whereas 40% participated exclusively in one or the other. 18/05/2016

29 NICE I (5) Several marks placed on the floor between three and ten meters away from the acquisition device Two distinct acquisition sessions performed each lasting two weeks and separated by an interval of one week. From the first to the second session, both the location and orientation of the acquisition device and artificial light sources were changed. 18/05/2016

30 18/05/2016 NICE I (6)

NICE I (7) The Protocol The submitted application executable can be written in any programming language and must run in standalone mode, in one of the operating systems: Windows XP, Service Pack

31 NICE I (7) The Protocol The submitted application executable can be written in any programming language and must run in standalone mode, in one of the operating systems: Windows XP, Service Pack 2 or Fedora Core 6. There will be no internet access during the NICE.I evaluation. Thus, the application executable will need to be installed and executed without access to the internet. 18/05/2016

32 NICE I (8) Evaluation 1. Let Alg denote the submitted executable, which performs the segmentation of the noise free regions of the iris. 2. Let I={I 1,,I n } be the data set containing the input close-up iris images. 3. Let O={O 1,,O n } be the output images correspondent to the above described inputs, such that Alg(I i )=O i. 4. Let C={C 1,,C n } be the manually classified binary iris images, given by the NICE.I Organizing Committee. It must be assumed that each C i contains the perfect iris segmentation and noise detection result for the input image I i. 5. All the images of I, O and C have the same dimensions: c columns and r rows. 6. The classification error rate (E 1 ) of the Alg participation on the input image I i (E i ) is given by the proportion of correspondent disagreeing pixels (through the logical exclusive-or operator) over all the image:: 18/05/2016

NICE I (9) The best 8 participants, that achieved the lowest test error rates were invited to publish their approach in a Special Issue on the Segmentation of

33 NICE I (9) The best 8 participants, that achieved the lowest test error rates were invited to publish their approach in a Special Issue on the Segmentation of Visible Wavelength Iris Images Captured At-a-distance and On-the-move Image, Elsevier Image and Vision Computing 28 (2010) 18/05/2016 Visione delle Macchine 2012

34 NICE I: the winning algorithm Casia Algorithm by T. Tan, Z. He and Z. Sun (the Chinese Academy of Sciences)

35 NICE I: the winning algorithm (2)

36 AGENDA 1. Biometric Overview 2. IRIS and Daugman 3. NICE I: Iris Segmentation (Detection) 4. NICE II: Iris Coding and Matching (Recognition) 5. Dataset Evaluations 6. MICHE: Mobile Iris Challenge Evaluation 7. Multibiometric Involving Iris 8. Conclusions 18/05/2016

37 NICE II Contest Chairs: NICE II Committees Hugo Proença, Department of Computer Science, SOCIA Lab., IT-Networks and Multimedia Group, University of Beira Interior. Luís A. Alexandre, Department of Computer Science, SOCIA Lab., IT-Networks and Multimedia Group, University of Beira Interior. Organizing Committee: David Carvalho, Department of Computer Science, University of Beira Interior. João Oliveira, Department of Computer Science, SOCIA Lab., University of Beira Interior. Gil Santos, Department of Computer Science, SOCIA Lab., University of Beira Interior. Sílvio Filipe, Department of Computer Science, SOCIA Lab., University of Beira Interior. 18/05/2016 Visione delle Macchine 2012

38 NICE II (2) NICE II received a total of 67 participants from over 32 countries. June 30 th, 2010: The deadline for the final submission of the participations July 15 th, 2010: The classification of the best participants is available. 18/05/2016 Visione delle Macchine 2012

39 NICE II (3)

40 NICE II (4) The Protocol The submitted application executable can be written in any programming language and must run in standalone mode, in one of the operating systems: Windows XP, Service Pack 2 or Fedora Core 6. 18/05/2016 There will be no internet access during the NICE.II evaluation. Thus, the application executable will need to be installed and executed without access to the internet.

41 18/05/2016 NICE II (5)

42 NICE II (6) The best 8 participants, that achieved the the best 8 results were invited to publish their approach in a Special Issue NICE II NOISY IRIS CHALLENGE EVALUATION PART II, Pattern Recognition Letters, (Elsevier),vol 33, n 8, /05/2016

43 NICE II: the winning algorithm (2)

44 AGENDA 1. Biometric Overview 2. IRIS and Daugman 3. NICE I: Iris Segmentation (Detection) 4. NICE II: Iris Coding and Matching (Recognition) 5. Dataset Evaluations 6. MICHE: Mobile Iris Challenge Evaluation 7. Multibiometric Involving Iris 8. Conclusions 18/05/2016

45 Comparison between the Receiver Operating Characteristics curves obtained by the best participants in the NICE II contest

46 NICE II: Sensitiveness to Segmentation

47 Correlation and Fusion of Results in the NICE II contest 1: CASIA 2: NU 3: UBI 4: BERC 5:Peihuag 6: Biplab 7: HLJUCS 8: TUL

48 AGENDA 1. Biometric Overview 2. IRIS and Daugman 3. NICE I: Iris Segmentation (Detection) 4. NICE II: Iris Coding and Matching (Recognition) 5. Dataset Evaluations 6. MICHE: Mobile Iris Challenge Evaluation 7. Multibiometric Involving Iris 8. Conclusions 18/05/2016

MICHE: Mobile Iris Challenge Evaluation The new frontier for iris recognition: mobile Detection,segmentation,coding,and matching are still quite challenging for iris recognition,in general,and for

49 MICHE: Mobile Iris Challenge Evaluation The new frontier for iris recognition: mobile Detection,segmentation,coding,and matching are still quite challenging for iris recognition,in general,and for their successful embedding on mobile devices, in particular. MICHE reflects on such challenges and promotes reproducible research so one can compare methods and assess progress. Data Fusion and Interoperability Samsung S4, IPhone5, Galaxy Tablet II

50 MICHE: Mobile Iris Challenge Evaluation (2)

51 MICHE: Mobile Iris Challenge Evaluation (3)

52 MICHE: Mobile Iris Challenge Evaluation (4)

53 AGENDA 1. Biometric Overview 2. IRIS and and Daugman 3. NICE I: Iris Segmentation (Detection) 4. NICE II: Iris Coding and Matching (Recognition) 5. Dataset Evaluations 6. MICHE: Mobile Iris Challenge Evaluation 7. Multibiometric Involving Iris 8. Conclusions 18/05/2016

54 FIRME: Face and Iris Recognition for Mobile Engagement Biometric application based on a multimodal recognition of face and iris Designed to be embedded in mobile devices Modular architecture Fusion of face and iris information at matching level Framework: OpenCV libraries for Android Application Scenario: Mobile Banking

55 FIRME Architecture

56 FIRME: Localization of Reference Points

57 FIRME: Detection of a Real Live User

58 FIRME: Best Sample Selection

FIRME: Communication between the mobile device and the Web Banking System 2 different protocols Both the acquisition of the image and the generation of the corresponding biokey are executed on the

59 FIRME: Communication between the mobile device and the Web Banking System 2 different protocols Both the acquisition of the image and the generation of the corresponding biokey are executed on the mobile device; in this case, after spoofing detection and best template selection, a verification operation is rather required, which only o entails a 1:1 matching; The acquisition of the image, spoofing detection, best template election and feature extraction are all carried out on the mobile device, afterwards the biokey is sent to the server that performs an identification operation (1:N matching) to recognize the user.

60 FIRME: Performance of face and iris biometrics with indoor probes

61 FIRME: Performance of face and iris biometrics with outdoor probes

62 AGENDA 1. Biometric Overview 2. IRIS and Daugman 3. NICE I: Iris Segmentation (Detection) 4. NICE II: Iris Coding and Matching (Recognition) 5. Dataset Evaluations 6. MICHE: Mobile Iris Challenge Evaluation 7. Multibiometric Involving Iris 8. Conclusions 18/05/2016

63 Conclusions and Trends Due to its potential use in very different settings, iris is the most promising biometric Performing in uncontrolled scenario and mobile device Operating in the near infrared spectrum and in the visible wavelength IrisCode is computationally efficient Data Fusion (periocular and other biometrics) How improve Iris recognition Performance Iris Lifetime Stability (??)

64 Some Readings and Links H. Proença, L. A. Alexandre (2012). Toward Covert Iris Biometric Recognition: Experimental Results From the NICE Contests. IEEE Transactions on Information Forensics and Security, vol. 7, no. 2, pp M. De Marsico, M. Nappi, D. Riccio (2012). Noisy Iris Recognition Integrated Scheme. Pattern Recognition Letters, vol. 33, no. 8, pp

IRIS recognition II. Eduard Bakštein,

IRIS recognition II. Eduard Bakštein, edurard.bakstein@fel.cvut.cz 22.10.2013 acknowledgement: Andrzej Drygajlo, EPFL Switzerland Iris recognition process Input: image of the eye Iris Segmentation Projection