woodbe
commented
6 years ago I think we need to consider a "meta" toolbox that defines some of the expectations for some of the tools to be used. I say this because some of the same comments I would have for the face I have for the iris. Things like what kind of paper, the printers used, etc, should be fairly standardized, but I'm also thinking about things like photo paper I can readily pick up in any office supply store, department store or Amazon.
What I'm thinking is that we provide something that lists "common" tools, like paper, printers, cameras, etc, and then using those defined terms in the modality-specific toolbox we can have less to worry about in terms of repeat-ability and also copying and updating them in all places to keep them in sync.
BrJu
nils-tekampe
n-kai
Preliminary draft of Mobile toolbox for iris
This document is a preliminary draft of mobile toolbox for iris that lists presentation attack test items for mobile iris unlock. Each test item will be refined further to develop attack descriptions. Any comments on this document would be welcomed.
Introduction
Presentation attack to iris verification is classified into following categories [1]
a) Print attack Present a genuine iris image printed on a paper to the mobile device b) Textured contact lenses attack Attacker wears the textured contact lenses to interfere with iris verification c) Synthetic Iris Images attack Present a synthetic iris image created from target user’s iris code d) Artificial eye attack Present an artificial eye made of, for example, plastic with counterfeit iris pattern to the device
b) is indeed effective for obfuscating one’s true biometric trait. However, the obfuscating is out of scope of the cPP. In theory, there exist no technical limitations of printing fully opaque cosmetic contact lenses with iris texture of a targeted person and wearing the customized lenses to successfully masquerade as someone else to unlock the device. However, there are no known real-world examples of successful presentation attack of this kind [2] and there are also no research papers to study such kinds of attack. So, b) is out of scope of our evaluation at this moment. c) is also out of scope because we assume that iris code is securely protected by the mobile device. Therefore, this document lists test items related to a) and d).
Presentation attack test items
The following lists the presentation attack test items. Number of attempts for each PAI and pass criteria should be discussed later, referring the relevant studies (e.g. [10] shows the state of art PAD technique can achieves ACER=2.78).
a) Attack 1 This attack assumes that attacker can get digital image of target user’s face taken by normal digital camera with visible light.
Attack method Attacker gets [digital image of target user’s face] taken by [digital camera] under [environment] and [preprocess the image], then print out eyes on [paper] with [printer] to [present] it to the mobile device.
Text enclosed in [ ] should be refined further. The following shows my current idea.
[digital image of target user’s face] Digital image should meet the following condition [3a]:
[digital camera] High resolution digital camera should be used to get clear image of the iris [3b]
[environment] Controlled environment (the background of the scene is uniform, the light in the office is switched on and the window blinds are down)
[preprocess the image] For example,
[paper] Normal paper, mat paper etc [3b], [6]
[printer] Laser printer or inkjet printer [3b], [6]. At least 1200dpi [7].
[present] Present the paper under the controlled environment as the same way the user enrolls the iris
Note Evaluator should take a picture of the paper using NIR camera to confirm that the NIR image of the paper contain enough quality of iris pattern before testing.
b) Attack 2 This attack assumes that attacker can take a photo of target user’s face using a digital camera using NIR.
Attack method Attacker takes [digital image of target user’s face] using [digital camera] under [environment] and [preprocess the image], then print out eyes on [paper] with [printer] to [present] it to the mobile device.
Text enclosed in [ ] should be refined further. The following shows my current idea.
[digital image of target user’s face] Same as Attack 1
[digital camera] Normal digital camera that can take NIR image. Attacker can modify the camera to have its IR-blocking filter replaced with a glass plate to take a NIR photo [8], [9]
[environment] Outdoor environment with sunlight that include NIR. No need to use NIR LED because such attack is considered by Attack 3.
[preprocess the image] For example,
[paper] Same as Attack 1
[printer] Same as Attack 1
[present] Same as Attack 1
Note Camera’s auto exposure (AE) may not work in the NIR [9]. Evaluator should carefully choose the right one to take a good NIR image referring relevant information in the web
c) Attack 3 This attack assumes that attacker can take a photo of target user’s face using NIR cameras including commercial iris scanners.
Attack method Attacker takes [digital image of target user’s face] using [NIR camera] under [environment] and [preprocess the image], then print out eyes on [paper] with [printer] to [present] it to the mobile device.
Text enclosed in [ ] should be refined further. The following show my current idea.
[digital image of target user’s face] Same as Attack 1
[NIR camera] Commercial iris scanner [10]
[environment] Controlled environment
[preprocess the image] Same as Attack 2
[paper] Same as Attack 1
[printer] Same as Attack 1
[present] Present the paper under the controlled environment as the same way the user enrolls the iris. Attacker may place a contact lens over the iris image to give the iris scanner the curvature it's expecting from a real eye [8]
d) Attack 4 Same as Attack 3 except that attacker attach the captured iris pattern to the artificial eyes.
Attack method Attacker takes [digital image of target user’s face] using [NIR camera] under [environment] and [preprocess the image], then print out eyes on [paper] with [printer] and attach the iris image to [artificial eye] to [present] it to the mobile device.
[digital image of target user’s face] Same as Attack 3
[NIR camera] Same as Attack 3
[environment] Same as Attack 3
[preprocess the image] Same as Attack 3
[paper] Same as Attack 3
[printer] Same as Attack 3
[artificial eye] Create one referring [1]
[present] Same as Attack 3
Reference
[1] Learning Hierarchical Visual Codebook for Iris Liveness Detection http://www.csis.pace.edu/~ctappert/dps/2011IJCB/papers/226.pdf [2] Contact lens detection in iris images http://jultika.oulu.fi/files/nbnfi-fe201801222182.pdf [3a] Hacking Putin's Eyes: How To Bypass Biometrics The Cheap And Dirty Way With Google Images https://www.forbes.com/sites/thomasbrewster/2015/03/05/clone-putins-eyes-using-google-images/#35160438214a [3b] Robust Scheme for Iris Presentation Attack Detection Using Multiscale Binarized Statistical Image Features http://ieeexplore.ieee.org/document/7031897/ [4] Cross-spectral Iris Recognition for Mobile Applications using High-quality Color Images https://pdfs.semanticscholar.org/349f/a8805e9d5f7d4096fb605453a90877f15ae9.pdf [5] LivDet Iris 2017 - Iris Liveness Detection Competition 2017 http://iab-rubric.org/papers/Yambay-IJCB17-LiveDET.pdf [6] Database of Iris Printouts and its Application: Development of Liveness Detection Method for Iris Recognition http://zbum.ia.pw.edu.pl/PAPERS/MMAR_2013_Czajka.pdf [7] Spoofing iris recognition technology with pictures http://www.biometricupdate.com/201503/spoofing-iris-recognition-technology-with-pictures [8] Samsung Galaxy S8 Iris Scanner Hacked In Three Simple Steps https://www.forbes.com/sites/ianmorris/2017/05/23/samsung-galaxy-s8-iris-scanner-hacked-in-three-simple-steps/#5151feb4ccba [9] The physics of near-infrared photography http://www.montana.edu/jshaw/documents/NIR%20Photography%20-%20Mangold%20et%20al%20-%20EJP2013.pdf [10] Presentation attack detection using a generalizable statistical approach for periocular and iris systems https://dl.gi.de/bitstream/handle/20.500.12116/1244/111.pdf?sequence=1&isAllowed=y