Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

How to Teach an Iris Scanner That the Eye It’s Looking at Is Dead

spectrum.ieee.org

Post reply on HN
Print view View original

41–50 of 54 posts

Re: How to Teach an Iris Scanner That the Eye It’s Looking at Is Dead

#41

Post by bsenftner »

tfigment wrote: post #34
dao- wrote: post #31

Earlier quoted context omitted.

> Also, another thing is that they should make sure that they train their system with dark eyed people. Often the iris is very easy to identify if people with light colored eyes. In people with dark eyes, it may be more difficult. Surely this is a concern (and presumably a solved problem) for any iris scanner, regardless of whether it tries to detect if the eye responds to stimuli.

You would think but its a valid concern. Facial recognition systems tend to fail more often with dark skinned people and I would argue that should have been considered the same way.

The mainstream press coverage of FR skin color failures has caused the FR industry to significantly enhance their algorithms. Any press still harping on this is beating a dead horse - or in this case a revived horse.

Re: How to Teach an Iris Scanner That the Eye It’s Looking at Is Dead

#42

Post by hutzlibu »

finnthehuman wrote: post #17
ebg13 wrote: post #2

> The method we developed images the eye multiple times to see whether it’s actually responding to changes in brightness This is new? In 2019? Really? I find that so hard to believe. > the process takes about 3 seconds WHAAAAT? WHY? > For comparison, even iris-recognition systems that require just one snapshot usually take about that much time to make an identification. I feel like Robin Williams fresh out of Jumanji…

>This is new? In 2019? Really? I find that so hard to believe. Yeah, sure, why? Who needs that level of security? Governments that can already afford to pay someone to stand by that door if it's so important?

But it is pretty expensive, if you have many doors to protect and also those humans sometimes turn out as unreliable .. so you need to pay people to check on them, etc.

(I worked briefly in security ... well, humans are really unreliable)

Re: How to Teach an Iris Scanner That the Eye It’s Looking at Is Dead

#43

Post by 3pt14159 »

RcouF1uZ4gsC wrote: post #5

> The method we developed images the eye multiple times to see whether it’s actually responding to changes in brightness Hopefully the person does not have Argyll Robertson pupils https://en.wikipedia.org/wiki/Argyll_Robertson_pupil which is seen in late stage syphilis and causes the pupil not to react to the light. Also, another thing is that they should make sure that they train their system with dark eyed people.…

Some people also lack eyes altogether, so these systems need a fallback either way, no?

Re: How to Teach an Iris Scanner That the Eye It’s Looking at Is Dead

#44

Post by meddlin »

JulianMorrison wrote: post #28
RcouF1uZ4gsC wrote: post #5

> The method we developed images the eye multiple times to see whether it’s actually responding to changes in brightness Hopefully the person does not have Argyll Robertson pupils https://en.wikipedia.org/wiki/Argyll_Robertson_pupil which is seen in late stage syphilis and causes the pupil not to react to the light. Also, another thing is that they should make sure that they train their system with dark eyed people.…

Surely at that point it's correctly detecting they're "mostly dead"?

Ah, you would need to blave the system...and as we all know "to blave" means "to bluff".

Re: How to Teach an Iris Scanner That the Eye It’s Looking at Is Dead

#46

Post by Qwertystop »

tyingq wrote: post #38
ebg13 wrote: post #2

> The method we developed images the eye multiple times to see whether it’s actually responding to changes in brightness This is new? In 2019? Really? I find that so hard to believe. > the process takes about 3 seconds WHAAAAT? WHY? > For comparison, even iris-recognition systems that require just one snapshot usually take about that much time to make an identification. I feel like Robin Williams fresh out of Jumanji…

Apparently pupils don't adjust terribly quickly: "Briefly, it was found that when light is flashed on a normal eye that is accommodated for the dark, there occurs a latent period of 0.1875 second before the pupil begins to contract. Then there follows a rapid primary contraction for 0.4365 second at the rate of 5.48 mm. per second. This is succeeded by a secondary contraction of 0.3125 second at the slower rate of 1.…

Well, the three times there make about a second for one light-change. So let's say it makes two light-changes, to be sure, for three images (it does say "multiple images", not "two"). That's two seconds, and then say there's another second for processing. Not too unreasonable, especially if (as it says) single-image systems also take about three seconds overall.

Re: How to Teach an Iris Scanner That the Eye It’s Looking at Is Dead

#48

Post by swyx »

haolez wrote: post #47

If I remember correctly, the eyes of live persons are always twitching a little. It's relevant when, for example, a device needs to take a picture from your retina. Maybe this should be used as an heuristic as well.

saccade https://en.m.wikipedia.org/wiki/Saccade

Re: How to Teach an Iris Scanner That the Eye It’s Looking at Is Dead

#50

Post by klyrs »

bsenftner wrote: post #41
tfigment wrote: post #34

Earlier quoted context omitted.

You would think but its a valid concern. Facial recognition systems tend to fail more often with dark skinned people and I would argue that should have been considered the same way.

The mainstream press coverage of FR skin color failures has caused the FR industry to significantly enhance their algorithms. Any press still harping on this is beating a dead horse - or in this case a revived horse.

That progress is not made all at once. It's a problem with a known fix; but just like buffer overflows, a known fix doesn't always get implemented everywhere. For example, this 2019 paper:

https://arxiv.org/pdf/1902.11097.pdf

Post reply on HN