[tsk1979]

Originally Posted by Samurai

Imagine the CPU power required to compare a set of fingerprints to to rest of the billion fingerprints to detect a duplicate.

Well, let's say it takes a millisecond for one comparison. And there are 10 fingers.

Total time taken check a duplicate (worst case) = 10 fingers x 1000000000 samples x .001sec = 115 days


Now apply that for every check: N(N+1)/2 times where N is a billion.

Why so?
A fingerprint can be stored as an ID number, lets say 12 digits(For our 100 billion+ population some day )
So lets say Mr. X comes and his finger print is scanned. Based on data points an ID is generated. Few seconds at most to compare data, thats about it.
Finger print technology uses "data points" which generate a unique hashtag. Think of it like this.
Lets say you have huge files. You want to know if any of them are duplicate. Any sane system will create a md5sum or something out of the files, and store the hashes which will be much faster to compare as opposed to comparing files individually.

[Samurai]

Originally Posted by tsk1979

Why so?
A fingerprint can be stored as an ID number, lets say 12 digits(For our 100 billion+ population some day )

Really, tell me about it. How would you convert a fingerprint into an unique 12 digit number? If the same finger is scanned elsewhere, how will it create the same unique number?

[tsk1979]

Originally Posted by Samurai

Really, tell me about it. How would you convert a fingerprint into an unique 12 digit number? If the same finger is scanned elsewhere, how will it create the same unique number?

Okay, I understand, you are getting to the fact that AFIS uses probability. So if two fingerprints are 99.9% match it will flag. However, with hashtags even if same print is scanned twice, you would need 100% accuracy.
That said, the system with FBI (100 million prints) takes 27 minutes for a response. So its much faster than 1ms/print.

[binand]

Originally Posted by tsk1979

A fingerprint can be stored as an ID number, lets say 12 digits(For our 100 billion+ population some day )

Therein lies the problem. A fingerprint can be stored only as a fingerprint. That is, as a picture. Current fingerprint comparison techniques rely on comparing such pictures (which is an error-prone system, as people don't place their fingers identical to the way they did it the first time).

[tsk1979]

Originally Posted by binand

Therein lies the problem. A fingerprint can be stored only as a fingerprint. That is, as a picture. Current fingerprint comparison techniques rely on comparing such pictures (which is an error-prone system, as people don't place their fingers identical to the way they did it the first time).

I realized my error. However, if FBI can do 100 million in 27 minutes I think we can do 1000 million in 1000 minutes, if not 270!

[Samurai]

I am sure some amount for ordering can be achieved so that the search can be narrowed. I am not privy to the advanced mathematics (fractal analysis, etc). But it is the N(N+1)/2 check that will bust any duplicate check. One set can be compared against rest, but checking every set?

Originally Posted by bblost

I don't think it will work that way, but then again you might be over simplifying things.

But if its just a comparison for validating a user its will take just a milliesecond. As you are comparing the sample with just the one stored against this sample's name.

We are talking about a case where same guy will enroll at different places using different names.

[carboy]

Originally Posted by binand

Therein lies the problem. A fingerprint can be stored only as a fingerprint. That is, as a picture. Current fingerprint comparison techniques rely on comparing such pictures (which is an error-prone system, as people don't place their fingers identical to the way they did it the first time).

This is not correct. There is a ISO format (forgot the number now) for fingerprints. The fingerprint gets scanned as a picture. However, after that, it gets converted to the template format (this is around 400 bytes for a fingerprint). The conversion involves converting the picture into characteristics of fingers prints. This makes the comparison fast. All comparison is done using the template format & not the picture. Also, I think comparison involves percentages rather exact matches - i.e. when you call the comparison API, you call it with a threshold parameter also.

Other than that, vendors also have their proprietary template formats which are supposed to be faster - at least that's what I heard from one of the Indian dealers of a 3Com fingerprint scanner.

Update - the ISO format is "ISO 19794-2".

Originally Posted by tsk1979

Lets say you have huge files. You want to know if any of them are duplicate. Any sane system will create a md5sum or something out of the files, and store the hashes which will be much faster to compare as opposed to comparing files individually.

This will not work for pictures because each time the picture will be a little different so the hash will be totally different. Even with templates I don't think it will work - I would assume that the templates would come out a little different every time & hence the hash would be totally different.
I feel hashing may not be a suitable technique for fingerprints at all (though I may be wrong).

Originally Posted by Samurai

Imagine the CPU power required to compare a set of fingerprints to to rest of the billion fingerprints to detect a duplicate.

Well, let's say it takes a millisecond for one comparison. And there are 10 fingers.

Total time taken check a duplicate (worst case) = 10 fingers x 1000000000 samples x .001sec = 115 days


Now apply that for every check: N(N+1)/2 times where N is a billion.

I don't know how it really works - but I think the search would work by narrowing based on fingerprint characteristics.
How do forensic teams check fingerprints against all known criminals - this used to work even 50 years back.

[abhijeetsng]

Originally Posted by msdivy

The problem is again biometrics. There is a discussion on this topic many pages back. Let me give the instance of Brandon Mayfield, a resident of Washington County, Oregon. After 2004 Madrid train bombings, FBI found the bombing suspect's fingerprint to match with Brandon Mayfield and so he was placed under surveillance. His house had been broken into at least twice, although nothing was stolen. His phones were wiretapped, his house was bugged, and his house was searched several times (with warrant). Finally he was arrested as material witness in connection with the Madrid attacks, and was held for over two weeks.
Mayfield was never charged, and FBI acknowledged serious errors in their investigation. Finally the U.S. government apologised with a $2 million settlement.

Nobody knows the exact false positives (wrong matches), in either fingerprinting or iris technologies. It is just assumed that they are unique. For crime detection it is ok. But in a database of billions? Even if the error rate 1 in a million, that is 1000 in our countries population.

Other problems with biometrics:
1) Human tissue change with time. So iris today may not be same as few years back. A valid UID holder 10 years back might not be recognised today. So every few years, UID must be re enrolled with new biometrics.

2) Scanners cannot distinguish between live & non live tissue. Fingerprints can easily be lifted & placed on the scanner. Except the high end ones, a regular iris scanner can be fooled with a photograph of the eyes.

In all likelihood, given an UID number, a person can be identified. But given a person, detecting his UID number will be challenging. I hope UID is not misused.

(Just imagine if FBI planted an evidence in Brandon Mayfield's house. He would have been doomed for life).

Exceptions do happen.
But this iris scan can change with time is new to me. Being from a science background, I am surprised. Still I may be wrong. Would request some expert to shed some light.

Yes, agreed that finger print scanners can be fooled but not with a different one.
For Iris scan, Please tell me, I am really keen to know how can one fool an iris scanner with the help of a photograph of eyes - new concept to me.

But fingerprints and iris scans don't change in a lifetime of a human being.
Burns, accidents are different issues.

Cheers!

[carboy]

Originally Posted by abhijeetsng

Yes, agreed that finger print scanners can be fooled but not with a different one.

What do you mean - not with a different one?

[whitewing]

Originally Posted by carboy

What do you mean - not with a different one?

I'm guessing that it should have meant "not with the iris scanner"

[msdivy]

Originally Posted by abhijeetsng

But this iris scan can change with time is new to me. Being from a science background, I am surprised. Still I may be wrong. Would request some expert to shed some light.

Study published in Nature magazine - http://www.nature.com/news/ageing-ey...cans-1.10722#/

For Iris scan, Please tell me, I am really keen to know how can one fool an iris scanner with the help of a photograph of eyes - new concept to me.

http://www.wired.com/threatlevel/201...ris-scans/all/

[Samurai]

Originally Posted by abhijeetsng

But fingerprints and iris scans don't change in a lifetime of a human being.
Burns, accidents are different issues.

People doing manual labour, barely have any fingerprint left. I have bio-metric security in my office. We had major problem registering the office handyman/electrician since he has no fingerprints in most fingers. Finally we registered his little finger, the only finger he doesn't use for work.

[abhijeetsng]

Originally Posted by carboy

What do you mean - not with a different one?

brother, I mean that say Mr X has to forge my finger prints, mr x can do so but not without an impression of my actual finger prints.

Moreover, even with newer techniques, a person can forge one, two three finger prints of a single person but not all 10. CIDR of UIDAI stores all 10 finger prints along with iris scan for an individual.

Hope this clears.

[carboy]

Originally Posted by abhijeetsng

brother, I mean that say Mr X has to forge my finger prints, mr x can do so but not without an impression of my actual finger prints.

How difficult is it to get someone's fingerprints?

Originally Posted by Samurai

People doing manual labour, barely have any fingerprint left. I have bio-metric security in my office. We had major problem registering the office handyman/electrician since he has no fingerprints in most fingers. Finally we registered his little finger, the only finger he doesn't use for work.

The 3Com vendor also told me something similar - i.e. for blue collar guys you need all 10 fingerprints whereas others you can make do with a 3-4 fingers.

[msdivy]

Originally Posted by abhijeetsng

CIDR of UIDAI stores all 10 finger prints along with iris scan for an individual.

Thats the belief - 10 fingers & iris scan will identify a person. Hope it works out. I am not confident because there is no study or evidence to support this belief.

But, UID must be commended to take up such a large scale project. Nobody else has attempted to cover even 10% of population what UID has achieved. Right now the data is circulated by companies with commercial interest in these biometric scanners. I wish UID shares the findings of the data openly or at least with academic community.