Can you do a follow-up video and explain: - how many points are used to a match - are these thresholds implemented differently or is there a standard - when some fp detector is faster than others, does it mean they lower the matching threshold - what is the likelihood of a false positive? They say our fp are unique but it sounds like there could be many overlaps when the fp matching algorithm is matching only a few characteristics.

@isaactriguero3155

5 жыл бұрын

Hi, thanks for your questions! I will see what I can do, and I will make another video! But for now: - There are studies that suggest that the minimum number of minutiae points to identify a person is about 12-13. - There is no standard! every fingerprint matching algorithm will provide different scores, and those thresholds are typically found empirically! - Not necessarily, it means that the feature extraction and matching processes are faster (and probably simpler) - That very much depends on the particular matching algorithm. Local approaches are typically faster but they may result in plenty of false positives (e.g. reference [1]), and more complex algorithms, such as MCC [2] will take longer, and their score will be more precise (therefore fewer false positives!) Thanks again for the great questions! hope my answers help! References: [1] X. Jiang, W. Yau, Fingerprint minutiae matching based on the local and global structures, in: Proceedings of the 15th International Conference on Pattern Recognition, vol. 2, IEEE, 2000, pp. 1038-1041. [2] R. Cappelli, M. Ferrara, D. Maltoni, Minutia cylinder-code: a new representation and matching technique for fingerprint recognition, IEEE Transactions on Pattern Analysis and Machine Intelligence 32 (2010) 2128-2141

@Biometrics

2 жыл бұрын

Any number of points can be used for matching but a higher number of points for matching reduces False Acceptance Match while on the other hand fewer points for matching could result in computations that lead to increased False Acceptance Match outcomes. A fingerprint match result outcome is usually a probabilistic score that based on a preset threshold in a matching algorithm determines whether the biometric fingerprint match score values was a match found or no match round. Every fingerprint sample from a fingerprint image capture is different from the next and the previous one hence the reason why minutiae counts from same finger will be different in various fingerprint sample images of the same finger. I did a video demo on extracting minutiae data from biometric fingerprints. You could check it out and see data that is extracted from a fingerprint image's bifurcations and ridge endings.

@sameergonal2496

2 жыл бұрын

Hhu

The only problem with fingerprent scanners is that you leave the key on everything you touch lol.

@colin-campbell

5 жыл бұрын

That’s why you use gloves.

@ThePositiev3x

5 жыл бұрын

Exactly. Even worse, when you are using face id, you show the key to everyone and every camera nearby.

@xericicity

5 жыл бұрын

Another problem with them is that they are hard to change once you have yours stolen.

@deoxal7947

5 жыл бұрын

@@mikeking568 Wouldn't it have to be a Time of Flight camera to do that?

@thesilenttraveller7

5 жыл бұрын

I don't know if some fp scanners do this, but they could check for pulse in addition to fp.

Great video! Nice job Brady and Dr Triguero!

Great video. Does anyone have some more resources (or tips how to find them) about the matching algorithm. Like finding a part of it and also when it's in another direction, how is it achieved?

nicely explained and well -presented too..!

Great explanation!

Oh man, more of Dr. Triguero please! :D

Nice to see a video on this. I am myself working on a small project that is on fingerprint recognition by geometric reasoning. Roughly how mine works; intensity isophotes (think of a heightmap) run tangent to the ridges of a fingerprint. The feature detector that I constructed is able to detect the high curvature of these isophotos at the ends of these ridges. Besides the high curvature, ends of ridges are also characterized by a high intensity gradient. Combining these factors will give a detector that recognizes the edges of the ridges (There are nice mathematical formulas that can be applied on a local region with appropriated scales that detect these). Interestingly, exactly the same procedure can be used for the detection of bifurcations, since these have exactly the same structure when the image intensity is inverted!

Storing a copy of biometric data creates a risk of unauthorized duplication of the original data which could be used to falsely incriminate someone by replicating and then planting a copy to frame an innocent person. A proper technique would use a one-way hash function to summarize the data so that it can be used to match later samples but can’t be used to fabricate an original copy.

@phy2sll

5 жыл бұрын

That may limit the system's ability to perform some of the more sophisticated "structure" (as opposed to element-wise) matching he describes. But otherwise, yes, you could create signatures by applying some locality sensitive hashing technique (minhash or simhash, perhaps?) and compare these instead.

@ibycus314

5 жыл бұрын

Not sure this is a ‘real’ problem. Lets assume I’m trying to frame you for a crime. I want a copy of your fingerprints. Do I (a) break in to your phone, retrieve some (likely encrypted) fingerprints, decrypt them, and then reconstruct a physical fingerprint from what I find, or (b) lift a copy from literally anything you’ve touched bare handed (including but certainly not limited to your phone)....

@tylisirn

5 жыл бұрын

The feature extraction is a function like that. You could use the extracted features to create a fingerprint that to a machine looks the same, but it would not be visually identical to the original fingerprint.

@alessandroceloria

5 жыл бұрын

In any device fingerprint data is acquired, stored, compared and managed by a separate coprocessor which is completely isolated form the rest of the system, to which it doesn't report anything but a simple "yes" or "no" to signal whether a match occurred or not. In aple devices it's called secure enclave and it also manages other security functions. About Android phones, I'm not sure, but they HAVE to have that processor in order to be certified so rest assured.

Here is an idea for a video: we're all familiar with long integers, double precision floats and whatnot, but how do computers deal with truly massive numbers, such as come from finding big primes?

@dankwarmouse6248

5 жыл бұрын

Great idea, I've personally ran into integer overflows without really trying that hard, so I've been curious how the people who are doing this non-recreationally deal with it.

@LaurenLagarde

5 жыл бұрын

The key words to look up are "arbitrary precison arithmetic." It's actually a pretty neat trick: they operate by simply placing multiple smaller integer types back-to-back and treating them as a single number. To add two of them together, you can start off by setting the result to be the width of larger of the two and then go through from least significant to most significant by the largest feasible integer type, taking the carry bit over each time. If your most significant bit overflows, add another bit of precision to the result. Multiplication would work similarly, doing the math long int by long int and adding additional precision as necessary and summing the results together at the end. Floating point would work a bit differently due to the lack of hardware support but it is still possible to do it in software slowly by using arbitrary precision ints to store the mantissa and exponent.

@schifoso

5 жыл бұрын

You can use a construct such as an array to hold the digits of the large number. Then you need only write routines for doing the mathematical operations on them. Programming languages, such as Java, have this type of thing. In Java it's called the BigInt class, or something like that.

@johnfrancisdoe1563

5 жыл бұрын

puskajussi37 Are you asking about bignum libraries as used for security software, or the techniques for doing extreme precision numbers such as setting records for largest prime or largest number of digits of Pi ?

@puskajussi37

5 жыл бұрын

@@johnfrancisdoe1563 Im just expressing a general topic. The details on how to make it into an intresting video I leave to the experts.

Is possible to change your fingerprints with a laser operation?

Is it possible to restore (or build) a fingerprint image (with the same characteristics of the original fingerprint) from the fingerprint characteristics file?

@brucedickie9394

2 жыл бұрын

Somewhat, one can reverse map the feature points of a fingerprint based on the template of the fingerprint and just fill in the blanks with unimportant information that won't be reused when you apply this fake fingerprint to another sensor. It likely won't fool a human or clever AI, but there are techniques then to add noise to the feature map so that the fingerprint can't be generated from the template

Please do iris scanning next.

Would be nice, if you could do a follow up on how the identification of interesting points on the fingerprint image works and how the matching algorithm works. Thanks for the video!

@isaactriguero3155

5 жыл бұрын

Thanks! I will see what I can do! Initially, I just wanted to give an overview of the process!

@drew295

5 жыл бұрын

@@isaactriguero3155 It was very nice and definitly sparked interest!

Is there a standard algorithm of sorts for rotational and scalable feature, or rather point, matching? i.e., given two sets of a couple of dots with coordinates, check if they are the same pattern, independent of rotation and scaling?

@daniellavery6667

5 жыл бұрын

Yes. Compute the Fourier-Mellin transform of each set of coordinates, then compute the cross-correlation of each transformed set of points. See, for example, "An FFT-based technique for translation, rotation, and scale-invariant image registration." B.S. Reddy, B.N. Chatterji. 1996.

@MoritzvonSchweinitz

5 жыл бұрын

@@daniellavery6667 Thanks! Will look into it! I'm a bit surprised that there doesn't seem to be a simpler, programming 201 class standard algorithm?

@isaactriguero3155

5 жыл бұрын

Hi, There is no standard algorithm (that I am aware of), but my experience with the two algorithms referenced below is quite positive. [1] X. Jiang, W. Yau, Fingerprint minutiae matching based on the local and global structures, in: Proceedings of the 15th International Conference on Pattern Recognition, vol. 2, IEEE, 2000, pp. 1038-1041. [2] R. Cappelli, M. Ferrara, D. Maltoni, Minutia cylinder-code: a new representation and matching technique for fingerprint recognition, IEEE Transactions on Pattern Analysis and Machine Intelligence 32 (2010) 2128-2141

I was wondering what would be the numerical data of a fingerprint, does it use hexadecimal numbers?

CGP Grey will be annoyed by the fact that someone is using pages to annotate an image However, that pencil is still enjoying a better life than Brady's, only used once as a weapon to deter a bee

@sharank

5 жыл бұрын

A Hello internet reference. I like it

@leonhard5841

5 жыл бұрын

When did he talk about not using Pages?

One question (maybe 4...), for some reason it become very hard to unlock my phone via fingerprint, it got so bad that I often made too many attempts and had to use a password instead. When I created a new profile it now unlocks near-instantly, so it wasn't an issue with the sensor. Why would this happen? I can't recall any injury to my them I haven't grown and my weight is stable.. Is this a common occurrence? Do people's fingerprints Change much with age? I'd imagine the answer is yes. How is this normally managed and what are the security implications with false positives from making the algorithm more accommodating? I've had the android phone for a little over two years now.

That is all very interesting but how does the sensor physically read your fingerprint before it starts to analyse it?

This is Computerphile, has no one thought to do a screen capture or mirroring or anything other than camera over the shoulder or at a crappy angle with no zoom?

@Schoko4craft

5 жыл бұрын

I actually like this style since the professional gets more highlighted than in a normal information video. Also they add pictures to clearify some things

@elmersbalm5219

5 жыл бұрын

This is actually preferable. feels like I’m there for the interview. some over-shoulders are a bit long though. they should be interleaved with quick switches to front camera

If you graze your fingers will finger print recognition still work ?

@isaactriguero3155

5 жыл бұрын

If you can still get enough minutiae points (around 13), it should do!

Miss Isaac so much🤭

I'd love to see an explanation on how this data is stored and how it's matched against a finger press, especially in the context of an encrypted device. So: how do you use a fingerprint to authenticate and decrypt the master decryption key?

Do not try to trick us, by putting your tablat on a stack of paper. You are violating the rules - pen and paper only!

when's the notchproX coming out?

6:50 Is that you, Jared Harris? :))

Fingerprints are easy to acquire and impossible to replace once they are leaked. I prefer passwords for now

@RealCadde

5 жыл бұрын

Some prefer both or even more.

how fingerprint information is verified code number

Is the fingerprint matching typically done on the device? Does that mean my bank trusts my device's scrutiny when I login via fingerprint? Or does the device hash these features somehow for my bank to scrutinize?

5 жыл бұрын

On device I think

@stekeln

5 жыл бұрын

I believe that they just send either the actual data or the minutia points to their authentication servers (encrypted of course) and let them verify the print. They might also be checking for whether the same exact data has been sent previously to ensure that it hasn't been duplicated, but that may or may not be worth it since it's very easily bypassed. I'm not very sure, but a traditional hashing-technique would make it very hard to compare the fingerprints if they don't somehow quantize the data before hashing.

@coolfrog23

5 жыл бұрын

There is a way to use the properties of finite fields to encrypt the fingerprint data so that you don't lose accuracy (from irregularities in the pictures). You can't just hash it because even a slight distortion of the picture will give you a different hash.

@antonvroemans5964

5 жыл бұрын

It definetly trusts the device on this one.

@stekeln

5 жыл бұрын

@@antonvroemans5964 If this actually is the case, I am never using anything that authenticates by fingerprint over the internet again. It would be like letting clients check their own passwords, just telling the servers whether they logged in successfully or not; like letting students grade their own exams. Edit: I guess you shouldn't be relying on fingerprint authentication in the first place, but for something so widely incorporated it would be a shame if its implementation is handled this poorly.

Correction: a whorl is a type of fingerprint pattern (the one that looks circular). The innermost place where the lines turn around in a loop pattern is only called core.

I always bite the skin off my thumb when I'm nervous and then I can't unlock my phone... Let's see how you fix that, technology!

They say that there's no two people on Earth exactly the same; no two faces, no two sets of fingerprints. But do they know that for sure? 'Cause they would have to get everybody in one huge space, and obviously that's impossible, even with computers.

@nathanieljasondemesa3135

5 жыл бұрын

Pure presumptions. So u cant negate

@isaactriguero3155

5 жыл бұрын

Agreed... as I said in the video. There is no proof of that! That's probably why 'real' authentication systems are typically a combination of multiple biometric features (i.e. face, iris, fingerprint, even palms!).

Where are all those hackers stealing fingerprint featured in this video?

Fingerprints should *never* be used for authentication. Only identification. If they are used for authentication, it's basically a password you are leaving everywhere an on everything you touch.

@Vekkq

5 жыл бұрын

Not even identification. Fingerprints are not unique.

@gizmoguyar

5 жыл бұрын

While that's true. Modern capacitive fingerprint sensors do require a 3D model of the fingerprint. It's still extremely difficult to recreate a 3D finger model from an image of a fingerprint.

@xericicity

5 жыл бұрын

Its like having your password printed on your shirt for everyone to copy and is incredibly hard to change once stolen.

@johnfrancisdoe1563

5 жыл бұрын

Mladen Mijatov It's not like a password. Biometric security doesn't check if you know the fingerprint but if you have the fingerprint on your finger. Which is why it is so important to check if it is seeing a living finger and not a rubber copy molded into a glove or other such trickery.

@randomizednamme

5 жыл бұрын

Authentication is identification 🧐 maybe you mean authorization?

Fingerprint sensors are great for usability and practicality, but they are a bit of a security nightmare. You can copy the fingerprint from a photo, this has been done.

hello, you could add subtitles in English to your videos, thanks are very interesting

@suds-p

5 жыл бұрын

Subtitles are in progress, hopefully they'll get approved soon

25643rd

But how does it work.

There have been cases of people unlocking their partner's phones whilst they are asleep by pressing their finger on the phone. Scary...

@johnfrancisdoe1563

5 жыл бұрын

dazimor Not as scary as stealing the finger with a knife!

I refuse to use the fingerprint or eyeprint features. If someone hacked a system, they could take the fingerprints/eyeprints and use them to do all sorts of damage with ID theft. Also feels way too personal. I will just stick with typing in my password or passcode and changing it. I can never change my fingerprints unless I were to file them off somehow. Also, what if you lost a digit? I guess you'll never get back into your account again lol

@kadblue2000

5 жыл бұрын

Nobody has to hack your system to get your fingerprint, they can just obtain it from any smooth surfaces you touch. Thats why passwords are much safer.

@noxello

5 жыл бұрын

That's the downside of biometric authentication, it's unique so it can be hard to replicate but if stolen you can't change it. That makes it so you have to care about the security of the system as you said (for example would you trust Google, Apple or Samsung enough to deal with your biometric data), personally I do, some people don't (including you it seems) and that's perfectly fine. Also if you get amnesia and forget your password too you wouldn't be able to log in with a password so that's not really an argument 😅

@Remls

5 жыл бұрын

You can record multiple fingerprints. Passwords are used as a fallback anyway. Losing digits is not an issue. Well, I do suppose it makes eating your cereal a wee bit harder.

@saintchuck9857

5 жыл бұрын

Biometric data is relatively easy to steal. A picture can give facial recognition data points and nearly anything you touch can give your fingerprints. I don't see identity theft being a valid argument here. It's also no less hack proof than a password.

@oscarchampion5842

5 жыл бұрын

Is this comment a troll? 1. The system doesn't probably store the finger print, it probably stores a hash of it 2. Even if you have someones finger pints and face, you can't steal there identity with it 3. If you lose a finger you can still use a password

When he's pointing out the areas of interest, show those, not his face, ffs.

Fun fact: fingerprints are actually not at all unique. In 2004, a lawyer from Oregon was arrested because his fingerprint was found at the scene after the Madrid train bombings, but it later turned out that his fingerprint matched the one of one of the terrorists responsible. In the USA there are on average 7 people with a 12-point match between their fingerprints, which means there are around 150 people around the world that have the same fingerprint as you (within the accuracy of a 12-point test).

@nm_scarecrow302

5 жыл бұрын

This seems to say more about the limitation of the test, than the "uniqueness" of the fingerprint(s).

@johnfrancisdoe1563

5 жыл бұрын

Nihal Mishra But 12 point match used to be considered irrefutable evidence in court cases around the world, with no other reason to accuse that person.

@isaactriguero3155

5 жыл бұрын

Interesting, I read that the minimum number of minutiae point should be around 12-13 (in some research papers). If you have a link to that news, I would really appreciate it!

'Notchprox' lol

The only problem with using your fingerprint as an example is someone can make a replica silicone fingertip. Not worth the effort of course, just saying.

Am I the only one looking at my fingerprints right now?

@mikevanin1

5 жыл бұрын

Yes.

@auto514

5 жыл бұрын

Yes.

@MrFram

5 жыл бұрын

No

@auto514

5 жыл бұрын

Census says 1/3 of viewers are looking at their fingertips.

H

Worst authentication idea

WTF!? How come?! Would You use fingerprints in authentication process when you leave them everywhere each time yuou touch anything? It's like leaving your plain-text password that cannot be changed everywhere. I would rather protect my fingerprints from stealing and not use it for auth. D'oh!

@RemikPi

5 жыл бұрын

On the other hand, how easy it is to scar your fingerprints, accidentally or purposely, so those cannot be recognized any more?

Sponsored by Apple.

i want to see the script not you ,

If you tell me that you cannot do it without neural nets, I'm going to assume that you're a genius and right.

@SuviTuuliAllan

5 жыл бұрын

btw is joke

@johnfrancisdoe1563

5 жыл бұрын

Suvi-Tuuli Allan I'd actually assume the opposite.

@SuviTuuliAllan

5 жыл бұрын

@@johnfrancisdoe1563 Oh yeah? Aren't we fancy!

the guy is so cute

Actually; fingerprint recognition and matching softwares using hashing techniques. They produce hashcode for each finger print and matching these hashcodes. This guy is not telling about hashing techniques used by fingerprint recognition softwares. Thus he is misleading the programmers with useless too much information.

@isaactriguero3155

5 жыл бұрын

Hashing techniques are typically used for identification purposes! What I explained in the video is called fingerprint verification! not quite the same problem.

@cmdody

5 жыл бұрын

@@isaactriguero3155 Actually; there is a hashing technique which converts each fingerprint to the unique hash code! Something like md5, sha256, ... But; I don't know the exact name of the hashing technique... It's a secret technique and mathematical formulation, which they don't share because of patent rights.

Fingerprints - and most other biometrics - are very poor security factors. They are too easily copied or worse, synthesised. And the authentic originals are too easily defaced.

Unfortunately I have not learned anything from this video :(

First?

@AlejandroV84

5 жыл бұрын

Looks that way mate

Your fingerprint is never stored on phones, is through mathematical representations that represent your fingerprint, if these fingerprints don't match the previous representation, it will not authenticate the unauthorised person trying to access it.

3:00 Looks like a delta? It looks like a gamma to me.

Delta: Δ δ Lamda: Λ λ

@xristoss.2037

5 жыл бұрын

Σωστος

@uF1G

5 жыл бұрын

yeah that caught my attention too but I guess he was referring not to the letter itself but more to the river-delta shape

20th

Also not first

Not first

I see a lot of desperate people here who probably fear their own shadow lol. It's simple, if you have anything that's important enough to be worth the time and money to spoof your fingerprint: don't store it behind a fingerprint lock. Otherwise stop bitching about it

@johnfrancisdoe1563

5 жыл бұрын

Samuel Godoy One thing many people are locking with the cheap fingerprint recognition on their phone is unrestricted access to their bank account with all their money. I guess that's worth stealing!

Last

First

Apple products. yuck

Not first