Biometrics and facial recognition

[Caute_cautim]

For those who want to avoid facial recognition systems currently - the Chinese have pointed out a 30 percent error rate for those with beards - an obvious message there.   Now the US have pointed out they are having problems with people with a dark complexion - which appears to have an error rate of 20 percent.  

https://www.wired.com/story/best-algorithms-struggle-recognize-black-faces-equally/?bxid=5cc9e0b62dd...

So for identification purposes alone, it is insufficient without additional controls.

Regards

Caute_cautim

[rslade]

> Caute_cautim (Community Champion) posted a new topic in Industry News on

> For those who want to avoid facial recognition systems currently - the Chinese
> have pointed out a 30 percent error rate for those with beards

And just why do you think greybeards *have* beards?

====================== (quote inserted randomly by Pegasus Mailer)
rslade@vcn.bc.ca slade@victoria.tc.ca rslade@computercrime.org
Bullet proof vest vendors do not need to demonstrate that naked
people are vulnerable to gunfire. Similarly, a security
consultant does not need to demonstrate an actual vulnerability
in order to claim there is a valid risk. The lack of a live
exploit does not mean there is no risk. - Ron Rivest
victoria.tc.ca/techrev/rms.htm http://twitter.com/rslade
http://blogs.securiteam.com/index.php/archives/author/p1/
https://is.gd/RotlWB

[Caute_cautim]

@rsladeI am sure there is a rationale behind your question about greybeards?

Either they are time honoured and highly distinguished people or the fact they could of the male persuasion ask suggested by Google and Gamer's links.

Or you have you have a great explanation?

Regards

Caute_cautim

[Steve-Wilme]

Isn't it just the data set these algorithms have been trained on that skews the results?

If the input dataset is hairless white caucasian almost exclusively then I'd expect beards (what's that!) and darker skin tones to result in a higher false positive identification rate.

[Steve-Wilme]

Isn't it just the data set these algorithms have been trained on that skews the results?

If the input dataset is hairless white caucasian almost exclusively then I'd expect beards (what's that!) and darker skin tones to result in a higher false positive identification rate.

[Caute_cautim]

@Steve-WilmeYes, in terms of the input data provided to the systems to learn and develop.   It also depends on the ethics and integrity of the scientists or developers creating the systems too - there can also be inherent bias, on the decisions made without them actually realising. 

Regards

Caute_cautim

[vt100]

Well, I'll get on with growing the beard and tanning then.

My wife and daughter are out of luck...

[rslade]

---

@vt100 wrote:

Well, I'll get on with growing the beard and tanning then.

My wife and daughter are out of luck...

---

They can still try tanning ...