Facial Recognition 101

By IPVM Team, Published Mar 18, 2020, 10:52am EDT (Info+)

Facial recognition interest, use and fear is increasing. This guide aims to teach you the fundamentals of facial recognition.

IPVM Image

****** ** *****:

  • **** *********
  • **** ***********
  • *:* ** *:*
  • *********** ** *************
  • ********** / ***** ************
  • ********** ****** ******* ** ****** ***********
  • ******** / *********** ****** *********** - Masks, ****, **********, ***.
  • ******* **** ** ***** **** ***********
  • ****** / ****-******* *********** ******
  • ******** *********
  • ********
  • ******* ********
  • **** *********** *********

Facial *********

****** **** *********** *** ** *********, a **** ***** ** ** ********. Typically, * *********** ****** ** ***** over *** ****. **** ** * relatively ****** ********, *** **** ****** for **** ******** **** **** ***********.

*********** ****** ** * *********** *******:

  • ***** ** *****: ***** ** ** 'easy' ** ****** * **** ******* directly ** *** ******, *********** *** vary ************* ********* ** *** * person ***** ***** **** (****, ****, right, ***.)
  • ******** ** ***** - ***** ** is '****' ** ****** * **** looking ******** ** *** ****** ** a ****-*** *****, *********** **** **** significantly ********* ** *** ******** ********** of *** ***** (*******, ********, *****, etc.).
  • ********* **** ** ****** ***** - Finding *** *********** **** ******* *** in * ***** *** ******* ***** objects *** * **** (******* ** a ****, * ***, * ***, a ******* ****, ***.) *** ** very *********** ***** **** ***** ************ devices (*.*., ** *******, ****) **** limited ********** *****.

* ****-*** **** ******* ******** ** the ****** **** ** **** ** detect:

***** *** **** ** * **** in *** ******** **** ** **** challenging:

Facial ***********

****** *********** ******* *** ******** "***** face ** ****?", ***** ** ********* information **** ************* * ***** ** pursuing * ********** ******.

IPVM Image

****** *********** ****** ****** ****** *** face *** ******* **** ********.

1:1 ** *:*

****** *********** ****** ******** * ****** to * ******** **** (*:*) ** identifies * ****** **** **** ***** (1:N).

*:* ******** ** ****** ************. * single ********* ***** ** ********* ** the **** *********** ******, *** ** determines ** *** **** ******** ******* that *****. *:* ******** ** ********** easy:

IPVM Image

*:* ******** ** ****** **************. **** photos *** ********* ** *** **** recognition ******, *** ** ********** ** the **** ******* *** ** ***** photos. *:* ******** ** ********** ****, much ****** **** *:*:

IPVM Image

***** ************ **** *********** ************ *** typically *:* *** **************.

Cooperative ** ************* **** ***********

*********** **** *********** ** ********** ****. This ** **** * ****** ***** and ******** ***** ******** ** *** camera (*.*., ****** *******, ******** ***********,******):

IPVM Image

************* **** *********** ** ********** **** because *** ****** *** ** ********* to *** ****** ** ***** ** worse *** **** *** ** ****** to ***** **:

********, ***** *** ****** ** ********* and ******** ********** *** ******** *** challenge, ********* ** ************* ***** ********.

Logistical ****** ************* **** ***********

******* **** ****** *** ******* **** face *********** ******* ***** *** ************ cameras *** ********* ********* **** ***** the ****** ** ***** **** *****, uncooperative **** *********** ** * *********** percentage ** ****-***** ************.

**** ***** **** ********* **** **** recognition ******** **** **** ************ ******* is **** ***********.

Resolution / ***** ************

**** *********** ******** ************* ****** ********** or ****** *** ****/***** **** ******* person ********* ** **** *********. **** manufacturers ******* ************ ** "****** ******* the ****" ** "****** *** ****". While *** ************** ****, **** ************* require ********* ******-***-**** (~*** ******-***-*****).

*** ******* ****-***** ************ *******, **** generally ***** ***** ********* *** ****** accurate ****** ***********:

IPVM Image

Mounting ****** / ****** ********

**** ************ ******* *** ********* ** ceilings ** ***** **' *** ****** so **** *** *** ** *****. This ***** **** **** ******* *** aimed ***** ** ** *** *** of * ******'* ****, ***** ***** face *********** *********** *** * ****** or *********:

IPVM Image

**** *** **** ** ********* ****-**** images, **** ****** *********** ******* **** specific ******** ************ **** ******* ******* or ******* ***** ** **** *******.

Angle ** ********* / ****** ********

******* **** ************ ******* *** ********* on ** **** ** *****, ** they *** ***** ** **** **** of * **** ** ********, **** do *** ********* ******* * ****** view ** *****, ***** ********* *** side ** * ******'* ****:

IPVM Image

***** **** ****** *********** ******* **** fail ** ***** ***** ** * high ***** ** *********, **** **** that ******* * ***** **** ** so **** ************* ***** ******** ** confidence ******** ** ****-**** ******.

Lighting ******

***** **** ************ ******* *** **** to ******* ****** ******* *** ****** detection ** *** *****, ****** *********** requires ************* ****** ******* *** ***** clarity.

**** ******* *** ***** ***-***** ****** techniques **** ****** *********** ** *** light **** ***********/**********:

**** ******* ******* ** **** ** illuminate **** ***** *** ******** ******* captured ** *** ******. *******,** *********** ****** ****************** *** ******** ******* *** ******** face *********** **********:

IPVM Image

Avoiding / *********** ****** *********** - *****, ****, **********, ***.

**** ******* ***** *** ****** ** actively ** ************* ********* *** *********** of ****** *********** *******:

  • **** *** **********
  • *****
  • ******* **** / ****
  • ******

Hats *** **********

********* ** *** ******** (******** ** exterior) *** ******, **** *** ********** are ******** **** ******* ********** ******* suspicions. *******, **** *** ********** ********** impact ****** ********* **********:

IPVM Image

***** **** ****** *********** ******* ****** their ******* ** ******* ****** *********** with ****, ********** *** **** *****, at ****, ** ****** ******* *** confidence ***** ** ******* *** ********* increases *** *********** ** ******.

*****

***** *** ******** **** ****** ** public ******, *** **** ** ******* facilities, ** *** ****** ** *********** continues.******* ***** ************* ******** ***** **********, **** ****** ********* *** ********** at ***:

IPVM Image

*** ******** ** **** *********** ********** also **** *** ******* *** *** decrease ** ***** **** *** ******** while ******* *****, ***** ***** ******* compound *** *********.

Looking **** ** ***** **** ***********

******* **** ** * ****** *** significant *** ** ********* ****** *********** systems. ** ******** ** *** ********* of ******* ****** ************ *******, **** people **** ****, ** ***** (*.*. looking ** ***********):

IPVM Image

***, ************* ********* ** ***** ****** who **** ** ***** *********, **** known *** ***** **** ******* **** makes ** ****** *** ****** ** facial *********** ******* ** ******** ****.

IPVM Image

Crowds / ****-******* *********** ******

**** ********* **** **** *********** ** marketed *** (*.*. ********, *******, ********) often ********** ***** ****** ** ****** that ***** ****** **** **** **** detection *** **** *********** *******.

** ******** ** *** **** ****** of ***** ******* ********, ***** ***** blocked/obstructed *** **** *********** *** ***** multiple ********** *** *** **** ****** in * ***** **** ******.

Liveness *********

****** *********** ** ***** ****** ******** as * ******** ** ******** ****** control. *******, **** ******* ** *** offer ******** *********, ***** ***** **** can ** ******* ** ******* * photo ** * ****** ** * real ******:

IPVM Image

******** ********* ** **** ******** ******* and *** ** ********* ** ******* specialized ********. *** ****, ***:****** *********** ******* **** ****** ******** Detection ****.

Accuracy *** ****** ***********

********* ******** *** ****** *********** ******* is **** ****. ******* **** ** market **** **********, **********-******** ******* **** 98.6% ********:

IPVM Image

*******, ********** ******* ****** *********** **** differently:

IPVM Image

***** **** ****** ** ****** *** scope ** * *** ************, *******, it ** ********* ** ********** **** these ****** ******* *** *********** ******** between ***** ********* (*.*., ******** *** wrong **** ** ******* ****) *** false ********* (******* ** ***** * face ** * ****** ** *** watchlist).

IPVM Image****** * *.* ********** ********* *** technology ********** **** ***** **** *********** algorithms ** **** ** ***** *********** *********** ****** ******* (****).

****** **** ******* *** ***** ** many ************* ** * ********* ****, however, ** *** ** ********* *** non-experts ** ********** *** ******* ** the *******, ***** *** ******** *** academic.

**** **** ********** ************, *** ******* still ****** **** ****** *********** ******** is ****** ********* ** *** ****** it ** ***** **** - *.*., how *** ***** *** ****** ** mounted *** **** ****** *** ******* or ***** **** *** ********* ***********.

Ethical ********

**** *********** *** ******** ********** ***** attention ******* ** *** ******* ******** of **** ***********, ******* **** **** collection, ******************. ********, **** *********** ******* *** be **** ** *********** ** ****** minorities *** ********* ******* *** **** detention *** ********* ************ (*,*,*).

***** ** ******** ** **** ********* *** *** ********* ** *** ** *** ** GDPR(******** *** ****** ********), ********* ************* ******* **** *********** ******* **** ****** media ********, ** ******** *** ***** of **** *********** *** ****** ******/****** users.

Face *********** ******** *********

***** ** * **** ** ******* face *********** *********, ****** **************:

  • ******: ****** *****-*****************, **** ************* **** ******* ***** their *** ****** *********** *******.
  • *********: **** ****-****** *** ************* ******* offers **** *********** *** ************, ****** control, *** ****** ************ (**** ****)
  • ********: ***-**-*** ************ ************, ******** ***** face *********** (**** ****)
  • ********: ***** ********** ********* *********, ******** added **** *********** (**** ****)
  • *****: ***** ***** ************, ****** **** recognition ******* *** **** (**** ****)
  • *********: *****'* ******* ************ ************, ****** face *********** ******* *** **** (**** ****)
  • ******/****++/*********: ***** **** *********** *********** ** China, **** ****** ******** ******* *****
  • ***: **** ******** **** *** ********** facial *********** ******** *** **** ***********, requires ********** *****.

Comments (9)

The truth is, whether we like it or not, that this particular genie is out of the bottle and is unlikely to ever be put back into it. Face recognition is becoming more and more mainstream, one only has to look at social media and the likes of Facebook to see how indifferent everyone is the identification of people in their own photos, or the photos of others.
Whatever is happening in the West, China is pushing forward with its "social credit" system, which appears from the outside to rely heavily on what we might call public space facial recognition. Yet, it does not seem that objections to this on privacy grounds are particularly vociferous. Doubtless this has a lot to do with government themselves, but the possibility must be considered that people honestly don't care.
Here in the US, certainly in my field, use of facial recognition systems in a casino environment is gaining considerable traction. From the end-user perspective the benefits of deploying such a system appear to outweigh all other factors.

Agree: 2
Disagree: 1
Informative
Unhelpful
Funny

I had not heard of liveness detection until this article. Interesting!

Agree
Disagree
Informative
Unhelpful
Funny

Connie, thanks! We have a report on this - Facial Recognition Systems Fail Simple Liveness Detection Test. It's an interesting read. I've added that link to the body of the report.

Agree
Disagree
Informative: 1
Unhelpful
Funny

Hi John,

I've been hearing more and more about Herta facial recognition and been hearing that their algorithms can handle more simultaneous reads than its competitors. Have you tested Herta?

Agree
Disagree
Informative
Unhelpful
Funny

We are actually talking to them right now about their newest announcement about facial recognition with masks on. We have not tested them yet but we will update with more information.

Btw, who are you hearing this about Herta?

Agree
Disagree
Informative
Unhelpful
Funny

A few places, some customers who heard about them and asked us. In direct talks with their sales people and a few from my company’s HQ, over in Europe. I haven’t seen or heard of any actual installs though and wanted to check here.

Agree
Disagree
Informative
Unhelpful
Funny

Is there a standard for setting up a camera specifically for facial recognition? I looking for mounting height and any other details related to this topic.

Agree
Disagree
Informative
Unhelpful
Funny

It depends on the face recognition technology, Some vendors limit the tilt to 30 degrees, ours can work well in 45, I guess each vendor got it's strong/weak points, you need to find the one that suit your needs, Some are very sensitive to light conditions so for them you need a better camera or improve the illumination

Agree
Disagree
Informative
Unhelpful
Funny

Great high level article IPVM. One nit: Verification need not be a single image for 1:1. It simply requires that the user is identified through some other modality (e.g., what you know or what you have). So in an access control application, a user may use a card or PIN first and then the face rec can stream video with lots of images of the person to verify the identity.

Agree
Disagree
Informative
Unhelpful
Funny
Login to read this IPVM report.
Why do I need to log in?
IPVM conducts reporting, tutorials and software funded by subscriber's payments enabling us to offer the most independent, accurate and in-depth information.
Loading Related Reports