NIST Facial Recognition Mask Accuracy Nov 2020 Results Analyzed

By Zach Segal, Published Dec 03, 2020, 09:51am EST

In July, NIST found that face masks negatively impacted face recognition created before coronavirus significantly. Now, NIST released a report on recent algorithms developed after Mid-March for testing on face masked subjects.

IPVM Image

Do these new algorithms perform better with face masks? IPVM investigates.

Executive *******

**** ***** ******** ** significantly ****** *** ******** of **** *********** ********** despite ************* ****** **** to ****** ********** *** masked ********. ***** *** an *********** *******, *** the *** ****-********** ********** were ********* ***-******** *** several *** ********** *** error ***** ***** ** 100%. **** ***** *** a *********** ********* *** face *********** ********** *** end-users ****** ** ***** that *** ******** ** their ******* *** ** greatly ******** ** ****.

Real-World ************

***-*****, ********** ***** ***** surveillance ******* ** ***** one-to-many ***********, *** ***-**-*** verification, **** ** ********** if ***** **** *********** solutions *** ****** **** widespread **** **** ***. This **** *** ******* than **** *** *****, because ******* **** ***, aimed ******** ** *****, and ******** ********** **** shots, *** ******** *** still ******* ******** **** some ********** ******* ** even ********* ****** ***** as *****.

IPVM Image

** ********, *** ********** submitted **** ** *********** on ********** ***** ** time *** *** *** necessarily ********** ********** ** real-world ********* *** ******* even ***** ** ****** subjects. ***-*****, ********** **** using **** *********** *** tasks ***** **** ****** control, ****** ** ****.

New ***** ********

************************** ******** *** ****** of ******* **** ***** on ********. **** ****~* ******* ****** ******** images ***** **** ******* to ******* **** ~* million **** *********** ****** for ************/ *:* ******** (i.e. ** *** ****** crossing ***** ******* *** visa *****). *** **** Application ****** **** ****-******* and ***** ** *********** offices **** * ***** background. **** *** ******* as ***/*** ***** *.*. JPEG *** *** *** x *** ******. *** border ******** ****** *** not ****** **** *** ISO/IEC *****-* ****-******* ***** standards *** ** **** cases **** *********. ***** images **** ***** ** trained ********* ** *********** conditions ******* ***** ******, which ***** **** *********** much ****** **** ************ use-cases *** **** ******* to ****** ******* ******* WDR ******.

IPVM Image

**** ***** ******* “*****” to *** ****** ******** images. **** ****** * colors ** *** ***** round (*****-**** *** *****) and ***** * **** in *** ****** (*** and *****), * ********** (wide *** *****), *** 3 ****** ** **** coverage (***/**** ************, ******/**** partially *******, ****/**** ****** up ** ***-*****, ********** covering ****).

IPVM Image

**** ****** ** *** one-to-one ************ ******** ** these **********/**** ***** **** **** the ***** ***** **** set ** */**,***. **** is **** ******* **** one-to-many (*.*., ***** ************) which ** ** ************* harder *******.

Not **** *** *******

**** *** ********** *** fairly **** *** ********** are ***** ******** ** masks, ** ****** ******* degrees, *** *** **** performer *** ** *** first *****. * ** the ** *** ********** had ********** ** **+% with *** ******-****-****-**** ***** compared ** **** * of *** ******** ** algorithms, *******, ***** * are ***** *** *** performers.**** ***** ********* ********** that **** ******* ********, like ******* *&* **** went **** **% ** 96% ********, *** ***** new ********** ********* ****** (up ** ***% ***** rates) ***, ** *** case ** ********* *+*, their *** ********* ********* slightly ****** **** ***** new ***.

IPVM Image

********

**** ***** ******** ******** to ******* *******, **** larger ***** ****** **** impact, *** ** **** cases, ****** ***** ****** a ****** ******.******** ******* ** *** cases, *** *** ****** is ****** ********. *** new **** *** ********* algorithm ******* **** **.*% to **.*% **** * mask, ******* **** **.*% to **.*%, *** *** Deepglint ********* **** *** first ***** **** ******* from **.*% ** **.*%.

IPVM Image

**** **** **** *********** on ****** ******** ** roughly ** ******** ** 2017 **** *********** ******* masks:

*** ******* *********** ** face *********** **** **** masks ** ********** ** the *****-**-***-*** ** ******** images ** ***-****

*** ******* ****** *** nose ********, ***** ** many *****, *********** ** over **** *** *** decrease ** **** ************ accuracy. ***** ***** *** a ****** ****** **** round **** (*** *****). Also, ***** *******, **** darker ***** (***** *** red) ****** * ******** larger ****** **** ******* (white *** ****) *******, this *** *** **********.

Changing ********** *** *****

******** ********** *** **** limit *** ****** ** face-masks ** ********** *** true ************ **** **** only ******* ******* ** false ***********, ** ***** in *** ******* *****.

IPVM Image

Exposing *****

** ** ** ****, it *** **** ***** for ***** ** ***** their ***** ******** ***** nose, ** **** *** greatly ******** *** **** verification ****.

*******

*** *** ********** *** not *********** ******** *** use, *** *** **** processing ** ******* **********, and ***** ** ******** to *** *********, ** the **** ** * new **** ********* ******* face ***** ***** **** a *********** ********* ** algorithm ********** **** **** not **** ****** ****** by *** **** ** the ***** ******* ** incentive ** ** **.

Comments (4)

********* **** ******* ** certainly ********, *** (** Zach *****) ***** ***...*****. I ******************** **** ********/********* *** choose ** ******* ***** using ***** *** ****, rather **** * ***** party's (****'*) ****. ** course, **** ******** ****'* going ** **** ******** of ******** **** *** going ** ** **** for **** *****.

*'* ******* ***** *** usual ******** **** **** is ********** ** **** readers. *** **** *******' database ***** ** *** hundreds? *********? ********? **** could *********** ****** *** threshold ******** **** *** use.

* **** **** **** continues ** ****** ***** test *******, ** *** same *** **** ** is ******** ***** **** test *******. ******* ** the ****** **** *** incorporate **** ***** ************ into *** **** *****, such ** ******** ********** power.

Agree
Disagree
Informative: 1
Unhelpful
Funny

** ******, **** ******** aren't ***** ** **** millions ** ******** **** are ***** ** ** used *** **** *****.

**** ** *** *****. It's * *** ****** to ** **** ** a ********* ********* ** a *** **** ** footage **** **** * long-continuous ********** **** **** times **** *****.

*'* ******* ***** *** usual ******** **** **** is ********** ** **** readers.

* ** ***. ***************** ** * ***** place ** *** ********* of *** ***** **** membership.

* **** **** **** continues ** ****** ***** test *******, ** *** same *** **** ** is ******** ***** **** test *******.

** ***** **** **** is********** ** **** ***** submitted **********. *** ** will ******** ** ****** our ******** ** *** testing.

Agree
Disagree
Informative
Unhelpful
Funny

******* ** **** ***** use * ******** **** Facebook, ****** ***** ** Shutterfly *** ** ****** across *** ***** ******** for ***** *******.

* **** * **** Google ****** ***** ***** algorithms *****, *** *** free.

Agree
Disagree
Informative
Unhelpful
Funny

**** *** ****** ** some ****** ******* *** extensive **** ******** ** the ***** ********** *** other ********** *************. **** don't **** **** ** do"**-**** **********", ******* *****, *********, and *******-******** *** **** to ** *******.

*** **** ** **** surveillance ***** *******: ***** from ***** **** ******, possibly ** * ********, with ******* ******** **********, etc. *** *** ******** is *** **** *** tests ******* *********** ***** from ***** **** * large ******** *****.

Agree
Disagree
Informative
Unhelpful
Funny
Read this IPVM report for free.

This article is part of IPVM's 7,264 reports and 968 tests and is only available to subscribers. To get a one-time preview of our work, enter your work email to access the full article.

Already a subscriber? Login here | Join now
Loading Related Reports