Search |  Contact |  SRI Home Do not follow this link, or your host will be blocked from this site. This is a spider trap. Do not follow this link, or your host will be blocked from this site. This is a spider trap. Do not follow this link, or your host will be blocked from this site. This is a spider trap.A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A ASRI International.  333 Ravenswood Avenue.  Menlo Park, CA 94025-3493. SRI International is a nonprofit corporation.

AIC Seminar Series

SRI face recognition from single or multiple images

Brian J BurnsArtificial Intelligence Center, SRI International[Home Page]

Date:  Thursday, December 1st 2016 at 4:00pm

Location:  EK255 (SRI E building)  (Directions)

Webex: 

Slides (SRI Wiki access required):

https://wiki.sri.com/download/attachments/4919820/JANUS-161201-SRI.pdf

Audio: 

PLEASE USE WEBEX AUDIO WHENEVER POSSIBLE
alternative audio (in case webex audio does not work):
VoIP on WebEx.
   Abstract

This talk covers the work done at SRI in the IARPA JANUS program. The objective was face recognition from one or more images taken under extreme ranges of viewing conditions. Our approach minimizes the use of face registration and normalization (e.g., frontalization). In other words, there is no explicit correction or compensation for view angle, lighting, expression or age, as this kind of processing is difficult to perfect and can generate unpredictable and disruptive transformations of the data. Instead, our method uses deep neural networks to localize faces and compute image features that are sensitive to subject identity and less so to viewing conditions. We also detect viewing contexts (classes of viewing conditions) and apply context-specific featurization strategies. We exploit multiple images of subjects by training person-specific classifiers and performing class model comparisons. Finally, the methods used to acquire and annotate training examples will be covered, as well as methods to train the system to handle difficult subclasses of problems.

   Bio for Brian J Burns

   On-line Resources