Upload File

image understanding lab detecting unspecified familiar faces

  • Home
  • Documents
  • Image Understanding Lab Detecting Unspecified Familiar Faces
1
Detecting Unspecified Familiar Faces Emily X. Meschke 1 *, Catrina M. Hacker 1 , Jordan J. Juarez 1 , Rafael S. Maarek 2 , Irving Biederman 1,3 1 Neuroscience Program, 2 Department of Biomedical Engineering, 3 Department of Psychology, University of Southern California Image Understanding Lab 1. Can an unspecified familiar celebrity—any celebrity—be detected among faces of non-celebrities at RSVP rates? –YES 2. Do Congenital Prosopagnosics (CPs) perform worse than controls on this task? –YES, but CPs perform remarkably well on this task—better than on non-speeded celebrity recognition tasks! 3. Is detection mediated by a general familiarity signal, not linked to the identification of the target? –NO. Detection is overwhelmingly associated with Identification, even for CPs. 4. How does such detection of celebrity faces compare to performance on negative detection of objects (e.g., an unspecified object that is not a tool)? –Objects are much more easily detectable *[email protected] http://geon.usc.edu/ Conclusions 1. Subjects were able to detect celebrity target faces—without the celebrity being specified prior to the sequence—at RSVP rates. This is in stark contrast to what is observed in the identification of familiar voices where performance nosedives to near chance levels with just a handful of possible voices. 2. In neither the RSVP nor the Doppelgänger task was there any evidence that the subjects could avail themselves of a familiarity signal. RSVP Task: When subjects made a positive detection they were nearly always able to provide an accurate identification of the target. Doppelgänger Task: Prosopagnosics were at or below chance in their celebrity discrimination accuracy. References Intraub, H. (1981). Identification and processing of briefly glimpsed visual scenes. In D.F. Fisher, R.A. Monty, & J.W. Senders (eds.), Eye movements: cognition and visual perception. Hillsdale, N.J.: Lawrence erlbaum associates. Tranel, D. & Damasio, A. R. (1985). Knowledge without awareness: an automatic index of facial recognition by prosopagnosics. Science, vol. 228, 1453-1455. Acknowledgments NSF BCS 0617699; Dornsife Research Fund. Subjects 7 “Moderate” Congenital Prosopagnosics (mCPs): USC students (Age 19 to 22 years, 5 female) selected as CPs based on their performance on: a) the faceblind.org survey (< 70), b) the Cambridge Face Memory Test (< 70), c) the USC Image Understanding Lab Celebrity Test (< 80), d) the PI20 (> 50), and unremarkable neurological history. 48 Controls: USC undergraduates, (Age 18 to 47 years, 35 female) with scores in the normal range on the above measures of face recognition ability. 1 “Extreme” Congenital Prosopagnosic (GJ): Male Age 33, faceblind.org survey: 18%, Cambridge Face Memory Test: 28%, USC Image Understanding Lab Celebrity Test: 25%, and PI20: 90. 1 “Extreme” Acquired Prosopagnosic (MJH): Male Age 53, faceblind.org survey: 3%, Cambridge Face Memory Test: 38%, USC Image Understanding Lab Celebrity Test: 26%, and PI20: 83. Also evidences some simultagnosia. Detection is almost always accompanied by identification There is thus no evidence for a familiarity signal (Tranel & Damasio, 1985) independent of identification. Celebrity detection improves with higher rated familiarity Prior to the RSVP and Doppelgänger Discrimination trials, subjects were given the names of the celebrities and rated their frequency of having seen the faces (1 = low; 5 = high). Doppelgänger Discrimination Task Subjects viewed 100 pairs of faces, one of which was a celebrity. The non-celebrity headshot was selected to be as similar as reasonably possible given the sampling of the headshot galleries of non- celebrities. The photogenic quality of the non-celebrity headshots were indistinguishable from the celebrities, and none of the faces underwent removal of hair or alteration of face shape as in the CFMT. No limit was imposed on the time required for a response. Correct Answer: Right Correct Answer: Left Percent Identification following Accurate Detection Controls Controls CPs CPs (Questions and Our Answers) The Detection of an Unspecified Familiar Face RSVP Tasks 200 RSVP sequences each with 30 images. A given sequence consisted entirely of either faces or objects. Both tasks could be regarded as “Negative Detection” (Intraub, 1981) in that the subject had to detect an object that was not an instance of a prespecified category (e.g., “Not an Article of Clothing”), and for faces, a celebrity—any celebrity—among non-celebrities. A target was present in half the sequences, though never in the first or last six images. The subject had to report what or who the target was, e.g., “an axe” or “Leonardo DiCaprio” or some identifying information, e.g., “The guy in Titanic.” Categories for objects were clothing, plants, tools, animals, and modes of transportation. On any given object trial, an image from any of these categories might be used as a target. Object Trials: All subjects performed near ceiling. Face Trials: Both Controls and mCPs performed well above chance on negative detection of faces, with the Controls superior to CPs. Objects RSVP Celebrity Detection and Negative Object Detection Task chance Faces Controls mCPs xCP (GJ) Presentation Rates Faces Objects RSVP Pros Doppelgänger Controls Doppelgänger Pros RSVP Controls Scan for a copy!

Upload: others

Post on 29-Dec-2021

3 views

Category:

Documents


0 download

Report

TRANSCRIPT

Page 1: Image Understanding Lab Detecting Unspecified Familiar Faces

Detecting Unspecified Familiar FacesEmilyX.Meschke1*,CatrinaM.Hacker1,JordanJ.Juarez1,RafaelS.Maarek2,IrvingBiederman1,31NeuroscienceProgram,2DepartmentofBiomedicalEngineering,3DepartmentofPsychology,UniversityofSouthernCalifornia

ImageUnderstandingLab

1. Cananunspecifiedfamiliarcelebrity—any celebrity—bedetectedamongfacesofnon-celebritiesatRSVPrates?–YES

2. DoCongenitalProsopagnosics(CPs)performworsethancontrolsonthistask?–YES,butCPsperformremarkablywellonthistask—betterthanonnon-speededcelebrity

recognitiontasks!

3. Isdetectionmediatedbyageneralfamiliaritysignal,notlinkedtotheidentificationofthetarget?–NO.DetectionisoverwhelminglyassociatedwithIdentification,evenforCPs.

4. Howdoessuchdetectionofcelebrityfacescomparetoperformanceonnegativedetectionofobjects(e.g.,anunspecifiedobjectthatisnotatool)?

–Objectsaremuchmoreeasilydetectable

*[email protected]

http://geon.usc.edu/

Conclusions1. Subjectswereabletodetectcelebritytargetfaces—withoutthe

celebritybeingspecifiedpriortothesequence—atRSVPrates.Thisisinstarkcontrasttowhatisobservedintheidentificationoffamiliarvoiceswhereperformancenosedivestonearchancelevelswithjustahandfulofpossiblevoices.

2. InneithertheRSVPnortheDoppelgängertaskwasthereanyevidencethatthesubjectscouldavailthemselvesofafamiliaritysignal.

RSVPTask:Whensubjectsmadeapositivedetectiontheywerenearlyalwaysabletoprovideanaccurateidentificationofthetarget.DoppelgängerTask:Prosopagnosicswereatorbelowchanceintheircelebritydiscriminationaccuracy.

ReferencesIntraub,H.(1981).Identificationandprocessingofbrieflyglimpsedvisualscenes.InD.F.Fisher,R.A.Monty,&J.W.Senders(eds.),Eyemovements:cognitionandvisualperception.Hillsdale,N.J.:Lawrenceerlbaumassociates.Tranel,D.&Damasio,A.R.(1985).Knowledgewithoutawareness:anautomaticindexoffacialrecognitionbyprosopagnosics.Science,vol.228,1453-1455.AcknowledgmentsNSFBCS0617699;DornsifeResearchFund.

Subjects7 “Moderate”CongenitalProsopagnosics(mCPs):USCstudents(Age19to22years,5female)selectedasCPsbasedontheirperformanceon:a)thefaceblind.org survey(<70),b)theCambridgeFaceMemoryTest(<70),c)theUSCImageUnderstandingLabCelebrityTest(<80),d)thePI20(>50),andunremarkableneurologicalhistory.

48Controls:USCundergraduates,(Age18to47years,35female)withscoresinthenormalrangeontheabovemeasuresoffacerecognitionability.

1“Extreme”CongenitalProsopagnosic (GJ): MaleAge33,faceblind.org survey:18%,CambridgeFaceMemoryTest:28%,USCImageUnderstandingLabCelebrityTest:25%,andPI20:90.

1“Extreme”AcquiredProsopagnosic (MJH): MaleAge53,faceblind.org survey:3%,CambridgeFaceMemoryTest:38%, USCImageUnderstandingLabCelebrityTest:26%,andPI20:83.Alsoevidencessomesimultagnosia.

Detectionisalmostalwaysaccompaniedbyidentification

Thereisthusnoevidenceforafamiliaritysignal(Tranel &Damasio,1985)independentofidentification.

Celebritydetectionimproveswithhigher

ratedfamiliarityPriortotheRSVPandDoppelgänger Discriminationtrials,subjectsweregiventhenamesofthecelebritiesandratedtheirfrequencyofhavingseenthefaces(1=low;5=high).

DoppelgängerDiscriminationTask

Subjectsviewed100pairsoffaces,oneofwhichwasacelebrity.Thenon-celebrityheadshotwasselectedtobeassimilarasreasonablypossiblegiventhesamplingoftheheadshotgalleriesofnon-celebrities.Thephotogenicqualityofthenon-celebrityheadshotswereindistinguishablefromthecelebrities,andnoneofthefacesunderwentremovalofhairoralterationoffaceshapeasintheCFMT.Nolimitwasimposedonthetimerequiredforaresponse.

CorrectAnswer:Right CorrectAnswer:Left

PercentIdentificationfollowing

AccurateDetection

Controls ControlsCPs CPs

(QuestionsandOurAnswers)TheDetectionofanUnspecifiedFamiliarFace

RSVPTasks200RSVPsequenceseachwith30images.Agivensequenceconsistedentirelyofeitherfacesorobjects.

Bothtaskscouldberegardedas“NegativeDetection”(Intraub,1981)inthatthesubjecthadtodetectanobjectthatwasnotaninstanceofaprespecified category(e.g.,“NotanArticleofClothing”),andforfaces,acelebrity—anycelebrity—amongnon-celebrities.Atargetwaspresentinhalfthesequences,thoughneverinthefirstorlastsiximages.Thesubjecthadtoreportwhatorwhothetargetwas,e.g.,“anaxe”or“LeonardoDiCaprio”orsomeidentifyinginformation,e.g.,“TheguyinTitanic.”Categoriesforobjectswereclothing,plants,tools,animals,andmodesoftransportation.Onanygivenobjecttrial,animagefromanyofthesecategoriesmightbeusedasatarget.

ObjectTrials:Allsubjectsperformednearceiling.

FaceTrials:BothControlsandmCPs performedwellabovechanceonnegativedetectionoffaces,withtheControlssuperiortoCPs.

Objects

RSVPCelebrityDetectionandNegativeObjectDetectionTask

chance

FacesControls

mCPs

xCP (GJ)

PresentationRates

Faces Objects

RSVPPros

DoppelgängerControls

DoppelgängerPros

RSVPControls

Scanforacopy!

An Image-based Approach for Detecting Faces Carved in

Eigen Representations: Detecting faces in images

Detecting and Aligning Faces by Image Retrieval · Detecting and Aligning Faces by Image Retrieval Xiaohui Shen1 Zhe Lin2 Jonathan Brandt2 Ying Wu1 1Northwestern University 2Adobe

Dynamic texture analysis for detecting fake faces in video ... · 1 Dynamic texture analysis for detecting fake faces in video sequences Mattia Bonomiy, Cecilia Pasquini and Giulia

DRAFT PMB BENEFIT DEFINITION GUIDELINE: SCHIZOPHRENIA Definition Project... · F25.9 Schizoaffective disorder, unspecified F28 Other nonorganic psychotic disorders F29 Unspecified

Detecting Hardware Trojans in Unspecified Functionality Using Mutation Testing Nicole Fern K.-T. Tim Cheng UC Santa Barbara 1

Direction Unspecified Direction Direction X

Detecting Faces in Impoverished Images · Detecting Faces in Impoverished Images Antonio Torralba and Pawan Sinha AI Memo 2001-028November 2001 CBCLMemo 208 ©2001 massachusetts institute

Growth Trends and Patterns Discussionbos.stafford.va.us/2017/02102017/9.pdf · 2017. 2. 10. · Widewater Heights MF Unspecified Unspecified 0 Unspecified 10/17/06 Griffis-Widewater

Detecting Visually Observable Disease Symptoms from Faces...detecting process running on testing dataset is introduced in section 3.3 Fig. 2 An example of data labeling process by

SIDE-EFFECTS, ADVERSE REACTIONS OR UNSPECIFIED REACTIONS

Detecting Faces in Images : A Survey

Detecting and Tracking Human Faces in Videos · Detecting and Tracking Human Faces in Videos Yadong Li, Ardeshir Goshtasby & Oscar Garcia Computer Science and Engineering Department

Representing Images Detecting faces in imagesbhiksha/courses/mlsp.fall2009/class7/slides.pdf · Representing Images Detecting faces in images Class 5. 15 Sep 2009 Instructor: Bhiksha

DON’T CARE CONDITIONS Functions that have unspecified output for some input combinations are called incompletely specified functions. Unspecified minterms

Detecting Masked Faces in the Wild With LLE-CNNsopenaccess.thecvf.com/content_cvpr_2017/papers/Ge...Detecting Masked Faces in the Wild with LLE-CNNs Shiming Ge1, Jia Li2∗, Qiting

1 Face Detection - Advanced Topics in Information Processing Junhong Liu Agenda Introduction of Face Detection Detecting Faces in A Single Image Face Image

MedStar Health, Inc. · C50.919 Malignant neoplasm of unspecified site of unspecified female breast C76.40 Malignant neoplasm of unspecified upper limb C76.50 Malignant neoplasm of

A novel method for detecting lips , eyes and faces in real time

Detecting faces in images: a survey - Pattern …engr.case.edu/merat_francis/EECS 490 F04/References/Face...Detecting Faces in Images: A Survey Ming-Hsuan Yang, Member, IEEE, David

Introduction Detecting Faces in a Single Image –Knowledge-Based Methods –Feature-Based Methods –Template Matching –Appearance-Based Methods Face Image

Detecting Photoshopped Faces by Scripting …openaccess.thecvf.com/content_ICCV_2019/papers/Wang...Detecting Photoshopped Faces by Scripting Photoshop Sheng-Yu Wang1 Oliver Wang2 Andrew

Eigen representations; Detecting faces in images

Detecting and Aligning Faces by Image Retrievalcourses.cs.washington.edu/courses/cse590v/13au/CVPR13... · 2013-10-03 · Detecting and Aligning Faces by Image Retrieval Xiaohui Shen1

ENCLOSURE 4: UNSPECIFIED GASTROINTESTINAL ILLNESS

Detecting Human Faces in Color Images_ElseVier_1999

Detecting Photoshopped Faces by Scripting Photoshop...Detecting Photoshopped Faces by Scripting Photoshop Sheng-Yu Wang1 Oliver Wang2 Andrew Owens1 Richard Zhang2 Alexei A. Efros1

Costa Rica/Eastern Central Pacific Longlines (unspecified)

Unspecified Call Drop - Test Case 2012

Hoe, E; Boelsen, LK; UNSPECIFIED; UNSPECIFIED (2015) Re ...researchonline.lshtm.ac.uk/2210888/1/pone.0129199.pdf · demonstrated aprotectiverole forIL-17Aagainst pneumococcal carriage

COVID-19 vaccine brand unspecified analysis print

3 · Web view Code Meaning Addinfo 1 permanent system error (unspecified) 2 temporary system error (unspecified) 3 unsupported search (unspecified) 4 Terms only exclusion (stop) words

Somatostatin Analogs · K52.9 Noninfective gastroenteritis and colitis, unspecified K56.1 Intussusception K56.2 Volvulus K56.50 Intestinal adhesions [bands], unspecified as to partial

Wheelchair Seating ICD-10 Reference - quantumrehab.com seating icd -10... · Hemiplegia and hemiparesis following unspecified cerebrovascular disease affecting unspecified side L89.130

Detecting Faces in Images: A Survey - CS Course Webpagescourses.cs.tamu.edu/rgutier/cpsc689_s07/yang2002faceDetection... · Detecting Faces in Images: A Survey By: Ming-Hsuan Yang,