Unformatted text preview:

MAIN MENUPREVIOUS MENU---------------------------------Search CD-ROMSearch ResultsPrintFace Recognition byHumans and MachinesA Tutorial SurveyCVPR’01 Short CourseInstructor:Baback Moghaddam1063-6919/03 $17.00 © 2003 IEEENote to RegistrantsAt press time these course notes were still in preparation andcontributed material from outside was still trickling in. As aconsequence this collection of slides is about 80% complete.Final version of the course slides can be found at:http://www.merl.com/people/moghaddam/cvpr01.htmlIf you have any questions contact me at [email protected] Outline• Brief History• Introduction to Key Problems• Face Perception in Humans• Automatic Face Recognition– face detection– neural network methods– features vs. templates– subspace methods– FERET test protocol– lighting/pose techniques– 2D/3D models• Future DirectionsA Brief History (1900-2000)Kaya 1972Kanade 1977Galton 1888Features Period Templates PeriodTurk & Pentland 1992Brunelli & Poggio 19921880 1970 1990FERET 19961980Neural PeriodBledsoe 19643D Models 2D Subspace ModelsCottrel & Fleming 1987Kohonen 1980Chellappa 1997Moghaddam 1995Cootes & Taylor 1997Belhumeur 1998Vetter 20002000Wiskott 1997Sir Francis Galton (1822-1911)• Face Research– “Personal identification and description,” Nature, 1888– “Numeralized profiles for classification and recognition,”Nature, 1910• Eugenics Research– "Hereditary talent and character." (Macmillan's 1865)– Hereditary Genius (1869)– "The possible improvement of the human breed under theexisting conditions of law and sentiment." (Annual Reportof the Smithsonian Institution, 1902)Face Recognition Surveys• Samal & Iyengar, “Automatic Recognition and Analysis of HumanFaces and Facial Expressions,” Pattern Recognition, vol. 25, 1992• Valentin, Abdi, O’Toole & Cottrell, “Connectionist Models of FaceProcessing: A Survey,” Pattern Recognition, vol. 27, 1994• Chellappa, Wilson & Sirohey, “Human and Machine Recognition ofFaces: A Survey,” Proc. IEEE, vol. 83, 1995.• Grudin, “On Internal Representations in Face Recognition Systems,”Pattern Recognition, vol. 33, 2000• Zhao, Chellappa, Rosenfeld & Phillips, “Face Recognition: ALiterature Survey”, UMD CS-TR-4167, 2000Aspects of Face Processing• Recognition– familiarity (membership)• Identification– who is it? (assign identity label)• Verification• Classification– expression, gender, race, age, etcApplications of Face Biometrics• financial transactions• check-in or boarding planes • crossing borders• casting votes• security or surveillance • identity fraud • criminal justice & law enforcement • access to facilities, databases or privileged information, etcFace Publications by Category(from F&G’95/96/98/2000)Recognition35%Detection34%Modeling29%Gender2%Course Outline• Brief History• Introduction to Key Problems• Face Perception in Humans• Automatic Face Recognition– face detection– neural network methods– features vs. templates– subspace methods– FERET test protocol– lighting/pose techniques– 2D/3D models• Future DirectionsWhy Face Recognition is Easy!• It is not general object recognition!• It is a single-class object recognition task– representation & matching can be optimizedWhy Face Recognition is Hard!“The variations between the images of the same face due to illumination and viewing direction are almost always larger than image variations due to change in face identity.” -- Moses, Adini, Ullman, ECCV ‘94Computational Face Models• Feature-based– fiducial points– distances, angles, areas, etc– geometrical• Template-based– holistic– appearance based, images– statisticalFeatures: ProfileFeatures: FrontalTemplatesORL database -- pose/expressionWhole Face RegionsBrunelli & Poggio (1993)Human Face Representation• Prototypes or “schemas”– [Goldstein & Chance 1980]• Feature-based or “configural”– [Roberts & Bruce 1988]• PCA is a good model of human memory[O’Toole et al 1994]• Distinctiveness relates to recall ability• Recognition is very hard with– line-drawings (with no shading)– luminance negatives– upside-down facesHuman Face RepresentationImpeded Face Perception“Mooney” facesThe “Face Inversion”effect (upside-downfaces) has been usedextensively as anexperimental tool!The Human BrainThe Visual System in Primates:Two Pathways: “What” and “Where”dorsalstream:“where”ventral stream:“what”Tommi PoggioHemispheric SpecializationActivation of the right fusiformarea (in the inferotemporalcortex) during face processing(Nakamura et al., 2000)• Right hemisphere is biased forface recognition• Left hemisphere better at feature-based processing (less at holistic)A particular brain wave (N200)occurs most strongly in fusiformregions of the right hemisphere whenindividuals view upright faces, but notwhen viewing inverted (or scrambled)faces (Allison et al., 1994).Face Perception in Humans• Cortical localization in IT/STS [Desimone et al. 1984]• Independent face modules [Bruce et al. 1986]ExpressionGenderRaceAgeFamiliarityIdentityGender PrototypesImages courtesy of University of St. Andrews Perception LaboratoryGender Shape PrototypesO’Toole et al (1998)The average head plus versus minus the first eigenvector for the head surface data is shown. The analysiswas performed on 65 female and 65 male heads. Individual face projections onto this eigenvector werehighly correlated to the gender of the face.“Sex classification is better with 3D head structure than with texture. “A.J. O'Toole, T. Vetter, N.F. Toje and H.H. Bülthoff, H. H. Perception, 26:75-84.Humans vs. MachinesIntelligence/Consciousness usually not (yet)Size of Elements meters meters610− 610− Number of Elements synapses gates1410810Power Consumption 30 W 30 W (CPU)Processing Speed 100 Hz 1 GHzComputational Style parallel / distributed serial / centralizedFault Tolerance yes noLearning Potential yes noCourse Outline• Brief History• Introduction to Key Problems• Face Perception in Humans• Automatic Face Recognition– face detection– neural network methods– features vs. templates– subspace methods– FERET test protocol– lighting/pose techniques– 2D/3D models• Future DirectionsHuman Face PerceptionBradley C. DuchaineVision Sciences LaboratoryDepartment


View Full Document

MIT 6 891 - Humans and Machines

Download Humans and Machines
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Humans and Machines and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Humans and Machines 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?