Face Recognition and Detection .


51 views
Uploaded on:
Description
CSE 576, Spring 2008. Face Recognition and Detection. 2. Acknowledgment issues. What is it?Object and scene recognitionWho is it?Identity recognitionWhere is it?Object detectionWhat are they doing?ActivitiesAll of these are order problemsChoose one class from a rundown of conceivable hopefuls.
Transcripts
Slide 1

Confront Recognition and Detection The "Margaret Thatcher Illusion", by Peter Thompson Computer Vision CSE576, Spring 2008 Richard Szeliski Face Recognition and Detection

Slide 2

Recognition issues What is it? Question and scene acknowledgment Who is it? Character acknowledgment Where is it? Protest discovery What are they doing? Exercises All of these are grouping issues Choose one class from a rundown of conceivable competitors Face Recognition and Detection

Slide 3

What is acknowledgment? An alternate scientific categorization from [Csurka et al. 2006]: Recognition Where is this specific question? Classification What sort of object(s) is(are) display? Content-based picture recovery Find me something that appears to be comparable Detection Locate all occasions of a given class Face Recognition and Detection

Slide 4

Readings C. Priest, "Neural Networks for Pattern Recognition", Oxford University Press, 1998, Chapter 1. Forsyth and Ponce, Chap 22.3 (through 22.3.2- - eigenfaces) Turk, M. what\'s more, Pentland, A. Eigenfaces for acknowledgment . Diary of Cognitive Neuroscience, 1991 Viola, P. A. also, Jones, M. J. (2004). Vigorous ongoing face location. IJCV , 57(2), 137–154. Confront Recognition and Detection

Slide 5

Sources Steve Seitz, CSE 455/576 , past quarters Fei-Fei, Fergus, Torralba, CVPR\'2007 course Efros, CMU 16-721 Learning in Vision Freeman, MIT 6.869 Computer Vision: Learning Linda Shapiro, CSE 576, Spring 2007 Face Recognition and Detection

Slide 6

Today\'s address Face acknowledgment and discovery shading based skin identification acknowledgment: eigenfaces [Turk & Pentland] and parts [Moghaddan & Pentland] location: boosting [Viola & Jones] Face Recognition and Detection

Slide 7

Face identification How to tell if a face is available? Confront Recognition and Detection

Slide 8

Skin identification Skin pixels have a particular scope of hues Corresponds to region(s) in RGB shading space Skin classifier A pixel X = (R,G,B) is skin in the event that it is in the skin (shading) area How to discover this locale? skin Face Recognition and Detection

Slide 9

Skin identification Learn the skin locale from cases Manually name skin/non pixels in at least one "preparing pictures" Plot the preparation information in RGB space skin pixels appeared in orange, non-skin pixels appeared in dim some skin pixels might be outside the area, non-skin pixels inside. Confront Recognition and Detection

Slide 10

Skin classifier Given X = (R,G,B): how to figure out whether it is skin or not? Closest neighbor find named pixel nearest to X Find plane/bend that isolates the two classes prominent approach: Support Vector Machines (SVM) Data displaying fit a likelihood thickness/conveyance model to every class Face Recognition and Detection

Slide 11

Probability X is an irregular variable P(X) is the likelihood that X accomplishes a specific esteem called a PDF likelihood dissemination/thickness work a 2D PDF is a surface 3D PDF is a volume ceaseless X discrete X Face Recognition and Detection

Slide 12

Probabilistic skin arrangement Model PDF/instability Each pixel has a likelihood of being skin or not Skin classifier Given X = (R,G,B): how to figure out whether it is skin or not? Pick understanding of most astounding likelihood Where do we get and ? Confront Recognition and Detection

Slide 13

Learning contingent PDF\'s We can compute P(R | skin) from an arrangement of preparing pictures It is basically a histogram over the pixels in the preparation pictures every canister R i contains the extent of skin pixels with shading R i This doesn\'t fill in too in higher-dimensional spaces. Why not? Approach: fit parametric PDF capacities basic decision is turned Gaussian focus covariance Face Recognition and Detection

Slide 14

Learning contingent PDF\'s We can ascertain P(R | skin) from an arrangement of preparing pictures But this isn\'t exactly what we need Why not? How to figure out whether a pixel is skin? We need P(skin | R) not P(R | skin) How would we be able to get it? Confront Recognition and Detection

Slide 15

Bayes lead as far as our issue: what we measure ( probability ) space information ( earlier ) what we need ( back ) standardization term What would we be able to use for the earlier P(skin)? Space information: P(skin) might be bigger in the event that we know the picture contains a man For a representation, P(skin) might be higher for pixels in the inside Learn the earlier from the preparation set. How? Confront Recognition and Detection P(skin) is extent of skin pixels in preparing set

Slide 16

Bayesian estimation Bayesian estimation Goal is to pick the name (skin or ~skin) that augments the back ↔ limits likelihood of misclassification this is called Maximum A Posteriori (MAP) estimation probability back (unnormalized) Face Recognition and Detection

Slide 17

Skin identification comes about Face Recognition and Detection

Slide 18

This same strategy applies in more broad conditions More than two classes More than one measurement General order Example: confront location Here, X is a picture area measurement = # pixels every face can be considered as a point in a high dimensional space H. Schneiderman, T. Kanade. "A Statistical Method for 3D Object Detection Applied to Faces and Cars". CVPR 2000 Face Recognition and Detection

Slide 19

Today\'s address Face acknowledgment and discovery shading based skin identification acknowledgment: eigenfaces [Turk & Pentland] and parts [Moghaddan & Pentland] location: boosting [Viola & Jones] Face Recognition and Detection

Slide 20

Eigenfaces for acknowledgment Matthew Turk and Alex Pentland J. Intellectual Neuroscience 1991

Slide 21

Linear subspaces Classification can be costly: Big pursuit prob (e.g., closest neighbors) or store substantial PDF\'s Suppose the information focuses are organized as above Idea—fit a line, classifier measures separation to line change over x into v 1 , v 2 arranges What does the v 2 facilitate measure? separation to line utilize it for characterization—close to 0 for orange pts What does the v 1 arrange measure? position along line utilize it to determine which orange point it is Face Recognition and Detection

Slide 22

Dimensionality diminishment Dimensionality lessening We can speak to the orange focuses with just their v 1 facilitates ( since v 2 directions are all basically 0) This makes it much less expensive to store and think about focuses A greater arrangement for higher dimensional issues Face Recognition and Detection

Slide 23

Linear subspaces Consider the variety along bearing v among the majority of the orange focuses: What unit vector v limits var ? What unit vector v boosts var ? Solution: v 1 is eigenvector of A with biggest eigenvalue v 2 is eigenvector of A with littlest eigenvalue Face Recognition and Detection

Slide 24

Principal part investigation Suppose every information point is N-dimensional Same method applies: The eigenvectors of A characterize another organize framework eigenvector with biggest eigenvalue catches the most variety among preparing vectors x eigenvector with littlest eigenvalue has minimum variety We can pack the information utilizing the main few eigenvectors compares to picking a "straight subspace" speak to focuses on a line, plane, or "hyper-plane" these eigenvectors are known as the essential segments Face Recognition and Detection

Slide 25

The space of countenances A picture is a point in a high dimensional space A N x M picture is a point in R NM We can characterize vectors in this space as we did in the 2D case = + Face Recognition and Detection

Slide 26

Dimensionality diminishment The arrangement of appearances is a "subspace" of the arrangement of pictures We can locate the best subspace utilizing PCA This resembles fitting a "hyper-plane" to the arrangement of confronts spread over by vectors v 1 , v 2 , ..., v K any Face Recognition and Detection

Slide 27

Eigenfaces PCA extricates the eigenvectors of A Gives an arrangement of vectors v 1 , v 2 , v 3 , ... Every vector is a course in face space what do these resemble? Confront Recognition and Detection

Slide 28

Projecting onto the eigenfaces The eigenfaces v 1 , ..., v K traverse the space of countenances A face is changed over to eigenface arranges by Face Recognition and Detection

Slide 29

Recognition with eigenfaces Algorithm Process the picture database (set of pictures with names) Run PCA—register eigenfaces Calculate the K coefficients for every picture Given another picture (to be perceived) x , figure K coefficients Detect if x is a face If it is a face, who is it? Find nearest marked face in database closest neighbor in K-dimensional space Face Recognition and Detection

Slide 30

Choosing the measurement K what number eigenfaces to utilize? Take a gander at the rot of the eigenvalues the eigenvalue reveals to you the measure of difference "in the course" of that eigenface overlook eigenfaces with low change i = K NM eigenvalues Face Recognition and Detection

Slide 31

View-Based and Modular Eigenspaces for Face Recognition Alex Pentland, Baback Moghaddam and Thad Starner CVPR\'94

Slide 32

Part-based eigenfeatures Learn a different eigenspace for every face include Boosts execution of consistent eigenfaces Face Recognition and Detection

Slide 33

Bayesian Face Recognition Baback Moghaddam, Tony Jebara and Alex Pentland Pattern Recognition 33(11), 1771-1782, November 2000 (slides from Bill Freeman, MIT 6.869, April 2005)

Slide 34

Bayesian Face Recognition Face Recognition and Detection

Slide 35

Bayesian Face Recognition Face Recognition and Detection

Slide 36

Bayesian Face Recognition Face Recognition and Detection

Slide 37

Morphable Face Models Rowland and Perrett \'95 Lanitis, Cootes, and Taylor \'95, \'97 Blanz and Vetter \'99 Matthews and Baker \'04, \'07

Slide 38

Morphable Face Model Use subspace to model versatile 2D or 3D shape variety (vertex positions), notwithstanding appearance

Recommended
View more...