lec21

# lec21 - Lecture 21 November 5 10 HW5 due Nov 10 Review...

This preview shows pages 1–6. Sign up to view the full content.

USC CS574: Computer Vision, Fall 2010 Copyright 2010, by R. Nevatia 1 Lecture 21: November 5, 10 • HW5 due Nov 10 •R e v i e w – Parametric classifiers – Cascade classifiers using AdaBoost Learning •T o d a y – PCA and LDA – Support Vector Machines

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
USC CS574: Computer Vision, Fall 2010 Copyright 2010, by R. Nevatia 2 Learning Framework of V&J L 1 L 2 3 samples face T T T F F F Learn weak classifiers from rectangle features Boost strong classifier Bootstrap negative samples Training samples Cascade classifier Weak classifier Layer Cascade
USC CS574: Computer Vision, Fall 2010 Copyright 2010, by R. Nevatia 3 0/1 AdaBoost • Given sample set S ={( x i , y i )}, x i is a feature vector, y i is its label • Initialize sample weights w 1 ,i =1/| S | • For t = 1 : T – For each feature in the feature pool, call the weak learner to get a weak classifier h : -> {0,1} – Select h t with the lowest error e t = E w [| h t ( x i )- y i |] – Update sample weights • Samples with higher errors get a higher weight – Normalize weights to a PDF (add to one) • Output the final ensemble classifier ti i -|h ( )-y | t+ ,i t,i t t t t w= w β , β =e ( -e ) 1 1 1 x   T tt t t t=1 H( )= sgn α h( )-b , α =-log β xx Default 0

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
USC CS574: Computer Vision, Fall 2010 Copyright 2010, by R. Nevatia 4 0/1 AdaBoost S W R 1 Feature Pool H= α 1 h 1 R 2 + 2 h 2
USC CS574: Computer Vision, Fall 2010 Copyright 2010, by R. Nevatia 5 Cascade Learning • Set the max FAR f

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### Page1 / 20

lec21 - Lecture 21 November 5 10 HW5 due Nov 10 Review...

This preview shows document pages 1 - 6. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online