A Comprehensive Guide to Real-Time Face Detection using AdaBoost and Integral Image, Slides of Electronics engineering

Download A Comprehensive Guide to Real-Time Face Detection using AdaBoost and Integral Image and more Slides Electronics engineering in PDF only on Docsity!

A Robust Real Time Face

Detection

Outline

 AdaBoost – Learning Algorithm

 Face Detection in real life

 Using AdaBoost for Face Detection

 Improvements

 Demonstration

Boosting

 The Horse-Racing Gambler Problem

• Rules of thumb for a set of races
• How should we choose the set of races in order

to get the best rules of thumb?

• How should the rules be combined into a single

highly accurate prediction rule?

 Boosting!

Boosting

AdaBoost - algorithm

∑

=

≠

= ×

−

T

t

t t

t

e h x y

t e h x y

t t

t

t t

t t i D t i i

t

m m i i

H x sign h x

Z

Z

D i D i

h X h x y

D

y T

D i m

x y x y x X y Y

i i i t

i i t

t

1

if ( )

if ( )

1

~

1

1 1

Output thefinalhypothesis:

where isanormalizationfactor

Update ( )

ln( 2

Choose

Get weakhypothesis : { 1 , 1 } witherror Pr [ ( ) ]

Select thebest weakclassifierusingdistribution

For 1 ..

Initialize ( ) 1 /

Given( , ),..,( , )where , { 1 , 1 }

α

α

example (^) decision

distribution

step

AdaBoost – training error

 Freund and Schapire (1997) proved that:

 AdaBoost adapts to the error rates of the

individual weak hypotheses.

• Therefore it is called ADABoost.

t t

err H e γ

T

t

t

ε

γ

= −

∑

≤

=

−

2

1

' ( ) , where

1

2

AdaBoost – generalization error

 The analysis implies that boosting will overfit

if the algorithm is run for too many rounds

 However, it was observed empirically that

AdaBoost does not overfit,

• even when run thousands of rounds.

 Moreover, it was observed that the

generalization error continues to drive down

long after training error reached zero

AdaBoost – generalization error

 An alternative analysis was presented by

Schapire et al. (1998), that suits the empirical

findings

y h x x y t

t t

m

d err H x y O

∑

∑

≤ ≤ + ≥

= α

α

( ) margin( , )

where :

2

Pr[ ( ) Pr'[margin( , ) ] ( ) ] 1 -

AdaBoost – different point of view

 Sometimes it is advantageous to minimize some

other (non-negative) loss function instead of the

number of classification errors

 For AdaBoost the loss function is

 This point of view was used by Collins, Schapire

and Singer (2002) to demonstrate that AdaBoost

converges to optimality

∑

=

−

n

i

i i

y f x

1

exp( ( )) α

Face Detection

(not face recognition)

Face Detection in Human

There are ‘processes of face detection’

Faces Are Special

We humans analyze faces in a ‘different way’

Faces Are Special

We analyze faces in a

‘different way’

Face Recognition in Human

We analyze faces ‘in a specific location’