1 chukwuemeka duruamaku. machine learning, a branch of artificial intelligence, concerns the...

17
CSCI 6175 ADABOOST ADABOOST FACE DETECTION ALGORITHM BASED ON CORRELATION OF CLASSIFIERS 1 CHUKWUEMEKA DURUAMAKU

Upload: gloria-flynn

Post on 18-Jan-2016

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

1

CSCI 6175

ADABOOSTADABOOST FACE DETECTION ALGORITHM BASED ON CORRELATION OF CLASSIFIERS

CHUKWUEMEKA DURUAMAKU

Page 2: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

2

Machine Learning: Definition Machine learning, a branch of artificial intelligence,

concerns the construction and study of systems that can learn from data.

Definition: A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.

EXPERIENCE (E) --—TASKS (T) ------ PERFORMANCE (P)

For example, a machine learning system could be trained on email messages to learn to distinguish between spam and non-spam messages. After learning, it can then be used to classify new email messages into spam and non-spam folders.

Page 3: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

3

Importance of Machine Learning

The amount of knowledge available about certain tasks might be too large for explicit encoding by humans (e.g., medical diagnostics).

Environments change over time hence learning a adaptability are essential.

New knowledge about tasks is constantly being discovered by humans. It may be difficult to continuously re-design systems “by hand”.

Page 4: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

4

BOOSTING

Boosting is a machine learning ensemble meta-algorithm for reducing bias primarily and also variance in supervised learning, and a family of machine learning algorithms which convert weak learners to strong ones.

Can a set of weak learners create a single strong learner?

Weak learner – is a classifier which is only slightly correlated with the true classification.

Strong learner – is a classifier that is arbitrarily well-correlated with the true classification.

 EXAMPLES: LPBoost, TotalBoost,BrownBoost, MadaBoost, LogitBoost,

Page 5: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

5

GENDER: MALE AND FEMALE

CLASSIFIERS EXAMPLE

• HEIGHT, BMI (WEIGHT), AGE, HAIR

LENGTH

Page 6: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

6

AdaBoost Algorithm AdaBoost, short for Adaptive Boosting, is

a machine learning algorithm, formulated by Yoav Freund and Robert Schapire. 

It is a meta-algorithm, and can be used in conjunction with many other learning algorithms to improve their performance.

AdaBoost is adaptive in the sense that subsequent classifiers built are tweaked in favour of those instances misclassified by previous classifiers.

Page 7: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

7

AdaBoost Algorithm

Instead of resampling, uses training set re-weighting

Each training sample uses a weight to determine the probability of being selected for a training set.

AdaBoost is an algorithm for constructing a “strong” classifier as linear combination of “simple” “weak” classifier

Final classification based on weighted vote of weak classifiers

Page 8: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

8

PROFESSORS

Examples of AdaBoost

• ESSAY, GRE, TOEFL, EXTRACURRICULAR

ACTIVITIES, GENDER

Page 9: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

9

Advantages of AdaBoost

Very simple to implement

Feature selection on very large sets of features

AdaBoost adjusts adaptively

Page 10: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

10

Adaboost Face Detection Algorithm Based on Correlation of

Classifiers In order to enhance the ensemble of the traditional

Adaboost algorithm and reduce its complexity, two improved Adaboost algorithms were proposed, which are based on the correlation of classifiers.

In the algorithm, Q-statistic is added in the training weak classifiers, every weak classifier is related not only to the current classifier, but also to previous classifiers as well, which can effectively reduce the weak classifier similarity.

Simulations result in CMU (Carnegies Mellon University) show that the algorithms are of better detection rate and lower false alarm rate, compared with traditional Adaboost algorithm and FloatBoost.

Page 11: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

11

Adaboost Face Detection Algorithm Based on Correlation of

Classifiers C1 – Adaboost

1) Define a set of weak classifiers 2) Train weak classifier by Adaboost; 3) Calculate the Q-statistic against previous

classifier 4) Repeat steps 2, 3, and form the strong

classifier finally. C2 – Adaboost

1) Same as above however Q-statistic is calculated against the all classifiers in the strong classifier ensemble.

Page 12: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

12

Adaboost Face Detection Algorithm Based on Correlation of

Classifiers The researchers presented a novel

approach to training weak classifiers by estimating the correlation between weak classifiers.

According to the value of correlation, they selected the weak classifiers with complementary characteristic and diversity. While eliminating the classifiers which had higher correlation Q.

Page 13: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

13

Dataset

5,000 faces and 5,000 non-faces which were selected from some face libraries and the internet, which including a variety of postures, facial expressions, brightness faces and be normalized to 24 24 u .

There are 11,843 features extracted from each sample.

150 features were trained by Adaboost algorithm based on correlation of classifiers on a 2.4GHz Pentium 4 processor.

A test set consisted of 1100 faces, which 400 faces from CMU Face library and 700 faces from world wide webs, is used to test the performance.

Page 14: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

14

Adaboost Face Detection Algorithm Based on Correlation of

Classifiers

Page 15: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

15

Adaboost Face Detection Algorithm Based on Correlation of

Classifiers

Page 16: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

16

TRADITIONAL ADABOOST VERSUS C2-ADABOOST

Page 17: 1 CHUKWUEMEKA DURUAMAKU.  Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data

17

THANK YOU