real-time face detection and recognition

1
Zelun Luo, Anarghya Mitra Mentor: Jia-Bin Huang. Advisor: Narendra Ahuja. Department of Electrical and Computer Engineering University of Illinois at Urbana-Champaign Make a robust system capable of identifying multiple faces with a learning algorithm for identifying faces not in its database. Integral Image Cascade Architecture Haar-like Features Very few faces in an image Most sub-windows rejected early since they are not faces Each successive classifier is trained only on those selected samples which pass through the preceding classifiers. s(1) = A s(2) = A + B s(3) = A + C s(4) = A + B + C + D The sum within D can be computed as: s(4)+s(1)-s(2)-s(3). (x, y) A B C D 1 2 3 4 The value of the integral image at point (x, y) is the sum of all the pixels above and to the left of x, y, inclusive: where s(x, y) is the integral image and i(x, y) is the original image. The area around the eyes is lighter than the eyes itself i.e. the nose is brighter than the eyes on a normalized graph. The area on top and below the eyes is lighter than the eyes. Our face can be used to control computers. One example is a game we can play with our nose acting as the mouse. Main ideas - Very few faces in an image => cascade structure of weak classifiers Evaluate haar-like features in O(1) time using integral image Select discriminative features with Adaboost [Viola & Jones, 2004] Viola, P. & Jones, M. J. (2004). Robust real-time face detection. International journal of computer vision, 57(2), 137154. [Yang, 2002] Yang, M.-H. (2002). Kernel eigenfaces vs. kernel fisherfaces: Face recognition using kernel methods. In Proceedings of the Fifth IEEE International Conference on Automatic Face and Gesture Recognition Original images Eigenfaces encode the variation in the training set. Each image can be represented using a linear combination of eigenfaces. Can be used in identity verification along with fingerprint and iris recognition systems. Reconstruction using eigenfaces

Upload: jia-bin-huang

Post on 10-Nov-2014

1.105 views

Category:

Documents


2 download

DESCRIPTION

Code: https://github.com/amitra93/FaceRecognition

TRANSCRIPT

Page 1: Real-time Face Detection and Recognition

Zelun Luo, Anarghya Mitra

Mentor: Jia-Bin Huang. Advisor: Narendra Ahuja.

Department of Electrical and Computer Engineering

University of Illinois at Urbana-Champaign

Make a robust system capable of identifying multiple faces with a

learning algorithm for identifying faces not in its database.

Integral Image

Cascade Architecture

Haar-like Features

• Very few faces in an image

• Most sub-windows rejected

early since they are not

faces

• Each successive classifier

is trained only on those

selected samples which

pass through the

preceding classifiers.

s(1) = A s(2) = A + B s(3) = A + C s(4) = A + B + C + D

The sum within D can be

computed as:

s(4)+s(1)-s(2)-s(3).

(x, y)

A B

C D

1 2

3 4

The value of the integral image

at point (x, y) is the sum of all

the pixels above and to the left

of x, y, inclusive:

where s(x, y) is the integral

image and i(x, y) is the original

image.

• The area around the

eyes is lighter than

the eyes itself – i.e.

the nose is brighter

than the eyes on a

normalized graph.

• The area on top and

below the eyes is

lighter than the

eyes.

Our face can be used to control

computers. One example is a

game we can play with our

nose acting as the mouse.

Main ideas -

• Very few faces in an image => cascade structure of weak

classifiers

• Evaluate haar-like features in O(1) time using integral

image

• Select discriminative features with Adaboost

[Viola & Jones, 2004] Viola, P. & Jones, M. J. (2004). Robust real-time face

detection. International journal of computer vision, 57(2), 137–154.

[Yang, 2002] Yang, M.-H. (2002). Kernel eigenfaces vs. kernel fisherfaces:

Face recognition using kernel methods. In Proceedings of the Fifth IEEE

International Conference on Automatic Face and Gesture Recognition

Original images

• Eigenfaces encode the variation

in the training set.

• Each image can be represented

using a linear combination of

eigenfaces.

Can be used in identity

verification along with

fingerprint and iris recognition

systems.

Reconstruction

using eigenfaces