Label the group photo locate and identify faces and label them.

Label the group photo

locate and identify faces and label them

Ramona CiulpanWebmaster

Label the group photo

locate and identify faces and label them

Kornel Toth SVM, Database

Label the group photo

locate and identify faces and label them

Mircea FocşaPPT Presentation

Label the group photo

locate and identify faces and label them

Krisztian Olle Project manager

Label the group photo

locate and identify faces and label them

Project Description

Label the group photo- locate and identify faces and label them

 • Input group photo ( for example 10 people)• Segment it to isolate people/faces• Number the faces• Extract the faces• Build of library of faces• From photos of similar faces try to find that person on

the group photo

Face DetectionFinding faces is complicated?

Possible solution

Neural Network Template matching Principal Component Analysis Support Vector Machine

Possible solution

Neural Network Template matching Principal Component Analysis Support Vector Machine

Support Vector Machines algorithm

Minimize W(Λ)=- ΛT 1 + 1/2 Λ T D Λ ΛSubject to

ΛT y = 0Λ-C1 ≤ 0- Λ ≤ 0

Face detection (I)• Create an images database

– 266 pictures: 150 faces + 116 non-faces

. . .

• Preprocessing– Gray scale transformation– Histogram equalization– Adjust resolution to 30x40 pixel

• Training the SVM based on that 266 vectors, using a polynomial kernel.

Face detection (II)

• Moving over the input image with a 30x40 pixel sub window

• Histogram equalization of a sub window• Classification by SVM• Removing intersections

Face recognition

• Training the SVM based on the people faces who want to recognize

• Classifying the detected faces• Labeling the known faces

Implementation (I)

Input group photo

Isolate people / faces

Number the faces

Implementation (II)

Input group photo

Isolate people / faces

Number the faces

Implementation (III)

Extract the faces

Implementation (IV)

Build of library of faces

Implementation (V)

Label the faces

Train the SVM with new set of vectors

Results

Image name Resolution# of

faces# of tests

# of found faces

Time (sec.)

False

Classific.

csoport.pgm 600x398 15 9600 13 11.45 6

team2.pgm 700x465 4 13020 4 15.077 0

team3.pgm 600x398 4 9600 4 14.671 0

team31.pgm 500x331 4 6700 4 10.499 0

team4.pgm 500x331 4 6700 4 10.515 0

team41.pgm 400x265 4 4240 4 5.984 0

test5.pgm 500x332 5 6700 4 9.937 1

Examples

Future Plans

• Multi-resolution image pyramid• Better face databases• Better face recognition databases• Improve the speed • Improve the masking technique

Thank You!How many faces ?

11

33

22

44 55 66

77

8899

1111

1010

References

• Open Source Computer Vision Library Reference Manual http://developer.intel.com/

• Guodong Guo, Stan Z. Li, and Kapluk Chan: “Face Recognition by Support Vector Machines” Proceeding of Fourth IEEE International Conference on Automatic Face and Gesture Recognition, 2000 Grenoble, France.

• Edgar Osuna, Robert Freund: “Training Support Vector Machines: an Application to Face Detection”. Proceeding of CVPR’97,  1997 Puerto Rico

• The Face Detection Homepage http://home.t-online.de/home/Robert.Frischholtz/face.htm