Quantitative Measurement of
Face Detection Algorithm Performance
Setiawan Hadi
Mathematics Department
Universitas Padjadjaran (UNPAD)
Bandung - INDONESIA
The 4th International Conference on
Outline
1
Background and Motivation
Background
Problem Definition
Motivation
Outline
1
Background and Motivation
Background
Problem Definition
Motivation
Why Human Face ?
Background
Why Human Face ?
Background
Sample Results of Face Detection Algorithms
Background
(a)
(b)
(c)
Outline
1
Background and Motivation
Background
Problem Definition
Motivation
Problem Definition
☞
Most face detection algorithms do not clearly define a
What constitutes successful face detection?
Problem Definition
(a)
(b)
(a) Test Image
Outline
1
Background and Motivation
Background
Problem Definition
Motivation
Motivation
☞
A uniform criterion should be adopted to define or to
Outline
1
Background and Motivation
Background
Problem Definition
Motivation
Measuring Accuracy Technique
Development Strategy
☞
Using bounding box determination (GTBB and DBB
Measuring Accuracy Technique
Development Strategy
☞
GTBB : Ground Truth Bounding Box
Measuring Accuracy Technique
Development Strategy
☞
GTBB : Ground Truth Bounding Box
Measuring Accuracy Technique
Development Strategy
D
=
1
4
(
d
TL+
d
TR+
d
BL+
d
BR)
(1)
d
TL,
d
TR,d
BLand
d
BRare Euclidean distances of GTBB and DBB.
d
TL=
q
(
GTBB
TLx−
DBB
TLx)
2+ (
GTBB
TLy−
DBB
TLy)
2(2)
d
TR=
q
(
GTBB
TRx−
DBB
TRx)
2+ (
GTBB
TRy−
DBB
TRy)
2(3)
d
BL=
q
(
GTBB
BLx−
DBB
BLx)
2+ (
GTBB
BLy−
DBB
BLy)
2(4)
Outline
1
Background and Motivation
Background
Problem Definition
Motivation
Characteristics
Implementation Scenario
☞
Using
DeWa
, a framework for multiple human face
detection in complex background digital image that utilized
multiaspect approach
☞
Algorithm for measuring single face and multiple faces
☞
Using single human face image databases and multiple
DeWa Framework
Human Face Databases
Mechanism
Outline
1
Background and Motivation
Background
Problem Definition
Motivation
Experimental Result
Using Single Face Image Databases
Database
Image
DR
TN
FP
Speed
FERET
735
90.63%
9.27%
0.0%
0.28 detik
DWI
347
89.78%
10.22%
0.0%
0.28 detik
Using Complex Face Image Databases
Database
Face
DR
TN
FP
Speed
mDWI
488
92.21%
7.79%
7.96%
0.19 s/f
VALID
298
91.28%
8.72%
7.67%
0.15 s/f
Conclusion
☞
A technique for quantitatively measuring face detection algorithm performance
has been proposed
☞
A face detection algorithm called DeWa is used for implementing the technique
☞
It has been tested successfully using two single face image database and two
multiple face image database
Future Works
☞
Using different bounding objects such as ellips and facial edge boundary
Selected References
1
Hadi, S, Ahmad, A.S, Suwardi, I.S., Wazdi, F., DeWa : A multiaspect approach
for multiple face detection in complex scene digital image, ITB Journal of
Information and Communication Technology, Vol. 1C No. 1, 2007
2
Hsu, R.L., Abdel-Mottaleb, M., Jain, A.K.,Face Detection in Color Images, IEEE
Transaction on Pattern Analysis and Machine Intelligence, Vol. 24 No.5 Pp.
696-706, 2002
3
Rowley, H. A., Baluja, S. dan Kanade, T., Neural Network-based Face Detection,
IEEE Trans. on Pattern Analysis and Machine Intelligence , Vol. 20 No. 1 Pp.
23-28, 1998