Abstract - The exact age estimation is often treated as a
classification problem; while it can be formulated as a
regression problem. In this article, three different classifiers based
on KNN classifier's concept for facial age estimation were
designed and developed to achieve high efficiency calculation of
facial age estimation. In the first classifier, we adopt KNN-distance
approach to calculate minimum distance between test face
image and all instances belong to the class that has the highest
number of nearest samples. Additionally, in the second
classifier a modified-KNN version was proposed and the
classifier scoring results interpolated to calculate the exact age
estimation. Furthermore, KNN-regression classifier as third
classifier that used to combine the classification and regression
approaches to improve the accuracy of the age estimation
system. Moreover, we compared age estimation errors under
two situations: case 1, age estimation is performed without
discrimination between males and females (gender unknown);
and case 2, age estimation is performed for males and females
separately (gender known). Results of experiments conducted
on well know benchmark FG-NET Database show the
effectiveness of the proposed approach.
31 ĐỀ THI THỬ VÀO LỚP 10 - TIẾNG ANH - FORM MỚI 2025 - 40 CÂU HỎI - BÙI VĂN V...
Three different classifiers for facial age estimation based on K-nearest neighbor
1. THREE DIFFERENT CLASSIFIERS
FOR FACIAL AGE ESTIMATION BASED
ON K-NEAREST NEIGHBOR
By
Alaa Tharwat
Electrical Engineering Department, Suez Canal University,
Fac. of Eng. Ismailia, EGYPT
ICENCO 28-29/12/2013 – Cairo Egypt
4. Introduction
•
•
Age estimation is the determination of
a person’s age based on biometric
features (2D Face image).
Facial aging effects are mainly
attributed to:
•
•
•
Bone growth
Skin related deformations associated with
the introduction of wrinkles (texture
changes)
Muscle strength
5. Introduction
Background of Facial Age
Estimation
Used FGNET, Morph, Own
database
Applications
•
Age-Based Access Control
•
Conventional classification and
feature extraction methods.
Age Adaptive Human Machine
Interaction (HCI)
•
Used Local, Global , or feature
fusion method.
Age Invariant Person
Identification
•
Data mining and organization
Classification or Regression.
8. The proposed age estimation
approach: Feature extraction and fusion
Local binary pattern (LBP) Features
Sub - Window
image
150
120
160
152
150
60
40
20
22
20
35
30
30
33
30
30
35
37
30
35
40
43
45
40
37
70
60
50
45
40
Thresholding
150
120
160
1
1
1
60
40
20
1
135
0
35
30
30
0
0
LBP Code
(10000111)2
=135
0
Illustration of LBP. Typically the binary codes
obtained by local thresholding are transformed into
decimal codes.
9. The proposed age estimation
approach: Feature extraction and fusion
Landmarks (Fiducial) Points
Some images of the FG-NET database with
landmarks
10. The proposed age estimation
approach: Feature extraction and fusion
Feature fusion
Advantage
the fusion in feature level contains richer information than classification level
Disadvantage
• The features may be incompatible, so it needs to normalization.
• The new feature vector needs more CPU time and memory (Dimensionality
problem), so it needs to dimensionality reduction techniques.
Local features (f1=[l1,…….,lm])
Normalization
(f’1)
Global features (f2=[g1,……..,gn])
Normalization
(f’2)
New Feature vector
fnew =[f’1 f’2]
=[l1,…….,lm,g1,……..,gn]
11. The proposed age estimation
approach: Three Classification
The first classifier
The second classifier
KNN-distance approach to calculate minimum
distance between test face image and all
instances belong to the class that has the
highest number of nearest samples.
A modified-KNN version was proposed and the
classifier scoring results interpolated to calculate
the exact age estimation.
The third classifier
KNN-regression classifier as third classifier that
used to combine the classification and
regression approaches to improve the accuracy
of the age estimation system
12. Experimental Results
[14] http://www.fgnet.rsunit.com/.
The FG-NET Aging Database [*] is used in the experiment. There are 1,
002 face images from 82 subjects in this database. Each subject has 618 face images at different ages. The ages are distributed in a wide
range from 0 to 69. Besides age variation, most of the age-progressive
image sequences display other types of facial variations, such as
significant changes in 3D pose, illumination, expression, etc.
16. Conclusions
Proposed classifiers achieved relatively good age
estimation from 2D face images
Proposed age estimation system based on three
proposed classifiers (KNN-Distance, ModifiedKNN, and KNN-Regression) gives good age
estimation process and estimating age when gender
is known
Estimating age from males achieves results better
than females.