An Aging Face Synthesis Method Based on Feature
R
egions
Morphing and Wavelet Image Enhancement
Weiping Hu
1,2
1 Intelligent Computing and Distributed Information Processing Laboratory, Guangxi University of Science and
Technology, Liuzhou, Guangxi, China
2Guangxi Colleges and Universities Key Laboratory of Intelligent Processing of Computer Images and Graphics, Guilin
University of Electronic Technology,Guilin, Guangxi, China
Keywords: Aging face synthesis, feature region morphing, wavelet image enhancement, Delaunay triangulation.
Abstract: Image morphing method based on trigonometric feature region was used to change contours of face images,
and method of wavelet decomposition and synthesis was used to transfer aging textures, so as to synthesize
aged face image. Experimental results show that better aged face images can be synthesized through our
method and that it has certain practical value.
1 INTRODUCTION
Face recognition has made great progress till now,
and it has been applied in some occasions such as
railway station and supermarket. However face
changes with age, which has a great influence on the
correct rate of face recognition. It is helpful to
improve the recognition effect of face recognition
system if face aging problem was solved. There are
few researches on face aging at present. Face aging
methods can be classified into methods based on
empirical knowledge and methods based on
statistical learning. Skulls and skins that varies with
age are considered to simulate aged face images in
methods based on empirical knowledge. Wu
developed a 3-layer facial structure to simulate the
aging process dynamically(Wu Y,1999). Wu
Xuefeng used active shape model algorithm to
extract children’s face features, and obtained aged
images by changing geometric and texture
features(Wu X F,2015).Large scale face databases
are studied to find the law of how face contours and
textures varied with age in methods based on
statistical learning. Liu et al proposed a method to
estimate aging pattern by aging increment
distribution for re-rendering of facial age effects, so
as to realize face aging(Liu J, 2007). Hu Weiping
combined face morphing algorithm based on the
feature line pairs and wavelet decomposition and
synthesis algorithm to obtain aged face images(Hu
W P,2016). Huang Fenglan used extending face
database and IBSDT algorithm to improve the face
prototype synthesis effect and adopted nonlinear
operator method to enhance face textures(Huang F L,
2017). Liu Zhenyu established a face aging model
through the gated recurrent unit to obtain aging face
smoothly(Liu Z Y,2018).However in general,
research on aging is still in the basic stage.
Considering that there are two distinct stages in
the process of face aging, that is mainly contours
change from children to young people, and skins and
textures mainly change from young to old age, this
study adopts a combination of two different
strategies.Firstly, face contours are morphed by
method of feature region deformation. Then wavelet
transform method is used to enhance facial aging
features. Finally, the aging characteristics are
synthesized.
2 THE COMPOSITION OF FACE
AGING SYNTHESIS SYSTEM
The system consists of four parts: image pre-
processing, contours morphing, extraction and
strengthening of aging characteristics and face aging
features synthesis, as shown in Figure 1.
The image pre-processing part is responsible for
pupils alignment, geometry normalization and
illumination normalization. The deflected face can
be turned into a positive face by the pupil alignment,
the resolution of face images becomes uniform by
geometric normalization, and the illumination
becomes relatively uniform by the histogram
equalization and the gray transformation.
Figure 1: System structure.
In the face morphing part, the feature region
deformation method is adopted. Firstly, the average
face of different age segments is obtained according
to the face database. Secondly the face images are
divided into several triangles using the Delaunay
triangulation method, and then each characteristic
triangle is deformed between the source image and
the target image, so as to obtain face images with
changed contours.
The aging feature extraction and strengthening
part are divided into two parts. The wavelet image
decomposition method is used in extraction part to
extractthe aging features such as wrinkles, eye bags
and decree lines. The following strengthening
module enhances the aging characteristics.
Wavelet image synthesis method is used in the
synthesis part to add the enhanced aging features to
the morphed images, so as to get the final synthetic
images.
The face pre-processing method is described in
Face illumination compensation algorithm based on
symmetrical blocks(Hu W P,2014), here we will not
go into details of them.
3 FACE FEATURE REGIONS
MORPHING METHOD
Face feature points can be manually punctuated and
also can be obtained by ASM algorithm. The FG-
NET aging face database of Cyprus University is
used which provided 68 face features with each
image. Considering the importance of hairstyle in
the aging process,22 extra feature points are added
to characterize hairstyle. In order to get the change
rule of feature points in face aging process, average
faces of each age group are calculated, and aged face
image can be obtained according to difference of
average faces, as shown in equation (1).
U
aging
= U
origin
+ (M
aging
M
origin
)
(1)
U
aging
, U
origin
, M
aging
, M
origin
are respectively the
contours of aged test image, origin test image, aged
average face and origin average face.
After obtaining the target face coutours,
Delaunay triangulation is used to get triangulated
face images, as shown in Figure 2.
Figure 2: Delaunay triangulation.
Then we use the deformation method shown in
Figure 3 to deform each feature triangle into each
corresponding feature triangle.For any point F in the
target triangle, we can get xy coordinates of D and E,
and calculate the proportion of AD to AB ,AE to AC
and DF to DE. We can calculate xy coordinates of
point d, e and f with the same proportions.To avoid
holes, we start from the target triangle area to find its
corresponding pixels in the source triangle area, and
use the bilinear interpolation method to determine
the gray value of each point. All triangles
transformed and merged, the morphed face image
can be obtained.
Figure 3: Feature region morphing.
4 EXTRACTION AND
ENHANCEMENT OF AGING
FEATURES
Considering the aging features such as wrinkles and
eyes bags are sudden changes while contours of face
are smooth changes, we use wavelet image
decomposition method to extract high-frequency
parts of face images to characterize aging features.

,
,
,
DWT
(2)
LL(I),HL(I),LH(I) and HH(I) are respectively the
low-frequency component, the horizontal, vertical
and diagonal high-frequency components of image I.
In order to enhance the aging features, we
change wavelet coefficients to highlight high-
frequency information and suppress low frequency
information. That is, for each data g
xy
in HL(I),LH(I)
and HH(I), we can transfer with formula (3).



2 

240

2

240
(3)
5 FACE AGING FEATURES
SYNTHESIS
In order to get the image after aging, we use the
method mentioned in part 2 to change the contours
of face image, and then increase the aging features.
We choose a typical aging face photo as aging
model. The model face and test face are decomposed
by two-level wavelet decomposition method. The
high-frequency part of model face is enhanced by
method mentioned in part 3.
The weighted average of the high frequency part
of typical aging image and target image is calculated
and considered as the high frequency part of target
image. Then we can obtain the final synthesis image
by two-layer wavelet synthesis method, as shown in
formula (4) to (9).

,
,
,
DWT
(4)

,
,
,
DWT
(5)
T
IDWT

,
,
,′
(6)
′


/2
(7)
′


/2
(8)
′

′
/2
(9)
LL(T), HL(T), LH(T), HH(T) and LL(M), HL(M),
LH(M), HH(M) are respectively the low-frequency
component, the horizontal, vertical and diagonal
high-frequency components of image I and model M.
HL’(M), LH’(M), HH’(M) are high frequency
partsenhanced by formula (3).
6 EXPERIMENT RESULTS AND
ANALYSIS
In order to verify the algorithm, we randomly
selected some face photos to test. The sample image
is pre-processed first, then the triangle feature region
method is used to change the contour, and the
wavelet decomposition and synthesis method is used
to increase the aging features. We compare it with
the method in the article Face illumination
compensation algorithm based on symmetrical
blocks(Hu W P,2014). The results show that
algorithm proposed in this paper has obvious aging
effect, and gain faster speed.
Figure 3: Experiment Results.
In Figure 3, (a) and (d) are two test images, (b)
and (e) are results of reference while (e) and (f) are
results of our algorithm. We can find in the figure
that our aging effect is better.
7 CONCLUSIONS
The face aging synthesis method proposed in this
paper combined the method of facial feature region
morphing and face texture transplantation to deal
with the changes of contours and textures in the
aging process. The aging effect is obvious, and the
execution speed is faster. It can be used to improve
the face recognition system easily.
In this system, the manual punctuation of face
feature points is adopted. In practical application,
ASM algorithm can be used to automatically
punctuate the point, which can greatly improve the
operation efficiency.
ACKNOWLEDGEMENTS
This work is supported by Guangxi Colleges and
Universities Key Laboratory of Intelligent
Processing of Computer Images and Graphics
(GIIP201508) and Guangxi Education Hall Project
of Improving the Basic Ability of Young
Teachers(KY2016YB252). It is also supported by
doctor funded project in Guangxi University of
Science and Technology (NO.15Z07).
REFERENCES
1. Wu Y, Kalra P, Moccozet L, et a1(1999). Simulating
Wrinkles and Skin Aging.Visual Computer,Vol.15:4,
pp.183-198.
2. Wu Xuefeng(2015).Design and Implementation of a
Child Face Aging Simulation System. Doctoral
Dissertations of Guangzhou Sun Yat-sen University.
3. Liu J,Zheng N,Chen B,et al(2007). Estimating Aging
Pattern by Aging Increment Distribution for Re-
rendering of Facial Age Effects.In International
Conference on Intelligent Computing,Springer Berlin
Heidelberg, pp.782-791.
4. Hu Weiping(2016). Construction of Face Aging
Model Based on Regression Analysis.Journal of
Guangxi University of Science and Technology, Vol.
27:3, pp.39-44.
5. Huang Fenglan(2017). The Research on the Image-
Based facial Aging Synthesis Method.Doctoral
Dissertations of People's Public Security University of
China.
6. Liu Zhenyu, Cui Zhi(2018).Research on Face Aging
Method based on RNN. Communications Technology.
Vol. 51:4, pp. 809-817.
7. Hu Weiping, Deng Huiwen(2014).Face illumination
compensation algorithm based on symmetrical blocks.
Journal of Computational Information Systems. Vol.
10:24, pp. 10801-10808.