Skin Youthfulness Index – A Novel Model Correlating Age with Objectively Measured Visual Parameters of Facial Skin

Abstract

A novel skin youthfulness index is intro-duced in this study to establish a mathema-tical model that correlates age with the visual properties of facial skin using an image analysis method. Images of 1,505 Asian female volunteers between the ages of 20 and 60 years were captured using a VISIA-CR® system under five different lighting conditions. Skin properties, such as wrinkles, hyperpigmentation, pores, color, translucency, ITA o , color evenness and surface texture parameters, were objec-tively measured from the images using Amway exclusively developed image analy-sis algorithms. Correlations between the measured parameters and the participants' chronological age were observed with sta-tistical significance. By defining and calcu-lating a set of weight factors five objec-tively measured visual parameters of skin were found to be most relevant to describe skin conditions influenced by the aging process. Combining these parameters in a mathematical model we have established a skin youthfulness index which has a range of 0 to 100 and is inversely correlated to people's chronological age (R 2 = 0.9959). The index allows us to accurately assess a person's apparent skin age based on the measured skin parameters. Of the various age groups tested, the largest difference between the actual and the calculated skin age was 2.4 years with a mean difference of 0.86 years. This model has potential for quantification of skin care product efficacy and thereby substantiation of new product claims.

Full text

IFSCC Magazine 3 - 2014 9
Skin Youthfulness Index – A Novel Model Correlang Age
with Objecvely Measured Visual Parameters of Facial
Skin
Di Qu and Yulia Park
Advanced Imaging and Measurement Laboratory, R&D
Amway Corporaon, 7575 Fulton St East, Ada, MI 49355, USA
This publicaon was presented as a podium presentaon at the 22nd IFSCC Conference, October 30 – November 01, 2013,
Rio de Janeiro, Brazil.
Keywords: skin youthfulness index, age correlaon, objecve measurement, skin aging parameters, image analysis.
ABSTRACT
A novel skin youthfulness index is intro-
duced in this study to establish a mathema-
cal model that correlates age with the
visual properes of facial skin using an
image analysis method. Images of 1,505
Asian female volunteers between the ages
of 20 and 60 years were captured using a
VISIA-CR® system under ve dierent
lighng condions. Skin properes, such as
wrinkles, hyperpigmentaon, pores, color,
translucency, ITAo, color evenness and
surface texture parameters, were objec-
vely measured from the images using
Amway exclusively developed image analy-
sis algorithms. Correlaons between the
measured parameters and the parcipants’
chronological age were observed with sta-
scal signicance. By dening and calcu-
lang a set of weight factors ve objec-
vely measured visual parameters of skin
were found to be most relevant to describe
skin condions inuenced by the aging
process. Combining these parameters in a
mathemacal model we have established a
skin youthfulness index which has a range
of 0 to 100 and is inversely correlated to
people’s chronological age (R2= 0.9959).
The index allows us to accurately assess a
person’s apparent skin age based on the
measured skin parameters. Of the various
age groups tested, the largest dierence
between the actual and the calculated skin
age was 2.4 years with a mean dierence
of 0.86 years. This model has potenal for
quancaon of skin care product ecacy
and thereby substanaon of new product
claims.
INTRODUCTION
Skin is the largest organ in the human body
and measuring the changes in its properes
with age is a primary topic in skin related
research [1]. Skin care researchers have stri-
ved to develop a comprehensive model cor-
relang mulple skin properes with age,
thereby providing an objecve and quanta-
ve measure of skin condions that will help
assess the ecacy of skin care products and
treatments [2].
Currently, most models reported in the lite-
rature use subjecvely measured skin para-
meters to assess skin aging. Guinot et al. [3]
introduced a skin age score (SAS) correlang
24 visual and tacle parameters of facial skin
with chronological age, concluding that SAS
could be generated from the evaluaon of
mulple discrete signs on facial skin and was
an informave tool for quanfying skin aging.
Vierkoer et al. [4] reported a skin aging
index (SCINEXA) which incorporates 23 clini-
cally graded intrinsic and extrinsic parame-
ters characterisc of skin aging. They con-
cluded that the model could be used to sepa-
rate the extrinsic and intrinsic eects of
aging. Nkengne et al. [5] established an index
of aging using clinically graded parameters
such as the degree of wrinkles, brown spots,
and sagging. They believed that their skin
aging index captured informaon relevant to
the visual transformaon of facial skin with
age and was meaningful when applied to
product ecacy evaluaons. Addional rese-
arch by Bazin and Doublet [6] and Bazin and
Flament [7] described linear correlaons with
mulple clinically assessed parameters for
Caucasian and Asian populaons, respec-
vely. While subjecve grading is the current
standard of clinical assessment, it is a com-
mon belief that the subjecvity of these as-
sessments carries the intrinsic possibility of
variaon between graders and inconsistency
in the grader’s percepon at dierent me
points.
Zedayko et al. [8] developed an instrumental
method to correlate age with skin brightness
of Caucasian subjects. While the measure-
ment was objecve and the correlaon was
good, the approach was rather simplisc in
that it only used skin color as the measurable
aspect of skin aging. A more complex measu-
rement was established by Dicanio et al. [9],
in which a linear funcon between age and
mulple skin parameters was constructed
using principal component analysis and mul-
variate regression. A total of 76 parameters
(10 clinical, 14 biophysical and 52 biochemi-
cal) were analyzed to idenfy 12 primary
variables for age esmaon. While the stas-
cal analysis method was sound, the physical
signicance of their results was sll open to
discussion. For example, both clinical and
instrumental measurements of the same skin
property (such as crow’s feet) were included
in the formula as two independent variables,
which was dicult to jusfy. In addion, both
glycaon (a biochemical parameter) and the
degree of wrinkles (a clinical parameter) we-
re included in their model. Since it is com-
monly believed that glycaon is the
molecular marker for the clinical signs of
aging [10, 11], lisng both of them as separa-
te independent variables in a linear equaon
could potenally impair the validity of the
model.
Over the past decade, sophiscated facial
imaging systems have been developed to
measure visual properes of skin using image
analysis [12]. Our skin research program uses
Amway-designed Facial Analysis Computer
Evaluaon System (F.A.C.E.S.) which has been
used to establish an infrastructure of more
than 20 facial imaging centers in various
countries and regions throughout the world.
Since 2007, we have collected images from
more than 30,000 people worldwide, re-
presenng a wide range of ages and ethnici-
es in both genders. This vast database has
allowed us to analyze mulple visual parame-
ters of facial skin and correlate them with
age.
Compared with the studies referenced
above, which used subjecve or mulple
instrumental methods to collect age related
data, exclusive use of image analysis for the
quancaon of the visual signs of aging has

10 IFSCC Magazine 3 - 2014
the advantage of being more simple than the
mul-instrument method and in the mean
me is more comprehensive than the single
measurement technique. In this report, we
describe a novel model correlang the chro-
nological age of female Asian consumers
with a list of objecvely measured visual
parameters of facial skin. This approach es-
tablishes a comprehensive funcon, the skin
youthfulness index (SYI), calculated using
image analysis to bridge age and the mea-
sured skin properes. A special focus was
placed on calculang meaningful weight
factors for each of the skin parameters in
order to improve the age correlaon and
more accurately predict skin age based on
visually displayed skin condions.
EXPERIMENTAL
Facial imaging system
A VISIA-CR® System (Caneld, U.S.A.) was
used to capture facial images under ve
dierent lighng condions (standard, at,
UV, cross polarized, and parallel polarized).
The system consisted of a facial imaging
booth with eight ashes placed at dierent
locaons for uniform illuminaon, a Nikon
200 SLE camera, and a set of standard color
plates. The camera sengs were ISO100,
f14, and “cloudy” for white balance. An Am-
way proprietary F.A.C.E.S. soware was used
to control the image capture process.
Study Design
From our image database of more than
30,000 parcipants collected around the
world, we used the images of 1,505 female
volunteers between the ages of 20 and 60
years covering four Asian countries in the
East, Far East, and South East regions. The
selecon was made to have exact or almost
exact ages in each of nine age groups (e.g.
volunteers in the 20, 25, 30, 35, 40, 45, and
50- year-old groups were exactly at the clai-
med age, while the 55-year-old group con-
sisted of volunteers 54 and 55 years old, and
the 60-year-old group consisted of volunteers
59, 60 and 61 years old). Our large image
database allowed us to make this ght age
selecon, which consequently improved the
predictability of the model by using ner age
ranges of study subjects. Currently, in the
literature the age ranges of study subjects
are usually wide, e.g., 20s (to represent sub-
jects aged 20 – 29 years), 30s (30 – 39), 40s
(40 – 49), etc. We believe more accurately
dened age ranges help to construct a more
accurate mathemacal model. Table I sum-
marizes the age and the count of the volun-
teer populaon included in this study. At
least 100 subjects in each age group were
included in this study, which provided a good
approximaon of skin property distribuon
of the general populaon.
All parcipants were conrmed by means of
wrien informed consent. Five front view
images of each study volunteer were taken
during the image collecon stage aer face
washing by a standardized cleansing proce-
dure. Using proprietary image analysis so-
ware, visual skin properes representave of
aging (wrinkles, pores, translucency, redness,
yellowness, ITAo, unevenness of skin tone,
and surface texture parameters) were quan-
ed from the set of captured images.
Image analysis
Parameters obtained through F.A.C.E.S. ana-
lysis
Amway’s F.A.C.E.S. and other proprietary
image analysis algorithms were used to ob-
jecvely quanfy facial skin properes such
as wrinkle score, hyperpigmentaon score,
pore count, skin color parameters, lightness
of skin tone, evenness of skin tone, skin
translucency, and surface texture properes.
All images were rst color corrected using
standard color plates embedded in each pic-
ture to achieve accurate measurements of
skin color and other visual skin properes.
The automac feature recognion algorithms
generate a facial mask that excludes eye-
brows, eyes, nostrils, mouth, and terminal
hair, rendering only the skin surface for accu-
rate wrinkle, pore, and subsurface hyperpig-
mentaon analysis. A representave graphic
output of the facial masks is shown in Figure
1.
Facial wrinkle analysis was performed in the
enre facial area. A wrinkle score was re-
ported which reected both the number of
wrinkles and wrinkle severity; therefore, a
deep wrinkle would be equivalent to mulple
smaller wrinkles lying on top of each other in
one locaon. Skin sub-layer hyperpigmenta-
on was measured from the UV images in
which areas with large amounts of melanin
deposion were quaned to produce a hy-
perpigmentaon score. Facial pores were
quaned in the selected regions of interest
that included the nose, upper lip, chin, cheek
areas close to the nose, and poron of the
forehead close to the eyebrows. The output
of the facial pore analysis included pore
count and pore area.
Facial skin color was measured using the
cross-polarized images. Regions of interest
on the cheeks and forehead were created
following automac detecon of the facial
features such as hairline, eyes, eyebrows,
nose, and mouth. Color parameters in the
RGB color space were obtained from the
regions of interest and converted to the L*,
a* and b* of the CIELAB color space using in-house
developed algorithms in ImageJ (Naonal
Instutes of Health). The skin individual typo-
logy angle (ITAo) was calculated using the
Figure 1: A representative graphic output (a screenshot) of the facial masks generated by
our F.A.C.E.S. analysis software showing quantified facial skin properties. A= wrinkles, B=
skin sub-layer hyperpigmentation, and C= facial pores.
Table I: Age and Count Distribution of the Study Participant Population

IFSCC Magazine 3 - 2014 11
measured L* and b* values. Unevenness of
skin tone (U) was measured as the variance
of pixel intensity in each region of interest.
Skin surface texture parameters were obtai-
ned from the region of interest using the
Gray Level Co-occurrence Matrix (GLCM)
stascal method, a built-in funcon of
ImageJ. Two GLCM parameters, entropy (E)
and inverse dierence moment (IDM), were
found to be the most relevant to describing
the age-related changes of skin texture pro-
peres. Entropy is a measure of the orderli-
ness of the surface texture paern. The skin
with more ne lines and wrinkles oen
shows a more regular parallel paern and
would therefore result in higher entropy
values. The IDM, on the other hand, indicates
the homogeneity of surface texture paern.
A uniform surface texture paern like that of
young skin would have a high IDM value.
Skin translucency, a quality of facial skin gre-
atly appreciated in Asian culture, was mea-
sured by using both cross-polarized and pa-
rallel-polarized images. Skin with high
translucency is perceived by consumers to
have a awless surface appearance, delicate
texture, subtle subsurface reecon, and a
rosy glow. Matsubara et al. [13] described an
image analysis method to quanfy facial skin
translucency. We employed a modied versi-
on of this method by quanfying skin
translucency through diuse reecon, as
opposed to specular reecon used in
Matsubara’s study, and dened a skin
translucency index based on the average
intensity value and its distribuon in each of
the RGB channels.
DATA ANALYSIS
Data type and range
Properes of the 10 objecvely measured
visual parameters of facial skin are summa-
rized in Table II. The extensive properes
such as wrinkles, sub-layer spots, and pores
were measured in the whole face area, while
the intensive properes such as color and
texture were measured in regions of interest
on both cheeks.
Statistical analysis
Stascal analysis was performed using JMP®
10.0.0 stascal soware (SAS Instute Inc.).
Distribuon and histogram analyses were
performed with the data normality test, and
ANOVA/Tukey-Kramer analysis was used for
comparisons of the means of skin properes
among various age groups.
Multiple regression analysis
A mulple regression analysis was performed
using JMP® to correlate parcipant age with
the objecvely measured skin parameters in
order to establish a linear equaon in the
following form:
where I= intercept; C= coecient; V= value
of an objecvely measured visual parameter;
and i= any specic parameter.
Skin youthfulness index
In addion to the mulple regression me-
thod, we established a new model, a skin
youthfulness index (SYI), by correlang the
age of the study parcipants with the mea-
sured parameters of their facial skin. The
following requirements were considered to
dene the SYI:
 A single comprehensive index that
indicates the youthfulness of facial skin
and is correlated inversely with people’s
Figure 2:
Flow chart for calculation of weight factors of the objectively measured visual parameters of
facial skin. r2: coefficient of determination, r: correlation coefficient, rC: critical value for linearity
test, SYI: skin youthfulness index, SigCo: significance factor, %MaxImp: maximum impact
factor, ImpactF: impact factor, W: weight factor and i: any of the objectively measured visual
parameter of facial skin.
Table II: Properties of objectively measured visual parameters of facial skin
1Equation AgePredicted
1
i
n
i
iVCI 



12 IFSCC Magazine 3 - 2014
chronological age (i.e., younger people
have a higher index value and older
people a lower value)
 An index that is aected by the measured
visual parameters in a linear composite
fashion through appropriately dened
weight factors
 The posive or negave eect of each
parameter on the index is reected (i.e.,
the value of a parameter that increases
with age would have a negave
inuence on SYI, whereas the value of a
parameter that decreases with
increasing aging would have a posive
eect).
 A target scale for the index of 0 – 100
Based on these consideraons, the following
linear composite funcon was proposed:
where W= weight factor; V= value of an ob-
jecvely measured visual parameter; and
i= any specic parameter type. The constants
N1 and N2 are factors to produce SYI values
on a scale of 0 – 100. The J term in Equaon
2 indicates whether a parameter has a posi-
ve or negave eect on SYI where J= 1 re-
presents a posive eect and J= -1 a negave
eect. For example, since a higher age indi-
cates a lower SYI value, an increasing wrinkle
score with increasing age would have a nega-
ve eect on SYI.
Weight factor calculation
Calculang the weight factor W for each
visual parameter was a key step in the deve-
lopment of the index and was performed as
outlined in the ow chart shown in Figure 2.
Basically, the coecient of determinaon, ri
2,
and the correlaon coecient, ri, were obtai-
ned from the age correlaon plots for each of
the 10 visual parameters (Figures 3A – 3J). A
linearity test was conducted by determining
a crical value for the correlaon coecient
[14]. If the correlaon of a parameter passed
the linearity test, the variable was considered
meaningful and included in the weight factor
calculaon. A signicance factor was then
dened, SigCo = (r2)2, which ranks the signi-
cance of the contribuon of the ten visual
parameters. Then a maximum impact factor
(%MaxImp) was dened emphasizing the
level of inuence a variable has as it changes
with age (i.e., a high %MaxImp indicates that
the parameter has a high impact on the SYI-
age correlaon). Then the impact factor,
dened as ImpactF, was calculated as the
product of SigCo and %MaxImp. Finally, the
weight factor was calculated by normalizing
the impact factor in a unit fracon form.
Age prediction from SYI
Aer a funcon of SYI is obtained, it can then
be correlated with the study parcipants’
chronological age to establish an SYI-age
curve. Such a curve enables us to examine
the goodness of t of Equaon 2 by compu-
ng the residual sum of squares , a stascal
parameter, between each group’s actual and
calculated age. In addion, this SYI-age corre-
laon allows us to calculate a person’s skin
age from the objecvely measured visual
parameters of facial skin, as discussed later in
the results and discussion.
Parameter optimization
To idenfy parameters that contribute most
meaningfully to SYI, we used RSS to compare
the goodness of t in the SYI-age correlaon.
Using Equaon 2, the individual eect of
each parameter was rst evaluated to iden-
fy the one which correlated the best with
age. The combined eects were then exa-
mined by adding other parameters one aer
another to Equaon 2. Their corresponding
RSS values were calculated and compared to
determine if the age correlaon was impro-
ved.
RESULTS AND DISCUSSION
Effect of age on the measured visual
parameters of skin
It has been well documented through clinical
grading that a person’s visual signs of aging
Figure 3: Correlation of age with each of the ten objectively measured visual
parameters. 3A: wrinkle score (Wr), 3B: count of sub-layer spots (S), 3C: pore count (P),
3D: skin translucency index (T), 3E: skin redness (a*), 3F: skin yellowness (b*), 3G: skin
tone lightness (ITAo), 3H: color unevenness (U), 3I: texture orderliness (E), and 3J: local
homogeneity (IDM).
 
2Equation ln
1
21 




 

i
n
i
iVWJNNSYI

IFSCC Magazine 3 - 2014 13
increase with age [15]. In this study, we ob-
served stascally a signicant age correla-
on for each of the ten objecvely measured
visual properes (wrinkles, pores,
translucency, redness, yellowness, ITAo, une-
venness of skin tone, and surface texture).
Figure 3 shows the average values of each of
the visual parameters ploed against the
parcipants’ chronological age.
The average wrinkle score increases expo-
nenally with increased age (Figure 3A). This
is, to our knowledge, the rst me such a
trend has been reported. Since our F.A.C.E.S.
algorithms include both number and severity
of facial wrinkles in the calculaon, a deep
and wide wrinkle is represented by mulple
single wrinkle lines as opposed to a single line
color-coded to dierenate it from other
smaller wrinkles, as seen in many commercial
wrinkle-analysis soware packages. We belie-
ve an exponenal increase in facial wrinkling
over age displays a meaningful progression of
the aging process of human facial skin.
The curve in Figure 3B indicates that the
amount of sub-layer spots increases with age.
This is due to the accumulave UV damage
acquired during life. Figure 3C shows the
average number of visible pores, which in-
creases steadily with age and reaches a pla-
teau aer age 45. While it is dicult to argue
that the pore number increases with age in a
physiological sense, we can conclude that,
due to changes in skin color and pore size,
facial pores become more easily detectable
with age, both visually and with image analy-
sis. Facial skin translucency decreases with
increasing age (Figure 3D) and levels o aer
age 45.
Younger people possess higher skin
translucency, as their skin looks less dull and
exhibits higher diuse reecon. Therefore,
the color components in the subsurface of
skin are more visible in younger people.
The lightness of skin tone, as dened by ITAo,
decreases steadily with increasing age (Figure
3G), indicang that older people have darker
complexions, which agrees with the trend of
changing facial skin color in a Caucasian po-
pulaon [8]. One component of ITAo, b*
which is a measure of skin yellowness, in-
creases with age (Figure 3F), indicang that
older people in general have more yellowish
skin tone. A similar trend is observed for skin
redness as shown by the a* values in Figure
3E.
Unevenness of facial skin tone increases with
age due to discoloraon, wrinkling, and other
physiological changes (Figure 3H). The local
homogeneity of skin texture (IDM) decreases
with age, while the orderliness of skin texture
(entropy) exhibited the opposite trend, as
shown in Figures 3J and 3I.
Age correlation by multiple regression
analysis
The values of ten objecvely measured visual
parameters were ed to Equaon 1 using
the mulple regression tool in JMP®. Aer
examining the outcome of the analysis, three
parameters (STI, b*, and IDM) which had p-
values larger than 0.05 were removed from
the correlaon. The nal linear equaons
obtained from the mulple regression analy-
sis correlated the parcipants’ age with se-
ven parameters with r2= 0.6277. The output
of the mulple regression is shown in Table
III. Inserng these values into Equaon 1 we
calculated the predicted age of the nine
groups of Asian female volunteers using the
average values of these visual parameters.
Figure 4 shows the correlaon between the
actual age and the predicted age of each
group with RSS= 215.32. Compared with the
best t line (diagonal), the predicted ages
deviated more in the lower and higher age
groups. The largest dierence between the
predicted and the actual age was 8.0 years.
Skin youthfulness index and its correla-
tion with age
To calculate skin youthfulness index (SYI), we
rst calculated the weight factors for each of
these ten visual parameters by following the
ow diagram described in Figure 2. The re-
sults are shown in Table IV from which we
can see that skin wrinkling has the most signi-
cant eect on the SYI. This is due to the fact
that it is closely correlated with age and its
change over age is the largest in the order of
magnitude. This agrees with the common
understanding that facial wrinkling is a signi-
cant marker for skin aging. By plugging the
objecvely measured visual parameters from
each of the 9 age groups into Equaon 2 we
could calculate a set of SYI values for the
corresponding age groups. Then by correla-
ng the SYI with the volunteers’ actual age a
linear funcon was obtained which allowed
us to back-calculate their apparent skin age
based on their objecvely measured visual
skin parameters. Table V summarizes these
results together with the dierence between
the predicted and the actual age of the study
volunteers. The goodness of t was calcula-
ted from this table and a RSS= 22.96 was
obtained, which is much beer than that of
the mulple regression method.
Using the parameter opmizaon method
described above, we further examined the
eect of each visual parameter and the com-
binaons of various parameters which contri-
bute to the SYI-age correlaon. This was do-
ne by nding the best age correlaon (the
least RSS) among the individual parameters
and then adding more parameters one aer
another to idenfy the best combinaon at
the next level. Among the ten individual visu-
al parameters, the eect of wrinkle score
correlated the best with age (RSS= 12.61).
Adding other parameters to the wrinkle score
and screening through all ten parameters at
various combinaons, we were able to obtain
Table III: Estimated parameters (I & Ci in
Equation 1) using multiple regression analysis
Table IV: Weight factors for Equation 2 Table V: Results of SYI and age Calculation
using equations 2 & 3

14 IFSCC Magazine 3 - 2014
Figure 4:
Correlation plot of the predicted and actual ages of nine female Asian groups
using the multiple regression method. RSS: residual sum of squares, a measure
of the discrepancy between the actual age and the age calculated from the
regression model.
Figure 5:
Effect of parameters and their combinations on the goodness of fit of the SYI model.
The RSS value for wrinkles alone was 12.61, high out of the chart area. As more
parameters are added to the wrinkle score, RSS starts to decrease until reaching a
minimum. More parameters added after that have negative effects on RSS and its
value starts to increase. The RSS value for the all 10 parameter combination was
22.96, high out of the chart area. The minimum point of a 5-parameter combination
represents the optimal condition for the SYI-age correlation.
Figure 6:
Correlation between age and skin youthfulness index. Solid dots: SYI
values calculated (using Equation 3) from the 5- parameter combination
from the facial skin of 20, 25, 30, 35, 40, 45, 50, 55, and 60-year-old age
groups. Hollow triangle: calculated SYI of the 28-year-old age group. Hollow
diamond: calculated SYI of the 38 year-old-age group.
Figure 8:
Correlation plot of predicted and actual ages using the skin youthfulness
index function. Solid dots: SYI values calculated from the five objectively
measured visual parameters of facial skin for the 20, 25, 30, 35, 40, 45, 50,
55, and 60-year-old age groups. Hollow triangle: calculated SYI of a 28-
year-old group. Hollow diamond: calculated SYI of a 38-year-old group.
Figure 7:
Distribution of the skin
youthfulness index (SYI) in
each of the nine age groups
of the Asian female
population. The numbers in
the legend indicate group
ages in years.
The SYI distribution shifts
from high to low regions with
increasing group age.

IFSCC Magazine 3 - 2014 15
the opmal parameter combinaon as shown
in Figure 5. From the chart we can see that
by combining more parameters with wrinkles
a beer age predicon was achieved with
decreasing RSS values unl a point where
adding more parameters started to inuence
the SYI-age correlaon in a negave way. This
opmal combinaon involved ve parame-
ters: wrinkle score, pores, skin translucency,
yellowness, and color unevenness. Their cor-
responding weight factors are listed in Table
IV under the column heading of “aer para-
meter opmizaon”.
With the above results, we obtained the nal
equaon for the SYI calculaon:
where T= translucency index, Wr= wrinkle
score, P= pore score, b*= yellowness, and U=
color unevenness.
Using Equaon 3 and the values of the objec-
vely measured visual parameters of facial
skin, we calculated SYI values from the
images of all 1,505 study parcipants in the
nine age groups indicated above. The average
SYI value of each age group was correlated
with the chronological age of the study par-
cipants, as shown by the solid dots and the
regression line in Figure 6 from which a
strong inverse linear correlaon was obser-
ved with r2= 0.9959. As expected, the youn-
ger groups have higher SYI values while the
opposite holds true for the older groups.
The correlaon in Figure 6 enables us to cal-
culate a person’s apparent age based on the
visual parameters objecvely measured from
her facial images. By apparent age we mean
the age of skin which has visual properes of
the facial skin of people in that specic age
group. This age might be dierent from the
perceived age, as the laer is subjecve in
nature and strongly inuenced by the percei-
ver’s knowledge, experience, preference, and
cultural background. Therefore, when we use
Equaon 3 to predict a person’s age based on
the measured visual parameters of facial skin,
we say the subject exhibits a skin age similar
to those people who typically possess the
same level of visual properes. Higher levels
of skin aging parameters shown in the facial
images would result in lower SYI values,
which would correspond to a higher apparent
age.
Figure 7 shows the normalized SYI distribu-
ons for each of the nine age groups. The SYI
values for all 1505 parcipants ranged appro-
ximately from 44 to 91, with higher
values corresponding to a more youthful skin.
From these plots we can see the SYIs for the
20-year-old group reside in the high value
region. With an increase in group age, the SYI
distribuons shied toward the lower value
region, diminishing the peak value from 72 to
53. These distribuon curves show how pe-
ople’s SYI, as well as their exhibited visual
properes of facial skin, change with age.
ANOVA/Tukey-Kramer tests were performed
to idenfy signicant dierences in SYI distri-
buons between the dierent age groups.
The dierences in SYI values between any
two adjacent age groups were stascally
signicant at a 95% condence level except
those between the 40 and 45 and the 55 and
60 age groups, as shown by the p values in
Table VI. Since we selected study parcipants
who are at the same exact age (or almost
same exact age) for each of the nine age
groups indicated above, the results of these
comparisons became meaningful. For examp-
le, from Table VI we can say with condence
that the skin’s visual properes and its youth-
fulness index are stascally dierent
between people 20 and 25 years old. They
are now measureable and disncve proper-
es of skin.
Validation
To validate the age predictability of Equaon
3, we selected two new data sets from a
Southeast Asian populaon. Facial images of
104 female volunteers 28 years of age and 70
females 38 years of age were analyzed. The
ve visual parameters were measured and
inserted into Equaon 3 along with their
corresponding weight factors shown in Table
IV. The resulng SYIs are shown by the hol-
low square and triangle, respecvely, in Figu-
re 6. While both of the validang data points
t into the model well, the average value of
SYI for the 38-year-old group lies almost right
on the regression line suggesng an excellent
model for this analysis.
Age prediction using measured visual
parameters of skin
From the results of these analyses we were
able to calculate skin age using the objec-
vely measured visual parameters of facial
skin. This was done by replong the data in
Figure 6 to show a dependence of age on
SYI. Fing the correlaon to a linear model,
we obtained the following relaon for the
predicon of a person’s apparent age:
where SYI= skin youthfulness index calculated
from Equaon 3, and Age= apparent age of
any study parcipant.
By inserng the average SYI values into the
equaon we were able to calculate the
average age of the nine age groups. Figure 8
is a correlaon plot in which the predicted
ages are ploed against the actual ages of
the nine test groups. An excellent correlaon
was obtained with RSS= 6.07. Comparing the
result of this age correlaon with that of the
mulvariate regression analysis (Figure 3),
we can see that the SYI method is much mo-
re eecve at predicng the skin age of the
populaon in this study than the conveno-
nal mulple regression method. The maxi-
mum age deviaon between the predicted
and the actual ages was 1.3 years, much
smaller than the 8.0 years from the mulple
regression method for the same populaon.
The results from the SYI analysis also show a
good age correlaon and can be used for
meaningful age predicon. Using the data
from the 28 and 38-year-old age groups used
for model validaon, we calculated the appa-
rent ages to be 25.6 and 38.9, respecvely.
As indicated in Figure 8, the dierences of
2.4 and 0.9 years between the actual and the
calculated ages for the 28 and 38-year-old
426.238.194 EquationSYIAge 
Table VI: p-Values of SYI Between Various Age Groups
3Equation
ln062.0*ln056.0ln174.0
ln627.0ln081.0(10
10 









UbP
WrT
SYI

16 IFSCC Magazine 3 - 2014
age groups, respecvely, suggest a fairly good
age prediction capability.
Concept application
The SYI-age correlaon described in this
study may provide a useful method for the
evaluaon of skincare product ecacy. For
any given clinical study, we would be able to
analyze both before and aer clinical images
to objecvely measure the ve visual para-
meters. If a product or skincare regimen
were to demonstrate a skin benet, such as
wrinkle reducon or increase in skin
translucency, it would be detected by image
analysis and show a posive change in the
corresponding measurement results. When
the improved values are inserted into Equa-
on 3, the corresponding SYI value would
show an increase as seen in Figure 6. This
increase in SYI would correspond to a skin
property of people of a younger age group,
i.e., a decrease in calculated skin age. Since
all measured parameters are the visual pro-
peres of facial skin, this decrease in calcula-
ted skin age aer product treatment could
be used to support a claim that the facial
skin of an individual appeared measurably
years younger aer product use. Our prelimi-
nary analysis of images before and aer a
laser resurfacing procedure indicated a very
promising reducon in the calculated age
aer treatment (unpublished data).
CONCLUSIONS
The large number of facial images obtained
from Asian female consumers through the
Amway-exclusive F.A.C.E.S. program allowed
us to objecvely measure ten dierent visual
properes of facial skin in nine age groups. A
stascally signicant age correlaon was
obtained for each of the measured visual
parameters of skin. Combining the objec-
vely measured parameters into a single
funcon enabled us to establish a novel index
of skin youthfulness (SYI), which quanta-
vely describes the aging of facial skin. An
excellent correlaon was obtained between
age and SYI, providing a potenally useful
applicaon to establish skin product ecacy
and substanate new product claims.
Acknowledgements
We thank the sta of Amway F.A.C.E.S. Sys-
tem implemenng group for their eort in
establishing and maintaining the global infra-
structure of this image collecng program.
We appreciate Ms. Valenna Kazlova’s
guidance on stascal analysis, and Mr. Brad
Richardson’s eort in data mining. We also
extend our thanks to Dr. Gopa Majmudar, Dr.
Rong Kong, and Ms. Barbara Olson for the
crical reading and eding of this manuscript.
REFERENCES:
[1] Yaar, M., Clinical and Histological Features
of Intrinsic Versus Extrinsic skin Aging, in:
Gilchrest, B. A. and Krutmann, J. (Eds.), Skin
Aging, Springer, Berlin, Heidelberg, 2006, pp.
9-21.
[2] Lange, N., and Weinstock, M., Stascal
Analysis of Sensivity, Specicity, and Predic-
ve Value of a Diagnosc Test, in: Serup, J.,
Jemec, G., and Grove, G. L. (Eds.), Handbook
of Non-Invasive Methods and the Skin, CRC
Press, 2006, 2nd Edion, pp. 53-62.
[3] Guinot, C., Malvy, D.J., Ambroisine, L.,
Latreille, J., Mauger, E., Tenenhaus, M., Mori-
zot, F., Lopez, S., Le Fur, I., and Tschachler, E.,
Relave Contribuon of Intrinsic vs Extrinsic
Factors to Skin Aging as Determined by a
Validated Skin Age Score, Arch. Dermatol.,
138 (2002) 1454-1460.
[4] Vierkoer, A., Ran, U., Kramer, U., Sugiri,
D., Reimann, V., and Krutmann, J., The SCINE-
XA: A Novel, Validated Score to Simultaneous-
ly Assess and Dierenate Between Intrinsic
and Extrinsic Skin Ageing, J. Dermatol. Sci., 53
(2009) 207-211.
[5] Nkengne, A., Roure, R., Rossi, A.B., and
Bern,C., The Skin Aging Index: a New Ap-
proach for Documenng An-aging Products
or Procedures, Skin Res. Technol., 19 (2013)
291-298.
[6] Bazin, R. and Doublet,E., Skin Aging Atlas,
Volume 1 Caucasian Type, Edions Med’Com,
Paris, 2007, pp. 32.
[7] Bazin, R., and Flament, F., Skin Aging Atlas
Volume 2 Asian Type, Edions Med’Com,
Paris, 2010, pp. 26.
[8] Zedayko, T., Azriel, M., and Kollias,N.,
Caucasian Facial L* Shis May Communicate
An-Ageing Ecacy, Int. J. Cosmet. Sci., 33
(2011) 450-454.
[9] D. Dicanio, D., R. Sparacio, R., L. Declercq,
L., H. Corstjens, H., N. Muizzuddin, N., J. Hi-
dalgo, J., P. Giacomoni, P., L. Jorgensen, L.,
and D. Maes, D., Calculaon of apparent age
by liner combinaon of facial skin parame-
ters: a predicve tool to evaluate the ecacy
of cosmec treatments and to assess the
predisposion to accelerated aging, Bioge-
rontology, 10 (2009) 757-772.
[10] Kollias, N., Gillies, R., Moran, M., Koche-
var, I.E., and Anderson, R.R., Endogenous Skin
Fluorescence Includes Bands that May Serve
as Quantave Markers of Aging and Photo-
aging, J. Invest. Dermatol., 111 (1998) 776-
780.
[11] Maillard-Lefebvre, H., Boulanger, E.,
Daroux, M., Gaxae, C., Hudson, B.I., and
Lambert, M., Soluble Receptor for Advanced
Glycaon End Products: a New Biomarker in
Diagnosis and Prognosis of Chronic Inamma-
tory Diseases, Rheumatology, 48 (2009) 1190
–1196.
[12] Hawkins, S., Computerized Image Analy-
sis of Clinical Photos, in: Serup, J., Jemec, G.,
and Grove, G. L. (Eds.), Handbook of Non-
Invasive Methods and the Skin, CRC Press,
2006, 2nd Edion, pp. 95-100.
[13] Matsubara, A., Dierences in the Surface
and Subsurface Reecon Characteriscs of
Facial Skin by Age Group, Skin Res. Technol.,
18 (2012) 29-35.
[14] Weathington,B., Cunningham, C., and
Pienger, D. (Eds.), Understanding Business
Research, John Wiley & Sons, Inc, Hoboken,
New Jersey, 2012, pp. 245-270.
[15] R. Bazin, and F. Flament, Skin Aging Atlas
Volume 2 Asian Type, Paris: Edions
Med’Com, 2010, pp. 28.
Corres po nding a ut ho r - em ail:
Di.Qu@amwa y.com