Skin Research and Technology 2014; 0: 1–12
Printed in Singapore
All rights reserved
doi: 10.1111/srt.12143
Skin texture aging trend analysis using dermoscopy
images
Y.-H. Choi1, D. Kim1, E. Hwang1 and B. J. Kim2
1
School of Electrical Engineering, Korea University, Seoul, Korea and 2Department of Dermatology, Chung-Ang University College of Medicine,
Seoul, Korea
Background/purpose: To date, the degree of skin damage
caused by diverse factors, such as aging and persistent sunlight
exposure, has been evaluated based on the personal experience
and knowledge of dermatologists because there is no standard
method for objective evaluation. If a standard method were avail-
able, patients could obtain more consistent information about
their skin condition, and hence perform more effective treatment
of the skin damage. In this paper, we demonstrate how to estab-
lish a standard method using dermoscopy images of subjects of
various ages. We focus on three body parts, specifically the face,
neck, and hands, and extract various skin texture features to
quantitatively and objectively represent the skin condition.
Method: We construct a model for skin damage evaluation
based on various skin texture features. To accomplish this objec-
tive, we consider various features from face, neck, and hand
dermoscopy images, including texture length, width and depth,
cell area, the number of cells in a fixed region, radius ratio of
inscribed and circumscribed circles of a wrinkle cell, and average
perimeter of a wrinkle cell. In this study, a wrinkle cell represents
the smallest skin region enclosed by textures. We then perform a
linear regression for texture features based on subject age.
the outermost part of the human
S KIN IS
body. It protects the body from infection,
injury, and water loss, while helping regulate
body temperature. Additionally, the skin main-
tains homeostasis and produces vitamin D. Skin
performance is impaired with age and visual
beauty is lost. This impairment and loss can
proceed more quickly because of various fac-
tors, such as persistent sunlight exposure,
smoking, and excessive drinking. Because of its
visibility and role, considerable attention has
been given to its condition, and substantial
efforts have been taken to mitigate skin aging
and damage progression.
To accurately access the skin condition, a
method to objectively evaluate the skin is
required. To date, this evaluation has been typi-
cally performed by dermatologists based on
their personal experience and knowledge.
© 2014 John Wiley & Sons A/S.
Published by John Wiley & Sons Ltd
Skin Research and Technology
Result: A dermoscopy image can be automatically analyzed
by extracting skin texture features. We demonstrate aging
trends by performing linear regression on these features.
Based on this result, a quantitative and objective evaluation of
the skin condition can be provided.
Conclusion: We proposed several new skin texture features
and developed algorithms to accurately extract them. We ana-
lyzed these features and demonstrated their age-related
change trends by using graphs and charts. We believe that
our result can be used as a standard method for evaluating
degrees of skin damage. Moreover, we believe that our pro-
posed method can be applied in various areas, such as perfor-
mance evaluation of certain skin products.
Key words: skin texture – skin texture analysis – skin feature
extraction – dermoscopy image analysis
Ó 2014 John Wiley & Sons A/S. Published by John
Wiley & Sons Ltd
Accepted for publication 5 January 2014
Therefore, evaluation results can differ depend-
ing on the dermatologist. The degree of skin
damage can be measured based on various fea-
tures, such as texture pattern, skin elasticity,
and moisture. Among these aspects, skin tex-
ture has the most visual effect and can be seen
with the naked eye. A standard metric to evalu-
ate the skin condition can be established using
the diverse features of skin texture. In this
paper, we propose a model for skin damage
evaluation based on diverse skin texture fea-
tures captured from dermoscopy images.
We believe that skin texture patterns and their
detailed shapes are closely related to the degree
of skin damage. Figure 1 presents texture
patterns from dermoscopy images of subjects of
various ages. As shown in the figure, skin texture
demonstrates some aging-related trends. There-
fore, we focus on the quantitative representation
1
Choi et al.
of various features extracted from the dermos-
copy images. We construct an evaluation model
that calculates skin age using average skin tex-
ture features that are the most similar to the skin
texture of subjects of the given age.
Since the 1950s, numerous methods have
been proposed for skin surface topography
analysis (1–9). According to the dimensional
property of the parameters measured, these
methods can be divided into two categories:
three-dimensional topography analysis (3DTA)
and two-dimensional image analysis (2DIA).
3DTA requires skin surface negative or positive
silicon replicas for further analysis by profilom-
etry (mechanical, optical, laser, transmission, or
interference). 3DTA mainly involves the follow-
ing roughness parameters: depth of roughness
(Rt), maximal depth of roughness (Rm), mean
depth of roughness (Rz), depth of smoothness
(Rp), and mean roughness value (Ra) (4, 5, 7–9).
Rt is the peak-to-valley roughness of a surface;
a single feature has a significant impact on the
derived value. The profile length is divided into
five equal segments. The peak-to-valley rough-
ness is measured in each segment, and the larg-
(a)
(c)
Fig. 1. Skin texture based on age – (a) Subjects in their 10s, (b) Subjects in their 20s, (c) Subjects in their 30s, (d) Subjects in their 40s.
2
est of these five values is Rm. Rz characterizes
the mean value of the peak-to-valley roughness
in each segment. The depth of smoothness, Rp,
is calculated as the mean distance of every
point of the curve from the highest peak. Ra is
the most useful international parameter of
roughness. To compute Ra, an average line is
generated to run through the center of the pro-
file; the area in which the profile deviates above
and below this line is determined. 3DTA is typ-
ically used in clinical investigations of the local
effects of cosmetic and medical substances.
However, it requires a relatively long duration
of replica hardening and measuring time (3, 4);
moreover, its accuracy can be degenerated
when the patient moves or if the replica is sha-
ken during the hardening process. Because of
these concerns, the replica should be handled
and measured by experts.
2DIA analysis, on the other hand, does not
require silicone replicas of the skin, and it can
quickly acquire a skin surface image. This latter
factor makes the method applicable to the rou-
tine detection of skin surface topography. 2DIA
primarily involves two parameters: tau and
(b)
(d)

density. Tau is the level of anisotropy; i.e. the
percentage of furrows oriented in a different
direction. The higher this level is, the greater is
the anisotropy of the surface (7). Density refers
to furrow density, which is expressed through
the occurrence of intersections between primary
wrinkles (5, 7).
Fujimura et al. (10) examined the relationships
between roughness values obtained from the
morphological structure of the corner of the eye
and age or wrinkle scores using Japanese
women aged 10-70 years in order to demonstrate
the practicality of using direct three-dimensional
in vivo measurement system (PRIMOS) to
evaluate wrinkles objectively in clinical tests of
wrinkle-smoothing ingredients or cosmetic sur-
gery. To detect skin wrinkles, Tanaka et al.
applied a cross-binarization method to a digital
skin image to obtain its binary image. A short
straight-line matching method was then per-
formed to detect wrinkles from the binary image
and to measure their lengths (11). For each base-
line in the cross-binarized image, if more than
70% of its pixels were marked black, then the
line was considered a wrinkle. The end of the
previous baseline was then extended to create a
new baseline. This step was repeated until the
end of the wrinkle or image was reached.
Hayashi et al. proposed an age and gender
estimation method based on the facial image
(12, 13). They first detected wrinkles from a
facial image using a Hough transform called
the Digital Template Hough Transform (DTHT).
Hatzis presented a profilometric method for
measuring the size and function of the wrinkles
(14). The replication was made using silicon rub-
ber similar to dental impression material. The
microtopography of wrinkles in negative replicas
was photographed by a scanner or stereomicro-
scope under suitable lighting to change the nega-
tive appearance of the replica into a positive one.
In our previous work, we proposed a scheme
to detect skin texture from dermoscopy images
and to estimate texture length and width (15).
In the scheme, skin textures were detected
using the watershed algorithm, which was first
introduced by Lantuejoul and Beucher for
image segmentation (16). The texture length
was estimated simply by counting the pixels on
the textures. To estimate the texture width, we
proposed a block thickening method. Neverthe-
less, we observed that there is ample capacity
for improvement in these methods.
Skin texture aging trend analysis
To improve accuracy, we proposed line siev-
ing and morphological region growing methods
for calculating texture length and width, respec-
tively (17). To acquire additional features, we
proposed a method for estimating texture depth
based on the color difference between texture
and non-texture regions. On the other hand,
Zou et al. detected the area mean of a superfi-
cial skin texture block formed by primary and
secondary lines crossing each other (18).
To evaluate the skin age of subjects, we
extracted the texture features from their der-
moscopy images and classified them into an
age group using a support vector machine
(SVM) (19).
In this paper, we demonstrate the aging
trends of various skin features, which include
texture features, cell information, and radius
ratios of inscribed and circumscribed circles of
each cell and cell perimeter. We have devel-
oped diverse image processing algorithms to
efficiently extract these age-related features, and
we briefly describe in the following section how
to extract the features. In the Result and Discus-
sion section, we present statistical values of the
extracted features and their aging trends.
Material and Methods
Feature extraction scheme
We have developed algorithms for extracting
various features from dermoscopy images. Our
feature extraction method consists of three
major steps: pre-processing, texture feature
extraction, and cell feature extraction. Table 1
shows the detailed tasks performed in each step.
Before performing analysis of these features, we
made the following assumptions, which are
based on common knowledge of skin.
• Texture length decreases with age.
• Texture width increases with age.
• Texture depth increases with age.
• Cell area increases with age.
• Number of cells decreases with age.
• Diameter ratio of inscribed circle and circum-
scribed circle of a cell decreases with age.
• Total length of lines connecting cross points
of a cell increases with age.
In our study, we show that our assumptions
about texture features are reasonable by con-
structing their related trends using regression
graphs.
3
Choi et al.
TABLE 1. Three major steps of feature extraction
Step 1. Pre-processing Region of interest (ROI) selection
Image normalization
Binarization and denoising
Image segmentation
Step 2. Texture Texture length estimation
feature extraction Texture width estimation
Texture depth estimation
Step 3. Cell feature Cell detection using PMDA
extraction Over-segmented cell merging
Cell area estimation
Cell counting
Inscribed/circumscribed circle evaluation
Length measurement for line connected
between contact points
Step 1. Pre-processing
Due to camera device limitations and light
source interference, captured images may exhi-
bit noise and vignetting. In the pre-processing
step, we remove these effects. Vignetting is a
reduction in brightness or saturation at the
periphery of an image compared to the image
center. The luminous source of a digital micro-
scope image is typically concentrated around its
center. Therefore, to remove the vignetting
effect, we first crop a region of interest (ROI)
from the image center, where the luminance is
the most homogeneous. We set its size to
300 9 300 pixels, which empirically provides a
minimal vignetting effect. Nevertheless, the
cropped image may contain slight vignetting.
Therefore, we apply a gradation masking
method to remove the vignetting residue. In
addition, we apply a contrast equalization
method to improve the contrast in the cropped
image. To eliminate noise in the image, we first
binarize the image using Otsu’s method and
perform the denoising algorithm. To obtain the
skeleton image, we apply a watershed algorithm
to the binarized image, which produces a single
pixel line along the center of the skin texture.
Step 2. Texture feature extraction
In this step, we extract from the skeleton image
diverse skin texture features, such as length,
width, and depth. We first estimate the length
by counting the pixels on the wrinkle texture,
which provides relative length. We convert the
measurement into a real length using an SI
unit. The idea is simple; a 1-mm line
enlarged by a 609 magnifier, which is used to
capture dermoscopy images, corresponds to
4
152 pixels. Therefore, the relative wrinkle
length in the pixel unit can be converted into a
real length in millimeters by applying this
relationship.
To measure the length, we use the line siev-
ing method. This method first counts the num-
ber of pixels on the horizontal and vertical
texture lines. It then counts the number of pix-
els along the diagonal lines, and it multiplies
their slope to estimate the actual wrinkle length.
In the case of single pixel islands on the image,
we simply count these islands and add the
number to the total length.
To measure wrinkle width, we must recover
the original wrinkle from its single pixel skele-
ton image. For this purpose, we use a morphol-
ogy-based region growing method. The method
basically involves applying a dilation process to
the wrinkle skeleton to recover its actual width.
The wrinkle area is identified in the dilated
region and gradually expanded. During this
expansion, we perform a cross-comparison of
the output image with the grayscale image to
recover the wrinkle region. This step is repeated
until the comparison fails or succeeds n times,
where n is a threshold value for determining
noise. That is, if the number of expanded pixels
is smaller than n, we consider it a noise. Equa-
tion 1. Depth measurement:
1 Xn
DðIk Þ ¼ maxðjIk ðSÞ  Ik ðci ÞjÞwhere
n i¼1
1
js  ci j ¼ avgwidth þ margin s.t.8s 2 Si ; 8ci 2 C
2
ð1Þ
Careful investigation of skin images reveals
that the color of wrinkle regions is darker than
normal skin texture. Furthermore, depending
on the depth of the wrinkle regions, their
degree of darkness can differ because of light
interference. Based on this difference, we esti-
mate relative wrinkle depth using Eq. 1. In the
equation, C represents a set of n randomly
chosen points along the skeleton line of a
wrinkle. For each point ci in C, we first iden-
tify set Si of all points whose distance from
point ci is r. We then calculate the maximum
color difference between points in Si and ci.
Distance r can be defined as half the average
width + the margin. Finally, if we add the
maximum color differences of all points in C,
their average becomes the depth of the wrinkle
region.
 Skin texture aging trend analysis

Step 3. Cell structure

Typical skin images reveal that skin texture is
composed of regions separated by wrinkles
called wrinkle cells. Boyer et al. (21) showed that
the skin of younger people has more cells in the
skin wrinkle image than that of older people.
Based on this observation, we propose a Poly-
gon Mesh Detection Algorithm (PMDA) for cal-
culating the number of cells and the average cell
area. This algorithm works in a recursive way.
Using numerous dermoscopy images of subjects
of various ages, we observed that the number of
cells and the average cell area are closely related
to subject age. We also found that as people age,
their skin cells enlarge with wider wrinkles. Fig. 2. PMDA image.

Algorithm 1 degrade the accuracy. Therefore, if the size of a

PMDA cell is not sufficiently large, we do not consider
Procedure: PMDA it. Algorithm 2 outlines detailed steps to con-
Input: binary skeleton image S,
boolean image B, struct a cell map to efficiently run PMDA. The
current position (x,y)
Output: current cell size A
algorithm first initializes the map with the skel-
1: B(x,y) := false eton image and enqueues all random coordi-
2: A:=1
3: for each adjacent points p(x’,y’)
nates. The algorithm then dequeues the random
4: if B(x’,y’) == true coordinates and calls the PMDA algorithm for
5: A=A+PMDA(S,B,(x’,y’))
6: end if cell detection.
7: end for
8: return A

Algorithm 2
Cell Map Construction Algorithm
Algorithm 1 outlines detailed PMDA steps. Procedure: Cell Map Construction Algorithm
As mentioned above, the skin image is repre- Input: binary skeleton image S
Output: average area Aavg, total cell count Ccell,
sented by a skeleton graph using the watershed fully enclosed cell count Cfcell
algorithm. For example, in Fig. 2, polygons are 1: for each pixel p(x,y) in boolean image B
2: if S(x,y) == 0
separated by skeleton lines, and each polygon 3: B(x,y) := false
represents one wrinkle cell. To obtain the num- 4: else
5: B(x,y) := true
ber of cells and the dimension of each cell, we 6: end if
7: end for
use a connected component detection method. 8: Ccell :, Cfcell , Atotal := 0
The method recursively searches for four neigh- 9: enqueue randomly all position pairs (x,y) in B into queue Q
10: while Q != empty
bors (top, down, left, and right) from the start 11: (x,y) := dequeue Q
point until a wrinkle line or image boundary is 12: if B(x,y) == true
13: (A,CAB) := PMDA(B, (x,y))
encountered. If an image boundary is met, we 14: Ccell := Ccell + 1
15: if CAB == false
then increase the total cell count and mark the 16: Cfcell := Cfcell + 1
cell as a border cell. We use the total cell count 17: Atotal := Atotal + A
18: else
as a feature; however, when we calculate the 19: if (A / border distance) < RatioD
average cell area, we divide the total area by 20: Cfcell := Cfcell + 1
21: Atotal := Atotal + A
the number of complete cells, which excludes 22: end if
23: end if
the border cells. If an image boundary is met, 24: end if
we then check whether the size of the cell is 25: end while
26: Aavg := Atotal / Cfcell
meaningful. If it is meaningful, we consider it a 27: return Aavg, Ccell, Cfcell
complete cell and count it. Otherwise, we dis-
card the cell. Because our scheme uses a small
segment of the image, incomplete cells, which Cell features are crucial for accurately evalu-
are border cells, could exist. Those cells tend to ating skin texture. However, at times cells turn

5
Choi et al.

(a) (b)

Fig. 3. Detected over-segmented and merged cell image – (a) Convex hulls and over-segmented cells, (b) Merged cell image.

Fig. 4. Detected inscribed/circumscribed circles.

(a) (b)

Fig. 5. Average connected-line length measurement – (a) Detected contact points, (b) Connected contact points.

out to be over-segmented for various reasons, cells should be eliminated before feature extrac-
such as light source reflection and shadows. tion. In this part, we describe our method for
Therefore, for better accuracy, over-segmented detecting over-segmented cells in the skeleton

6

Fig. 6. User interface.
skin image and for removing them by merging
them with neighboring cells.
To detect over-segmented cells in the skeleton
image, we construct a convex hull for each cell.
The convex hull represents a cell by using a
minimum set of points connected with straight
lines. After constructing a convex hull for each
cell, we check whether there are any cells
whose center points of gravity are inside
another cell’s convex hull. If so, we consider the
cell to be over-segmented. For example,
Fig. 3(a) shows convex hulls constructed for the
wrinkle cells in the sample image. Over-seg-
mented cells are marked in blue.
After detecting all over-segmented cells, we
eliminate them by merging. For efficient merg-
Algorithm 3
Contact point detection
Procedure: ContactPointDetection
Input: binary skeleton image S(x,y)
Output: contact point contactPoints CP
1: for each point p(x,y) in S(x,y)
2: cnt = 0
3: for each adjacent points padj(x,y)
4: if padj(x,y) is not equal with adjace
5: cnt++
6: end if
7: end for
8: if cnt >= 6
9: insert p(x,y) into contactPoints CP
10:
11: end for
end if
Skin texture aging trend analysis
ing, we maintain three lists: detected cell, over-
segmented cell, and merged cell. The detected
cell list contains diverse information about cells
in the image. The over-segmented cell list
maintains all over-segmented cells with their
neighboring cell. Finally, the merged cell list
contains information about merged cells.
For each over-segmented cell pair in the list,
we first check whether the area of the inner cell
is less than the average cell area. If it is, we
merge the inner cell into the outer cell. This
information is recorded in the merge cell list.
The merge cell list will eventually contain all
merged cells. Finally, we update the detected
cell list based on the merged cell list to obtain
the final set of cells. For example, Fig. 3(b)
shows the result of merging over-segmented
cells for the image in Fig. 3(a). After cell merg-
ing, we calculate the new contour and area of
the merged cells. We also adjust the average
cell area and cell count.
To extract inscribed/circumscribed circles of
a wrinkle cell, we use the morphological region
growing method. The method starts from the
gravity center point of the cell and grows the
region until it meets the wrinkle contour. The
method records the coordinates of first and last
meeting points.
The first meeting point becomes the radius of
the inscribed circle, and the last meeting point
7
Choi et al.
becomes the radius of the circumscribed circle.
Finally, we calculate their radius ratio. We later
use the top 15 ratios for skin-aging trend analy-
sis. As shown in Fig. 4, inscribed and circum-
scribed circles are represented as bold and thin
lines, respectively.
Finally, we calculate the total length of all
lines connecting the wrinkle line contact points
on the cell. A contact point is the point where
Fig. 7. Face skin texture aging trend.
8
three or more wrinkle lines meet. As shown in
Fig. 5(a), all contact points are marked in red.
To determine the contact points, we individu-
ally consider each wrinkle pixel, and we search
the pixel along eight adjacent neighbors. Cir-
cling from the top pixel of the current point to
the right side, the algorithm searches for the
color changing point. The color changing point
is where the adjacent two points have differing

color values. If there are four color changing
points, the current point is on the skin texture,
and the scheme proceeds to the next point.
However, if there are six color changing points,
then there are three texture lines, and the cur-
rent point is the contact point. Therefore, the
algorithm detects if there are six or more points.
This detecting contact point procedure is
explained in Algorithm 3. Figure 5(b) exempli-
fies a connection of all the contact points based
on the original skeleton image.
Fig. 8. Neck skin texture aging trend.
Skin texture aging trend analysis
Result and Discussion
Experimental setup
Above, we have described our scheme for
extracting various skin texture features from
dermoscopy images. Based on this scheme,
we have implemented a prototype system and
performed various experiments to determine
skin-aging trends. Figure 6 shows the user
interface of our prototype system. As men-
tioned earlier, we used a 300 9 300 pixel image
9
Choi et al.
at the center of the image to avoid the vignett-
ing effect. We have constructed a dataset of
face, neck, and hand dermoscopy images from
approximately 200 subjects. We then grouped
them into five classes depending on subject age.
Feature extracting result
Figures 7–9 show skin texture features of each
subject and the overall face, neck, and hand skin-
Fig. 9. Hand skin texture aging trend.
10
aging trends. We again extracted seven texture
features, including skin texture length, width
and depth, average cell area, cell count, radius
ratio of inscribed/circumscribed circles, and
average length of lines between contact points.
The first graph in each of these figures shows
the distribution of total lengths of skin textures
in the respective image and their linear regres-
sion. From information in these graphs, it is evi-
dent that the total length of skin texture
 Skin texture aging trend analysis

TABLE 2. Analytic results of statistical values (face)

Avg. of total Avg. texture Avg. depth Avg. cell Avg. of total Avg. radius Avg. length
length (mm) width (pixel) (color diff.) area (pixel) no. of cells (count) ratio (ratio) of lines (mm)

10s 36.09 7.97 111.12 903.77 88.12 0.5456 0.1601

20s 33.42 8.04 113.50 1021.01 77.15 0.5322 0.1661
30s 30.41 8.51 115.03 1124.576 68.06 0.5203 0.1695
40s 29.94 8.81 121.86 1312.821 63.46 0.5146 0.1699
50s 28.25 8.99 120.91 1710.794 60.75 0.5120 0.1721

TABLE 3. Analytic results of statistical values (neck)

Avg. of total Avg. texture Avg. depth Avg. cell Avg. of total Avg. radius Avg. length
length (mm) width (pixel) (color diff.) area (pixel) no. of cells (count) ratio (ratio) of lines (mm)

10s 36.57 9.79 116.64 882.27 91.44 0.5672 0.1655

20s 34.43 10.57 118.81 997.20 80.20 0.5564 0.1755
30s 33.07 11.40 118.09 1151.77 73.83 0.5400 0.1787
40s 32.51 11.95 126.22 1153.29 73.25 0.5368 0.1805
50s 31.83 12.08 125.75 1230.70 70.91 0.5310 0.1835

TABLE 4. Analytic results of statistical values (hand)

Avg. of total Avg. texture Avg. depth Avg. cell Avg. of total Avg. radius Avg. length
length (mm) width (pixel) (color diff.) area (pixel) no. of cells (count) ratio (ratio) of lines (mm)

10s 26.81 9.87 117.52 1418.32 52.09 0.5145 0.1930

20s 23.45 11.49 118.60 1700.91 42.38 0.5033 0.2065
30s 21.08 11.89 118.48 2425.09 35.50 0.4952 0.2123
40s 20.04 13.07 127.86 2680.59 32.73 0.4899 0.2232
50s 18.33 13.99 127.26 2741.33 28.80 0.4514 0.2352

decreases with age. Another interesting obser-
vation is that the average cell count decreases
with age and the cell area increases. In fact,
these observations are closely related to each
other. Because the cell size increases, both the
number of cells in the region and the total tex-
ture length decrease.
Another feature of interest is texture width
and depth. We generally believe that wrinkles
become wider and deeper with age. The graphs
in Figs 7–9 clearly show this trend.
The next feature that we evaluate is the
radius ratio of inscribed/circumscribed circles.
We consider this feature because it is commonly
believed that skin texture becomes stretched in
one direction with age. In fact, as cells become
stretched, their circumscribed circles enlarge.
As a result, the radius ratio decreases. The
graph is well matched to our analysis. In the
graph, we used the top 15 ranked ratios among
the cells in the image because they showed this
tendency very well.
The last feature we consider is the number
of contact points and the average length of
lines connecting them. According to the graph,
the number of contact points decreases and
the average length of lines increases. This ten-
dency is explained by the fact that wrinkle
cells enlarge with age. As the cell size
increases, there are fewer cells in the image.
Therefore, the number of contact points
decreases and the line connecting them length-
ens. Because of this tendency, the graph shows
an increasing trend.
Tables 2–4 show the average value of each
texture feature for the five age groups: 10s,
20s, 30s, 40s, and 50s. The tables show a simi-
lar trend as the graphs in Figs 7–9.
Conclusion
In this paper, we proposed a scheme for skin tex-
ture aging trend analysis in terms of diverse skin
texture features. We considered several features,
including skin texture length, width and depth,
the number of cells and their average areas, the
radius ratio of inscribed/circumscribed circles,
and the length of lines between contact points.
The feature trend graphs are well matched with
what is typically understood about skin texture.

11
Choi et al.

For example, with age, wrinkle cells enlarge and

skin cell count decreases. In sum, a single two-
dimensional dermoscopy image contains abun-
dant information for estimating skin texture
aging, and our proposed scheme is highly effec-
tive in achieving this estimation. Our scheme can
be used for establishing a standard method for
measuring the degree of skin damage or aging.
We expect that our scheme can be used in vari-
ous commercial and non-commercial skin-
related applications.
Acknowledgements
This research was supported by the MSIP(Min-
istry of Science, ICT and Future Planning),
Korea, under the IT/SW Creative research pro-
gram supervised by the NIPA(National IT
Industry Promotion Agency) (NIPA-2013-
H0502-13-1119).

References 8. Lagarde JM, Rouvrais C, Black D. 16. Beucher S, Lantuejoul C. Use of

1. Purba MB, Kouruis-Blazos A,
Wattanapenpaiboon N, Lukito W,
Rothenberg E, Steen B, Wahlqvist
ML. Can skin wrinkling in a site
that has received limited sun expo-
sure be used as a marker of health
status and biological age? Age
Ageing 2001; 30: 227–234.
2. Nita D, Mignot J, Chuard M, Sofa M.
3-D profilometer using a CCD linear
image sensor: application to skin
surface topography measurement.
Skin Res Technol 1998; 4: 121–129.
3. Fischer TW, Wigger-Alberti W, Els-
ner P. Direct and nondirect measure-
ment techniques for analysis of skin
surface topography. Skin Pharmacol
Appl Skin Physiol 1999; 12: 1–11.
4. Lagarde JM, Rouvrais C, Black D,
Diridollou S, Gall Y. Skin topography
measurement by interference fringe
projection: a technical validation.
Skin Res Technol 2001; 7: 112–121.
5. Akazaka S, Nakagawa H, Kazama
H, Osanai O, Kawai M, Takema Y,
Imokawa G. Age-related changes
in skin wrinkles assessed by a
novel three-dimensional morpho-
metric analysis. Br J Dermatol
2002; 147: 689–695.
6. Leveque JL, Querleux B. SkinChip,
a new tool for investigating the
skin surface in vivo. Skin Res Tech-
nol 2003; 9: 343–347.
7. Edwards C, Heggie R, Marks R. A
study of difference in surface
roughness between sun-exposed
and unexposed skin with age.
Photodermatol Photoimmunol Pho-
tomed 2003; 19: 169–174.
Topography and anisotropy of the
skin surface with ageing. Skin Res
Technol 2005; 11: 110–119.
9. Ilive D, Hinnen U, Elsner P. Skin
roughness is negatively correlated
to irritation with DMSO, but not
with NaOH and SLS. Exp Derma-
tol 1997; 6: 157–160.
10. Fujimura T, Haketa K, Hotta M,
Kitahara T. Global and systematic
demonstration for the practical
usage of a direct in vivo measure-
ment system to evaluate wrinkles.
Int J Cosmet Sci 2007; 29: 423–436.
11. Tanaka H, Nakagami G, Sanada H,
Sari Y, Kobayashi H, Kishi K, Ko-
nya C, Tadaka E. Quantitative
evaluation of elderly skin based on
digital image analysis. Skin Res
Technol 2008; 14: 192–200.
12. Hayashi J, Yasumoto M, Ito H,
Koshimizu H. Age and gender esti-
mation based on wrinkle texture
and color of facial images. Int Conf
Pattern Recognit 2002; 1: 405–408.
13. Hayashi J, Koshimizu H, Hata S.
Age and gender estimation based
on facial image analysis. KES 2003;
LNAI 2774: 863–869.
14. Hatzis J. The wrinkle and its mea-
surement: a skin surface profilo-
metric method. Micron 2004; 35:
201–219.
15. Choi Y, Kim K, Hwang E.
WASUP: A wrinkle analysis A
using microscopic skin image. Pro-
ceedings of Int’l Conference on
Ubiquitous Information Technolo-
gies & Application, Hochiminh,
Vietnam, (December 2008).
watersheds in contour detection.
In Proceedings of the International
Workshop on Image Processing,
Real-time edge and motion detec-
tion/estimation, Rennes, France,
(September 1979).
17. Kim K, Choi Y, Hwang E. Wrinkle
feature-based skin age estimation
scheme. In Proceedings of the Int’l
Conference on Multimedia and
Expo, New York, USA, (June
2009).
18. Zou Y, Song E, Jin R. Age-depen-
dent changes in skin surface
assessed by a novel two-dimen-
sional image analysis. Skin Res
Technol 2009; 15(4): 399–406.
19. Choi Y, Tak Y, Rho S, Hwang E.
Skin feature extraction and proces-
sing model for statistical skin age
estimation. Multimed Tools Appl
2013; 64(2): 227–247.
20. Burges CJC. A tutorial on support
vector machines for pattern recog-
nition. Data Min Knowl Discov
1998; 2: 121–167.
21. Boyer G, Laquieze L, Le Bot A,
Laquieze S, Zahouani H. Dynamic
indentation on human skin in vivo:
ageing effects. Skin Res Technol
2009; 15: 55–67.
Address:
E. Hwang
School of Electrical Engineering,
Korea University, Seoul, Korea
Tel: +82 2 3290 3784
Fax: +82 2 921 0544
e-mail: ehwang04@korea.ac.kr

12